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
There is evidence that lipids can be allosteric regulators of membrane protein structure and activation . However , there are no data showing how exactly the regulation emerges from specific lipid-protein interactions . Here we show in atomistic detail how the human β2-adrenergic receptor ( β2AR ) – a prototypical G protein-coupled receptor – is modulated by cholesterol in an allosteric fashion . Extensive atomistic simulations show that cholesterol regulates β2AR by limiting its conformational variability . The mechanism of action is based on the binding of cholesterol at specific high-affinity sites located near the transmembrane helices 5–7 of the receptor . The alternative mechanism , where the β2AR conformation would be modulated by membrane-mediated interactions , plays only a minor role . Cholesterol analogues also bind to cholesterol binding sites and impede the structural flexibility of β2AR , however cholesterol generates the strongest effect . The results highlight the capacity of lipids to regulate the conformation of membrane receptors through specific interactions . G protein-coupled receptors ( GPCRs ) are versatile signaling proteins that mediate diverse cellular responses . With over 800 members , GPCRs constitute the largest family of integral membrane proteins in human genome and represent roughly half of all drug targets in modern medicine ( Gilchrist , 2010 ) . The human β2-adrenergic receptor ( β2AR ) is one of the best-characterized GPCRs . It is expressed in pulmonary and cardiac myocyte tissues and is a therapeutic target for asthma and heart failure ( Lefkowitz , 2000 ) . The functional diversity of β2AR is associated with its structural dynamics ( Manglik and Kobilka , 2014; Kobilka , 2013 ) . Recently found structures of β2AR in the inactive and active states have provided valuable insights into the structure-function relationship of β2AR ( Cherezov et al . , 2007; Hanson et al . , 2008; Rasmussen et al . , 2011 ) . Subsequent biophysical and biochemical studies have provided direct evidences of multiple distinct conformational states for specific GPCRs , such as β2AR ( Manglik and Kobilka , 2014; Kobilka , 2013; Nygaard et al . , 2013 ) . Meanwhile , molecular dynamics ( MD ) simulations have depicted the dynamic behavior of β2AR and have significantly enhanced our understanding of the activation mechanism of GPCRs ( Dror et al . , 2009; Ozcan et al . , 2013; Dror et al . , 2011 ) . Intriguingly , it is now evident that the activation of GPCRs is modulated by lipids ( Oates and Watts , 2011 ) . The lipid raft concept ( Lingwood and Simons , 2010; Allen et al . , 2007 ) essentially states that cell membranes include functional nanoscale domains where the function emerges from proteins whose structure and activation are modulated by lipids . However , despite a large body of research data , direct substantiation of lipid-induced protein modulation remains limited . Contreras et al . showed that the COPI machinery protein p24 is recognized by a specific sphingomyelin ( Contreras et al . , 2012 ) . Coskun et al . showed that monosialoganglioside GM3 influences the activation of the epidermal growth factor receptor ( Coskun et al . , 2011 ) , however the mechanism is not known . Lipid modulation also holds to GPCRs ( Oates and Watts , 2011; Neale et al . , 2015; Dawaliby et al . , 2016 ) in particular through cholesterol ( Oates and Watts , 2011; Paila and Chattopadhyay , 2009; Gimpl et al . , 1997; Paila et al . , 2011; Muth et al . , 2011 ) , which changes the physical properties of cellular membranes and supports the dynamic assembly of nanoscale membrane domains ( Simons and Ikonen , 2000 ) . The best known case is β2AR , which is a prototype of cholesterol-interacting GPCRs . β2AR belongs to the family of class A GPCRs . GPCRs belonging to this class show a high structural similarity and functional diversity . The literature reporting on the specific functional role of cholesterol and other lipids is extensive ( Pucadyil and Chattopadhyay , 2006; Gimpl , 2016 ) . It has been experimentally shown that cholesterol affects the conformation ( Muth et al . , 2011; Casiraghi et al . , 2016 ) and function ( Gimpl et al . , 1997; Paila et al . , 2011; Pucadyil and Chattopadhyay , 2006; Casiraghi et al . , 2016; Jafurulla et al . , 2014 ) of many GPCRs . Based on X-ray crystal structures cholesterol has specific contacts with β2AR ( Cherezov et al . , 2007; Hanson et al . , 2008 ) , suggesting that β2AR has binding sites for cholesterol . Spectroscopic ( Gater et al . , 2014 ) and MD simulation ( Cang et al . , 2013; Prasanna et al . , 2014; Lee et al . , 2012 ) studies have reported direct interactions between cholesterol and GPCRs , including β2AR . Experimental data show that cholesterol binding to β2AR changes its structural properties ( Hanson et al . , 2008; Zocher et al . , 2012 ) . Cholesterol is also necessary in crystallizing β2AR ( Cherezov et al . , 2007; Hanson et al . , 2008 ) , and cholesterol and its analogue cholesteryl hemisuccinate ( CHS ) have been exhibited to improve β2AR stability ( Zocher et al . , 2012; Loll , 2014 ) . Since the structure and function of GPCRs are closely related , cholesterol binding specifically to β2AR is also expected to change the functional properties of the receptor . Indeed experimental studies indicate that cholesterol has a functional role in β2AR ( Paila et al . , 2011; Pontier et al . , 2008; Xiang et al . , 2002 ) . Further , inhibition of β2AR-associated signaling has been observed with increasing membrane cholesterol content ( Pontier et al . , 2008 ) . However , as with GPCRs in general , the atomic-scale mechanism cholesterol uses to regulate β2AR is not known . Does cholesterol modulate β2AR activity through membrane-mediated effects by altering the physical properties of the membrane ? Alternatively if regulation takes place through specific direct interactions , then what is the atom-scale mechanism ? We performed extensive atomistic MD simulations ( totaling >100 μs , Table 1 ) to clarify the mechanism responsible for the modulatory role of cholesterol on β2AR . In essence , we show that as cholesterol concentration reaches ~10 mol% , the conformational distribution of β2AR is drastically altered . The mechanism of action is based on the binding of cholesterol at specific high-affinity sites of the receptor . 10 . 7554/eLife . 18432 . 003Table 1 . Descriptions of systems simulated: β2AR in bilayers with varying lipid compositions . ‘Chol’ stands for cholesterol . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 003SystemsInitial lipid arrangement around β2AR Lipids Sterol mol % No . of repeats†Time ( μs ) ‡DOPCRandomDOPC033×2 . 5 DOPC-activeRandomDOPC033×2 . 5 C H O L Chol2RandomDOPC + Chol233×2 . 5 R A N D O M Chol5RandomDOPC + Chol533×2 . 5 Chol10RandomDOPC + Chol1033×2 . 5 Chol25RandomDOPC + Chol2522×2Chol40RandomDOPC + Chol4033×2 . 5 Chol40-activeRandomDOPC + Chol4033×2 . 5 C H S CHS10RandomDOPC + CHS1022×2CHS40RandomDOPC + CHS4022×2CHSA10[A for anionic]RandomDOPC + CHSA1012CHSA40RandomDOPC + CHSA4012O X Y S T E R O L 27-OH-CholRandom[16 mol % Chol was randomly replaced by 27-OH-Chol]DOPC +Chol +27-OH-Chol25 ( 4 mol% 27-OH-Chol + 21 mol% Chol ) 32 + 1 + 1 4β-CholRandom[16 mol% Chol was randomly replaced by 4β-OH-Chol]DOPC +Chol +4β-OH-Chol25 ( 4 mol% 4β-OH-Chol + 21 mol% Chol ) 31 + 1 + 1 Chol-Bound§8 cholesterols bound at sites predicted by simulationsDOPC + Chol1 . 933×2 . 5 B O U N DChol-IC12 Chol bound at IC1DOPC + Chol<1 22×2CHS-IC12 CHS bound at IC1DOPC + CHS<1 12CHSA-IC12 CHSA bound at IC1DOPC + CHSA<1 12PC-20:0–22:1 c13 [Double bond at carbon 13]RandomPC-20:0–22:1 c13 033×1 . 5 Pyrene20RandomDOPC +20 mol% pyrene033×1 . 5 In the DOPC-active and Chol40-active systems , we used the active-state conformation of the receptor as the starting structure; for all the other systems , we used the inactive conformation . †For systems with no sterols initially bound to β2AR , i . e . , the systems which started with a random distribution of lipids , a number of different repeat simulations for each lipid composition were performed with different initial lipid arrangements around the receptor . For systems with sterols initially bound to β2AR ( seed and BOUND ) , different replicas were generated with different starting velocities . ‡Listed are the simulation times of production simulations; the equilibration time of the systems ( 100 ns ) is not included . §In the Chol-Bound system , eight cholesterol molecules were initially ( at time zero of the simulation ) bound at eight binding sites predicted by the present simulations , while the rest of the system had no cholesterol at all . We first studied the impact of cholesterol on the conformational distribution of β2AR by systematically increasing the cholesterol concentration from 0 to 40 mol% in a DOPC ( 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ) bilayer . Crystallographic studies and previous biophysical and biochemical studies have shown that helices 5–6 ( H5-H6 ) ( Figure 1A ) constitute a highly dynamic region of β2AR ( Kobilka , 2013 ) . Upon activation , the most dramatic conformational change , which is conserved among many GPCRs , is a 7–14 Å outward movement of the intracellular end of H6 from the heptahelical core of the receptor ( Manglik and Kobilka , 2014; Kobilka , 2013 ) . The large rearrangement in the G protein-coupling interface is accompanied by a comparatively subtle change in the ligand-binding pocket . In a conformational change from the inactive to the active state β2AR , H5 ( around S2075 . 46 ) has been found to move inward by 2 Å to establish an optimal interaction between the agonist and the two anchor sites ( D1133 . 32/N3127 . 39 and S2035 . 42/S2045 . 43/S2075 . 46 ) on the receptor ( Kobilka , 2013 ) . 10 . 7554/eLife . 18432 . 004Figure 1 . Conformational dynamics of β2AR . ( A ) The distances between the Cα atoms of D1133 . 32–S2075 . 46 ( distance defined as LL ) and R1313 . 50–E2686 . 30 ( LG ) pairs used to measure the fluctuations at the ligand and G-protein binding sites , respectively . ( B–C ) The conformational distributions of β2AR in membranes with 0 and 10 mol% cholesterol ( Chol ) as a function of LL and LG . The gray dotted lines represent the corresponding LL and LG values in the inactive crystal structure of β2AR ( Hanson et al . , 2008 ) . The cartoon diagram shows the fluctuations of LL and LG at the ligand and G-protein binding sites of the receptor , respectively . ( D–E ) The time evolution of LL ( light red ) and LG ( light blue ) in systems with 0 and 10 mol% cholesterol . Corresponding 50-point running averages are shown in dark colors . ( F ) Standard deviation for the distribution of the distance between the intracellular ( IC ) ( or extracellular ( EC ) ) end of H5 and its average position , and its dependence on whether the given end of H5 is in contact with cholesterol or not; similarly for H6 . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 00410 . 7554/eLife . 18432 . 005Figure 1—figure supplement 1 . Conformational distributions of β2AR in lipid bilayers with various cholesterol ( Chol ) concentrations . In panels ( A–F ) the distributions are plotted as a function of LL ( distance between the Cα atoms of D1133 . 32 and S2075 . 46 ) at the ligand binding site and LG ( distance between the Cα atoms of R1313 . 50 and E2686 . 30 ) at the G protein-binding site . ( A–E ) Starting from situations where no cholesterol molecules were initially bound to β2AR , distributions are plotted over all independent trajectories of a given system , where the equilibration time ( the first 100 ns ) was discarded from the analysis . ( F ) β2AR conformational distribution in control simulations , where cholesterol molecules were initially bound at the eight interaction sites of β2AR predicted by our simulations , but no further cholesterol was in the membrane ( total ( average ) cholesterol concentration 1 . 9 mol% ) . Here , there is reason to keep in mind the rapid migration of cholesterols away from the receptor surface when the cholesterol concentration is low ( see main text and Figure 5 ) , implying that panel ( F ) corresponds to cholesterol-rich conditions in the vicinity of the receptor at very short times but to cholesterol-poor conditions at long times . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 005 In the present work where we started from the inactive structure of β2AR ( Manna et al . , 2015 ) , we calculated the distance between the Cα atoms of D1133 . 32 and S2075 . 46 ( referred to as LL ) to measure the displacement of H5 in the ligand-binding site , and the distance between the Cα atoms of R1313 . 50 and E2686 . 30 ( referred to as LG ) to determine the displacement of H6 in the G protein-binding site ( Figure 1A ) ; the position of H3 does not change noticeably ( RMSD < 0 . 8 Å ) during the simulations . These two parameters ( LL and LG ) have been used in many previous studies to monitor changes in β2AR conformation ( Manglik and Kobilka , 2014; Kobilka , 2013; Nygaard et al . , 2013; Dror et al . , 2009; Ozcan et al . , 2013; Dror et al . , 2011; Manna et al . , 2015 ) , thus here we discuss the conformational distribution of the receptor as a function of LL and LG ( Figure 1B , C and Figure 1—figure supplement 1 ) . In the inactive crystal structure , the LL and LG values are 12 . 07 and 11 Å , respectively ( Hanson et al . , 2008 ) . In a cholesterol-free DOPC bilayer , we find β2AR to adopt a wide range of conformations with LL varying between ~11 . 5–17 . 5 Å and LG ranging between ~7 . 5–12 . 5 Å ( Figure 1B ) . The receptor populates two major conformational states . One of them has a relatively open G protein site ( LG being 10–12 Å ) and a smaller ligand-binding site ( LL ~ 13 ± 1 Å ) . The other conformation is characterized by a shift of ~3–4 Å from the intracellular end of H6 towards the receptor core that blocks the G protein interface ( LG ~ 8 . 5 Å ) . At the same time , the ligand-binding pocket expands as the extracellular part of H5 moves ~ 4–5 Å away from H3 ( LLnow ~16 ± 1 Å ) . This conformation represents an alternative inactive structure of the receptor , as both changes occur in the opposite direction compared to the case of agonist binding ( Kobilka , 2013 ) ; we do not observe any transition to the active state of β2AR . Figure 1D shows the receptor oscillating between the different inactive conformations during 2 . 5 µs . The closing of the intracellular G protein-binding surface is found to correlate with the opening of the extracellular ligand-binding pocket , and vice-versa ( Figure 1D ) . The conformational correlation between the two distal sites supports the view of allosteric regulation in GPCRs ( Kobilka , 2013; Ozcan et al . , 2013 ) . In the presence of cholesterol , the picture changes quite dramatically . With a cholesterol concentration of 10 mol% , the conformational flexibility of β2AR reduces significantly ( Figure 1C ) . The receptor stays predominantly in one conformation and no further opening of the ligand-binding site or the opening/closing of the G protein-binding site is observed , unlike in a cholesterol-free membrane . As shown in Figure 1E , LL and LG fluctuate around ~13 and~9 . 5 Å , respectively . The slowing down of the movements of H5 and H6 correlates with the observed high-density spots of cholesterol at these helices ( IC2 and EC1 in Figure 2 discussed in detail below ) . To further quantify this , Figure 1F depicts the standard deviation for the fluctuations of the intracellular and extracellular ends of H5 and H6 , when these ends are bound or unbound to cholesterol . The data show that the deviations of these helices from their respective average positions are much smaller when they are bound to cholesterol . The effect is particularly strong for the extracellular end of H5 at the ligand-binding site and for the intracellular end of H6 at the G protein-binding site . 10 . 7554/eLife . 18432 . 006Figure 2 . Cholesterol interaction sites on β2AR . ( A–B ) 2D number densities of cholesterol ( Chol ) around β2AR . The data are averaged over all independent trajectories for a given cholesterol concentration ( Table 1 ) and normalized with respect to the maximum density for that particular cholesterol concentration . The intracellular ( IC ) and extracellular ( EC ) bilayer leaflets are depicted separately . The major cholesterol interaction sites ( IC1 , IC2 and EC1 ) are marked in the density plots . The IC and EC sides of the transmembrane regions ( H1–H7 ) of β2AR are shown in gray scale ( the darker the color , the higher is the number density ) and numbered accordingly . ( C–D ) Cartoon representation of three main cholesterol interaction sites in β2AR . IC1 ( H1–H4 ) and IC2 ( H5–H6 ) are located on the intracellular side , and EC1 comprised of two closely placed cholesterols between H5-H6 and H6-ECL3-H7 is located on the extracellular side of β2AR . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 00610 . 7554/eLife . 18432 . 007Figure 2—figure supplement 1 . Residues of β2AR involved in cholesterol binding , and cholesterol interaction sites on β2AR . Panels ( A–B ) ( top ) : Cholesterol occupancy time per residue of β2AR described in terms of the normalized time fraction , where a value of one stands for a contact throughout the simulation trajectory and zero means no contact . Results are given for ( A ) 10 and ( B ) 40 mol% of cholesterol . The residues of β2AR are defined to be in contact with cholesterol when any non-hydrogen atom of the residue is within ≤0 . 5 nm of any heavy atom of cholesterol . The data show that there are several hot spots ( blue ) as cholesterol binding sites . These plots were averaged from all independent simulations for a given cholesterol concentration , where the equilibration time ( the first 100 ns of the simulation ) was disregarded from the analysis . Panels ( C ) ( bottom ) : Interaction sites as obtained from our simulations , are shown from two perspectives around the protein . EC and IC stand for extracellular and intracellular , respectively . Interaction sites at the intracellular ( IC ) side: IC1 ( dark green ) between helices ( H ) 1–4 , IC2 ( red ) between H5 and H6 , IC3 ( magenta ) between H3 and H5 , and IC4 ( orange ) between H1 and H8 . Interaction sites at the extracellular ( EC ) side: EC1 comprised of two closely placed cholesterol molecules between H5 and H6 ( cyan ) and in space surrounded by H6-ECL3-H7 ( green ) , EC2 ( purple ) between H3 and H4 , and EC3 ( blue ) between H1-H2-ECL1 ( where ECL stands for the extracellular loop ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 00710 . 7554/eLife . 18432 . 008Figure 2—figure supplement 2 . Sequence alignment of β2AR orthologues around the cholesterol-binding site IC1 . The residues that play a major role ( contact fraction ≥ 0 . 4 , where one stands for maximum contact and zero for no contact ) in cholesterol binding are highlighted . Here for IC1 , the residues in the cholesterol consensus motif are highlighted in red . Following sequence alignment , shown are the contact fraction per residue ( tan bars ) and its occurrence in the set of sequences ( cyan bars ) [one stands for 100% and zero for no occurrence] . The occurrence represents the extent a particular residue is conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 00810 . 7554/eLife . 18432 . 009Figure 2—figure supplement 3 . Sequence alignment of β2AR orthologues around the cholesterol-binding site IC2 . The residues that play a major role ( contact fraction ≥ 0 . 4 , where one stands for maximum contact and zero for no contact ) in cholesterol binding are highlighted . Following sequence alignment , shown are the contact fraction per residue ( tan bars ) and its occurrence in the set of sequences ( cyan bars ) [one stands for 100% and zero for no occurrence] . The occurrence represents the extent a particular residue is conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 00910 . 7554/eLife . 18432 . 010Figure 2—figure supplement 4 . Sequence alignment of β2AR orthologues around the cholesterol-binding site EC1 . The residues that play a major role ( contact fraction ≥ 0 . 4 , where one stands for maximum contact and zero for no contact ) in cholesterol binding are highlighted . Following sequence alignment , shown are the contact fraction per residue ( tan bars ) and its occurrence in the set of sequences ( cyan bars ) [one stands for 100% and zero for no occurrence] . The occurrence represents the extent a particular residue is conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01010 . 7554/eLife . 18432 . 011Figure 2—figure supplement 5 . Cholesterol density around the receptor at low cholesterol concentrations . Two-dimensional ( 2D ) averaged and normalized number densities of cholesterol around β2AR shown at low cholesterol concentrations ( 2 and 5 mol% ) . The intracellular and extracellular leaflets are depicted separately . The intracellular and extracellular sides of the transmembrane regions of β2AR are shown in gray scale ( the darker the color , the higher is the number density ) , and they are numbered accordingly to show the locations of the individual helices ( H1–H7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01110 . 7554/eLife . 18432 . 012Figure 2—figure supplement 6 . Structure of cholesterol analogues and properties of sterol-containing bilayers . ( A ) The different cholesterol analogues used in the current study . ( B–D ) Average lipid chain order parameter SCD of DOPC bilayers with different concentrations of cholesterol or cholesterol-analogues . ( E–G ) Average bilayer thickness in DOPC bilayers with different concentrations of cholesterol or cholesterol-analogues . Error bars for order parameter and thickness are less than 0 . 02 and 0 . 005 Å , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01210 . 7554/eLife . 18432 . 013Figure 2—figure supplement 7 . Interactions of cholesterol and cholesterol-like molecules with β2AR . The average interaction energies for van der Waals ( vdW ) and electrostatic interactions are determined separately . Error bars are in the range of 0 . 1–1 kJ/mol . The lower panel represents the oxysterol-containing systems , where a fraction of cholesterol is replaced by 4β-OH-Chol and 27-OH-Chol , respectively , resulting in bilayers with 4 mol% oxysterol and 21 mol% cholesterol . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01310 . 7554/eLife . 18432 . 014Figure 2—figure supplement 8 . Densities of sterols around β2AR . Normalized 2D average number densities around β2AR: ( A–B ) CHSA ( the deprotonated form of cholesteryl hemisuccinate ( CHS ) ) ; ( C–F ) CHS . Densities of sterols in mixed sterol-containing bilayers with other molecules: ( G–H ) 4β-hydroxy-Chol ( 4β-OH-Chol ) ; ( K–L ) 27-hydroxy-Chol ( 27-OH-Chol ) . The densities of 4β-OH-Chol and 27-OH-Chol are shown separately: ( I–J ) 4β-OH-Chol; ( M–N ) 27-OH-Chol . For descriptions of models , see Table 1 . For each system , the intracellular and extracellular bilayer leaflets are depicted separately . The intracellular and extracellular sides of β2AR transmembrane regions ( H1–H7 ) are shown in gray scale ( the darker the color , the higher the number density ) and numbered accordingly . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01410 . 7554/eLife . 18432 . 015Figure 2—figure supplement 9 . Conformational distributions of β2AR in lipid bilayers with different cholesterol analogues . ( A–B ) Oxysterol-containing systems having 4 mol% of oxysterol ( 27-OH-Chol or 4β-OH-Chol ) and 21% cholesterol . ( C–D ) DOPC bilayer with 10 mol% and 40 mol% of CHS . Conformational distributions are calculated over all independent trajectories of a given system , where the equilibration time ( 100 ns ) is disregarded from the analysis . The CαD1133 . 32–CαS2075 . 46 ( defined as LL ) and CαR1313 . 50–CαE2686 . 30 ( LG ) distances represent the fluctuations in the ligand and the G protein-binding sites , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01510 . 7554/eLife . 18432 . 016Figure 2—figure supplement 10 . IC1 interaction site . Specific cholesterol binding site in β2AR with the cholesterol consensus motif displayed with side chain positions of the conserved amino acid residues , as found in ( A ) the crystal structure ( ref . 17 ) and ( B ) during our simulation . In the simulation snapshot , residues are colored according to their strength of interaction with cholesterol ( red represents the weakest and blue represents the strongest interaction ) . ( C ) As to the time-dependent distance between H4 and its average position , as the H4 helix fluctuates around its average location , shown here are results for the standard deviation of the distance fluctuations . Data are given for cases , where IC1 is occupied ( blue ) or unoccupied ( orange ) by cholesterol . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 016 The restricted dynamics of β2AR is also observed at higher cholesterol concentrations ( 25 and 40 mol%; Figure 1—figure supplement 1D , E ) . In these cases , the receptor samples a similar conformational space as observed with 10 mol% cholesterol . At lower concentrations ( 2 and 5 mol% ) , the distribution of the receptor’s conformation is much wider ( Figure 1—figure supplement 1A , B ) . Particularly when the membrane contains a very small percentage of cholesterol ( 2 mol% ) , the range of conformations accessible to β2AR is almost comparable to that of a cholesterol-free membrane . A broad conformational distribution ( Figure 1—figure supplement 1F ) is also observed in control simulations , where eight cholesterol molecules were initially placed at the cholesterol-binding sites of β2AR predicted by our simulations ( see below ) , and this receptor-cholesterol complex was then embedded in a cholesterol-free membrane . Here ( Figure 1—figure supplement 1F ) the concentration of cholesterol in the annular region is therefore high in the beginning of the simulation , while it is zero elsewhere . Cholesterols dissociate from β2AR during the course of the simulation ( discussed in detail below ) and at long times the system corresponds to a dilute ( cholesterol-poor ) system , where the total average cholesterol concentration is low ( 1 . 9 mol% ) . One finds that as the data are averaged over the simulation period , the conformational behavior ( Figure 1—figure supplement 1F ) translates from cholesterol-rich ( Figure 1—figure supplement 1E ) to cholesterol-poor behavior ( Figure 1—figure supplement 1A , B ) . Further , we studied the effect of cholesterol on the active conformation of β2AR in its apo form in the absence of the G protein ( Rasmussen et al . , 2011 ) . In the active state , the intracellular end of H6 is splayed outward from the helical bundle , providing room for the G protein ( Figure 3A ) . We observe inward swinging of H6 towards H3 in the absence of cholesterol ( which occurred in two out of three replica simulations ) . As shown in Figure 3B , E , the intracellular end of H6 spontaneously approaches H3 with LG dropping from 18 . 97 Å in the starting active conformation to ~11 . 5 Å that is comparable to the crystallographically observed inactive conformation of β2AR ( LG ~ 11 Å ) ( Hanson et al . , 2008 ) . Such spontaneous deactivation of the receptor in the absence of the intracellular binding partner and cholesterol is in agreement with recent simulations ( Dror et al . , 2011; Neale et al . , 2015 ) and experimental studies ( Rosenbaum et al . , 2011 ) . Meanwhile , with 40 mol% cholesterol , we observe that the active-like open conformation is stable during the simulations ( Figure 3—figure supplement 1 ) . As shown in Figure 3C , E , the LG value remains stable around 16 . 5 Å and no deactivation is observed unlike in cholesterol-free systems . Interestingly , here again we found a high cholesterol density at the intracellular segments of H5-H6 ( IC2 in Figure 3C , D , F as discussed in detail below ) . 10 . 7554/eLife . 18432 . 017Figure 3 . Effect of cholesterol on the active conformation of β2AR . Cytosolic view of β2AR ( A ) in the beginning of a simulation ( active state ) as well as in representative simulation snapshots in ( B ) a DOPC bilayer and ( C ) in the presence of 40 mol% cholesterol . The dotted line represents the distance between the Cα atoms of R1313 . 50–E2686 . 30 ( defined as LG ) , used to measure the fluctuation at the G protein-binding site . ( D ) Simulation snapshot ( in the presence of 40 mol% cholesterol ) showing cholesterol binding at the interaction sites of β2AR . ( E ) The time evolution of LG in systems with 0 ( light red ) and 40 mol% cholesterol ( light blue ) . Corresponding 50-point running averages are shown in dark colors ( red , blue ) . ( F ) 2D number densities of cholesterol around β2AR ( cytosolic view ) . The individual transmembrane helixes of β2AR are numbered and shown in gray scale ( as in Figure 2A , B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 01710 . 7554/eLife . 18432 . 018Figure 3—figure supplement 1 . Conformational distribution of β2AR starting from the active state . The conformational distributions of β2AR in ( left ) a DOPC bilayer and ( right ) a DOPC bilayer with 40 mol% cholesterol ( Chol ) as a function of LL and LG . The gray dotted lines represent the corresponding LL and LG values in the initial active crystal structure of β2AR . The distribution is averaged over the different replicas of each system . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 018 These results show that cholesterol restricts the intrinsic conformation dynamics of β2AR and governs changes between different conformational states , thereby modulating its function . In all of the simulations ( Table 1 ) , cholesterol is observed to diffuse spontaneously to the receptor’s surface . Time-averaged two-dimensional ( 2D ) number density maps demonstrate that there are preferred cholesterol positions around β2AR ( Figure 2A , B ) . Localized cholesterol hot spots are often used as an indicator of potential cholesterol binding sites . We identify three such cholesterol interaction sites – two on the intracellular side ( IC1 and IC2 ) and one on the extracellular side ( EC1 ) ( Figure 2 , Figure 2—figure supplement 1A , B ) . Here we call them high-affinity sites since they reproducibly exhibit high cholesterol densities ( normalized number density above 0 . 7 ) at different cholesterol concentrations ( Figure 2A , B ) and also have large lifetimes as the below discussion shows . IC1 is a shallow groove formed by the intracellular parts of transmembrane helices H1-H4 and coincides well with the location of cholesterol observed in the crystal structure of β2AR ( Cherezov et al . , 2007; Hanson et al . , 2008 ) . In IC2 cholesterol penetrates deep into the cleft between H5 and H6 on the intracellular side . A high density of cholesterol is observed at IC2 not only in the inactive but also in the active β2AR conformation ( Figure 3C , D , F ) , which suggests that this site is biologically important . EC1 is comprised of two closely spaced cholesterol hot spots located in the extracellular part of H5-H6 and H6-ECL3-H7 ( where ECL stands for the extracellular loop ) . The occupancy of two cholesterol molecules at EC1 is in good agreement with the crystal structure of the adenosine receptor A2AAR ( Liu et al . , 2012 ) , while IC2 is so far unidentified among the experimentally determined structures ( Gater et al . , 2014 ) . Notably , the cholesterol binding residues of the three interaction sites are conserved to a large degree among β2AR orthologues ( Figure 2—figure supplement 2 , Figure 2—figure supplement 3 , Figure 2—figure supplement 4 ) , indicating that these sites have conserved during the evolution of the receptor . In addition , a few comparatively low-affinity cholesterol binding sites ( IC3-4 , EC2-3 ) with 10 and 40 mol% cholesterol are observed ( Figure 2—figure supplement 1 ) . When cholesterol concentration is lowered below 10 mol% , many of the interaction sites , particularly IC1 and EC1 , are occupied by cholesterol at concentrations as low as 5 mol% ( Figure 2—figure supplement 5 ) . A few sites ( IC2 and EC1 ) are visited , though transiently , by cholesterol even at 2 mol% ( Figure 2—figure supplement 5 ) . In addition to the above-discussed cholesterol hot spots , we observed two sites with comparatively weak cholesterol occupancies ( reproducible at both 10 and 40 mol% cholesterol concentrations ) : IC3 between H3 and H5 , and IC4 between H1 and H8 , both on the intracellular side ( Figure 2A , B and Figure 2—figure supplement 1 ) . IC4 recaptures the predicted cholesterol position at the dimerization interface of β2AR found by X-ray crystallography ( Cherezov et al . , 2007 ) . Besides these , another site with a low cholesterol density was observed near the extracellular part of H3-H4 ( EC2 ) in the 10 mol% cholesterol system , and a high-density site was observed on the extracellular side of H1-H2-EC1 ( EC3 ) in the 40 mol% cholesterol system ( Figure 2A , B ) . Concluding , we find cholesterol to bind to β2AR in specific binding sites . These sites are in agreement with those found in the crystallographic structures of GPCRs ( Cherezov et al . , 2007; Hanson et al . , 2008; Gimpl , 2016; Warne et al . , 2011; Liu et al . , 2012; Gater et al . , 2014 ) . Is it possible that the effects we observed on β2AR conformation could be due to cholesterol-induced changes in membrane properties , and the changes in β2AR would hence not be due to the specific direct binding of cholesterol in the hot spots ? To unlock this issue , we study the conformational properties of β2AR in cholesterol-free membranes whose physical properties ( thickness , order , diffusion ) match those of membranes with a large concentration of cholesterol . Summarizing , the changes in physical membrane properties , similar to those induced by cholesterol , do not restrict the conformational dynamics of β2AR . We conclude that the cause of the observed changes in β2AR conformation and dynamics is the specific binding of cholesterol to β2AR . When cholesterol is specifically bound to β2AR , how stable is the binding ? Figure 5 depicts the time-correlation function of cholesterol binding in the three main binding sites ( IC1 , IC2 , EC1 ) on β2AR and shows that at low cholesterol concentrations ( 2–5 mol% ) the binding lifetime is short , of the order of 100 ns or less . However , at ~10 mol% there is a clear transition to longer lifetimes ( see Video 1 and Video 2 ) given that the lifetime of binding increases to the microsecond time scale for 10 and 40 mol% cholesterol . 10 . 7554/eLife . 18432 . 021Figure 5 . Binding time of cholesterol . ( A–C ) Time-correlation function of cholesterol ( Chol ) at the three major interaction sites ( IC1 , IC2 , EC1 ) on the β2AR surface . Initially cholesterol is bound to the site ( distance ≤ 0 . 5 nm ) and the correlation function describes the probability that cholesterol remains bound to the given site for increasing time . Data are shown for DOPC-cholesterol membranes with 2 , 5 , 10 , and 40 mol% of cholesterol . ( D–E ) Schematic representation showing the transition from fast to slow exchange as cholesterol concentration increases from 2 to 40 mol% . Color code: β2AR ( blue ) , DOPC ( thin grey lines ) , cholesterol molecules bound to the interaction sites ( purple ) , and other cholesterol molecules not bound to the receptor ( yellow sticks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 02110 . 7554/eLife . 18432 . 022Figure 5—figure supplement 1 . Interaction of cholesterol with β2AR . Time development for the distances of cholesterol molecules from the β2AR surface , where these cholesterol molecules were initially bound at the eight binding sites identified in this study ( cholesterol-bound , see Table 1; Figure 2—figure supplement 1 ) . Here , EC1-A and EC1-B stand for the two cholesterol molecules in the EC1 binding site . The rest of the membrane was initially cholesterol-free . Shown here are the data based on the three independent repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 02210 . 7554/eLife . 18432 . 023Video 1 . Spontaneous binding/unbinding of cholesterol at the three main cholesterol interaction sites of β2AR during a 2 . 5-μs simulation with 10 mol% of cholesterol . Cholesterols interacting at the cholesterol-binding sites are highlighted ( yellow at IC1; green at IC2; and blue and red at EC1 ) . Other cholesterols are shown in gray . For clarity , other lipids in a membrane are not shown . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 02310 . 7554/eLife . 18432 . 024Video 2 . Spontaneous binding/unbinding of cholesterol at the three main cholesterol interaction sites of β2AR during a 2 . 5-μs simulation with 40 mol% of cholesterol . Cholesterols interacting at the cholesterol-binding interaction sites are highlighted ( yellow and green at IC1; red , blue and orange at IC2; and pink , purple and cyan at EC1 ) . Other cholesterols are shown in gray . For clarity , other lipids in a membrane are not shown . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 024 In three control simulations where cholesterols were initially bound at the eight cholesterol-binding sites identified in our simulations and no further cholesterol was in the bilayer ( Figure 5—figure supplement 1 ) , cholesterols underwent rapid unbinding from the majority of the binding sites in a timescale of tens to hundreds of nanoseconds ( Figure 5—figure supplement 1 ) , similarly to the short binding lifetime observed for cholesterol-poor systems ( 2 mol% , Figure 5 ) . However , at a few sites cholesterol stayed for the entire simulation time ( IC1 and IC2 in two out of three simulations ) or dissociated in the μs timescale ( IC3 and EC3 in one simulation ) . The results show that the lifetime of cholesterol is of the order of microseconds in the high-affinity binding sites , where the lifetime at large cholesterol concentrations is largely independent of cholesterol concentration . We next explore how cholesterol analogues , in comparison to cholesterol , interact with β2AR . We focus on four different analogues ( Table 1 ) : ( i-ii ) cholesteryl hemisuccinate ( CHS ) and its deprotonated form ( CHSA ) , and ( iii-iv ) two oxysterols , 4β-hydroxycholesterol ( 4β-OH-Chol ) and 27-hydroxycholesterol ( 27-OH-Chol ) , oxidized at the cholesterol ring and tail , respectively ( Figure 2—figure supplement 6A ) . As compared to cholesterol , CHS is a more water-soluble cholesterol ester and is widely used in structural biology and biophysical studies as a cholesterol analogue ( Zocher et al . , 2012; Loll , 2014 ) . Oxysterols , on the other hand , are derivatives of cholesterol with additional oxygen-containing substitutions at different positions of cholesterol ( Olkkonen and Hynynen , 2009; Kulig et al . , 2015a; Neuvonen et al . , 2014 ) . Due to the structural similarities with cholesterol , these analogues mimic cholesterol as to the effects on membrane properties ( e . g . , increasing bilayer order and thickness ) , although to different extents ( Figure 2—figure supplement 6 ) ( Kulig et al . , 2015a , 2015b ) . CHSA is found to interact strongly with β2AR due to the enhanced electrostatic coupling resulting from its negatively charged head-group ( Figure 2—figure supplement 7 ) , however it favors to reside around the receptor at locations different from those of cholesterol ( Figure 2—figure supplement 8A , B ) . Meanwhile , CHS closely mimics the behavior of cholesterol ( Figure 2—figure supplement 7 ) . Among the three major cholesterol interaction sites observed in our simulations , we find a very high CHS density at IC2 ( Figure 2—figure supplement 8C–F ) . High occupancy of CHS is also observed near IC1 ( at 40 mol% CHS concentration ) but not at all at EC1 . Occupancy of CHS at IC1 is consistent with the crystal structure of β1AR ( Warne et al . , 2011 ) . 4β-OH-Chol interacts only weakly with β2AR ( Figure 2—figure supplement 7 ) . Almost all of the interaction sites on the receptor surface are occupied by cholesterol rather than 4β-OH-Chol ( Figure 2—figure supplement 8G–J ) . As a result , the average density maps , showing the lateral arrangement of these sterols around β2AR , are similar to those of 10 and 40 mol% cholesterol systems ( Figure 2A ) , and reproducible . Unlike 4β-OH-Chol , 27-OH-Chol prefers to interact with the receptor directly ( Figure 2—figure supplement 7 ) . For the IC1 site , 27-OH-Chol competes , though weakly , with cholesterol , while at EC1 and IC3 , 27-OH-Chol exhibits preference over cholesterol ( Figure 2—figure supplement 8K–N ) . Altogether , our results show that also other cholesterol-like molecules interact with β2AR and may occupy the same binding sites on the receptor surface as cholesterol . However , the effects of cholesterol-analogues on β2AR are weaker compared to those induced by cholesterol ( Figure 2—figure supplement 9 ) . All the cholesterol analogues studied here have a rigid ring structure , yet their slightly different chemical compositions influence their occupancy as well as the strength of binding to the cholesterol-binding sites ( Table 2 ) . This is assessed here in terms of the van der Waals energy , which as a short-range interaction reflects how strongly two molecules are in contact and therefore serves as an appropriate measure for the gravity of lipid-protein binding in the binding site . 10 . 7554/eLife . 18432 . 025Table 2 . Interactions* of sterols at the three high-affinity cholesterol-binding sites . DOI: http://dx . doi . org/10 . 7554/eLife . 18432 . 025Cholesterol/Cholesterol analogueHigh-affinity cholesterol interaction sites IC1 IC2 EC1 vdW interaction energy ( kJ/mol ) No . of contactsvdW interaction energy ( kJ/mol ) No . of contactsvdW interaction energy ( kJ/mol ) No . of contactsCholesterol†−138 . 04 ± 0 . 20 141 . 02 ± 0 . 22 −95 . 06 ± 0 . 12 90 . 65 ± 0 . 16 −129 . 51 ± 0 . 29104 . 38 ± 0 . 28 CHS−29 . 63 ± 0 . 14 28 . 78 ± 0 . 16 −98 . 75 ± 0 . 11 96 . 30 ± 0 . 16 --27-OH-Chol−32 . 17 ± 0 . 30 34 . 95 ± 0 . 33 −22 . 69 ± 0 . 23 28 . 41 ± 0 . 28 −132 . 85 ± 0 . 27 120 . 20 ± 0 . 30 4β-OH-Chol----−41 . 80 ± 0 . 48 33 . 41 ± 0 . 42 * Shown are the total van der Waals ( vdW ) interaction energy and the number of contacts between cholesterol and β2AR , when cholesterol is in the IC1 , IC2 , or EC1 binding site ( and similarly for the cholesterol analogues ) . † Calculations are based on systems having ≥10 mol% cholesterol . Shown here are the average values over different trajectories . The results in Table 2 show that among the three major interaction sites , the binding of CHS at IC1 is much weaker than that of cholesterol . At IC2 the strength of interaction of CHS and cholesterol is comparable . Meanwhile , the extracellular EC1 site remains unoccupied by CHS indicating the binding energy to be low . As to the two oxysterols , 4β-OH-Chol interacts with β2AR only at EC1 and the interaction is weak , while 27-OH-Chol binds at EC1 as tightly as cholesterol , but its interaction at the two other binding sites ( IC1 and IC2 ) is much weaker than in the case of cholesterol . Concluding , CHS interacts at IC2 as strongly as cholesterol but its interactions at IC1 and EC1 are negligible compared to those of cholesterol . The oxysterol 27-OH-Chol interacts at EC1 as strongly as cholesterol but its interactions at IC1 and IC2 are negligible compared to those of cholesterol . The oxysterol 4β-OH-Chol does not interact with β2AR to a significant degree . These data can be considered in the context of molecular structures . In CHS , the difference compared to cholesterol is the additional chain bridged to the cholesterol structure via an ester bond ( Figure 2—figure supplement 6A ) . This additional chain does not interfere binding at IC2 , but it does alter the binding at IC1 and EC1 . In 27-OH-Chol , the oxidation has taken place in the short acyl chain that is the terminal subunit of the molecule . This does not interfere the binding at EC1 but does alter the binding at IC1 and IC2 . Finally , in 4β-OH-Chol , the oxidation has occurred in the rigid steroid moiety , making the α-side of the molecule rougher . In cholesterol , the α-side is exceptionally flat . Given this change in surface roughness , and the importance of the surface-surface contact in lipid-β2AR binding in the binding site , it is quite obvious why this oxysterol does not bind to any of the cholesterol binding sites ( IC1 , IC2 , EC1 ) . The results support the view that the restriction of β2AR dynamics arises from specific lipid binding to the receptor binding sites: the tighter the binding , the more is the receptor dynamics suppressed , and cholesterol induces the strongest effect . Our results show that cholesterol has a preference to bind to β2AR at specific locations on its surface . We identified three high-affinity cholesterol interaction sites in β2AR ( Figure 2C , D ) : IC1 ( at the cleft of H1-H4 on the intracellular side ) , IC2 ( H5-H6 on the intracellular side ) , and EC1 ( the H5-H6-ECL3-H7 region on the extracellular side ) . IC1 and EC1 are in agreement with the locations of cholesterol found in GPCR crystal structures ( Cherezov et al . , 2007; Hanson et al . , 2008; Liu et al . , 2012 ) . IC1 contains a cholesterol consensus motif that predicts cholesterol binding for 44% of human class A receptors ( Hanson et al . , 2008 ) . Moreover , these binding sites appear to be evolutionarily conserved in β2AR , which suggests their possible allosteric role in receptor function . A recent simulation study reported a correlation between cholesterol occupancy at IC1 and β2AR dimerization ( Prasanna et al . , 2014 ) . However , not much is known about the functional relevance of cholesterol binding to the other sites of β2AR . The present work for the inactive conformation of β2AR shows that cholesterol binding at IC2 and EC1 ( Figure 2C , D ) strongly influences the conformational dynamics of β2AR ( Figure 1 ) . In a cholesterol-free membrane the receptor samples multiple conformational states ( Figure 1B ) accounting for the high basal activity of β2AR ( Manglik and Kobilka , 2014; Kobilka , 2013 ) . Our results show that the presence of cholesterol in high densities around H5-H6-H7 impedes the dynamic nature of the receptor . In cholesterol-containing ( ≥10 mol% cholesterol ) membranes ( Figure 1C and Figure 1—figure supplement 1D , E ) , the overall structural flexibility of the receptor is significantly reduced to one predominant conformation . We observed that in the presence of strongly bound cholesterol , H5 and H6 undergo much smaller displacements from their average positions as compared to the situation without cholesterol ( Figure 1F ) . Cholesterol analogues that occupy the same interaction sites also restrict the β2AR conformation ( Figure 2—figure supplement 9 ) , although their effects are weaker compared to those of cholesterol . Cholesterol or cholesterol-like molecules bound at these inter-helical clefts can thus confine the movement of the respective helices to a substantial degree , thus dampening the overall conformational dynamics of the receptor . At IC2 of inactive β2AR , cholesterol pushes the intracellular end of H6 more towards the core of the helical bundle and prevents the outward movement of H6 required for G protein binding . The restriction of H6 movement by cholesterol is a potentially important allosteric effect , which can be used to modulate the receptor activity . Interestingly , our study on the active-state β2AR also exhibits a high cholesterol density at IC2 ( Figure 3D , F ) . Here cholesterol bound at IC2 acts as a spacer between H5-H6 and restricts the movement of H6 , thereby stabilizing the open active-like conformation of the receptor ( Figure 3D ) , while in the absence of cholesterol the receptor is more prone to undergoing spontaneous deactivation ( Figure 3E; Figure 3—figure supplement 1 ) . This result supports the postulate that cholesterol restricts the conformational dynamics of the receptor by binding at specific interaction sites and governs changes between different receptor states , therefore modulating its function . Moreover , cholesterol binding at IC2 in both inactive and active states of β2AR as found in our simulations highlights the biological relevance of this interaction site in allosteric regulation of the receptor conformation . The highly conserved IC1 site shows no major influence on the mobility of H5-H6 . On the other hand , IC1 exerts a stabilizing effect on H4 ( Figure 2—figure supplement 10 ) , in agreement with experiments ( Hanson et al . , 2008 ) . As H4 is one of the weakest points of the β2AR fold , its decreased mobility may account for the enhanced stability of the receptor . Cholesterol modulates the physical properties of membranes by increasing the bilayer thickness and order , and slowing down the dynamics . These general membrane effects can also influence the dynamic nature of a membrane protein ( Manna and Mukhopadhyay , 2011 ) . However , here we found that membrane-mediated interactions do not affect β2AR conformation to a significant degree ( Figure 4 ) . GPCRs are signaling machines that function by toggling between multiple conformers ( Latorraca et al . , 2016 ) . The dynamic nature of GPCRs has made their crystallization process extremely challenging ( Kobilka , 2013 ) . Besides techniques like protein engineering and use of detergents to increase the intrinsic stability of the receptor ( Loll , 2014 ) , cholesterol/CHS has emerged as a necessary component for crystallization of many GPCRs , including β2AR ( Cherezov et al . , 2007; Hanson et al . , 2008; Zocher et al . , 2012; Loll , 2014 ) . Our work shows that in the presence of more than ~10 mol% cholesterol , inactive β2AR partly loses conformational variability and populates just one major conformation . Achieving conformational homogeneity is the key to crystallize membrane proteins ( Loll , 2014 ) . In agreement with our results , a recent experimental study showed that CHS impacts the conformational dynamics of a GPCR leading to a restricted conformational space ( Casiraghi et al . , 2016 ) . Earlier it was experimentally reported that cholesterol induces a more compact conformational state of the oxytocin receptor ( Muth et al . , 2011 ) . Our results are also in agreement with a recent dynamic single-molecule force spectroscopic study , which showed that CHS strengthens interactions that stabilize the structural segments in β2AR and thereby considerably increase the kinetic , energetic , as well as the mechanical stability of the receptor ( Zocher et al . , 2012 ) . In addition , the function of adrenergic receptors is known to be modulated by cholesterol: cholesterol depletion enhances β2AR-associated signaling , while increased cholesterol content inhibits signaling ( Paila et al . , 2011; Pontier et al . , 2008 ) . To our knowledge , the results presented in this work provide the first atomic-scale picture of how lipids can govern the conformation of membrane receptors through direct lipid-protein interactions in specific lipid binding sites , and hence dictate the state of a receptor . The receptor-cholesterol interactions , such as those observed in our simulations for β2AR , can conceivably govern the signaling of many GPCRs in the given protein family . All simulations were performed using the GROMACS 4 . 6 . x package ( Berendsen et al . , 1995; Hess et al . , 2008 ) . The all-atom OPLS-AA ( optimized potentials for liquid simulations ) force field was used to parameterize the protein , ions , and pyrene ( Jorgensen et al . , 1996; Kaminski et al . , 2001 ) . Force field parameters for cholesterol , cholesteryl hemisuccinate , and oxysterols were taken from previously published papers ( Manna et al . , 2015; Kulig et al . , 2015a , 2015b , 2014 ) . For the studied phosphatidylcholines ( DOPC and PC-20:0/22:1 c13 ) , we used new torsional and Lennard-Jones parameters derived for saturated ( Maciejewski et al . , 2014 ) and unsaturated hydrocarbons ( Kulig et al . , 2015c , 2016 ) and the torsional potential developed for the glycerol backbone and the phosphatidylcholine head group ( Maciejewski et al . , 2014 ) . The TIP3P model , which is compatible with the OPLS parameterization , was used for water molecules ( Jorgensen et al . , 1983 ) . All simulations of the systems considered in this work ( Table 1 ) were performed under the isobaric-isothermal ( NpT ) ensemble . A time step of 2 fs was used for integrating the equations of motion . Periodic boundary conditions were applied in all three directions of the system . The temperature of the system was maintained at 310 K by employing the v-rescale ( stochastic velocity rescaling ) thermostat ( Bussi et al . , 2007 ) with a time constant of 0 . 1 ps . The temperatures of the receptor , lipids , and solvent molecules were controlled independently . The pressure of the system ( 1 bar ) was maintained semi-isotropically using the Parrinello–Rahman barostat ( Parrinello and Rahman , 1981 ) with a 1 ps time constant . The LINCS algorithm was applied to preserve hydrogen covalent bond lengths ( Hess et al . , 1997 ) . Lennard-Jones interactions were cutoff at 1 . 0 nm . The particle mesh Ewald ( PME ) method ( Essmann et al . , 1995 ) was employed for long-range electrostatic interactions using a real space cutoff of 1 . 0 nm , β-spline interpolation ( order of 6 ) , and a direct sum tolerance of 10−6 . The initial coordinates of β2AR were taken from our recently published work ( Manna et al . , 2015 ) , in which the structural modifications made for crystallization of the inactive β2AR structure [PDB id: 3D4S] ( Hanson et al . , 2008 ) were reverted back to its original sequence . This inactive crystal structure of β2AR bound to the partially inverse agonist timolol was heavily engineered to facilitate crystallization ( Hanson et al . , 2008 ) . We reverted all the structural modifications from the experimentally determined structure , i . e . , we removed mutations ( E1223 . 41W on the transmembrane helix H3 and the N1875 . 26E mutation on the extracellular loop 2 ) , removed the T4-lysozyme attached between the transmembrane helices 5 and 6 , and replaced it with the missing intracellular loop 3 . We did not attempt to model the unresolved N-terminal ( 32 residues ) and C-terminal ( 71 residues ) parts . The details of the procedure used to prepare the receptor model for our simulations are described elsewhere ( Manna et al . , 2015 ) . In the present work , we considered the apo-receptor ( without a ligand ) , as we were interested in the intrinsic dynamics of β2AR . For simulations with the active-state β2AR conformation , the starting structure was taken from the crystal structure of the receptor bound to an agonist and a Gs protein ( Rasmussen et al . , 2011 ) . Here again we considered the apo-form of the receptor without the ligand and the G protein . Additionally , we removed the lysozyme and modeled the missing loop regions ( A176-H178 and F240-F264 ) , but the mutations were kept as such . We simulated β2AR embedded in a number of lipid bilayers ( Table 1 ) with varying lipid composition . The lipid contents used in the studies were as follows: The lipid bilayers ( without β2AR ) were constructed using in-house scripts , and they were subsequently solvated with water . These lipid bilayers were then equilibrated for 100–200 ns . Next , β2AR was placed into the above-mentioned pre-equilibrated bilayers in such a manner that the lipid arrangement around the receptor was completely random and that there was no cholesterol or cholesterol analogue initially bound to β2AR . For incorporating the receptor into a pre-equilibrated lipid bilayer , we followed our recently published method , where the receptor was pushed into a lipid membrane from its side by applying a high lateral pressure on the system ( Javanainen , 2014 ) . Each system contained one β2AR and 256–512 lipids . Each of the systems was explicitly solvated by water . In all cases , counterions ( 8 Cl– ions for β2AR , and additional Na+ counter ions for bilayers containing the anionic CHSA ) were added to maintain electroneutrality of the systems . NaCl salt was added to achieve the physiological salt concentration of 150 mM . Subsequently each system was energy minimized and then equilibrated in two stages with position restraints first on protein heavy atoms and then on the backbone . Following equilibration ( 100 ns ) , all restraints were released and the equilibrated systems were subjected to microsecond length ( 1–2 . 5 μs ) production simulations . Multiple independent simulations were performed for each lipid composition , either by starting from a different lipid arrangement around β2AR ( for systems with no sterols initially bound to the receptor ) or starting with different initial velocities ( for systems with sterols initially bound to the receptor ) . Additional simulations were performed where cholesterol or its analogues were initially attached to certain locations on the surface of the receptor , and this receptor-lipid complex was then embedded to a cholesterol-free DOPC bilayer . Here we performed two sets of control simulations . In one set of simulations , two cholesterol or CHS ( neutral or anionic ) molecules were bound at the cleft formed by the intracellular side of the transmembrane helices H1-4 as predicted from the crystal structure ( Hanson et al . , 2008 ) . In another set of control simulations , cholesterol molecules were initially bound at the eight interaction sites of β2AR predicted by our simulations ( see discussion in the main article ) . The simulation conditions ( as to counterions and salt , release of restrains , simulation times , etc . ) were as described above . The systems investigated in this study are summarized in Table 1 . The total simulation time for the atomistic systems studied in this work covers a period of more than 100 μs . For calculation of deviations of helix ends , we first calculated their time series of X , Y , and Z coordinates . The coordinates were then divided into two groups based on whether the upper and lower halves of the helixes ( backbone atoms ) were in contact ( ≤ 0 . 5 nm ) with cholesterol ( heavy atom ) or not . Separately in each group , the distance from the average point of the group at each time frame ( say ith frame ) was calculated by:di2= ( xi−xg ) 2+ ( yi−yg ) 2+ ( zi−zg ) 2 , where xi , yi , zi were the coordinates of the ith frame , and xg , yg , zg were the average values . The standard deviation of each group was then calculated by:σ= 1Ng∑i=1Ngdi2 The average standard deviation of different simulations was calculated as a weighted average depending on the number of frames ( Ng ) of the group in each simulation . The 2D number density maps were calculated using the g_densmap tool of GROMACS . The two bilayer leaflets were calculated separately . The output was then processed ( using an in-house script ) to normalize the maximum number density to one . We calculated the 2D number densities of cholesterol ( non-hydrogen atoms ) and β2AR ( backbone atoms of transmembrane region ) separately . A residue of β2AR was considered to be in contact with cholesterol , when any of its non-hydrogen atoms was within ≤0 . 5 nm of any heavy atom of cholesterol . The total occupancy time was then normalized over the entire length of a simulation , i . e . , an occupancy time of one means that the particular residue of β2AR was in contact with cholesterol throughout the simulation , whereas a value of zero means no contact . The calculated total occupancy time per residue of β2AR was mapped onto the receptor’s surface to highlight the regions of β2AR involved in cholesterol binding . We analyzed amino acid sequences of β2AR orthologues from the available databases . We used the PhylomeDB server ( http://phylomedb . org/ ) ( Huerta-Cepas et al . , 2014 ) for finding orthologues and Clustal Omega ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) ( Sievers et al . , 2011 ) for sequence alignment . The amino acid residues of β2AR segments constituting the cholesterol binding sites as obtained from our simulations were used for the set of sequences obtained ( Figure 2—figure supplement 2 , Figure 2—figure supplement 3 , Figure 2—figure supplement 4 ) . The sequences in question belong to diverse species , such as insects , fish , birds , reptile , mammals , etc . The order parameter of lipid acyl chains was calculated using :SCD=⟨32 ( cos2 ⁡θi ) − 12⟩ where θi is the angle between a C-D bond ( C-H in simulations ) of the ith carbon atom and the bilayer normal . The angular brackets denote averaging over time and molecules in a bilayer . Bilayer thickness was defined as the distance between the average planes formed by phosphorous atoms in the two bilayer leaflets . We used the g_lomepro tool ( Gapsys et al . , 2013 ) to generate the 2D distribution of bilayer thickness . For the calculation of the lifetime of cholesterol bound to the cholesterol interaction sites on the receptor surface , we first monitored the binding/unbinding events of each individual cholesterol molecule along the simulation trajectory . A cholesterol molecule was considered bound when any of its heavy atoms came within ≤0 . 5 nm from an interaction site . To define the three major interaction sites on the β2AR surface , we used the amino acid residues ( with contact fraction ≥ 0 . 4 ) as shown in Figure 2—figure supplement 2 . The time series was then additionally smoothed ( over one ns time windows ) to discard very rapid ‘leave and return’ motions of cholesterol that take place due to thermal fluctuations . Given that lateral diffusion of lipids at the protein surface is very slow , and the lipids essentially do not move at all during a 1-ns time window , these fluctuations were then taken care of by the smoothing procedure . We then calculated the normalized time correlation function ( to describe the time-dependent probability of cholesterol that is next to the receptor to stay in contact with the receptor ) over all individual cholesterol binding/unbinding events occurred in all independent simulation trajectories for all cholesterol molecules present in a system at a given cholesterol concentration ( Arnarez et al . , 2013; Horn et al . , 2014 ) . For all analysis to measure time-averaged properties , the first 100 ns of production simulations were excluded from the calculation . Error bars were estimated through standard error , calculated by dividing the standard deviation of a given data set with the square root of its sample size ( Manna et al . , 2015; Kulig et al . , 2014 ) . We used the g_analyze tool of GROMACS for error estimation .	Proteins known as G protein-coupled receptors , or GPCRs for short , detect and respond to hormones and other signaling molecules found outside cells . A signaling molecule activates a GPCR by binding to it and causing the receptor to change its shape . This triggers a cascade of signals inside the cell that leads to the cell responding in a particular way . There are over 800 different GPCRs in human cells , making them the largest family of cell surface proteins . GPCRs span the membrane that surrounds each cell . This membrane is made of molecules called lipids and previous studies have shown that many lipids are able to bind to GPCRs and influence their shape and activity . Lipids can cause these changes via so-called ‘allosteric’ regulation , in which the lipid binds to a site on the receptor that is separate to where the signal molecule binds . Lipid binding can either enhance or inhibit the activity of the receptor . Human β2-adrenergic receptor is one of the best-studied GPCRs . It responds to a hormone called epinephrine ( also known as adrenaline ) , which plays important roles in many organs in the body , including the heart and lungs . A lipid called cholesterol , which is plentiful in the cell membrane , can also bind to this receptor and influence its shape , but how this happens was not fully understood . Manna et al . now use computer simulations to analyze the interaction between cholesterol and β2-adrenergic receptor in more detail . The simulations reveal that cholesterol makes the β2-adrenergic receptor less flexible so that it can only adopt certain shapes . This helps to stabilize both the inactive and active states of the receptor so that it is not as easy for the receptor to switch between them . The cholesterol molecules bind to specific sites on the receptor within the region of the protein that crosses the cell membrane . The new findings of Manna et al . provide detailed insights into how cholesterol governs the shape and activity of the β2-adrenergic receptor . The next step is to extend this analysis to other types of lipids and GPCRs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2016	Mechanism of allosteric regulation of β2-adrenergic receptor by cholesterol
The Notch pathway is integrated into numerous developmental processes and therefore is fine-tuned on many levels , including receptor production , endocytosis , and degradation . Notch is further characterized by a twofold relationship with its Delta-Serrate ( DSL ) ligands , as ligands from opposing cells ( trans-ligands ) activate Notch , whereas ligands expressed in the same cell ( cis-ligands ) inhibit signaling . We show that cells without both cis- and trans-ligands can mediate Notch-dependent developmental events during Drosophila oogenesis , indicating ligand-independent Notch activity occurs when the receptor is free of cis- and trans-ligands . Furthermore , cis-ligands can reduce Notch activity in endogenous and genetically induced situations of elevated trans-ligand-independent Notch signaling . We conclude that cis-expressed ligands exert their repressive effect on Notch signaling in cases of trans-ligand-independent activation , and propose a new function of cis-inhibition which buffers cells against accidental Notch activity . Canonical Notch signaling begins when the Notch receptor receives a stimulus from a DSL-type ligand ( Delta [Dl] or Serrate [Ser] in Drosophila ) in an adjacent cell , which leads to γ-secretase-dependent cleavage of Notch , and translocation of the intracellular domain—NICD— into the nucleus to act as a transcriptional co-activator ( de Celis , 2013 ) . Notch may also be activated in a non-canonical , DSL-ligand independent manner ( Hori et al . , 2012 ) . DSL ligands can cis-inhibit ligand-dependent Notch activation when expressed in the same cell as the receptor ( Micchelli et al . , 1997; Del Álamo et al . , 2011 ) . However , the possibility of a relationship between DSL-ligand independent Notch activation and cis-expressed ligands has not been explored . The developing Drosophila egg chamber is a convenient model for dissecting the effects of Notch ligands in cis and in trans , as Dl is the sole signaling source and the signal sending and receiving cells can be easily distinguished ( Deng et al . , 2001; López-Schier and St Johnston , 2001 ) . ( Figure 1—figure supplement 1 provides a brief schematic depiction of the stages of early oogenesis . ) At oogenesis stage 7 , Notch signaling is activated in the somatic follicle cells by a robust germline Dl upregulation , which leads to the expression of Hindsight ( Hnt ) , downregulation of Cut , and the polyploidization of the follicle cells ( Deng et al . , 2001; López-Schier and St Johnston , 2001; Sun and Deng , 2005 , 2007 ) ( Figure 1A ) . When Dl germline mutant clones were generated ( i . e . , trans-activation was removed ) , the follicle cells failed to downregulate Cut expression , which persisted past stage 7 , indicative of a failure to activate Notch ( Figure 1B ) . In contrast , Dl follicle cell mutant clones show precocious Cut downregulation at stage 6 attributable to the relief of cis-inhibition ( Poulton et al . , 2011 ) ( Figure 1C ) . Surprisingly , Dl mutant clones in the follicle cells bordering Dl mutant clones in the germline ( i . e . , a germline with no signaling source , herein referred to as Dl-/Dl- cells ) show correct Hnt and Cut expression from stage 7 ( Figure 1D , E , Figure 1—figure supplement 2A , B ) . These Dl-/Dl- clones also correctly transit into the endocycle , as their nuclear volumes are similar to wild-type follicle cells in the later stages of oogenesis after polyploidization ( Figure 1F , G ) , whereas cells neighboring Dl-/Dl- follicle cell clones ( retaining a cis-ligand but without a trans-ligand ) are comparable to wild-type cells before entry to endocycle ( Figure 1F , G ) . Removal of both cis- and trans-Dl through knockdown of Dl by RNA interference ( RNAi ) simultaneously in the germline and soma confirmed this finding ( Figure 2—figure supplement 1A , B ) . Together , these observations provide evidence that follicle cells without both cis- and trans-ligand sources can still enter the endocycle stages of oogenesis . This back-up route to the endocycle is not a co-option of Ser in place of Dl , as DlRevF10SerRx82 double clones recapitulated the Dl-/Dl- phenotype ( Figure 1E , Figure 1—figure supplement 2A ) . 10 . 7554/eLife . 04415 . 003Figure 1 . Follicle cells without DSL ligand bordering germline cells without DSL ligand show proper Notch activation and downstream differentiation . Illustrations legend: active Notch = white cytoplasm , inactive Notch = red cytoplasm , WT cell = grey nuclei , mutant clone = white nuclei . ( A–E ) . Follicle cells downregulate Cut at stage 7 of oogenesis ( A ) . DlrevF10 mutant germline cells cause late Cut expression in follicle cells ( B ) . DlrevF10 mutant follicle cells downregulate Cut early ( C ) . DlrevF10 follicle cell clones bordering DlrevF10 germline clones show proper Cut downregulation ( D ) . DlrevF10SerRx82 mutant follicle cell clones bordering DlrevF10SerRx82 germline clones also show proper Hnt ( E ) . See Figure 1—figure supplement 2 for a z-series image for 1D and 1E . These germline/follicle cell clones ( D and E ) show increased nuclear size comparable to wild-type ( WT ) follicle cells which have entered the endocycle ( n = 8 for each stage/genotype ) ( F and G ) . For ( G ) , Welch t-tests were done to assess significance between each condition . The only comparisons that were not significant were between WT stage 10B and Dl-/Dl- clones and between WT stage 6 and Dl germline clones , indicating nuclear size in germline clones alone is similar to that of cells before the endocycle , whereas Dl-/Dl- clonal nuclei are more similar in size to cells that have entered the endocycle . Scale bars represent 20 μm , except in F , where the scale bar represents 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 00310 . 7554/eLife . 04415 . 004Figure 1—figure supplement 1 . A schematic depiction of the early stages of Drosophila oogenesis . Oogenesis begins in the germarium , where germline stem cells divide four times , producing a 16-cell germline cyst which is encapsulated by somatic follicle cells ( FCs ) . When the FCs complete encapsulation and bud from the germarium , this is termed a stage 1 egg chamber . The egg chamber then grows and the FCs undergo mitosis until stage 6 , and during these stages Cut is expressed and cells remain diploid . At stage 5 , Dl is strongly upregulated in the germline . The transition from stage 6 to stage 7 is defined by activation of Notch , upregulation of Hnt , repression of Cut , and the endocycling of the FCs . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 00410 . 7554/eLife . 04415 . 005Figure 1—figure supplement 2 . Z-stacked images of Dl-/Dl- clones and quantification of Cut staining in egg chamber clones . Z series confocal images of DlrevF10SerRx82 ( A ) or DlrevF10 ( B ) germline/follicle cell clones from Figure 1D , E stained for Hnt ( A ) or Cut ( B ) . Notice Hnt staining in the anterior end of ( A ) is owing to the formation of a partial germline clone containing both wild-type ( WT; anterior , left ) and DlrevF10SerRx82 ( posterior , right ) nurse cells , and therefore the anterior-most WT follicle cells have a ligand source to induce normal Hnt expression . Quantification of Cut expression in Dl-/Dl- clones induced by RNAi or by DlrevF10 homozygous mutant cells , and the effect of loss of Notch or Su ( H ) ( C ) ( n = 30 for Dl-/Dl- MARCM , n = 25 for Dl-/Dl- RNAi , n = 38 for Dl-/Dl- MARCM + N RNAi , and n = 24 for Dl-/Dl- RNAi + Su ( H ) 47 MARCM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 005 To determine whether the entry into the endocycle in Dl-/Dl- follicle cells still requires the function of Notch , we implemented the mosaic analysis with a repressible cell marker ( MARCM ) system ( Lee and Luo , 2001 ) . The MARCM system enables us to create mutant clones while driving expression of a UAS transgene specifically in those clonal cells . Dl-/Dl- clones driving expression of NotchRNAi show a significantly higher proportion ( p < 0 . 0001 ) of late Cut-expressing cells than the Dl-/Dl- clones alone , indicating that Notch is still required for the mitotic-to-endocycle switch ( Figures 1D and 2A , Figure 1—figure supplement 2C , Figure 2—figure supplement 1C , Supplementary file 1 ) . Likewise , MARCM clones for the null allele of Suppressor of Hairless ( Drosophila Notch transcriptional effector ) , Su ( H ) 47 , in RNAi-induced Dl-/Dl- clones also show late Cut expression ( p < 0 . 0001 ) ( Figure 2B , Figure 1—figure supplement 2C , Supplementary file 1 ) in comparison with RNAi-induced Dl/Dl- clone controls ( Figure 2—figure supplement 1A , D ) . A Notch activity reporter , Notch Responsive Element ( NRE ) -green fluorescent protein ( GFP ) ( Stempfle et al . , 2010 ) was also upregulated in Dl-/Dl- clones as early as stage 2 , and this expression persisted beyond stage 6 ( Figure 2C , D ) , suggesting that NRE-GFP is probably more sensitive to Notch activation than Hnt in follicle cells . Together , these results suggest that Notch activity occurs independently of canonical ligands when both cis- and trans-ligands are removed , resulting in normal downstream developmental events in the follicle cells . Consistently , DlRevF10SerRx82 double mutant clones in the wing and eye discs show a slight cell-autonomous upregulation of NRE-GFP in the clone center , which would only occur if cis-inhibition blocked a DSL-independent mode of Notch activity , as interior cells have no access to trans-ligand ( Figure 2E , F ) . This NRE-GFP upregulation was spatially variable in the wing disc , having the highest prevalence in the notum region ( 25% incidence ) , a low incidence in the dorsal pouch ( 8% ) , whereas in the ventral pouch region it was never seen ( n = 80 ) ( Supplementary file 1 ) , perhaps owing to the differential regulation of Notch degradation throughout the wing disc ( Hori et al . , 2011 ) . As reported previously , most wing disc clones showed a higher NRE-GFP upregulation in the clone boundary where there is access to trans-ligand , indicating that the ligand-independent Notch activity observed occurs at a rather low level . 10 . 7554/eLife . 04415 . 006Figure 2 . Cis-ligand represses ligand-independent Notch activity in the follicle cells and imaginal discs . DlrevF10 mutant MARCM germline/follicle cell clones co-expressing NotchRNAi show prolonged Cut expression ( A ) . Su ( H ) 47 MARCM mutant germline/follicle cell clones co-expressing DlRNAi show failure to enter the endocycle ( B ) . Germline clones are shown by late Cut expression in wild-type follicle cells ( A , B , see arrowheads ) . See Figure 2—figure supplement 1A for control DlRNAi-induced germline follicle cell clones . Notch Responsive Element-green fluorescent protein ( NRE-GFP ) is upregulated beginning from stage 2 ( C ) and through later stages ( D ) in DlRevF10 germline and follicle cell clones . NRE-GFP is also upregulated cell-autonomously in DlRevF10SerRx82 mutant clones in eye ( E ) and wing ( F ) imaginal discs . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 00610 . 7554/eLife . 04415 . 007Figure 2—figure supplement 1 . Control experiments relating to Figure 2 . The Dl-/Dl- phenotype can also be recapitulated using DlRNAi , which knocks down Dl in both the germline and soma using the FLP-out method ( A and B ) . See the arrowhead in ( B ) for wild-type ( WT ) Dl staining . Again , germline clones are evidenced by aberrant Cut expression in WT follicle cells . MARCM-induced clones expressing only NotchRNAi show late Cut expression ( C ) . Su ( H ) 47 mutant clones created using the MARCM system also show late Cut staining and smaller nuclei ( D ) . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 007 Drosophila S2 cells are reported to have no Dl expression and a very low level of Ser expression , which had no effect on Notch signaling ( Fehon et al . , 1990; Graveley et al . , 2011 ) ( Figure 3—figure supplement 1 ) , and have been used as a model to study ligand-independent Notch activity ( Hori et al . , 2011 ) . Upon transfection with pMT-NFL , a CuSO4-inducible full-length Notch construct , Notch activation was increased by a factor of 5 . 13 compared with the control cells , as indicated by a NRE-firefly luciferase reporter gene ( p < 0 . 0001 ) ( Figure 3C ) . Notch activation in S2 cells is at least partially dependent on endosomal trafficking , as double-stranded ( ds ) RNA against early endosome component , Rab5 , or multivesicular body sorting protein , hrs , reduced the levels of Notch activation ( Figure 3A , B ) . This is consistent with the in vivo studies indicating that ligand-independent Notch activation relies heavily on receptor trafficking ( Hori et al . , 2012 ) ( Rab5 p = 0 . 00623 , hrs p = 0 . 0159 ) , and our observation that Notch accumulates in Dl-/Dl- clones ( Figure 3—figure supplement 2 ) . A requirement for trafficking is consistent with the results of others who have demonstrated aberrant Notch activation in follicle cell mutants for trafficking components ( Wilkin et al . , 2004; Vaccari et al . , 2008; Schneider et al . , 2013 ) , such as tsg101 mutant clones , which show early Notch activation in the follicle cells ( Figure 3—figure supplement 3 ) . Furthermore , co-transfecting pMT-NFL with pMT-GAL4 and pUASt-Serdel3 , a form of Ser that cannot activate Notch , but only cis-inhibit , ( Fleming et al . , 2013 ) almost entirely abolished the Notch activation detected when NFL was transfected alone ( p = 0 . 0048 ) ( Figure 3C ) . These results suggest that if Notch is expressed in a cell free of cis- and trans-ligands , DSL ligand-independent activity will occur and that cis-inhibition is extremely efficient in preventing this ‘accidental’ Notch activity as it travels through the endosomal pathway en route to degradation . 10 . 7554/eLife . 04415 . 008Figure 3 . DSL-ligand-independent Notch activity in S2 cells is buffered by cis-ligand . Trafficking is important for Notch activation in S2 cells , as treatment with Rab5 dsRNA ( A ) or hrs dsRNA ( B ) significantly decreases the amount of Notch activated in S2 cells as shown by Notch-responsive luciferase activity ( NRE-firefly ) in relative light units ( RLU ) . Transfecting only pMT-NotchFL into S2 cells causes a 5 . 13-fold increase in Notch activation , which is almost entirely reduced ( 1 . 34-fold from the negative control ) by co-transfection of pMT-GAL4 and pUASt-Serdel3 ( C ) . Each experiment was carried out with two technical replicates and three biological replicates . Means of the technical replicates were used to carry out a paired t-test ( n = 3 ) for each comparison . Error bars represent standard deviation ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 00810 . 7554/eLife . 04415 . 009Figure 3—figure supplement 1 . Addition of Ser dsRNA had no effect on the Notch activation in S2 cells in comparison with cells treated with control green fluorescent protein ( GFP ) dsRNA , indicating that the small amount of Ser expression is either not translated or does not significantly contribute to Notch activation upon transfection with pMT-NFL . This validates our assumption that the Notch activation which occurs in S2 cells is by a DSL-ligand-independent mechanism . Dl was not tested , as studies have already shown a lack of Dl mRNA and protein in S2 cells ( Fehon et al . , 1990; Graveley et al . , 2011 ) . Again , experiments were carried out with two technical replicates and three biological replicates , with means of the technical replicates used for a paired t-test to assess significance . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 00910 . 7554/eLife . 04415 . 010Figure 3—figure supplement 2 . Notch accumulates in Dl-/Dl- clones . Staining either Notch extracellular domain ( A ) or intracellular domain ( B ) showed increased Notch levels in DlrevF10 mutant germline/follicle cell clones . This could be seen as early as stage 2 where Notch protein seemed membrane localized ( A ) , but by stage 5 it no longer localized to the membrane and appeared as a somewhat cloudy cytoplasmic accumulation ( B ) . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 01010 . 7554/eLife . 04415 . 011Figure 3—figure supplement 3 . Follicle cells mutant for ESCRT component tsg101 show early Notch activity in the follicle cells ( Vaccari et al . , 2008 ) . tsg101111019 clones show early Cut downregulation . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 011 We next explored whether cis-inhibition can also block ligand-independent Notch activity induced in aberrant genetic backgrounds . The Notch target , Wingless ( Wg ) is normally expressed along the dorsoventral boundary of the wing disc ( Figure 4A ) . Lethal giant disc ( lgd ) homozygous mutant ( lgdd7 ) larvae display overgrown imaginal discs and ubiquitous ligand-independent Notch activation in the wing pouch region , as shown by upregulation of Wg ( Figure 4B ) . Notch activation in lgd mutant cells is caused by a defect in Notch trafficking and degradation , as the receptor is aberrantly transported to the limiting membrane of the lysosome which facilitates production of NICD ( Childress et al . , 2006; Gallagher and Knoblich , 2006; Jaekel and Klein , 2006; Schneider et al . , 2013 ) . Using dpp-GAL4 to misexpress UAS-Dl along the anterior–posterior axis of the wing disc in lgdd7 homozygous larvae , Wg expression was considerably reduced along the dpp expression domain , indicating that cis-inhibition can block the ligand-independent Notch activity observed in this situation ( Figure 4C ) . Overexpression of Deltex ( Dx ) , an E3 ubiquitin ligase that stimulates Notch monoubiquitination and promotes its trafficking to the lysosomal limiting membrane , has also been shown to induce ligand-independent Notch activation specifically in the ventral wing pouch region ( Matsuno et al . , 2002; Hori et al . , 2004; Wilkin et al . , 2008; Schneider et al . , 2013 ) ( Figure 4D ) . We used patched ( ptc ) -GAL4 to drive expression of UAS-Dx with either UAS-Dl or UAS-Serdel3 , whose ectopic expression leads to a reduction of Wg staining along the dorsoventral boundary ( Micchelli et al . , 1997; Fleming et al . , 2013 ) ( controls in Figure 4—figure supplement 1A , E ) . Co-expression of Dx and Dl led to a decrease in Wg expression in the ventral ptc domain as compared with expression of Dx alone ( Figure 4E ) . When UAS-Dx and UAS-Serdel3 were co-expressed , there was a small but noticeable , albeit variable , decrease in Dx-induced Notch activation ( Figure 4—figure supplement 1B–D ) . This incomplete reduction was probably due to the previously noted , slightly compromised , cis-inhibitory potential of UAS-Serdel3 ( Fleming et al . , 2013 ) ( Figure 4—figure supplement 1A ) . Taken together , these results provide evidence that cis-ligand has a negative effect on the raised levels of DSL-ligand independent Notch activation incurred in genetically abnormal cells . 10 . 7554/eLife . 04415 . 012Figure 4 . Notch ligand buffers against genetically induced DSL-independent activation . Wing discs were stained with Wg antibody and illustrations are colored red where Wg is expressed ( A–E ) . A wing disc with regions of interest is labeled and WT Wg staining shown ( A ) . lgdd7/lgdd7 wing discs show ubiquitous Wg expression in the wing pouch as a result of DSL-ligand-independent Notch activity ( B ) . Misexpression of UAS-Dl in lgdd7/lgdd7 discs causes a reduction in Wg staining along the anteroposterior boundary of the pouch ( C ) . ptcGAL4 drives UAS-Dx causing ectopic Notch activity in the ventral wing pouch ( D ) . Co-expression of Dx with Dl reduces Wg staining in the ptc domain ( E ) , although , as in lgdd7/lgdd7 discs , the reduction is not complete towards the dorsoventral boundary . Cis-ligand also decreases Notch activation caused by genetic defects in S2 cells ( F–H ) . Co-transfection with pMT-NFL and pMT-Dx caused a significant increase in Notch luciferase reporter expression , and adding Serdel3 significantly reduced this Dx-induced activation ( F ) . Cells treated with lgd dsRNA ( G ) or ESCRT-III component , shrub , dsRNA ( H ) also caused significant increases in Notch reporter activity , either of which could be blocked by addition of Serdel3 . For each of the S2 cell experiments , means were taken for technical duplicates and used for a paired t-test for three biological replicates . Error bars represent SD . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 01210 . 7554/eLife . 04415 . 013Figure 4—figure supplement 1 . Co-expression of UAS-Dx and UAS-Serdel3 has a variable effect on DSL-independent Notch activation . Wing discs were stained with Wg antibody ( A–E ) . Illustrations show Wg staining in red , with lower intensities of Wg presence being shown in pink ( A–E ) . ptcGAL4 driving green fluorescent protein ( GFP ) and UAS-Serdel3 along the anteroposterior ( AP ) boundary ( A ) . This caused an incomplete reduction in Wg staining along the dorsoventral ( DV ) boundary . The slight increase in the red channel along the AP boundary is because the UAS-Serdel3 construct is tagged with tomato and bleeds into our ‘red’ secondary antibody confocal channel . For the rest a different channel was used . Coexpression of UAS-Serdel3 with UAS-Dx showed a variable effect on the Dx-induced aberrant Notch activity ( B–D ) . Sometimes Dx-induced Notch activity was completely abolished ( B ) , sometimes only partially reduced ( C ) , and sometimes remained unchanged ( D ) . All UAS-Serdel3/UAS-Dx discs are from the same round of antibody staining and taken with the same scale and settings on confocal microscopy . ptcGAL4 driving UAS-Dl caused a complete reduction of Wg at the DV boundary and elicited aberrant Wg expression on the boundary of the ptc domain . ( E ) Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 01310 . 7554/eLife . 04415 . 014Figure 4—figure supplement 2 . Endogenous DSL-independent Notch activity in crystal cells is reduced by cis-inhibition . Lz-GAL4-driven green fluorescent protein ( GFP ) expression is an efficient marker of crystal cells which show a low incidence of bursting ( A and E ) . Misexpressing UAS-Serdel3 increased the frequency of witnessing bursting crystal cells ( see arrowheads in B ) ( B and E ) . Lymph glands were counted for each genotype ( n = 14 for lz > GFP , n = 12 for lz > GFP; UAS-NotchRNAi , and n = 14 for lz > GFP; UAS-Serdel3 ) . Welch's t-test was used to assess significance between wild-type ( WT ) lymph glands and each of the experimental groups ( p = 0 . 043 and p = 0 . 029 , respectively ) . To determine whether Serdel3-misexpression induced ‘bursting’ was caused by the cis-inhibitory effect of Ser on ligand-independent Notch activation , we used the Notch activity reporter E ( spl ) :mβ-CD2 . We focused our analysis on larger crystal cells , which enter the endocycle as part of their differentiation ( Terriente-Felix et al . , 2013 ) , and therefore are the ones most probably undergoing ligand-independent Notch activation . For illustrations , all GFP-positive cells were outlined and were filled in with differing shades of red corresponding to Notch reporter staining intensity . E ( spl ) mβ:CD2 is expressed in 55% of crystal cells and 77 . 4% of mature crystal cells ( C and F ) . Misexpression of UAS-SerWT significantly ( p < 0 . 0001 for mature crystal cells , p = 0 . 0457 if all crystal cells were taken into account ) reduced the fraction of crystal cells which show E ( spl ) mβ:CD2 expression , with 34 . 4% of all cells showing expression and 20 . 7% of mature cells showing CD2 expression ( D and F ) . For this analysis , the total number of lz > GFP cells were counted , taking into account their size , lz > GFP intensity , and E ( spl ) CD2 intensity . Mature crystal cells were defined as cells that were both large and had intense lz > GFP . We then took the proportion of either all cells , or mature cells which had E ( spl ) CD2 staining for each lymph gland ( n = 12 for lz > GFP , n = 13 for lz > GFP;UAS-Ser ) . Grubbs' outlier test was used , which removed one data point ( p < 0 . 05 ) from the control , which had an unusually small number of crystal cells . Then Welch's t-test was used to assess significance between the mean proportions of crystal cells which showed Notch activity . All error bars represent SD . These observations indicate that increased ligand expression in crystal cells decreases cell survival by blocking Notch ligand-independent activation , and therefore the buffering role of cis-expressed ligand can be extended to endogenous cases of DSL-independent Notch activity . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 01410 . 7554/eLife . 04415 . 015Figure 4—figure supplement 3 . Reduced Notch reporter activity in crystal cells was not caused by indirect effects on early ligand-dependent Notch signaling in prohaemocytes . Normal Hnt expression in crystal cells expressing green fluorescent protein ( GFP ) driven by lz-GAL4 ( A ) and in lz-GAL4 driving expression of UAS-SerWT ( B ) . There was no noticeable effect on the proportion of cells expressing Hnt when UAS-SerWT was misexpressed . Illustrations show outlines of crystal cells with either no Hnt expression ( white filling ) or Hnt expression ( red filling ) . Scale bars represent 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04415 . 015 To quantify this effect , we co-transfected pMT-Dx with pMT-NFL , causing an increase by a factor of 4 . 21 ( p = 0 . 0021 ) in the Notch activation compared with transfecting pMT-NFL alone ( Figure 4F ) . Transfection of pMT-NFL , pMT-Dx , pMT-GAL4 , and pUASt-Serdel3 significantly ( p = 0 . 0194 ) reduced the level of Notch activation ( Figure 4F ) . We next treated cells with dsRNA for either lgd or shrub ( a component of the ESCRT-III complex ) . Lgd dsRNA induced an increase in Notch activation by a factor of 1 . 73 compared with GFP dsRNA-treated cells ( p = 0 . 00286 ) ( Figure 4G ) . Likewise , shrub dsRNA caused a 3 . 93-fold increase ( p < 0 . 0001 ) in Notch activation in S2 cells ( Figure 4H ) ( Thompson et al . , 2005 ) . Expression of Serdel3 in both situations led to a significant decrease in the amount of Notch activated in comparison with Notch-expressing cells treated with control dsRNA ( lgd p = 0 . 0093 , shrub p = 0 . 0257 ) ( Figure 4G , H ) . To explore whether cis-acting ligands might block endogenous raised levels of ligand-independent Notch activation , in addition to the raised levels induced by genetic defects , we examined the effect of increased ligand expression in crystal cells in the larval lymph gland , which have recently been shown to have ligand-independent Notch activation ( Mukherjee et al . , 2011 ) . Notch activity in crystal cells promotes cell survival , and decreased Notch activity leads to a ‘bursting’ phenotype ( Mukherjee et al . , 2011 ) ( Figure 4—figure supplement 2B , E ) . Evidence for this bursting phenotype is provided by the disorganization of membrane-associated GFP ( Mukherjee et al . , 2011 ) . Using Lozenge ( Lz ) -GAL4 , a crystal cell lineage-specific driver ( Terriente-Felix et al . , 2013 ) to misexpress UAS-NotchRNAi or UAS-Serdel3 led to a significantly higher proportion of cells showed the ‘bursting’ phenotype than wild-type crystal cells ( NotchRNAi p = 0 . 0434 , Serdel3 p = 0 . 0286 ) ( Figure 4—figure supplement 2A , B , E ) . Furthermore , overexpression of UAS-SerWT led to a significant decrease of the Notch reporter E ( spl ) :mβ-CD2expression in mature crystal cells ( Figure 4—figure supplement 2C , D , F ) . Reduced Notch reporter activity was not caused by indirect effects on early ligand-dependent Notch signaling in prohaemocytes , as Hnt , a Notch target in differentiating crystal cells , ( Terriente-Felix et al . , 2013 ) was unaffected by ligand misexpression ( Figure 4—figure supplement 3A , B ) . These observations indicate that increased ligand expression in crystal cells decreases cell survival by blocking Notch ligand-independent activation , and therefore the buffering role of cis-expressed ligand can be extended to endogenous cases of DSL-independent Notch activity . In this study , we show that cells devoid of DSL ligands activate Notch sufficiently to stimulate reporter activity , and in the ovarian follicle cells the level of activation is above the threshold required to mediate normal Notch-induced downstream developmental events . During development , this type of noncanonical Notch activity is normally prevented by cis-expressed DSL ligands in numerous tissues . Cis-inhibition can also attenuate DSL-ligand independent Notch activity both in endogenous and genetically induced situations . Mechanistically , this could be explained if DSL ligands sequestered Notch at the membrane , made Notch more sensitive to degradation , or increased the stability of the heterodimer as it travels through the endosomal pathway . As we and others ( Fiuza et al . , 2010 ) have shown that increasing or decreasing ligand has variable effects on receptor distribution among tissues , and given that we observe a consistent effect among tissues on Notch activation upon cis-ligand removal , we prefer the stability hypothesis . Fiuza et al . ( 2010 ) show that ligand affects Notch stability during Notch activation by EDTA , giving support to the stability hypothesis as the most parsimonious explanation ( Fiuza et al . , 2010 ) . It is suggested that retaining a pool of translated Notch receptor keeps the pathway in a condition capable of almost instant activation ( Sprinzak et al . , 2010 ) . Therefore , we propose that a role of cis-ligands might be to keep the Notch pathway in a state of readiness by buffering against unintentional stochastic Notch activity resulting from normal processing through the endosomes . Endogenously , this may aid the ability of a cell to mediate future Notch-dependent developmental events that have strict temporal regulation . The following fly stocks were used for Drosophila crosses . hs-flp122;;FRT82B RFP ( Poulton et al . , 2011 ) , FRT82B DlRevF10 ( Haenlin et al . , 1990 ) , FRT82B DlRevF10SerRx82 ( BDSC #6300 ) , hs-FLP122; act-GAL4 UAS-GFP;FRT82B Gal80 , UAS-NotchRNAi ( VDRC #1112—no expression in germline cells ) , UAS-DeltaRNAi ( BDSC #34322—able to express in germline cells ) ; hsFLP GFPstau; act > y+ > GAL4 , UAS-GFP , hs-flp122; Gal80 FRT40A; tubGAL4 UASGFP , Su ( H ) 47FRT40A ( Morel and Schweisguth , 2000 ) , NRE-EGFP ( BDSC #30727; Stempfle et al . , 2010 ) , ubx-FLP;;FRT82B RFP , patched-GAL4 UAS-GFP ( Hinz et al . , 1994 ) , UAS-DlMyc ( a gift from Marc Muskavitch ) , tsg101111019 from Kyoto stock center , UAS-SerWT ( BDSC #5815 ) , UAS-Serdel3−tom ( a gift from Robert J Fleming ) ( Graveley et al . , 2011 ) , UAS-Deltex ( a gift from Martin Baron ) , lgdd740A ( BDSC #25087 ) , dppGAL4 ( BDSC #7007 ) , lz-GAL4 UAS-GFP ( BDSC #6314 ) . To create FRT82B , DlRevF10 germline/follicle cell clones by the FLP/FRT or MARCM methods ( Golic and Lindquist , 1989; Lee and Luo , 2001 ) ( e . g . , Figures 1B , D–F , 2A , C–D , Figure 1—figure supplement 2A–B ) , crossed flies were subjected to a 2 hr heat shock at 37°C for two consecutive days while in the mid-pupal to late-pupal stages . Flies were sorted three days after eclosion , and then kept for an extra three days at 25° before an additional 1-hr heat shock and incubation at 29°C with yeast paste for two more days before dissection . FLP-out-induced DlRNAi germline/follicle cell clones ( e . g . , Figure 2B , Figure 2—figure supplement 1A , B ) were produced by two consecutive 50-min heat shocks , followed by incubation at 25°C for a week and then transfer to yeasted vials in the 29°C incubator for dissection two days later . Evidence for MARCM and FLP-out-induced germline clones was provided by small nuclei and late Cut expression , as the UASt-GFP transgene does not reliably express in the germline . Follicle cell clones alone were produced by two 50-min heat shocks , followed by two days’ incubation at 29°C ( e . g . , Figure 1C and Figure 2—figure supplement 1C–D ) . Imaginal disc FLP-FRT-induced mutant clones were produced either by a ubx-FLP or a 1-hr heat shock with hs-FLP122 two days after egg laying . All other crosses were kept at 25°C unless otherwise noted . In lymph gland studies , Grubbs' test was used to identify significant ( p < 0 . 05 ) outliers , which were omitted from further analyses . Ovaries , imaginal discs , or lymph glands were dissected in phosphate-buffered saline ( PBS ) , fixed in 10% formaldehyde , washed three times in PBS + Triton-X ( PBT ) , and then blocked for at least 1 hr in PBT with goat serum . Tissues were then either stained overnight with mouse anti-Cut ( DSHB 2B10 , 1:30 ) , mouse anti-Hindsight ( DSHB 1G9 , 1:15 ) , mouse anti-NICD ( DSHB C179C6 , 1:15 ) , mouse anti-NECD ( DSHB C4582H , 1:15 ) , mouse anti-Wingless ( DSHB 4D4 , 1:20 ) , mouse anti-Dl ( DSHB C594 . 9B , 1:15 ) , rabbit anti-βGal ( MP Biomedical , Santa Ana , CA . SKU #08559761 ) , or rabbit anti-GFP ( abcam , Cambridge , UK . ab290—NRE-GFP was co-stained with this antibody to increase reporter sensitivity ) primary antibodies . Tissues were mounted on slides after PBT washes and secondary antibody incubation . 4' , 6-Diamidino-2-phenylindole ( DAPI ) was used to stain nuclei . Samples were then analyzed with a Zeiss 510 or Leica SP2 confocal microscope and after analysis with the Image J software . Nuclear volume quantification was done with the Volumest plug-in for ImageJ . S2 cells were grown under standard conditions and passaged once every three days in serum-free Gibco media ( Invitrogen , Waltham , MA ) supplemented with antibiotics . In preparation for transfection 106 cells per milliliter were seeded into either 24-well plates or 96-well plates for experiments with or without dsRNA treatment , respectively . Transfections were carried out with Qiagen Effectene ( Qiagen , Netherlands ) transfection reagent according to the manufacturer's instruction . Plasmids used for transfection were pMT-NotchFL ( a gift from Renjie Jiao ) , pMT-GAL4 ( DGRC #1042 ) , pUASt-Serdel3 ( a gift from Robert J Fleming ) , pMT-Deltex ( a gift from Spyros Artavanis-Tsakonas ) , NRE-firefly luciferase ( a gift from Sarah Bray ) , or Renilla luciferase ( a gift from Sarah Bray ) . Aliquots ( 75 ng for 24-well plates or 50 ng for 96-well plates ) of each non-luciferase plasmid were added and , where applicable , 10 ng of each luciferase plasmid . DNA concentration between transfections was kept constant with an empty vector . For experiments without dsRNA treatment , CuSO4 was added to a concentration of 500 µM 24 hr after transfection , and cells were assayed 24 hr later . dsRNA was transcribed in vitro using the RiboMAX large-scale RNA production system-T7 kit ( Promega , Madison , WI ) . The following primers were used to amplify genomic DNA taken from a single male fly from the NRE-GFP stock: Cells were transfected with plasmids of interest together with an NRE-driving firefly luciferase expression and a constitutively activated Renilla luciferase to control for transfection efficiency . Luciferase measures were inspected with the Dual-Luciferase Assay Kit ( Promega ) in 96-well luminometer plates . Each transfection was performed in duplicate and repeated several times . Student's t test was used to test for statistical significance .	Many biological processes require cells to send messages to one another . Typically , this is achieved when molecules are released from one cell and make contact with companion molecules on another cell . This triggers a chemical or biological reaction in the receiving cell . One of the most common examples of this is the Notch pathway , which is used throughout the animal kingdom and plays an important role in helping cells and embryos to develop . The Notch protein itself is a ‘receptor’ protein that is embedded in the surface of a cell , and relays signals from outside the cell to activate certain genes inside the cell . In fruit flies , two proteins called Serrate and Delta act as ‘ligands’ for Notch—by binding to Notch , they can change how this receptor works . If Serrate or Delta are present on the outside of one cell , they can activate Notch ( and hence the Notch signaling pathway ) in an adjacent cell . However , if the Serrate or Delta ligands are present on the surface of the same cell as Notch they turn the receptor off , rather than activate it . Notch can also work without being activated by Serrate or Delta , but whether the ligands can inhibit this ‘ligand-independent’ Notch activation if they are on the surface of the same cell as the Notch receptor was unknown . Palmer et al . study Notch signaling in the fruit fly equivalent of the ovary , in cells that are naturally deficient in Serrate and from which Delta was artificially removed . The Notch protein was activated when these ligands were not present . Furthermore , the developmental processes that are activated by Notch were able to proceed as normal when triggered by ligand-independent Notch signaling . In total , Palmer et al . investigated three different types of fruit fly cell , and found that ligand-independent Notch signaling can occur in all of them . Reintroducing Delta to the same cell as Notch turns the receptor off , suggesting that ligands on the surface of the same cell as the receptor can inhibit ligand-independent Notch activity . Many genetic diseases and cancers have been linked to Notch being activated when it should not be; therefore , understanding how Notch is controlled could help guide the development of new treatments for these conditions .	[ "Abstract", "Main", "text", "Materials", "and", "methods" ]	[ "developmental", "biology", "short", "report", "cell", "biology" ]	2014	Cis-interactions between Notch and its ligands block ligand-independent Notch activity
The characterization of the transcriptome and proteome of Plasmodium falciparum has been a tremendous resource for the understanding of the molecular physiology of this parasite . However , the translational dynamics that link steady-state mRNA with protein levels are not well understood . In this study , we bridge this disconnect by measuring genome-wide translation using ribosome profiling , through five stages of the P . falciparum blood phase developmental cycle . Our findings show that transcription and translation are tightly coupled , with overt translational control occurring for less than 10% of the transcriptome . Translationally regulated genes are predominantly associated with merozoite egress functions . We systematically define mRNA 5′ leader sequences , and 3′ UTRs , as well as antisense transcripts , along with ribosome occupancy for each , and establish that accumulation of ribosomes on 5′ leaders is a common transcript feature . This work represents the highest resolution and broadest portrait of gene expression and translation to date for this medically important parasite . The transcriptome of the intraerythrocytic developmental cycle ( IDC ) of P . falciparum is characterized by a continuous cascade wherein the expression of the majority of genes is maximally induced once per cycle and their timing correlates well with the timing for the respective protein's biological function ( Bozdech et al . , 2003 ) . The apparent lack of dynamic transcriptional regulation suggested that complementary post-transcriptional mechanisms could play an important role in the regulation of parasite gene expression ( Hughes et al . , 2010 ) . This is a reasonable assumption , given that global or gene-specific translational regulation of gene expression is a mechanism that allows fast adaptations during drastic changes in environmental conditions as well as during rapid transitions in developmental programs . Indeed a few examples of translational control in Plasmodium have been reported . In sporozoites present in the mosquito salivary gland , phosphorylation of the eukaryotic translation initiation factor eIF2α by the kinase IK2 , inhibits translation and causes accumulation of mRNAs into granules . Translational repression is alleviated by eIF2α phosphatase during the transition into the mammalian host , allowing parasites to transform into the liver stages ( Zhang et al . , 2010 ) . Similarly , PK4 kinase activity leads to the reduction of global protein synthesis through phosphorylation of eIF2α in schizonts and gametocytes and is essential for the completion of the parasite's erythrocytic cycle ( Zhang et al . , 2012 ) . Gene-specific translational regulation has also been observed in P . falciparum and is mediated by cis-acting sequences in combination with RNA-binding proteins . For example , dihydrofolate reductase-thymidylate synthase ( DHFR-TS ) binds within the coding region of its own cognate mRNA to repress translation ( Zhang and Rathod , 2002 ) and antifolate treatment has been shown to relieve this repressive effect without alteration of mRNA levels ( Nirmalan et al . , 2004a ) . In Plasmodium berghei , storage of translationally repressed mRNAs prior to fertilization is mediated by mRNA binding via the RNA helicase DOZI and the Sm-like factor CITH ( Mair et al . , 2006 , 2010 ) . Upstream open reading frames ( uORFs ) found on 5′ UTRs of transcripts have been reported to regulate the translation of specific genes ( Morris and Geballe , 2000 ) . In P . falciparum , the only uORF described and functionally characterized to date is a 120 codon region upstream of the var2csa ( PFL0030c ) coding region , a unique variant of the surface antigen PfEMP1 that mediates adhesion to placenta in pregnant women ( Amulic et al . , 2009 ) . In this case , translation of the uORF modulates repression of var2csa translation . Aside from these examples , the extent to which global and gene-specific translational control operates in P . falciparum during the IDC remains sparse . Since the P . falciparum genome was fully sequenced ( Gardner et al . , 2002 ) , several large-scale studies have provided detailed insights into the expression of genes and proteins across the parasite's life cycle . Parallel mass spectrometry-based proteomics and genome-wide expression profiling revealed differences between mRNA abundance and the accumulation of the corresponding protein , supporting the notion that post-transcriptional regulation of gene expression is at play in this parasite ( Le Roch et al . , 2004; Nirmalan et al . , 2004b; Foth et al . , 2011 ) . These methods , however , are limited in their ability to measure low abundance proteins and do not capture the underlying relationship between transcriptional activity and translational efficiency . More recently , polysome profiling was used to monitor discrepancies between polysome-associated and steady-state mRNAs in 30% of the P . falciparum blood stage transcriptome ( Bunnik et al . , 2013 ) ; however , this approach does not reveal the precise localization of the ribosomes , and thus can not be used to accurately assess the translational efficiency of a given mRNA ( Ingolia , 2014 ) . Here , we adapted the ribosome profiling technique ( Ingolia et al . , 2009 ) to describe the translational dynamics of the P . falciparum asexual blood stage transcriptome . We simultaneously evaluate mRNA abundance , gene structure , ribosome positioning , and translational efficiency for genes expressed through five stages of the IDC . We demonstrate that the data are highly reproducible , and we find that the translational efficiency of the majority of mRNAs expressed follows a narrow distribution , exhibiting a tight coupling between transcription and translation . Only 10% of the genes expressed deviate from this trend and are translationally up- or down-regulated . We found a surprising amount of ribosome density associated with 5′ leaders of transcripts particularly in genes with functions associated with merozoite egress and invasion . Overall , the precision and depth of the dataset presented herein add significantly to our understanding of P . falciparum gene expression by linking transcriptional and translational dynamics throughout the blood stages . To create whole-genome , high-resolution profiles of mRNA abundance and translation during in vivo blood stage development of P . falciparum , we adapted the ribosome profiling technique described by Ingolia et al . ( 2009 ) . Ribosome profiling is based on the deep sequencing of ribosome protected mRNA fragments obtained by nuclease digestion of polysomes , cycloheximide-arrested ribosomes bound to mRNA . These fragments represent the exact location of the ribosome at the moment the sample was harvested . Five stages representative of the 48-hr IDC of P . falciparum were harvested for both mRNA and polysome isolation; ring , early trophozoite , late trophozoite , schizont stages , and purified merozoites . To assess the reproducibility of the data , we harvested independent biological replicates of each stage . Polysomes were isolated in the presence of the translation elongation inhibitor cycloheximide , then nuclease digested to produce monosomes , and sedimented by centrifugation on a sucrose gradient ( Figure 1 and Figure 1—figure supplement 1 ) . To minimize isolation of RNA fragments bound by proteins other than 80S ribosomes , RNA was extracted only from the fractions of the sucrose gradient containing the monosome peak . The resulting ∼30 nt fragments of RNA , corresponding to ribosome footprints , were processed into strand-specific deep sequencing libraries in parallel with the mRNA samples , fragmented to ∼30 nt for consistency . Despite the unusually high AT content of the P . falciparum genome , over 92% of all 30 nt sequenced reads , derived from coding sequences ( CDSs ) , mapped uniquely to the genome ( Figure 1—source data 1 and Figure 1—figure supplement 2 ) . 10 . 7554/eLife . 04106 . 003Figure 1 . Ribosome profiling of the P . falciparum asexual blood stages , experimental outline . ( A ) Synchronized parasite cultures were maintained in hyperflasks at 5% hematocrit and maximum 15% parasitemia . Cycloheximide-treated cultures containing ∼1010 parasites were harvested at ring , early trophozoite , late trophozoite and schizont stages ( 11 , 21 , 31 , and 45 hpi , respectively ) for total RNA or polysome isolation . Merozoites were purified through magnetic isolation of late stage schizonts ( see ‘Materials and methods’ ) . Nuclease treated polysomes were fractionated on a sucrose gradient . Ribosome footprints ( ∼30 nt ) derived from the monosome peak ( dashed red line ) or chemically fragmented polyA purified mRNA ( ∼30 nt ) were used to build sequencing libraries . mRNA and ribosome footprint samples were processed in parallel to create deep sequencing libraries compatible with the Illumina platform . ( B ) Sucrose gradient A260 absorbance profile of polysome extracts derived from late trophozoites treated with micrococcal nuclease ( green , +MNase ) or untreated controls ( gray , No treatment ) . Red arrow indicates the 80S monosome peak collected for ribosome footprint library preparation . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 00310 . 7554/eLife . 04106 . 004Figure 1—source data 1 . Illumina sequencing mapping statistics against P . facliparum W2 SNP-corrected genome . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 00410 . 7554/eLife . 04106 . 005Figure 1—figure supplement 1 . Polysome profiles of the P . falciparum asexual blood stages . Sucrose gradient A260 absorbance profiles of polysome extracts treated with micrococcal nuclease ( green , +MNase ) and untreated controls ( gray , No treatment ) . Red dotted line indicates monosome peak harvested for ribosome footprint library generation . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 00510 . 7554/eLife . 04106 . 006Figure 1—figure supplement 2 . Read size influence on mappability . Single nucleotide sliding windows ranging from 10 to 50 nt were used to generate in silico libraries of the P . falciparum W2 SNP corrected genome . These were uniquely aligned , allowing no mismatches , to either the whole genome ( gray , WG ) or the coding sequences ( blue , CDS ) using Bowtie ( Mortazavi et al . , 2008 ) and the percentage of aligned reads were calculated for each window size . The analysis was repeated using sliding windows generated from the coding genome only ( red , CDS to WG ) for a more representative mappability estimate of an RNA-seq data set . Read sizes of ≥20 nt asymptotically approach maximum mappability percentages . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 00610 . 7554/eLife . 04106 . 007Figure 1—figure supplement 3 . Reproducibility and coverage threshold determination using two fully independent biological replicates . mRNA abundance measurements ( A ) and ribosome footprint densities ( B ) in terms of rpkM in two fully independent biological replicates of the late trophozoite timepoint . Genes with at least ≥32 total mRNA reads counted ( rM ) are highly reproducible ( r ≥ 0 . 9 ) across replicas ( A and B red dots , and Figure D ) whereas low read counts have a negative effect on rpkM reproducibility ( A and B blue dots , and C ) . ( C ) Genes were binned based on their rM in replica 1 . In each bin Pearson correlations of rpkM values of replica 1 and replica 2 were calculated . At 32 rM , r values were consistently above 0 . 9 indicating that rpkMs calculated for genes with ≥32 rM are highly reproducible across replicates , and this is independent of the number of genes in the bin . r = Pearson correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 007 To quantitatively obtain mRNA abundance and ribosome footprint density measures , we calculated rpkMs ( reads per kilobase of exon model per million reads mapped , as in Mortazavi et al . ( 2008 ) for each gene . We established the minimum number of mRNA reads sequenced per coding region ( rM; reads per million reads mapped ) required to confidently include genes in downstream analyses , to be ≥ 32 rM ( ‘Materials and methods’ , Figure 1—figure supplement 3 ) . Using this conservative threshold , 3605 genes qualified for further analysis . Between biological replicates , Pearson correlation values were consistently high , ranging from r = 0 . 94 to r = 0 . 99 ( Figure 2A ) , highlighting the quality and reproducibility of our data . In addition , we compared the RNA-seq transcriptome of the five stages sampled to our previously published transcriptome data set , originally generated using long oligo microarrays ( Bozdech et al . , 2003 ) . The RNA-Seq transcription profiles of the set of genes shared by the two data sets ( n = 1829 ) were highly correlated ( average r = 0 . 7 ) to the corresponding 11 , 21 , 31 , 45 , and 2 hr post merozoite invasion time points of the microarray data set , despite the use of different methodologies ( microarray vs RNA-seq ) and the use of different P . falciparum strains ( HB3 vs W2 , respectively ) . Because of the higher sensitivity of RNA-seq , we were able to accommodate an additional 743 genes into the cascade-like transcriptome extending it to a total of 3110 genes ( Figure 2B , Figure 2—source data 1 ) . The remaining 495 genes in our RNA-seq data set lacked sufficient variation over the five time points for inclusion within the phaseogram . These genes , referred to as non-phasic genes , are nevertheless included in all analyses . 10 . 7554/eLife . 04106 . 008Figure 2 . Ribosome profiling through the P . falciparum IDC . ( A ) Reproducibility among biological replicates . Two fully independent biological replicas of each stage were sampled for RNA-seq ( left panels , blue ) and ribosome profiling ( right panels , green ) . Each dot represents the log2 rpkM measured for each gene in each stage . r = Pearson correlation coefficient . ( B ) Gene expression and translation are tightly coupled during the P . falciparum IDC . Phaseograms of mRNA ( left heatmap ) and ribosome footprint density ( right heatmap ) as a function of development for 3110 phasic and 495 non-phasic genes organized in the same order in the left and right heatmap . Data represent mean centered log2 mRNA and ribosome footprint rpkM values for each gene ( rows ) in each sampled stage ( columns ) . R = rings , ET = early trophozoites , LT = late trophozoites , S = schizonts , M = merozoites . ( C ) log2 rpkM of mRNA abundance vs ribosome footprint density for all genes expressed ( rM ≥ 32 ) across the IDC . Pearson correlation coefficients r ≥ 85 . n = total number of genes . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 00810 . 7554/eLife . 04106 . 009Figure 2—source data 1 . P . falciparum ribosome profiling data set . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 009 While RNA-Seq reveals the abundance and architecture of individual mRNAs , ribosome profiling provides a complementary and quantitative measure of mRNA translation . Ribosome occupancy along the CDS results in a profile that indicates the timing and magnitude of translation of a given mRNA , thus quantitatively delineating regions of each mRNA molecule that are actually bound by 80S ribosomes ( Ingolia et al . , 2009 ) . To inspect translation on a genome-wide scale , ribosome density values of each gene expressed in the data set were organized in the same order as the transcriptome . The translational profile of each gene displayed a cascade-like quality strikingly similar to the transcriptome ( Figure 2B ) . Much like mRNA abundance , translation of phasic genes reaches a single maximum and a single minimum during the IDC . To determine the exact level of correlation between transcription and translation , we directly compared mRNA and ribosome footprint density measurements ( Figure 2C ) . In general , translation is tightly correlated with transcription for all phasic and non-phasic genes in rings ( r = 0 . 85 ) , early trophozoites ( r = 0 . 93 ) , late trophozoites ( r = 0 . 91 ) , schizonts ( r = 0 . 89 ) , and purified merozoites ( r = 0 . 86 ) . This indicates that when an mRNA is detected in one stage it is associated proportionally with ribosomes within the same stage . An example pair of genes is shown in Figure 3A . Here , mRNA abundance profiles of eukaryotic translation initiation factor eIF2 gamma subunit ( PF14_0104 ) and the conserved protein PF14_0105 , show that peak mRNA abundance for these two genes occurs at two different stages , early and late , respectively . Examination of ribosome occupancy of both genes reveals a ribosome density accumulation profile within the coding sequence that mirrors their respective mRNA profiles . As for the majority of genes , ribosome footprint density and mRNA abundance for these two genes are highly correlated ( r = 0 . 98 and 0 . 93 for PF14_0104 and PF14_0105 , respectively ) , indicating that mRNA translation occurs proportionally during the same stages at which these genes are transcribed ( Figure 3B; Supplementary file 1 ) . Globally , 77% of genes expressed in at least three stages of the IDC display high Pearson correlation ( r ≥ 0 . 7 ) between mRNA abundance and translation ( Figure 3—figure supplement 1 ) . Thus , our genome-wide analysis of translation establishes that for the majority of genes expressed during the IDC , transcription and translation occur proportionally . 10 . 7554/eLife . 04106 . 010Figure 3 . Transcription and translation are highly correlated . ( A ) Ribosome footprint ( green ) and mRNA ( blue ) coverage profiles of two neighbor genes , the eIF2 gamma subunit ( PF14_0104 ) and the conserved protein PF14_0105 ( CDS , white boxes; HMM-defined UTRs , black lines ) in rings ( R ) , early trophozoites ( ET ) , late trophozoites ( LT ) , schizonts ( S ) , and merozoites ( M ) . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . ( B ) mRNA and ribosome footprint density of the genes in ( A ) correlate during development . r = Pearson correlation coefficient between ribosome footprint density and mRNA abundance of each gene . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01010 . 7554/eLife . 04106 . 011Figure 3—figure supplement 1 . mRNA abundance and ribosome footprint density are highly correlated for the majority of genes expressed during the IDC . Pearson correlation of mRNA abundance and ribosome footprint density of every gene expressed in at least three stages ( 2412 genes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 011 Ribosome profiling allows the monitoring of translation rates through the simultaneous quantitative measure of mRNA abundance and ribosome density on mRNAs . The ratio of the footprint rpkM to the mRNA rpkM for any given gene represents its relative translational efficiency ( TE ) ( Ingolia et al . , 2009 ) . To assess the dynamics of translational control and detect variations in control within and between developmental stages , we calculated the relative TE of all expressed genes in our data set ( Figure 4A ) . The shape and the range of TE distributions obtained for each stage sampled is comparable to those seen in other eukaryotes ( Ingolia et al . , 2009; Dunn et al . , 2013 ) . Absolute mean translational efficiencies in all stages ( log2TE µRings = −0 . 43 , log2TE µE . trophs . = −0 . 56 , log2TE µL . trophs . = −0 . 31 , log2TE µSchizonts = −0 . 16 and log2TE µMerozoites = −0 . 68 ) had a maximum difference of 1 . 47-fold observed between early trophozoites and schizonts . Translational efficiencies display a roughly 100-fold range in absolute values in each of the stages with the exception of the ring and merozoite stages , which exhibit more extreme values . In these stages , the distribution of absolute TE values displays an approximately fourfold larger spread than in early trophozoites , late trophozoites , or schizonts ( Figure 4A , Figure 2—source data 1 ) . In rings the gene with the largest TE is the merozoite surface protein 9 ( PFL1385c , log2TE = 4 . 1 ) and the gene with the lowest TE is the FIKK family serine/threonine protein kinase ( PF14_0734 , log2TE = −5 . 1 ) . In merozoites the largest and lowest TE values correspond to the serine repeat antigen 5 ( SERA5 , PFB0340c , log2TE = 4 . 0 ) and the alpha adenylyl cyclase ( PF14_0788a-c , log2TE = −4 . 7 ) , respectively . 10 . 7554/eLife . 04106 . 012Figure 4 . Genome-wide measurements of translation . ( A ) Translational efficiency distributions in each stage . Rings and merozoites have most extreme TE values; ± 2 SD above ( yellow bars ) and below ( blue bars ) the mean . TE values of translationally up-regulated merozoite surface protein ( MSP6 ) and the eukaryotic initiation factor 2 alpha kinase 1 ( IK1 ) ( blue arrowhead ) across the time course remain high and low , respectively . μ = mean log2TE , n = total number of genes . ( B ) mRNA abundance and translational efficiency heatmap of translationally up- and down-regulated genes ( upper panel and lower panel , respectively ) . Note TE is independent of changes in mRNA abundance for all genes including MSP6 and IK1 ( C ) . R = rings , ET = early trophozoites , LT = late trophozoites , S = schizonts , M = merozoites . n = number of genes , μ = mean , SD = standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 012 To determine the contribution of translational efficiency to the dynamic range of gene expression , we examined the genes lying at the extremes of the TE distribution . For the purpose of this analysis , genes with a translational efficiency of two standard deviations above or below the mean in any of the stages were considered translationally up- or down-regulated , respectively . A total of 301 genes , 8 . 3% of the transcriptome , are translationally regulated by this metric , with 124 genes translationally down-regulated and 177 genes translationally up-regulated ( Figure 4B , Figure 2—source data 1 ) . The timing of maximum mRNA expression does not influence TE for either of these two groups . Translational efficiencies remain high for the translationally up-regulated and low for the translationally down-regulated genes in all the stages at which they are expressed , regardless of the stage of peak mRNA abundance , suggesting that translational efficiency is largely , but not completely , programmed by the mRNA sequence itself , rather than global factors . For example , translational efficiency of the merozoite surface protein 6 ( MSP6 , PF10_0346 ) remains high ( log2TE ≥ 2 ) across all stages irrespective of variations in its mRNA abundance . In contrast TE values for the eukaryotic initiation factor 2alpha kinase 1 ( IK1 , PF14_0423 ) are among the lowest measured despite high mRNA abundance across all stages ( Figure 4C ) . An examination of the 124 translationally down-regulated genes yielded some expected , and in some cases , unexpected findings . As would be expected , two pseudogenes , the ring-infected erythrocyte surface antigen 2 ( RESA-2 , PF11_0512 ) and reticulocyte binding protein homologue 3 ( PfRh3 , PFL2520w ) , represent a clear example of low translational efficiency . The PfRh3 pseudogene ribosome profile shows that translation of the 5′ end of this transcript occurs up until the encounter of several in-frame stop codons , causing the reduction in ribosome density from this point on ( Figure 5A , Figure 5—figure supplement 1 ) . This suggests that a truncated version of the PfRh3 protein is being produced in the W2 strain studied here . Evidence for peptides corresponding to the 5′ end of PfRh3 has been found in gametocytes and sporozoites ( however not during the asexual stages ) using mass spectrometry ( Florens et al . , 2002; Lasonder et al . , 2002 ) . We note that low levels of ribosomes can still be detected along the full length of this transcript in schizonts and merozoites . Whether these footprints derive from a low level of stop-codon read-through or accumulate via another unknown mechanism remains to be determined . 10 . 7554/eLife . 04106 . 013Figure 5 . Translationally down-regulated genes have decreased CDS ribosome density . ( A ) Ribosome footprint ( green ) and mRNA ( blue ) profiles of the PfRh3 pseudogene ( PFL2520w ) in merozoites ( M ) . In the detail the bars above the gene model indicate AUG , stop , and any other codon , in green , red , and gray , respectively . Boxes indicate the mapping location of peptides identified by mass spectrometry in gametocytes and sporozoites ( Florens et al . , 2002; Lasonder et al . , 2002 ) . Reduction of ribosome footprint coverage occurs upon encounter of consecutive stop codons ( extended red lines ) . ( B ) eIF2α kinase ( PF14_0423 ) gene in rings ( R ) showing ribosome footprint accumulation on the 5′ leader , 3′ UTR , and low translational efficiency of the CDS . ( CDS , white boxes; HMM-defined UTRs , black lines . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01310 . 7554/eLife . 04106 . 014Figure 5—figure supplement 1 . Translation of a truncated form of PfRh3 during the IDC . Ribosome footprint ( green ) and mRNA ( blue ) profiles of the PfRh3 pseudogene ( PFL2520w ) in rings ( R ) , early trophozoites ( ET ) , late trophozoites ( LT ) , schizonts ( S ) , and merozoites ( M ) . ( A ) Translation of PfRh3 occurs until ribosomes dissociate upon the encounter of several consecutive in-frame stop codons ( visible in B ) . ( B ) The vertical bars above the gene model indicate AUG , stop , and any other codon , in green , red , and gray , respectively . Boxes indicate the mapping location of peptides identified by mass spectrometry in gametocytes and sporozoites ( Florens et al . , 2002; Lasonder et al . , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01410 . 7554/eLife . 04106 . 015Figure 5—figure supplement 2 . Translationally down-regulated genes have decreased CDS ribosome density . ( A ) Ribosome footprint ( green ) and mRNA ( blue ) profiles of the ring-infected erythrocyte surface antigen 2 , RESA2 pseudogene ( PF11_0512 ) in rings ( R ) and merozoites ( M ) . Both , the annotated isoform from PlasmoDB version 7 . 1 and the gene model for the alternate isoform inferred using ribosome profiling and W2 genomic DNA sequencing data from this study is depicted ( CDS , white boxes; HMM-defined UTRs , black lines ) . The red star indicates a homopolymeric tract in which a single base deletion causes a premature stop codon ( red triangle ) , which coincides with the site of ribosome drop off . ( B ) Ribosome footprint and mRNA profiles of erythrocyte vesicle protein 1 , EVP1 ( PFD0495c ) . This gene is transcribed in all stages yet translational efficiencies are relatively low , as evidenced by a depletion of ribosomes on the CDS of the gene particularly in early trophozoites ( log2TE = −2 . 6 , −2 . 9 , −1 . 0 , −2 . 1 , −1 . 4 in rings , early trophozoites , late trophozoites , schizonts , and merozoites , respectively ) . ( CDS , white boxes; HMM-defined UTRs , black lines . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 015 Ring-infected erythrocyte surface antigen 2 , RESA2 ( PF11_0512 ) was first described as a pseudogene based on the presence of an internal stop codon ( Cappai et al . , 1992 ) . Since then , transcription of this gene has been demonstrated both in vivo ( Vazeux et al . , 1993 ) and in vitro ( Bozdech et al . , 2003 ) . RESA-2 is transcribed but poorly translated in rings , early trophozoites and merozoites ( log2TE −3 . 2 , −2 . 7 , −2 . 9 , respectively ) . Accordingly , the ribosome profile of this gene in merozoites shows a general depletion of ribosomes along the CDS ( Figure 5—figure supplement 2 ) . In rings , ribosome density diminishes at the second exon . To validate the RESA2 gene model , we used genomic DNA sequencing data derived from the P . falciparum W2 strain used in this study . We found that 69% ( n = 151 ) of reads mapping to this locus support a single base deletion that creates a premature stop codon exactly at the site of ribosome footprint drop-off ( Supplementary file 2 ) . These data suggest that RESA2 is transcribed and actually translated into a shorter protein product of 461 amino acids . Whether or not the protein product is functional or undergoes post-translational degradation remains to be determined . In addition to expected instances of translational regulation , our data permit the discovery of previously uncharacterized translational regulation , especially at the extremes of the TE distributions . One of the most notable examples of translational silencing is the eIF2α kinase IK1 ( PF14_0423 ) for which ribosome footprints accumulate at the 5′ leader and 3′ UTR but not on the CDS , resulting in an extremely low translational efficiency ( log2TE = −3 . 6 ) despite relatively high transcript abundance across all stages ( Figure 5B ) . The mechanism by which this gene is maintained in a translationally down-regulated state is unknown . Another example is the erythrocyte vesicle protein 1 ( EVP1 , PFD0495c ) for which abundant transcript levels can be detected across all stages , with peak mRNA abundance occurring in rings and schizonts ( Figure 5—figure supplement 2 ) . Protein levels , however , have been shown to be undetectable ( Tamez et al . , 2008 ) . Here , we find that translational efficiencies of this gene were low across all stages and lowest in rings and early trophozoites ( log2TE = −2 . 6 and −2 . 9 , respectively ) demonstrating that post-transcriptional regulation at the level of translation is , at least in part , responsible for its scarcity as a protein . Thus , our ribosome profiling data set highlights instances of translational control of genes that may not be detected by proteomic methods . Indeed a search for mass spectrometric data showed no evidence for ∼70% of genes in this category ( Aurrecoechea et al . , 2009 ) . Including the aforementioned examples , our data set describes a total 124 translationally down-regulated genes ( listed in Figure 2—source data 1 ) for which translational efficiency values lie at the lower extremes of the distribution . Protein products of translationally up-regulated genes are likely to be abundant and readily detected using mass spectrometry . Previous proteomic studies show protein evidence in the blood stages for almost all ( 171 of 177 ) well-translated genes identified here ( Aurrecoechea et al . , 2009; Pease et al . , 2013 ) . Mass spectrometric evidence for the remaining six genes is either absent ( PFL0245w , PFL2510w , PF11_0204 ) or has only been found in sporozoites ( PFE1615c , MAL7P1 . 300 , PF13_0069a ) . Despite the lack of proteomic data , our data indicate that these genes are both transcribed and translated during the blood stages of the parasite . Whether post-translational control points exist for these proteins is unknown . Among the top ten most highly translated genes are proteins involved in merozoite egress and invasion MSP3 , 6 , 7 , and 9 ( merozoite surface proteins PF10_0345 , PF10_0346 , PF13_0197 , and PFL1385c ) , serine repeat antigen 5 ( SERA5 , PFB0340c ) , and RAP1 , 2 , and 3 ( PF14_0102 , PFE0080c , and PFE0075c , respectively ) ( Figure 2—source data 1 ) . Interestingly , 73 ( 41% ) of all translationally up-regulated genes can be assigned to the repertoire of canonical functions for merozoite egress and invasion described to date ( Yeoh et al . , 2007; Blackman , 2008; Hu et al . , 2010; Farrow et al . , 2011 ) . Strikingly , for all genes in this set , maximum mRNA abundance is found during the late stages of the IDC ( 69 schizont and 4 merozoite stage mRNAs ) yet for the majority ( 50 , 70% ) peak translational efficiency occurs in rings . Consistent with this , peptides for most of these merozoite function proteins ( 58 of 73 ) have been detected in rings ( Oehring et al . , 2012; Pease et al . , 2013 ) . This mode of translational regulation whereby late stage transcripts are highly translated in rings was not exclusively limited to genes related to merozoite egress and invasion . We found evidence for an additional 14 genes with this profile , including , aquaglyceroporin ( PF11_0338 , log2TE = 3 . 8 ) , tubulin beta chain ( PF10_0084 , log2TE = 1 . 8 ) , and early transcribed membrane protein 2 ( PFB0120w , log2TE = 2 . 5 ) . Taken together these data demonstrate that transcription and translation are tightly correlated for the majority of genes expressed during the asexual life cycle of P . falciparum with few exceptions . These apply to a small subset of translationally down- and up-regulated genes for which translational efficiencies appear to be inherent properties of the mRNA , independent of changes in mRNA abundance . Genes in this category , especially those that exhibit high translational efficiencies , are enriched with functions associated with merozoite egress and invasion during the transition from late stages into rings . Ribosome profiling provides position specific information along each transcript allowing the detection of changes in ribosome distribution on the mRNA and their relationship to translational efficiency . To look for ribosome occupancy features beyond the CDS of transcripts , we first took advantage of the deep coverage and strand specificity of our RNA-seq data to identify 5′ leaders and 3′ UTRs of the P . falciparum transcriptome . We constructed a hidden Markov model ( HMM ) to automatically delineate the boundaries of both 5′ leaders and 3′ UTRs for known gene models ( see ‘Materials and methods’ ) . Within the limits imposed by our data , we were able to describe 5′ mRNA leaders and/or 3′ UTRs for 3569 genes in at least one of the stages ( Figure 6—figure supplement 1 , Figure 2—source data 1 ) . 5′ leaders are on average longer than 3′ UTRs in each of the stages and median lengths across stages vary to a larger degree for 5′ leaders ( from 607 to 1040 nt ) than for 3′ UTRs ( 518–622 nt ) . The longest 5′ mRNA leader was measured in late trophozoites ( 8229 nt ) for the Ap2 transcription factor , PF11_0404 , and the longest 3′ UTR stretched 4773 nt for 60S ribosomal protein L7-3 , PF14_0231 , in rings . An example pair of genes with mapped 5′ leaders and 3′ UTRs is shown in Figure 6 . Here , our HMM predicts a 636 nt and a 781 nt 5′ leader and a 468 nt and 423 nt 3′ UTR for the Myb2 transcription factor ( PF10_0327 ) and the bromodomain protein ( PF10_0328 ) , respectively . These genes , encoded on opposite strands , share a 1536-nt intergenic sequence; however , the span between the region delimited by their 5′ leader sequence is only 120 nt and presumably harbors their respective promoters . 10 . 7554/eLife . 04106 . 016Figure 6 . Example of extended transcript annotations using the HMM . 5′ leaders and 3′ UTRs of the gene pair Myb2 ( PF10_0327 ) and bromodomain protein ( PF10_0328 ) were defined using the HMM designed ( see ‘Materials and methods’ ) . The sizes of 5′ leaders and 3′ UTRs of these genes in the schizont stage are indicated . The intergenic region is 1536 nt and the spanning distance separating the 5′ leaders is 120 nt . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01610 . 7554/eLife . 04106 . 017Figure 6—figure supplement 1 . HMM-defined 5′ leader and 3′ UTR characteristics . 5′ leader ( A ) and 3′ UTR ( B ) length distribution and their statistics ( C ) per stage . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 017 Next , using mRNA boundaries derived from our data , we analyzed ribosome distribution along each transcript during life cycle progression . More than 80% of the ribosome footprints in rings , early trophozoites , late trophozoites , and schizonts , were mapped to CDS regions of the genome , except in merozoites , where only 68% were mapped to the CDS ( Figure 7A ) . On average less than 1% of all reads obtained were mapped to 3′ UTRs in each stage , and most transcripts had no observed footprints past the stop codon . In contrast , footprints were far more common in 5′ leaders ( 9 . 1% , 4 . 8% , 7 . 5% , and 4 . 8% in rings , early trophozoites , late trophozoites , and schizonts , respectively ) particularly in merozoites ( 23% ) . Footprint enrichment is specific to 5′ leaders and not due to non-specific background since this would result in an increase of footprints mapping evenly along the length of the transcript , including the 3′ UTR , and not just the 5′ leader . Furthermore , these footprints most likely represent ribosomes because they derive from the 80S monosome fraction of the sucrose gradient , and their footprint read length distributions are equal to those of CDS mapping footprints , whereas they are significantly divergent from rRNA or tRNA read length distributions ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 04106 . 018Figure 7 . Transcripts accumulate ribosome density within the 5′ leader . ( A ) Proportion of mRNA or ribosome footprint reads mapping to CDS , to HMM-defined 5′ leaders and 3′ UTRs , antisense to annotated coding genes or to other regions of the genome such as mitochondria , plastid , tRNA , rRNA , ncRNA , and 5′ leader and 3′ UTR regions not defined by the HMM . ( B ) Proportion of ribosome footprints mapping inside or outside predicted uORFs in the HMM-defined 5′ leaders . ( C ) Ribosome footprint ( green ) and mRNA ( blue ) profiles of the EBA-175 ( MAL7P1 . 176 ) gene in rings ( R ) showing ribosome footprint accumulation on the 5′ leader . In the detail , the bars above the gene model indicate AUG , stop , and any other codon , in green , red , and gray , respectively and in all three possible frames . Gray bars indicate the 9 uORFs present in the 5′ leader , starting with an AUG ( green line ) and ending with a stop codon ( red line ) . Black bar inside CDS indicates a deletion specific to the W2 strain used in this study . CDS , white boxes; HMM-defined UTRs , black lines . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01810 . 7554/eLife . 04106 . 019Figure 7—source data 1 . Predicted uORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 01910 . 7554/eLife . 04106 . 020Figure 7—figure supplement 1 . 5′ leader footprints are derived from ribosomes . Ribosome footprint read length distributions for reads mapping either to CDSs , 5′ leaders , 3′ UTRs , antisense , rRNAs or tRNAs are plotted . Read lengths of rRNA and tRNA mapping footprints are significantly different than those mapping the 5′ leader , the CDS , or the 3′ UTR of transcripts in all stages . KS = Kolmogorov–Smirnov test . D = KS test statistic . R = rings , ET = early trophozoites , LT = late trophozoites , S = schizonts , M = merozoites . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02010 . 7554/eLife . 04106 . 021Figure 7—figure supplement 2 . Distribution of uORF coverage on 5′ leaders of genes expressed during the IDC . The proportion of ribosome footprints mapping inside predicted uORFs was calculated for each gene expressed in each stage . The median of each of these distributions is ∼0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02110 . 7554/eLife . 04106 . 022Figure 7—figure supplement 3 . uORFs present on 5′ leaders have no effect on TE . Translational efficiency ( log2TE ) for all genes expressed in each stage is plotted against the proportion of reads mapping within uORFs , the number of predicted uORFs , or the length of predicted uORFs in the 5′ leader . No direct relationship between these parameters can be observed . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02210 . 7554/eLife . 04106 . 023Figure 7—figure supplement 4 . Detection of ribosome density on uORFs . ( A ) Ring stage mRNA ( blue ) and ribosome footprint ( green ) profiles of VAR2CSA ( PFL0030c ) are shown . There is virtually no ribosome density on transcript CDS ( log2TERings = −4 . 2 ) . Ribosomes do accumulate on the previously described ( Amulic et al . , 2009 ) 360 nt uORF ( white box ) . This region is depicted in more detail in the panel below where the amino acids , AUGs and stop codons of each of the three reading frames are denoted with gray , green , and red bars , respectively . Note that ribosomes start accumulating upstream of the previously described uORF . Mappability at the 3′ end of this antigenic variation gene is poor and therefore no mRNA read coverage can be detected here . ( B ) Ring stage mRNA ( blue ) and ribosome footprint ( green ) profiles of PFE1550w ( unknown function ) are shown . Translational efficiency of the CDS is log2TERings = −3 . 6 in rings . 90% of ribosome footprints that map to the 5′ leader of this gene accumulate on one of the six predicted uORFs ( detailed figure below ) . The predicted uORF is 168 nt ( 56 aa ) . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02310 . 7554/eLife . 04106 . 024Figure 7—figure supplement 5 . uORFs present on 5′ leaders have no effect on TE . Ribosome density on the 5′ leader ( log25′RD ) for all genes expressed in each stage is plotted against the proportion of reads mapping within uORFs , the number of predicted uORFs , or the length of predicted uORFs in the 5′ leader . No direct relationship between these parameters can be observed . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02410 . 7554/eLife . 04106 . 025Figure 7—figure supplement 6 . 5′ ribosome density can be found on 5′ leaders devoid of AUGs . Ring stage mRNA ( blue ) and ribosome footprint ( green ) profiles of ( A ) aquaglyceroporin ( PF11_0338 ) and ( B ) PFC0486c ( unknown function ) are shown . Both genes display high ribosome density on their 5’ leaders and these are devoid of AUGs . Mappability = mappability score at that position; range 0 ( white ) to 30 ( black ) . rM = coverage ( reads per million reads mapped ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 025 Upstream open reading frames are a major source for 5′ leader ribosome density found from yeast to humans ( Ingolia et al . , 2009; Brar et al . , 2012 ) , and these have been shown to play a role in translational regulation of the downstream ORF in a few well-studied examples ( Morris and Geballe , 2000 ) . In P . falciparum , ribosomes have been suggested to accumulate on 5′ leaders of genes displaying a delay in translation presumably due to long uORFs ( Bunnik et al . , 2013 ) . We defined 36 , 086 possible uORF regions in the 5′ leaders of genes expressed during the P . falciparum IDC using a liberal definition that includes any stretch of at least two amino acids , starting with an AUG codon ( Figure 7—source data 1 ) . Regardless of stage , half of the total ribosome footprint coverage in 5′ leaders , in aggregate , or on a gene-by-gene basis did not overlap with these predicted uORFs ( Figure 7B , Figure 7—figure supplement 2 ) . We could find no significant correlation between the number of uORFs per gene , the uORF lengths , or the degree to which ribosome density was enriched in uORFs with translational efficiency ( Figure 7—figure supplement 3 ) . For example , erythrocyte binding antigen-175 ( EBA-175 , MAL7P1 . 176 ) is well translated in rings ( log2TE = 1 . 4 ) and displays a large amount of 5′ leader ribosome occupancy . Half ( 49% ) of the reads map within the nine predicted uORFs on the 5′ leader of this gene , the other half maps outside these uORFs ( Figure 7C ) . Using this liberal definition of an uORF , the data do not support an association between ribosome occupancy in these regions , nor does it support an association between the presence of these regions and translational efficiency . Nonetheless , there exist at least two clear exceptions . First , we were able to validate translation of the reported uORF present in the 5′ leader sequence of the var2csa mRNA which is expressed only in rings ( Amulic et al . , 2009 ) . The majority of ribosome footprint density localizes to this uORF , and to a second one just upstream , while the var2csa coding sequence is largely devoid of footprints ( log2TERings = −4 . 2 , Figure 7—figure supplement 4 ) , consistent with its translational repression during growth in the absence of plancental tissue . Second , another striking example of uORF translation was found on PFE1550w ( unknown function ) for which the ratio of uORF to total 5′ leader mapping reads is 0 . 9 ( Figure 7—figure supplement 4 ) . Indeed , ribosome footprint density is concentrated on one of the 6 uORFs predicted in the 5′ leader of this gene , 56 amino acids long . This gene is also translationally down-regulated in all stages ( log2TE = −2 . 7 on average ) . These two genes represent exceptional cases for which uORF translation negatively correlates with translation of the downstream ORF . Aside from these two exceptions , for the vast majority of genes , ribosome occupancy appears spread along 5′ leaders and not preferentially concentrated within possible uORFs . For this reason , we calculated 5′ leader ribosome density ( 5′RD ) for each gene expressed during the IDC , defined as upstream ribosome occupancy normalized for mRNA expression level and size of the leader sequence ( 5′ leader ribosome footprint rpkM/5′ leader mRNA rpkM ) ( Figure 2—source data 1 ) . No positive correlation exists between the number of uORFs per gene , the uORF lengths , or the degree to which ribosome density is enriched in uORFs and 5′RD , reinforcing the notion that uORFs are not a requisite for ribosome association to 5′ leaders ( Figure 7—figure supplement 5 ) . In fact 5′ ribosome density can be found on transcripts with 5′ leaders completely devoid of AUGs , and thus , without uORFs by definition , such as the highly translated aquaglyceroporin ( log2TE = 3 . 8 and log25’RD = 2 . 9 in rings ) , and PFC0486c ( unknown function , log2TE = 1 . 6 and log25′RD = 1 . 1 in rings ) ( Figure 7—figure supplement 6 ) . Overall , rings and merozoite stage parasites were found to express transcripts with the highest 5′RD ( mean log25′RD −0 . 03 , and 0 . 11 , respectively ) relative to early trophozoites , late trophozoites , and schizonts ( mean log25′RD −1 . 11 , −0 . 26 , −0 . 83 ) , where the range of 5′RD values is also narrower ( Figure 8A ) . Interestingly , among genes at the extremes of the 5′RD distributions ( mean ± 1 SD ) , we also found many of our identified translationally up- and down-regulated transcripts ( 66% and 40% , respectively ) . On average , 5′RD was enriched on translationally up-regulated transcripts ( mean log25′RD = 0 . 83 ) and depleted for translationally down-regulated transcripts in all stages ( mean log25′RD = −1 . 11 ) , suggesting the possibility that 5′RD is a byproduct of translational efficiency itself ( Figure 8B ) . 10 . 7554/eLife . 04106 . 026Figure 8 . 5′ ribosome density is commonly found on genes expressed during the IDC . ( A ) 5′ RD distributions in each stage . Transcripts in rings and merozoites have on average higher 5′ RD values; ± 2 SD values lie outside gray shade . μ = mean log25′RD , n = total number of genes . ( B ) 5′RD values of the translationally up-regulated set of genes ( yellow boxes ) are relatively higher ( average log25′RD R = 1 . 73 , ET = −0 . 26 , LT = 0 . 78 , S = 0 . 30 , M = 1 . 16 . ) than the rest ( white boxes ) or the set of down-regulated ( blue buxes ) genes . ( C ) 5′RD weakly correlates with translational efficiency . The translationally up-regulated gene set ( yellow circles ) is associated with high 5′RD , particularly in rings . The translationally up-regulated genes merozoite surface protein ( MSP6 ) , aquaglyceroporin ( AQP ) , serine repeat antigen ( SERA5 ) , and the reticulocyte binding protein homologue 3 ( PfRh3 ) are pointed out . r = Pearson correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 026 In order to determine whether a direct relationship between 5′RD and translational efficiency of the downstream ORF exists , we compared these values for each gene . 5′RD positively correlates , albeit moderately , with translational efficiency in all stages , particularly in rings and merozoites ( r = 0 . 51 and 0 . 49 , respectively ) . We focused on the subset of genes with highest and lowest 5′RD values ( mean ± 2 SD ) and found that only a fraction of the translationally up- and down-regulated genes overlap with this category of extreme 5′RDs in each stage ( Figure 8C ) . The largest overlap occurred in rings where the highest 5′RD values were found in 43% ( 31 genes ) of the translationally up-regulated genes , including MSP6 , AQP , and SERA5 . These results indicate that while in general a correspondence between 5′RD and translational efficiency exists , one is not necessarily predictive of the other and exceptions apply . This is the case , for example , of the translationally down-regulated transcript of the pseudogene PfRh3 , which in rings has the second highest 5′RD value ( log25′RD = 5 . 1 ) . In summary , our data establish ribosome accumulation on 5′ leaders as a common feature of transcripts expressed during the IDC . Ribosome density is not restricted to predicted uORFs present within these regions and , with few exceptions , the uORF number , length , or coverage level , is not a requirement for 5′ ribosome density and has no measurable effects on the translation of the downstream ORF . Even though 5′RD is more commonly found on 5′ leaders of highly translated transcripts , this is not a universal trend since only a moderate correlation exists between 5′RD and the translational efficiency of the downstream ORF . While our data showed 3′ UTRs to be relatively depleted of ribosomes , we searched for rare cases of high 3′ UTR ribosome density , possibly arising from stop codon read-through , alternative stop codon usage , or re-initiation of downstream ORFs ( Dunn et al . , 2013; Guydosh and Green , 2014 ) . We systematically searched for transcripts for which coverage , in a sliding window of 30 nt , was greater in the 3′ UTR than the CDS , and found 19 genes meeting this criterion . These genes could be qualitatively divided into two categories: 14 with putative stop codon read-through and/or alternate stop codon usage and 5 genes for which the origin of the 3′ UTR density is unclear ( listed in Figure 9—source data 1 ) . An example of stop codon read-through is the conserved plasmodium protein ( PF13_0160 ) , shown in Figure 9A . Ribosomes not only extend beyond the annotated stop codon of this transcript but also skip subsequent in-frame stop codons present on the predicted 644 nt 3′ UTR . Interestingly , ribosome footprints accumulate in a single large peak approximately 130 nt downstream of the annotated stop codon . On the 1290 nt 3′ UTR of the sodium-dependent phosphate transporter ( MAL13P1 . 206 ) , two large peaks of ribosome footprint density , one approximately 560 nt and the other 860 nt from the stop codon , can be observed ( Figure 9B ) . The origin of these footprints is unclear , and it is possible that these are the product of nuclease protection by RNA-binding proteins that co-sediment with the 80S monosome . To confirm that 3′ UTR mapping reads are derived from ribosome footprints , we compared their cumulative read length distributions against a typical CDS footprint read length distribution ( Figure 9—figure supplement 1 ) . For the 16 of the 19 genes we observed no significant difference in footprint size distributions localized to the CDS compared with the 3′ UTR . For the remaining three genes , the sodium-dependent phosphate transporter ( MAL13P1 . 206 ) , the acyl-Coa synthetase ( PFD0085c ) , and the conserved plasmodium protein ( PF13_0160 ) , 3′ UTR footprint size distributions were divergent from those on the CDS , implying that footprints found on these genes' 3′ UTRs may be produced by nuclease protection of these regions by factors other than ribosomes that co-sediment with 80S ribosomes . 10 . 7554/eLife . 04106 . 027Figure 9 . 3′ UTR ribosome density . ( A ) Late trophozoite stage mRNA ( blue ) and ribosome footprint ( green ) profiles of the conserved plasmodium protein , PF13_0160 . Ribosomes can be detected up to ∼130 nt beyond the stop codon on the 3′ UTR and accumulate in a single large peak . Red lines indicate in-frame stop codons on the 3′ UTR . ( B ) Two large peaks of ribosome footprint density can be detected 560 nt and 860 nt downstream from the stop codon in the 3′ UTR of the sodium-dependent phosphate transporter , MAL13P1 . 206 . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02710 . 7554/eLife . 04106 . 028Figure 9—source data 1 . Genes with 3′ UTR ribosome occupancy . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 02810 . 7554/eLife . 04106 . 029Figure 9—figure supplement 1 . 3′ UTR ribosome footprint size distribution . Cummulative read length distributions of all reads mapping to the 3′ UTR of the 19 genes with 3′ ribosome density identified compared to the read length distributions of reads mapping to all CDSs in the late trophozoite stage ( black line ) . Footprint length distributions for MAL13P1 . 206 , PF13_0160 , and PFD0085c are least similar to the ribosome footprints that map to the CDS . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 029 Antisense transcription plays an important role in gene regulation from bacteria to humans , and while its role is increasingly studied in these organisms ( Faghihi and Wahlestedt , 2009 ) , less is known about its relevance in P . falciparum . Previous serial analysis of gene expression ( SAGE ) ( Patankar et al . , 2001 ) , nuclear run-on experiments ( Militello et al . , 2005 ) , and more recently antisense splicing events detected by RNA-seq ( López-Barragán et al . , 2011; Sorber et al . , 2011 ) , suggest that antisense RNAs are synthesized by RNA pol II and may constitute up to ∼12% of the erythrocytic-stage steady-state RNA ( Gunasekera et al . , 2004 ) , yet their presence and biological role , if any , remains unclear . A more recent study found no correlation between natural antisense transcript levels and protein abundance ( Siegel et al . , 2014 ) . The 30 nt fragmentation and RNA-ligase-based library preparation method employed here affords exquisite strand specificity by minimizing artifacts associated with random priming during reverse transcription . As evidence of this specificity , the highest expressed gene in our data set , histone h2a ( PFF0860c ) , yielded a total of 765 , 510 reads on the sense strand , and only two reads on the antisense strand , corresponding to a sense:antisense ratio greater than 105 . Furthermore , our HMM mapping of 5′ leaders and 3′ UTRs facilitates the differentiation between independently transcribed antisense RNA and transcripts that occur by virtue of being part of an adjacent gene . We took advantage of the nature of our data set to identify antisense transcripts and looked for effects on sense mRNA translation . For this analysis only , we relaxed our stringent coverage threshold from ≥32 rM to ≥16 rM for inclusion of antisense transcripts . We based our threshold on the presence of an antisense transcript to the sexual stage specific gene pfs16 ( PFD0310w ) confirmed by strand-specific RT-PCR ( Figure 10—figure supplement 1 ) . This antisense is predicted by the HMM to be ∼4 kb , extending over the complete coding sequence and beyond and with a coverage level of 23 rM over the sense CDS . Using the 16 rM threshold , we detected 84 antisense transcripts to several known ORFs ( listed in Figure 10—source data 1 ) , including the nucleoside transporter pfNT4 ( PFA0160c ) depicted in Figure 10A . The merozoite stage contained the highest number of antisense transcripts ( 46 ) , and the fewest ( 13 ) were found in early trophozoites . Manual inspection revealed that in 63% of these instances , the putative antisense transcript actually emanates from the 5′ leader or 3′ UTR of a neighboring gene ( not defined by the HMM ) . Antisense reads for the para-hydroxybenzoate polyprenyltransferase ( PFF0370w ) , for example , are actually derived from the 3′ UTR of the neighboring conserved protein PFF0375c ( Figure 10B ) . 10 . 7554/eLife . 04106 . 030Figure 10 . Strand-specific libraries can distinguish antisense from sense gene transcription . ( A ) Schizont stage mRNA ( blue ) and ribosome footprint ( green ) profiles of the nucleoside transporter pfNT4 ( PFA0160c ) . The antisense transcript covers the full extent of the sense transcript and displays ribosome density . ( B ) An example of antisense reads originating from a neighboring UTR in the schizont stage . The antisense reads in the para-hydroxybenzoate polyprenyltransferase ( PFF0370w ) stem from the 3′ UTR of the neighboring conserved plasmodium protein ( PFF0375c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 03010 . 7554/eLife . 04106 . 031Figure 10—source data 1 . Antisense transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 03110 . 7554/eLife . 04106 . 032Figure 10—figure supplement 1 . Strand-specific RT-PCR detection of the antisense to Pfs16 . Read coverage on the plus and minus strand of the stage-specific protein precursor Pfs16 ( PFD0310w ) locus . The gene is encoded on the plus strand and the antisense transcript covers and extends beyond the sense transcript ∼3 . 7 kb . The strand-specific primer was used for both reversetranscription and as forward primer for the PCR ( blue arrowhead ) . The 5 PCR primers ( black arrowhead ) and the expected amplicon sizes are shown next to the strand-specific RT-PCR results . 18S rRNA primers were used in the control reactions . a . u . = arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 04106 . 032 We next interrogated the impact of this set of antisense transcripts . Overall , antisense transcripts showed no effect on mRNA abundance and translational efficiencies of the cognate sense transcript . These observations parallel those described for antisense transcripts in yeast ( Brar et al . , 2012 ) . Thus , at first approximation antisense transcripts do not appear to play a role in translational regulation . However , these observations could be confounded due to the small number of genes in this set , and we cannot exclude the possibility of sense/anti-sense heterogeneity at the single cell level , obscured here at the population level . Herein we present , for the first time , a comprehensive view of the coupled transcriptional and translational dynamics of the P . falciparum IDC by determination of transcript abundance and architecture together with ribosomal density and positioning . The quality of our data relies on several critical features: ( 1 ) high temporal specificity and reproducibility of fully independent biological replicas of five strictly staged cultures; ( 2 ) purified merozoites to allow discrete measurements in this stage without confounding contributions from schizonts or rings; ( 3 ) monosome isolation from sucrose gradients to specifically enrich for ribosome-derived footprints and avoid potential complications that can arise with methods like sucrose cushions which are prone to mRNA contamination; ( 4 ) sufficient sequencing depth of biological replicates to set a statistical threshold for minimum read coverage and to demonstrate reproducibility; ( 5 ) stringent strand specificity to facilitate an HMM for the description of transcript boundaries and the detection of antisense transcription . Previous studies of the transcript abundance in the malaria blood stages revealed a periodic cascade of gene expression , whereby the majority of expressed genes exhibit one peak of expression per cell cycle ( Bozdech et al . , 2003 ) . The global profile of transcriptional expression was subsequently found to be highly stereotypical across strains and appeared to lack dynamic responses to perturbation ( Llinás et al . , 2006; Ganesan et al . , 2008 ) . It has been suggested that translational control of protein expression could compensate for the lack of transcriptional dynamics . Proteomic studies described delays in peak mRNA and corresponding protein abundance implicating translational or post-translational mechanisms in the modulation of gene expression ( Le Roch et al . , 2004; Foth et al . , 2011 ) . Our ribosome profiling results reveal a tight coupling of transcription and translation for the majority of expressed genes , indicating that most protein products are translated with highly similar timing and in proportion to their corresponding mRNA transcripts . Synthesized proteins are likely to exert their functions immediately upon translation but post-translational regulation , not captured by our data , could still be at play . Direct correlations of translational efficiencies measured in this study along with proteomic data sets are hampered by the reduced sensitivity of the latter and differences in temporal resolution and staging of the parasites between data sets . However , the available proteomic evidence is largely consistent with the results presented here , particularly for highly translated proteins . The simultaneous capture of mRNA abundance and translation is expected to be a more accurate proxy for protein levels than measurements of mRNA abundance alone ( Ingolia et al . , 2009 ) and provides a critical resource for the identification of instances of post-translational regulation of gene expression . However , we note that this data set only provides a direct measure of relative changes in translational efficiencies rather than changes in bulk transcription and translation . While no up- or down-regulation of global translation efficiencies were observed in any particular stage , more extreme translational efficiencies were measured in subsets of genes expressed in rings and merozoites . We find 177 translationally up-regulated genes with functions predominantly related to merozoite egress and invasion , with peak mRNA in schizonts and peak translational efficiency in rings . It is likely that the genes with unknown functions , regulated in an analogous way during the merozoite to ring transition , are also associated with this process . Our data support a model whereby the transcripts of proteins necessary for merozoite structure and function are made in the previous stage in large abundance , are translationally up-regulated during the invasion process , and remain highly translated well into the ring stage despite rapid mRNA decay during this stage ( Shock et al . , 2007 ) . Whether the accumulation of 5′ leader ribosome density is a mechanism that assists in this process or is it merely a byproduct of more efficient ribosomal initiation on these templates remains to be tested . With the emergence of genome editing tools such as CRISPR/Cas9 ( Ghorbal et al . , 2014 ) , it may be possible to create versions of genes with altered cis-acting sequences to test for modulation of 5′ ribosome density and its effect on translational efficiency . The global nature of ribosome accumulation within the 5′ leader sequences of many transcripts during the IDC and the lack of an association between 5′RD and the number or length of uORFs suggests that ribosomes accumulate on 5′ leaders through means other than a uORF model . For comparison , in yeast under starvation conditions the fraction of ribosome footprints derived from 5′ leaders is increased by sixfold and in some cases no single uORF can account for the entire distribution of ribosomes on the 5′ leader of a gene ( Ingolia et al . , 2009 ) , much like P . falciparum . What mechanism could account for global ribosome accumulation in the 5′ leader ? The presence of apparent 80S ribosomes within the 5′ leader sequence , regardless of whether they cover uORFs or not , suggests an engagement mode in which the fidelity of start codon recognition is altered or suspended . Current models propose that the 43S pre-initiation complex loads onto the mRNA with the assistance of other initiation factors near the 5′ cap and proceeds to scan down the length of the mRNA until it encounters an AUG codon . This is followed by the assembly of the 48S preinitiation complex and then finally the 80S complex ( for review , see Hinnebusch , 2011 ) . The AUG that is ultimately chosen is not always the first one encountered , and its sequence context is important for recognition . The factors eIF1 , eIF1A , and eIF5 have been implicated in recognition of the ‘correct’ AUG ( Aitken and Lorsch , 2012 ) . In the case of P . falciparum , differential regulation or modification of these factors could plausibly result in altered start codon selection and 80S assembly . Whether prematurely initiated complexes are able to scan without synthesizing a peptide or are required to assemble and reassemble until encountering the right start codon remains an open question . Large 5′ ribosome accumulation on translationally up-regulated genes in the ring stage suggests that premature initiation on these transcripts is not detrimental . The development of an in vitro translation system that recapitulates upstream 80S assembly on P . falciparum 5′ leaders will allow direct testing of premature initiation and its effect on translational efficiency in this parasite . Our ribosome profiling data add an important component to the rich compendium of genome-wide data , including transcript abundance ( Bozdech et al . , 2003 ) , mRNA decay ( Shock et al . , 2007 ) , splicing ( Sorber et al . , 2011 ) , and proteomics for this parasite ( Le Roch et al . , 2004; Foth et al . , 2011 ) . Features such as 5′ leaders , 3′ UTRs , introns , and antisense transcripts are clearly visible and often well delineated . While experimental validation of transcriptional start sites , terminators , and promoters is required , spanning regions between transcripts , such as the one shown in Figure 6 , can be used for the search and identification of such functional sites in a reduced sequence space . The data are available at NCBI GEO ( accession #GSE58402 ) to facilitate future queries and normalized read coverage plots for all 5 timepoints are available packed as a single Mochiview file ( Caro et al . , 2014 ) . Together our results describe a simplified regulatory architecture of gene translation , albeit one that includes peculiar and potentially unique mechanisms specialized for its highly structured and coordinated lifecycle within erythrocytes . Further biochemical dissection of translational initiation mechanisms and determinants of translational efficiency unique to Plasmodium may reveal weaknesses that could be exploited for possible therapeutic intervention . W2 strain cultures were maintained in Hyperflasks ( Corning , Corning , NY ) in 500 ml RPMIc ( RPMI 1640 media supplemented with 0 . 25% Albumax II ( GIBCO , Grand Island , NY ) , 2 g/L sodium bicarbonate , 0 . 1 mM hypoxanthine , 25 mM HEPES ( pH 7 . 4 ) , and 50 μg/L gentamycin ) , at 37°C , 5% O2 , and 5% CO2 , to maximum 10–15% parasitemia at 5% hematocrit ( HC ) and frequent media changes ( at least every 6–8 hr ) . Cells were synchronized by two consecutive sorbitol treatments for three generations , for a total of six treatments . Maximum invasion , point at which half of the culture is either rings or schizonts , was defined as hour zero and independent time points containing ∼1010 parasites were harvested 11 , 21 , 31 , and 45 hr later . Cultures were incubated for 5 min in 500 ml 37°C RPMIc , 100 µg/ml cycloheximide ( Acros Organics , Bridgewater , NJ ) and harvested by centrifugation for 8 min at 3 . 65×g at room temperature . An aliquot was removed and flash frozen in liquid nitrogen for total RNA purification , followed by poly ( A ) -purification and chemical fragmentation with Zn2+ to ∼30 nt for consistency in mRNA-Seq library preparation . The remaining culture was treated with ice-cold 0 . 1% saponin in 1X PBS , 100 µg/ml cycloheximide , for RBC lysis . Parasites were resuspended in ice-cold parasite lysis buffer ( 15 mM KOAc , 15 mM MgOAc , 10 mM Tris HCl pH 7 . 4 , 0 . 5 mM DTT , 0 . 5% Triton X-100 , 100 μg/ml cycloheximide ) and dripped into a conical tube filled with , and immersed in , liquid nitrogen . Frozen cells transferred placed in liquid nitrogen pre-chilled chambers and pulverized for 3 min at 15 Hz , on a Retsch MM301 mixer mill . Pulverized cells were thawed on ice , and cell debris was removed by centrifugation at 4°C , 16 , 000×g for 10 min . The supernatant was treated with 2 . 88 U/µg Micrococcal nuclease for 30 min at room temperature and immediately loaded onto sucrose gradients for ultracentrifugation at 35 , 000 rpm for 3 hr at 4°C in a L8-60 M Beckman centrifuge . Monosome fractions only , were collected to generate ribosome footprint libraries for deep sequencing using the HiSeq 2000 ( Illumina , San Diego , CA ) , as described ( Ingolia et al . , 2009 ) . Late stage schizonts ( 40–44 hpi ) were magnetically purified using MACS LD columns ( Miltenyi Biotec , San Diego , CA ) and resuspended in RPMIc without blood addition . After reaching maximum invasion ( 1:1 schizont to ring ratio ) , cultures were harvested by centrifugation at 1500 rpm at room temperature for 5 min . Pelleted cultures were resuspended into fresh RPMIc and placed at 37°C . Merozoites in the supernatant were treated with 100 µg/ml cycloheximide for 5 min at room temperature , harvested at 4000 rpm at 4°C for 5 min and resuspended in RPMIc and passaged again through a MACS LD column . Parasite lysis buffer was added to the merozoite-enriched flow-through and flash frozen in liquid nitrogen . This procedure was repeated three times every 45 min using the original culture . The same procedure described above was used for RNA extraction and library preparation . W2 strain genomic DNA was isolated from >90% synchronized ring stage cultures . Paired end libraries were constructed using the Nextera DNA Sample Prep Kit ( Epicenter Biotechnologies , Madison , WI ) according to the manufacturer's instructions reducing PCR cycles from nine to six and using 80% A/T dNTPs . Libraries were sequenced using the HiSeq 2000 ( Illumina ) . Reads were aligned to the P . falciparum PlasmoDB 3D7 version 7 . 1 genome using Bowtie 0 . 12 . 1 ( Langmead et al . , 2009 ) with parameters –v1 –m 1 ( one mismatch allowed , unique mapping ) . A SNP was called when five or more W2 reads supported , with over 90% agreement , a different base than the one found in the P . falciparum 3D7 7 . 1 genome . 19401 SNPs ( 0 . 08% of total bases ) were detected and used to produce the SNP-corrected W2 genome based on the 3D7 genome . Fastq files are available for download at NCBI SRA , accession #SRP042946 . Quality-filtered ribosome footprints and mRNA sequencing reads were trimmed to remove library adapter sequence , filtered for P . falciparum rRNA using blast , and aligned uniquely to the W2 SNP-corrected genome using Bowtie 0 . 12 . 1 ( Langmead et al . , 2009 ) allowing no mismatches . The percentage mappability was calculated using an in silico library of the P . falciparum W2 SNP-corrected genome created using a single nucleotide sliding window of 30 nt . The in silico library was uniquely aligned to the genome allowing no mismatches . The mappability score is given by the number of 30 nt sequences covering each nt position in the genome , such that any position has a score that ranges from 0 to 30 . Both mRNA and ribosome footprint rpkMs were calculated as in Mortazavi et al . ( 2008 ) , excluding the first 50 bases of each gene to eliminate bias introduced by the observed ribosome accumulation peak near the start codon . Genes with fewer than 80% mappable bases ( 248 genes ) or any overlapping non-CDS feature on the same strand ( 77 genes ) were excluded from this calculation . Data are available for download at NCBI GEO , accession #GSE58402 . MochiView genome browser data tracks are available in Supplementary file 1 ( Homann et al . , 2010 ) . The genes of the RNA-seq transcriptome obtained in this work were listed in the same phaseogram order as the previously published microarray transcriptome ( Bozdech et al . , 2003 ) . The criteria for inclusion of a gene into the phaseogram was mRNA ≥ 32 rM , >2 peak to trough ratios , and Pearson correlation coefficient >0 . 8 with the expression profiles of the two neighboring genes . The HMM was built using RNA-Seq data obtained in this study and two states: transcript ( t ) or intergenic ( i ) with three possible emissions: ( 1 ) read present , ( 2 ) read not present but position is unmappable , ( 3 ) read not present but the position is mappable . Both state and emission probabilites were calculated using ∼30 kb training set of manually identified transcript and non-transcript regions . The initial probabilities were set to 0 . 5 . Transition probabilities were estimated from the median length of intergenic regions of ( 1252 nt ) and median lengths of CDS regions ( 2545 nt ) , where the Pt->i = ( 1/2545 ) , P t->t = ( 2544/2545 ) , Pi->t = ( 1251/1252 ) , and Pi->i = ( 1/1252 ) . We applied the Viterbi algorithm to predict the optimal path of transcript tracks per time point with a 10 nt window resolution . HMM-defined 5′ leader and 3′ UTR coordinates are available for download at NCBI GEO , accession #GSE58402 . Total RNA from late stage parasites was isolated and reverse transcribed using SuperScript III ( Invitrogen , Carlsbad , CA ) according to manufacturer's instructions , using either an antisense-specific primer to Pfs16 ( PFD0310w ) or a random nonamer . cDNA was amplified using the Pfs16 antisense-specific primer as a forward primer in combination with one of five reverse primers ( Supplementary file 3 ) . 18S rRNA primers were used in the control reactions with the random nonamer-derived cDNA .	The genome of an organism includes all of the genes or information necessary to build , maintain , and replicate that organism . However , cells with the same genome—such as a skin cell and a liver cell from the same person—can look and behave very differently depending on which of the genes in their genomes they express , and to what extent . For a gene to be expressed , its DNA is ‘transcribed’ to make an RNA molecule , which is then ‘translated’ to make a protein . Efforts to measure the transcription and translation processes in diseased cells , or in the microorganisms that cause infections , may lead to new treatments and preventative medicines . Such work is currently ongoing in the global effort to treat and prevent malaria . Malaria is both preventable and curable , yet over 600 , 000 people are estimated to die from this disease each year . The disease is caused by a single-celled parasite called Plasmodium . Mosquitoes carry the parasites in their salivary glands , and when a mosquito bites a human , these parasites are injected into the bloodstream with the mosquito's saliva . Plasmodium parasites then travel to and infect the liver , before bursting out of this tissue into the bloodstream . Here , the parasites infect red blood cells and undergo rounds of replication during which the symptoms of the disease are manifested . It is also during this bloodstream phase that parasites can develop into forms capable of infecting another mosquito and continuing the transmission cycle . The genes , RNA molecules , and proteins of the Plasmodium falciparum parasite—which causes the most serious cases of malaria in humans—have been cataloged to better understand the biology of this parasite . However , the processes that control how , and when , an RNA transcript is translated into a protein are not well understood . Now Caro et al . have uncovered which RNA molecules are being translated , and by how much , during Plasmodium development within the blood . The transcription and translation of genes in this parasite were found to be tightly linked processes; the expression of only a few genes was controlled more by the translation process than by transcription . These translationally regulated genes were found mainly to be those that encode proteins involved in the parasite's exit from the red blood cells and spread throughout the bloodstream . Caro et al . discovered that genetic regulation of the malaria parasite resembles a preset genetic program , rather than a system that responds to changes and external signals . As such , these findings suggest that targeting such a genetic program within Plasmodium and preventing its implementation could prove an effective strategy to curb the spread of malaria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2014	Genome-wide regulatory dynamics of translation in the Plasmodium falciparum asexual blood stages
Crucial roles for T-box3 in development are evident by severe limb malformations and other birth defects caused by T-box3 mutations in humans . Mechanisms whereby T-box3 regulates limb development are poorly understood . We discovered requirements for T-box at multiple stages of mouse limb development and distinct molecular functions in different tissue compartments . Early loss of T-box3 disrupts limb initiation , causing limb defects that phenocopy Sonic Hedgehog ( Shh ) mutants . Later ablation of T-box3 in posterior limb mesenchyme causes digit loss . In contrast , loss of anterior T-box3 results in preaxial polydactyly , as seen with dysfunction of primary cilia or Gli3-repressor . Remarkably , T-box3 is present in primary cilia where it colocalizes with Gli3 . T-box3 interacts with Kif7 and is required for normal stoichiometry and function of a Kif7/Sufu complex that regulates Gli3 stability and processing . Thus , T-box3 controls digit number upstream of Shh-dependent ( posterior mesenchyme ) and Shh-independent , cilium-based ( anterior mesenchyme ) Hedgehog pathway function . The T-box gene family encodes transcription factors that play critical roles during embryonic development , organogenesis , and tissue homeostasis . Mutations in TBX genes in humans cause multiple developmental dysmorphic syndromes and disease predispositions ( Naiche et al . , 2005; Showell et al . , 2004 ) . Heterozygous mutation of TBX3 causes Ulnar-mammary syndrome ( UMS ) , initially described as a constellation of congenital limb defects , apocrine and mammary gland hypoplasia , and genital abnormalities ( Pallister et al . , 1976 ) . Recently , heart and conduction system defects have also been described in mice and humans with abnormal Tbx3 ( mice ) and TBX3 ( humans ) function ( Bakker et al . , 2008; Frank et al . , 2012; Linden et al . , 2009; Meneghini et al . , 2006; Mesbah et al . , 2008 ) . Germline deletion of Tbx3 in mice results in embryonic lethality with heart , limb , and mammary defects ( Davenport et al . , 2003; Frank et al . , 2012; 2013 ) . Tbx3 also regulates pluripotency and cell fate in early development ( Cheng et al . , 2012; Han et al . , 2010; Kartikasari et al . , 2013; Niwa et al . , 2009; Weidgang et al . , 2013 ) . TBX3 transcriptional repression controls expression of cell proliferation and senescence factors ( Brummelkamp et al . , 2002; Kumar et al . , 2014a ) ; abnormal TBX3 expression occurs in multiple cancers ( Liu et al . , 2011; Lu et al . , 2010; Peres and Prince , 2013 ) . TBX3 also regulates splicing and RNA metabolism ( Kumar et al . , 2014b ) . Although these studies highlight the important pleiotropic molecular functions of TBX3 , little is known about the core pathways it regulates in developing structures that require its function , such as the developing limb . UMS limb phenotypes are variable ranging in severity from hypoplasia of digit 5 to complete absence of forearm and hand ( OMIM #181450 ) . Mouse Tbx3tm1Pa/tm1Pa ( Davenport et al . , 2003 ) and Tbx3Δfl/Δfl ( Frank et al . , 2013 ) mutant forelimbs lack posterior digits and the ulna . Hindlimbs of Tbx3tm1Pa/tm1Pa and Tbx3Δfl/Δfl null mutants have only a single digit , but Tbx3Δfl/Δfl mutants also have pelvic defects ( Frank et al . , 2013 ) . Embryonic lethality of both types of mutants has prevented elucidation of Tbx3’s limb-specific roles . The Hedgehog pathway is a key regulator of limb development . Shh signaling in posterior mesenchyme promotes digit development and prevents processing of full length Gli3 ( Gli3FL ) to its repressor form , Gli3R , which constrains digit number ( Litingtung et al . , 2002 ) . The balance of Gli transcriptional activation and repression is critical for proper digit number and patterning ( Cao et al . , 2013; Hill et al . , 2007; Litingtung et al . , 2002; te Welscher et al . , 2002; Wang et al . , 2000; 2007a; Zhulyn et al . , 2014 ) . In mammals , the limited , partial proteolytic processing of Gli3FL to Gli3R requires functional primary cilia , the ciliary protein Kif7 ( Goetz and Anderson , 2010; Liu et al . , 2005 ) , as well as balanced activity of Sufu and the ubiquitin ligase adaptors βTrCP and Spop ( Chen et al . , 2009; Wang and Li , 2006; Wang et al . , 2010; Wen et al . , 2010 ) In this study , conditional ablation of Tbx3 reveals discrete roles for Tbx3 during limb initiation and compartment-specific functions during later limb development to regulate digit number . We discovered a novel molecular function of Tbx3 in the primary cilia where it interacts directly with Kif7 and is in a complex with Gli3 . Loss of Tbx3 decreases Kif7-Sufu interactions , resulting in excess Gli3 proteolysis and decreased levels of both Gli3FL and Gli3R . The resulting preaxial polydactyly phenocopies limb defects seen in Gli3 null heterozygotes and in mutants with abnormal structure or function of the primary cilia ( Cheung et al . , 2009; Endoh-Yamagami et al . , 2009; Goetz and Anderson , 2010; Haycraft et al . , 2005; Liem et al . , 2009; Liu et al . , 2005; Ocbina et al . , 2011; Putoux et al . , 2011 ) . Our findings reveal a novel mechanism where Tbx3 in the anterior mesenchyme is required for proper function of the Kif7/Sufu complex that regulates Gli3 stability and processing . Tbx3 is expressed in discrete anterior and posterior mesenchymal domains in the limb buds from embryonic day ( E ) 9 . 5 ( Figure 1A , C , E , Figure 1—figure supplement 1 ) . To assess the role of these domains during limb development , we generated conditional mutants using our Tbx3flox allele ( Frank et al . , 2012; 2013 ) and the Prx1Cre transgene ( Logan et al . , 2002 ) ( genotype Tbx3flox/flox;Prx1Cre , henceforth referred to in the text as Tbx3;PrxCre mutants ) . This driver initiates Cre activity at ~14-somite stage ( ss ) in the forelimb-forming region of the lateral plate mesoderm ( LPM ) ( Hasson et al . , 2007 ) . Its activity in the hindlimb is irregular , so our analysis focuses on the forelimb . In situ hybridization and immunohistochemistry confirm complete ablation of Tbx3 mRNA and protein in Tbx3;PrxCre mutant forelimb mesenchyme by E9 . 5 ( Figure 1B–F , Figure 1—figure supplement 1 ) . Expression in the apical ectodermal ridge ( AER ) is preserved ( Figure 1B , D , F ) . We previously reported the specificity of the custom anti-Tbx3 antibody used here and loss of limb mesenchymal protein production in Tbx3;PrxCre mutant forelimbs ( Frank et al . , 2012; 2013 ) . 10 . 7554/eLife . 07897 . 003Figure 1 . Tbx3 regulates anterior and posterior digit development . ( A ) Tbx3 expression assayed by mRNA in situ hybridization in E9 . 5 forelimb bud ( black line from a-p shows anterior-posterior axis ) . Red arrow points to Tbx3 expression in apical ectodermal ridge ( AER ) . Red ellipse encloses posterior mesenchymal expression domain . ( B ) Tbx3 transcripts are absent in the limb bud mesenchyme of E9 . 5 Tbx3fl/fl;PrxCre mutants . Tbx3 expression persists in the AER ( red arrow ) and adjacent posterior-lateral body wall ( black arrowhead ) . ( C , D ) As in A and B except limb buds are E10 . 5 . Red ellipses enclose anterior and posterior mesenchymal expression domains which are Tbx3 negative in the mutants . Red arrows highlight expression in AER . ( E , F ) Tbx3 immunohistochemistry on sectioned E10 . 5 limb . Tbx3 protein is lost in mesenchyme of Tbx3fl/fl;PrxCre mutants ( F , red ellipses ) but AER staining persists as expected ( white arrowhead ) . Please see also Figure 1—figure supplement 1 . ( G–J ) Skeleton preparations reveal preaxial polysyndactyly ( duplicated/fused digit 1 , red bracket , H , H’ , J ) and postaxial oligodactyly ( absent digit 5 , red arrows in H’ and J ) in Tbx3fl/fl;PrxCre mutants at E15 . 5 ( H , H’ ) and E19 . 5 ( J ) . Note delayed ossification of the humerus ( H , black arrowhead ) , loss of deltoid tuberosity ( J , black arrowhead ) and short , bowed ulna ( J , black arrow ) in mutant . s , scapula; h , humerus; oc , ossification center; dt , deltoid tuberosity r , radius; u , ulna; digits numbered 1–5 ( K–N ) Sox9 mRNA expression shows evolving skeletal defects are already evident in Tbx3fl/fl;PrxCre mutants at E10 . 5- E11 . 5 . Digit condensations are numbered . Bracket in L shows broadening of digit 1 forming region; red arrows highlight indentation in digit 5 forming region ( L , N ) and absence of Sox9 digit 5 condensation ( N ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 00310 . 7554/eLife . 07897 . 004Figure 1—figure supplement 1 . Ablation of Tbx3 with PrxCre eliminates anterior and posterior mesenchymal protein production . From Frank et al . , PLoSOne 2013 , with permission . Confocal micrographs of sectioned E10 . 0 forelimb buds after fluorescent immunohistochemical detection of Tbx3 using custom antibody to its C-terminus . ( C1–C4 ) Tbx3+/+ limb bud . C1 ) Merged color view of DAPI and FITC channels at 10X magnification . ( C2–C4 ) 60X magnification of white boxed region in C1 . ( C2 ) DAPI channel showing DNA immunoreactivity . ( C3 ) FITC channel showing Tbx3 immunoreactivity . ( C4 ) Merged view . ( D1–C4 ) Tbx3fl/fl;PrxCre limb bud . ( D1 ) Merged color view of DAPI and FITC channels at low magnification . ( D2–C4 ) 60X magnification of white boxed region in D1 . ( D2 ) DAPI channel showing DNA immunoreactivity . ( D3 ) FITC channel showing lack of Tbx3 immunoreactivity in nucleus and cytoplasm . ( D4 ) Merged view . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 00410 . 7554/eLife . 07897 . 005Figure 1—figure supplement 2 . Increased severity of limb phenotypes in Tbx3 null mutants ( Tbx3Δfl/Δfl ) compared to Tbx3;PrxCre is independent of Tbx3 in the AER . ( A–D ) Skeleton preparations comparing control ( A: Tbx3Δfl/+ , C:Tbx3fl/fl ) , Tbx3Δfl/Δfl ( B , null ) , Tbx3fl/fl and Tbx3fl/fl;PrxCre ( D , conditional mutant ) forelimbs . Note single digit , absent ulna , and shortened humerus in Tbx3Δfl/Δfl mutant ( B ) compared to preaxial polysyndactyly and absent digit 5 in Tbx3;PrxCre mutants ( D ) . s , scapula; h , humerus; r , radius; u , ulna; digits are numbered; red arrowhead highlights loss of digit 5 . ( E , F ) X-gal stained E10 . 0 ( E ) and E11 . 5 ( F ) RosaLacZ/+;Fgf8mcm/+embryos after the administration of tamoxifen at E8 . 5; black arrow indicates staining indicative of previous Cre activity in the AER . ( G–J ) mRNA in situ for Tbx3 expression shows the absence of signal in the AER of Tbx3fl/fl;Fgf8mcm/mcmE9 . 5 and E10 . 5 mutants ( H , J , respectively ) compared to controls ( G , I ) . White arrows point to AER in G–J; note persistent mesenchymal Tbx3 expression as expected . ( K–N ) Skeleton preparations comparing E15 . 5 control ( K , M ) , and Tbx3 fl/fl;Fgf8mcm/mcm ( L ) and Tbx3fl/fl;RarCre ( N ) mutants . Forelimbs of Tbx3 fl/fl;Fgf8mcm/mcm mutants are normal ( L ) , while defects in Tbx3fl/fl;RarCre ( N ) phenocopy those of Tbx3;PrxCre mutants ( compare panel N to D and also to Figure 1 , panel H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 005 Unlike mid-gestational lethality seen in constitutive Tbx3 mutants ( Davenport et al . , 2003; Frank et al . , 2013 ) , Tbx3;PrxCre mutants survive to adulthood with forelimb defects ( Video 1 ) : 100% have bilateral preaxial polysyndactyly of digit 1 ( called PPD1 in humans [Materna-Kiryluk et al . , 2013] ) , and 70% lack digit 5 ( Figure 1G–J ) . Loss of digit 5 was bilateral in 6/18 and in the remaining , only affected the left forelimb ( Table 1 ) . This is not due to asymmetric activity of PrxCre because it is also observed in Tbx3Δfl/Δfl mutants ( Δfl = recombined floxed conditional allele ) where no Cre activity is involved ( Frank et al . , 2013 ) . Delayed ossification of the humerus ( Figure 1H ) and loss of the deltoid tuberosity ( Figure 1J ) were also observed . Abnormal limb bud morphology is evident by E10 . 5 ( Figure 1L ) and evolving skeletal defects at E11 . 5 by the altered pattern of Sox9 expression ( Figure 1N ) . 10 . 7554/eLife . 07897 . 006Table 1 . Increased severity of left limb defects in Tbx3;PrxCre mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 006Skeletal phenotypes: E13 . 5-adultLoss of digit 5BilateralLeft onlyRight onlyTbx3 fl/+ or fl/fl000Tbx3fl/fl;PrxCre6120Molecular phenotypes: gene expressionExpression pattern or levelLeft = RightLeft >RightRight > LeftTbx3 fl/+ or f/flcontrolcontrolcontrolTbx3fl/fl;PrxCre 1930110 . 7554/eLife . 07897 . 007Video 1 . Adult Tbx3;PrxCre mutant mouse is healthy and mobile despite forelimb deformities . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 007 Germline Tbx3 null mutants ( genotype Tbx3Δfl/Δfl ) have more severe forelimb defects than Tbx3;PrxCre conditional mutants: of the few Tbx3Δfl/Δfl mutants that survive to E13 . 5 , 100% have agenesis of the ulna and digits 3–5 ( Figure 1—figure supplement 2B ) . Their hindlimbs have a single digit and no fibula ( Frank et al . , 2013 ) , phenocopying Shh and Hand2 null mutants ( Galli et al . , 2010 ) . Variable timing of Tbx3 loss of function by PrxCre may account for the disparate forelimb phenotypes of Tbx3Δfl/Δfl and Tbx3;PrxCre mutants , however , our skeletal data and phenotypes of Tbx3Δfl/fl;PrxCre mutants indicate that such variability manifests as incomplete penetrance of the ulnar and digit 5 defects ( Figure 1 H , H' , J , and Colesanto et al . , in preparation ) . The AER is a critical signaling center , and Tbx3 expression is preserved in the AER of Tbx3;PrxCre mutants ( Figure 1B , D , F; Figure 1—figure supplement 1 , Frank et al . , 2013 ) . We tested whether AER Tbx3 has a required function using two Cre drivers: RarbCre ( active in AER and mesenchyme from E9 . 0 [Moon and Capecchi , 2000] ) , and a novel Fgf8mcm allele , which produces tamoxifen-inducible Cre in Fgf8 expression domains ( Moon et al . , in preparation ) . Tamoxifen induction at E8 . 5 induces robust Cre activity in the AER in RosaLacZ/+;Fgf8mcm/+ embryos ( Figure 1—figure supplement 2E , F ) and ablates Tbx3 from forelimb AER by at least E9 . 5 ( Figure 1—figure supplement 2G–J ) . Tbx3fl/fl;Fgf8mcm/mcm mutants have normal limbs ( Figure 1—figure supplement 2L ) and phenotypes of Tbx3fl/fl;PrxCre and Tbx3fl/fl;RarbCre are indistinguishable ( Figure 1—figure supplement 2D versus N ) . The results with both Cre drivers indicate that the severe phenotypes of Tbx3Δfl/Δfl mutants are not due to a required function of Tbx3 in the AER . We next tested whether discrepant forelimb phenotypes in Tbx3Δfl/Δfl and Tbx3;PrxCre mutants reflect a role for Tbx3 in an earlier expression domain than affected by RarbCre or PrxCre . Limb initiation in the LPM requires Tbx5 expression in the prospective forelimb territory as early as the 8ss ( Minguillon et al . , 2005 ) , upstream of Hand2 ( Agarwal et al . , 2003 ) . Tbx3 is expressed in the LPM from E7 . 5 ( Figure 2A–C’ ) . Lineage tracing with a novel Tbx3mcm allele ( Thomas et al . , in preparation ) revealed that Tbx3-expressing progenitors in the LPM at E8-8 . 5 give rise to most E10 forelimb mesenchyme in Tbx3mcm/+;RosaLacZ/+ embryos ( Figure 2D ) . Consistent with a role for Tbx3 in limb initiation , Tbx3Δfl/Δfl mutants have decreased LPM expression of Tbx5 ( Figure 2E , F ) , visible defects in forelimb initiation and early limb bud morphology ( Figure 2G , H ) , and disrupted expression of Hand2 ( Figure 2I , J ) . In contrast , early stage Tbx5 expression and limb bud initiation are unaffected in Tbx3;PrxCre mutants ( Figure 2—figure supplement 1B , B' ) 10 . 7554/eLife . 07897 . 008Figure 2 . Tbx3 is required for normal limb bud initiation . ( A , B ) Tbx3 expression at E7 . 5 ( A ) and E8 . 5 ( B ) . Anterior on left ( A ) , posterior on right ( P ) . NM ( black arrow ) indicates nascent mesoderm exiting primitive streak in panel A . ( C , C’ ) Tbx3 expression in the LPM ( black arrow ) of sectioned E8 . 5 embryo . Plane of section indicated by line in B . Panel D is magnification of red-boxed area in C . ( D ) X-gal stained E10 . 0 Tbx3MCM/+; Rosa LacZ/+ embryo after Cre induction at E8 . 5 . FL , forelimb; HL , hindlimb . ( E , F ) 21 somite stage ( ss ) embryos assayed for Tbx5 mRNA . White arrows denote forelimb bud . Left sided view . ( G , H ) Dorsal view of budding forelimbs of 24 ss embryo forelimbs ( neural tube stained for Shh expression ) . Note abnormal shape and size of Tbx3Δfl/Δfl mutant forelimb buds indicative of disrupted initiation; white brackets are of equal size in both panels . ( I , J ) 22 ss embryos assayed for Hand2 expression . Black arrows denote emerging forelimb bud . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 00810 . 7554/eLife . 07897 . 009Figure 2—figure supplement 1 . Early Tbx5 expression is normal in Tbx3fl/fl;PrxCre mutants . ( A–B’ ) In situ hybridization for Tbx5 mRNA on E9 . 5 control ( A , A’ ) embryo versus Tbx3;PrxCre mutant ( B , B’ ) . Left sided views in A and B and dorsal views of dissected torsos with forelimbs in A’ , B’ . Tbx5 expression and limb initiation are normal after conditional loss of Tbx3 in the limb bud mesenchyme . ( C–D’ ) In situ hybridization for Hand2 mRNA on E9 . 5 control ( A , A’ ) embryo versus Tbx3;PrxCre mutant ( B , B’ ) . Left- sided views in C and D and dorsal views of dissected limbs in C’ , D’ . Hand2 expression is affected by conditional loss of Tbx3 in limb bud mesenchyme , but not as severely as in Tbx3Δfl/Δfl mutants shown in Figure 2J . Boxed area in D encloses forelimb forming region shown from dorsal view in D’ . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 009 Our data reveal required functions for Tbx3 in limb initiation ( demonstrated by Tbx3Δfl/Δfl mutants ) and in later limb bud morphogenesis ( demonstrated by Tbx3;PrxCre mutants ) . Most Tbx3;PrxCre mutants lack digit 5 , whose specification and formation depend on 'early phase' Shh signaling beginning at E9 . 5 ( Harfe et al . , 2004; Scherz et al . , 2007; Zhu and Mackem , 2011; Zhu et al . , 2008 ) . Hand2 protein is required to activate Shh expression in the limb bud ( Benazet and Zeller , 2009; Galli et al . , 2010 ) . We found that Hand2 transcripts are reduced in E9 . 5 and E10 . 5 Tbx3;PrxCre mutant forelimb buds ( Figure 2—figure supplement 1D , D'; Figure 3A , A' , F ) as is Shh expression ( Figure 3B , B' , F; Figure 3—figure supplement 1 ) . Expression of two targets and effectors of Shh signaling , Ptch1 and Grem1 , is also markedly reduced ( Figure 3C' , E'; Figure 3—figure supplement 2 ) . Tbx3 expression in posterior limb mesoderm begins earlier than in the anterior compartment ( Figure 1A , C ) and is required for normal Hand2 in posterior mesoderm ( Figure 3A , Figure 2J and Figure 2—figure supplement 1D’ ) ( Rallis et al . , 2005 ) . Thus , intact Tbx5 expression in Tbx3;PrxCre E9 . 5 forelimbs ( Figure 2—figure supplement 1B’ ) indicates that post-initiation , Tbx3 functions downstream of Tbx5 and upstream of Hand2 . 10 . 7554/eLife . 07897 . 010Figure 3 . Loss of mesenchymal Tbx3 disrupts Shh signaling in the posterior limb bud and decreases Gli3 protein stability . ( A–E’ ) In situ hybridization of control and mutant forelimb buds with probes and at embryonic stages as labeled . ( F ) qPCR of E10 . 75 ( 36-39ss ) limb buds for transcripts listed confirms findings by detected by in situ . ( G–I’ ) In situ hybridization for Zic3 , Epha3 and Hoxd13 transcripts in forelimb buds of Tbx3 fl/+controls ( K–M ) and Tbx3;PrxCre mutants ( G’–I’ ) at ages noted on panels . J ) qPCR assay of Zic3 , Epha3 , Hoxd13 transcript levels confirms findings detected by in situ . ( K–L’ ) Representative images of E10 . 5 forelimb buds stained for DAPI ( blue ) , pHH3 ( green ) , TUNEL ( red ) . K is Tbx3 fl/+ control and K’ is digital zoom of posterior mesenchymal boxed area in K . Panel L is Tbx3;PrxCre mutant and L’ is digital zoom of boxed area in L . This experiment is representative of data obtained from five biologic replicates . ( M ) Quantification of proliferating cells in anterior and posterior mesenchymal regions encompassing digit 1 and digit 5 progenitors from 20 control and 15 mutant sections . p=0 . 02 . Control anterior limb ( CA ) , Tbx3;PrxCre mutant anterior limb ( MA ) , control posterior ( CP ) , and mutant posterior ( MP ) . ( N–O’ ) Representative images of E11 . 5 whole mount forelimb buds stained for DAPI ( blue ) and pHH3 ( green ) . N is Tbx3 fl/+ control and N’ is digital zoom of boxed area . Panel O is Tbx3;PrxCre mutant and O’ is digital zoom of boxed area in O . Note decreased pHH3+ cells in mutants , particularly cells in prophase and anaphase , which have the faint and speckled patterns compared to the bright staining of highly condensed S-phase chromatin . ( P ) Quantification of proliferating cells in anterior and posterior mesenchymal regions encompassing digit 1 and digit 5 progenitors from 50 control and 44 mutant sections at E11 . 5 . p<0 . 1 . Control anterior limb ( CA ) , Tbx3;PrxCre mutant anterior limb ( MA ) , control posterior ( CP ) , and mutant posterior ( MP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01010 . 7554/eLife . 07897 . 011Figure 3—figure supplement 1 . Decreased Shh expression in E10 . 5 forelimb buds of Tbx3;PrxCre mutants . ( A , B ) Whole mount in situ hybridization for Shh transcripts on E10 . 5 embryos; left -sided views . ( C , D ) Dissected limbs from whole mount in situ hybridization for Shh transcripts on E11 . 0 embryos; dorsal views of left limb buds are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01110 . 7554/eLife . 07897 . 012Figure 3—figure supplement 2 . No evidence of ectopic hedgehog pathway activity in Tbx3;PrxCre mutant forelimbs . ( A , B ) In situ hybridization for Ptch1 at E11 . 5 in control ( A ) versus Tbx3;PrxCre mutant ( B ) limb buds . Ptch1 expression is decreased in posterior mesenchyme , consistent with results in Figure 3 . There is no ectopic Ptch1 signal in anterior mesenchyme . Note decreased size of Ptch1 negative zone in posterior mesenchyme ( white ellipses ) , consistent with loss of digit 5 progenitors which are unresponsive to Shh signaling at this stage ( Scherz et al . , 2004 ) Ahn and Joyner , 2004 ) . a , anterior; p , posterior ( C , D ) . In situ hybridization for Grem1 at E11 . 5 reveals decreased expression throughout limb , consistent with qPCR and in situ results shown in Figure 3 at E10 . 75 . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01210 . 7554/eLife . 07897 . 013Figure 3—figure supplement 3 . Microdissection of E11 forelimb buds into anterior and posterior compartments for gene and protein expression analyses . ( A ) Intact left limb bud after in situ hybridization for Tbx3 mRNA . Anterior ( ant ) at top , posterior ( post ) at bottom . ( B , C ) Microdissected anterior and posterior Tbx3+ compartments and list of example genes whose expression is confined to , or enriched in , each compartment . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01310 . 7554/eLife . 07897 . 014Figure 3—figure supplement 4 . qPCR of additional key transcripts in anterior and posterior forelimb compartments at E10 . 5–10 . 75 ( 36–39 somite stages ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01410 . 7554/eLife . 07897 . 015Figure 3—figure supplement 5 . Fgf8 expression and downstream in Tbx3;PrxCre mutant forelimb buds . ( A–F ) In situ hybridization for the transcripts listed on panels at ages specified . ( A , B ) View of AER stained for Fgf8 . ( C–F ) Dorsal view of left forelimb buds stained for Erm ( Etv5 ) and Pea3 ( Etv4 ) transcripts which are regulated by FGF signaling in the limb bud . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 015 To obtain a more comprehensive view of the transcriptional consequences of loss of Tbx3 in limb bud mesenchyme , we assayed gene expression of E10 . 25 limb buds ( 32–34 somite stage ) by microarray . Shh and other hedgehog pathway members and target genes ( Lettice et al . , 2003; Probst et al . , 2011; Vokes et al . , 2008; McGlinn et al . , 2005 ) ( Lewandowski et al . , 2015 ) were present among the significantly dysregulated transcripts ( Supplementary file 1 ) consistent with the previous report of decreased Shh expression in Tbx3tm1Pa/tm1Pa mutant limb buds ( Davenport et al . , 2003 ) . In addition to decreased levels of Shh- activated transcripts ( Gli1 , Hand2 , Osr1 , Dkk1 , Tbx2 , Cntfr , Pkdcc ) , we noted increased Rprm , Zic3 , Hand1 , and Gli3 transcripts and confirmed these with qPCR and in situ hybridization at E10 . 5–11 ( Figure 3 , Figure 3—figure supplement 4 ) . The increase in expression of these putative targets of the Gli3 repressor ( Lettice et al . , 2003; Probst et al . , 2011; Vokes et al . , 2008; McGlinn et al . , 2005 ) was intriguing as it suggested the possibility of alterations in both Gli3 activator and repressor function in Tbx3;PrxCre mutant forelimbs . The transcriptional profiles of anterior and posterior limb mesenchyme are quite different: alterations in gene expression in either compartment in response to Tbx3 could mask some changes in the other if assayed simultaneously . Thus , we proceeded to microdissect anterior and posterior limb segments and assayed them independently using RNA-sequencing on 38–42 somite stage ( ~E11; this later stage was needed in order to obtain sufficient RNA from accurately microdissected limb segments ) wild type and Tbx3;PrxCre mutant forelimb buds ( Figure 3—figure supplement 3 ) . The resulting RNA-Seq data confirmed accurate dissection with the expected distribution of known compartment-specific transcripts ( Figure 3—figure supplement 3 and Supplementary file 2 ) . Shh , Fgf4 and anterior Hoxd family transcripts were over-represented in the posterior compartment , and Alx and Pax family members in the anterior . Largely consistent with the previous microarray findings , we found evidence of aberrant cell differentiation/fate of posterior mesenchyme with downregulation of Shh-activated targets ( Osr1 , Dkk1 , Tbx2 , Cntfr , Ptch2 , Supplementary file 3 ) that validated by qPCR ( Figure 3—figure supplement 4 ) . It is known that Gli3 expression increases with decreased Shh activity ( Wang et al . , 2000 ) , as we see here ( Figure 3D’ , F ) . Although decreased levels of Hand2 , Shh , Ptch1 and Grem1 are clearly evident by in situ and qPCR at this stage ( Figure 3 and figure upplements ) , they were not detected on the RNA-Seq analysis for unclear reasons . Shh signaling is required for proliferation to ensure sufficient cell numbers to form the normal complement of digits , and loss of Shh results in an increase in the number of cells in G1 arrest ( Zhu et al . , 2008 ) . Assay of cell proliferation in E10 . 5 and E11 . 5 limb buds using anti-phosphohistone H3 immunohistochemistry revealed that at E10 . 5 there was a statistically significant decrease in the fraction of proliferating cells in the posterior mesenchyme ( Figure 3K–M ) . At E11 . 5 , proliferation was significantly decreased in both the anterior and posterior mesenchyme , indicating a global reduction in the number of mitotic cells in mutants ( Figure 3N–P ) . This suggests that 5th digit agenesis is attributable , at least in part , to decreased cell number , as opposed to decreased proliferation specifically in digit 5 progenitors . Assay for apoptosis using TUNEL showed normal levels of anterior AER cell death at E10 . 5 , as we have previously reported in this region ( Moon and Capecchi , 2000 ) ( Figure 3K , L ) . Proliferation of limb mesenchyme depends on activity of FGF8 and FGF4 from the AER ( Boulet et al . , 2004; Moon and Capecchi , 2000; Sun et al . , 2002 ) and integrity of this structure requires Shh activity in posterior mesenchyme ( Chiang et al . , 2001 ) . Despite decreased Shh expression in Tbx3;PrxCre mutants , Fgf4 transcripts were increased in posterior mesenchyme while decreased in anterior ( detected by RNA-Seq , Supplementary files 3 , 4 and qPCR , Figure 3—figure supplement 4 ) , the latter consistent with an expanded digit 1 region . qPCR detected increased Fgf8 expression in the posterior AER ( Figure 3—figure supplement 4 ) . Despite these changes in transcript levels , there was no evidence of altered downstream FGF signaling as expression of Etv4 , Etv5 , Dusp6 and Sprys was unchanged ( Figure 3—figure supplement 5 , note these transcripts are not listed in Supplementary files 3 or 4 because they did not meet criteria for differential expression ) . We conclude that despite the decrement in Shh pathway activity in posterior mesenchyme of Tbx3;PrxCre mutants , the level is sufficient to maintain ectodermal FGF signaling , consistent with preserved limb outgrowth . To understand the cause of the anterior PPD phenotype in Tbx3;PrxCre mutants , we pursued molecular mechanisms known to cause this defect in humans and mice: ectopic Hedgehog pathway activity ( Hill et al . , 2007; 2003; Lettice et al . , 2003 ) ; decreased Gli3R activity ( Hill et al . , 2009; Naruse et al . , 2010; Wang et al . , 2007a ) ; and abnormal composition or function of the primary cilia ( Goetz and Anderson , 2010 ) . RNA-Seq analysis of control versus mutant anterior compartments ( Supplementary file 4 ) showed no evidence of ectopic hedgehog activity in Tbx3;PrxCre mutants , and this was confirmed by in situ hybridization and qPCR for Shh and Ptch1 ( Figure 3B–C' , Figure 3—figure supplements 1 and 2 ) . Although Gli3 transcripts were increased in mutant limb buds ( Figure 3D' , F; Supplementary file 1 ) , targets of Gli3R transcriptional repression such as Zic3 , Epha3 , Hoxd13 ( McGlinn et al . , 2005; Vokes et al . , 2008 ) were overexpressed when assayed by microarray , RNA-Seq , qPCR and in situ hybridization ( Figure 3 G-J , Supplementary files 1 , 3 , 4 ) . The discrepancy between Gli3 RNA levels and increased expression of some repressor targets prompted examination of Gli3 protein levels . Gli3R constitutes the vast majority of Gli3 protein species in the anterior limb bud ( Wang et al . , 2000 ) and Figure 4A ) . Gli3R protein was markedly decreased ( 7 . 4 fold on this representative immunoblot ) with multiple bands of lower molecular weight than Gli3R present specifically in Tbx3;PrxCre mutant anterior mesenchyme ( mutant anterior compartment: MA , Figure 4A , red box ) . Gli3FL was virtually undetectable in mutant anterior mesenchyme ( Figure 4A’ ) . This finding is not due to poor sample quality because it was reproducible ( N=3 ) , no degradation was present in simultaneously prepared posterior compartment lysates ( mutant posterior , MP ) , and the β−tubulin control was intact . These findings indicate that Tbx3 is required for stability of Gli3FL and Gli3R proteins in the anterior limb mesenchyme , and are consistent with the PPD phenotype observed here and in other models of Gli3R deficiency ( Hill et al . , 2009; Naruse et al . , 2010; Wang et al . , 2007a ) . 10 . 7554/eLife . 07897 . 016Figure 4 . Loss of Tbx3 results in Gli3 protein instability and aberrant localization of Kif7 in limb bud cilia . ( A ) Representative immunoblot ( N=3 ) blot of E10 . 75 forelimb bud lysates prepared from microdissected Tbx3fl/+ control anterior limb ( CA ) , Tbx3;PrxCre mutant anterior limb ( MA ) , control posterior ( CP ) , and mutant posterior ( MP ) probed for Gli3 and βtubulin loading control . Note decreased level of Gli3FL and Gli3R , and multiple bands of lower molecular weight than Gli3R in MA sample . Densitometry of Gli3R bands in red box in N revealed that in this representative experiment , the level of Gli3R was 7 . 4 fold decreased in mutant anterior relative to control anterior . ( A’ ) Longer exposure of top of blot shown in panel A to examine Gli3FL band . The control ( CA ) Gli3FL band is 31 fold more intense than mutant ( MA , virtually undetectable ) . ( B ) Immunoblot of lysates from E10 . 5 forelimb buds immunoprecipitated ( IB ) with antibodies listed at top and immunoblotted ( IB ) for Tbx3 . Lane 5 shows that immunoprecipitation with anti-Kif7 antibody co-IPs Tbx3 . ( C ) As in panel B , but assayed for Kif7 . ( D ) Co-IP assay of Myc-tagged Tbx3 and Flag-tagged GFP overexpressed in HEK293 cells . IP was performed with antibodies listed at top and immunoblotted for Tbx3 . Myc-tagged Tbx3 co-IPs with Flag-tagged Kif7 . Input lane was 5 s exposure ( 5” exp ) while other lanes were 15 s ( 15” exp ) . ( E ) As in D , but in this case , blot probed for Kif7; confirms interaction of tagged , overexpressed proteins . ( F , G ) Representative images of anterior mesenchyme in sectioned forelimbs of control ( F , Tbx3fl/fl ) and mutant ( G , Tbx3fl/fl;PrxCre ) E10 . 5 embryos stained for the ciliary marker Arl13b ( red ) , Kif7 ( green ) and DAPI ( DNA , blue ) . White arrowheads highlight cilia with multiple punctae or streak of ciliary Kif7 immunoreactivity ( yellow ) indicating translocation of Kif7 within the cilia . Please also see Figure 5—source data 2 for z-stacks of additional Kif7 stained limb section . ( H ) Quantification of Kif7 staining pattern from multiple limb sections and three embryos of each genotype scored blinded to genotype . 10% fewer cilia have evidence of Kif7 translocation ( multiple punctae or streak of Kif7 immunoreactivity ) in Tbx3fl/fl;PrxCre mutants . N=1785 and 1792 cilia scored in controls and mutants , respectively . * p<0 . 001 There was no difference in the number of Kif7- cilia . Insets show digital zoom of cilia with representative pattern used for scoring . ( I ) Immunoblot assaying for Kif7 and b tubulin loading control in control anterior ( CA ) , mutant anterior ( MA ) , control posterior ( CP ) and mutant posterior ( MP ) e10 . 5 forelimb bud lysates . This is representative of four such experiments . ( J ) Western blot assaying for Sufu protein and b tubulin loading control in eE10 . 5 forelimb buds . Like Kif7 mutants , Tbx3;PrxCre mutants have increased Sufu protein . Sufu is increased 1 . 8 fold when normalized to loading control in this representative immunoblot; N=4 . K ) Western blot assaying for Spop protein and actin loading control in E10 . 5 forelimb buds . Spop is increased 1 . 5 fold when normalized to loading control in this representative immunoblot; N=3 . Both Sufu and Spop protein levels are increased in mutant forelimbs although their transcript levels are unchanged ( Figure 3—figure supplement 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01610 . 7554/eLife . 07897 . 017Figure 4—source data 1 . CZI file containing z-stack of E10 . 5 sectioned limb shown in Figure 4—figure supplement 1 . Kif7 is green , Arl13b red , DNA blue . Gli3 signal can be viewed if desired in the violet channel ( channel 2 ) . The entire z stack can be viewed using the free download of Zen software: http://www . zeiss . com/microscopy/en_de/downloads/zen . html . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01710 . 7554/eLife . 07897 . 018Figure 4—figure supplement 1 . Kif7 is also present in cytoplasm and nucleus . 100 X confocal maximum image projection of E10 . 5 sectioned limb stained for Kif7 ( green ) , Arl13b ( red ) and DNA ( blue ) . Please also see Figure 5—source data 2 for z-stack . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 01810 . 7554/eLife . 07897 . 019Figure 4—figure supplement 2 . Anterior mesenchymal limb cilia are bigger in Tbx3;PrxCre mutants compared to controls . ( A ) Distribution of cilia volumes in control and mutant limb buds . ( B ) Average cilia volumes and 95% confidence intervals . ( C ) Average surface area to volume ratios over the range of cilia volumes measured . ( D ) Calculated of surface area and volume of both WT and mutant cilia fit the equation: Surface=8 . 01 X Volume0 . 69DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 019 In an independent experiment to identify Tbx3 interacting partners , we performed Tbx3 co-immunoprecipitation ( co-IP ) on E10 . 5 mouse embryo lysates , followed by mass spectrometry ( Kumar et al . , 2014b ) . Surprisingly , Kif7 was among the Tbx3 interacting proteins identified . Kif7 is a ciliary protein that modulates activity of the Hedgehog pathway ( Hui and Angers , 2011 ) . It is required for proper formation of a 'cilium tip' compartment that regulates Gli function ( He et al . , 2014 ) , and for the regulated , partial proteolytic processing of GliFL to Gli3R ( Chen et al . , 2009; Cheung et al . , 2009; Endoh-Yamagami et al . , 2009; Law et al . , 2012; Ryan and Chiang , 2012 ) . As in Tbx3;PrxCre and Gli3+/- mutants ( Hill et al . , 2009 ) , human and mouse Kif7 mutants have PPD ( Cheung et al . 2009; Endoh-Yamagami et al . 2009; Putoux et al . 2011 ) . Immunoprecipitation of protein lysates from E10 . 5 forelimb buds showed that Tbx3 co-immunoprecipitates ( co-IPs ) with Kif7 , confirming interaction of Kif7 and Tbx3 in the developing limb ( Figure 4B; specificity and efficiency of Kif7 IP in this experiment is shown in Figure 4C ) . We next tested whether these proteins directly interact by overexpressing Flag-tagged Kif7 and Myc-tagged Tbx3 in HEK293 cells and immunoprecipitating for either Flag or Myc , followed by immunoblotting for Tbx3 ( Figure 4D ) or Kif7 ( Figure 4E ) . Both experiments confirmed direct interaction of the tagged proteins . The interaction of Tbx3 with Kif7 , and the shared PPD phenotypes of Kif7 and Tbx3;PrxCre mutants , suggest that Tbx3 may be required for normal Kif7 function in the anterior limb bud , and that loss of Tbx3 may disrupt Gli3 stability and processing in part via a Kif7-dependent mechanism . We examined Kif7 localization in E11 control and mutant forelimb buds , co-staining for the cilia marker Arl13b . Confocal fields spanning the anterior mesenchyme of controls and mutants were imaged and Figure 4F and G are representative 40X fields ( a higher magnification image and confocal z-stack , which also show Kif7 in the cytoplasm and nucleus are shown in Figure 4—figure supplement 1 and Figure 4—source data 1 ) . Blinded to genotype , fields were scored for ciliary Kif7 immunoreactivity as a single puncta , multiple punctae/streak , or none ( N=1785 and 1792 cilia scored in controls and mutants , respectively ) . In 16% of control cilia , Kif7 was detected in two punctae ( presumed base and tip ) or as a streak along the cilia , but this was only the case in 6% of mutant cilia ( Figure 4H , p<0 . 001 ) . There was no significant difference in the number of Kif7+ cilia rather , there were more single puncta cilia in mutants than controls ( Figure 4H , 84% vs 71% , p<0 . 001 ) . We did not detect any difference in the amount of Kif7 protein in control versus mutant limb bud compartments by western blot ( Figure 4I , representative of four4 separate experiments ) . Levels of Kif7 mRNA in the anterior limb mesenchyme were unaffected by loss of Tbx3 ( Figure 3—figure supplement 4 ) . There was a decrease in the transcripts posterior compartment but as shown , the protein level was unchanged . One feature of Kif7-/-mutants is excess Sufu due to increased protein stability ( Hsu et al . , 2011 ) . Transcript levels of Sufu were unchanged in mutant forelimbs ( Figure 3—figure supplement 4 , no difference was detected by microarray or RNA-Seq ) . However , increased amounts of Sufu ( and Spop ) protein were present in mutant limb buds ( Figure 4J , K; increased 1 . 5 fold in both cases ) . Humans and mice with Kif7 mutations have abnormally long cilia because Kif7 reduces the rate of microtubule growth in the ciliary axoneme ( He et al . , 2014; Putoux et al . , 2011 ) . With 3D images obtained from 100X confocal z-stacks of Arl13b stained limb sections , we used the 3D object counter from ImageJ to calculate the volume and surface area to volume ratio to derive the length of cilia . Consistent with aberrant , but not absent , Kif7 function , we found that while the range of cilia volumes detected were the same between mutants and controls , the distribution was not: mutants have an increased fraction of larger cilia and an average volume 18% greater ( Figure 4—figure supplement 2A , B; 475 and 575 cilia assayed in controls and mutants , respectively ) . Mutants and controls had superimposable surface area/volume ratios ( Figure 4—figure supplement 2C ) , indicating that the shape of cilia was not different thus , the derived length was 6% greater in mutants . Together , these findings indicate that loss of Tbx3 results in aberrant ciliary localization of Kif7 and are consistent with abnormal Kif7 function . Kif7 and other proteins required for Gli3 processing and function are present in , or translocate to , primary cilia in response to Hedgehog pathway activity ( Goetz and Anderson , 2010; Ryan and Chiang , 2012 ) . Dual immunostaining for Tbx3 and Arl13b on whole mount optically sectioned E10 . 5 forelimbs shows that Tbx3 is present in control limb anterior mesenchymal cilia ( Figure 5 A-E , Figure 5—source data 1 ) . Specificity of Tbx3 staining was confirmed by loss of signal in mesenchymal cilia of Tbx3;PrxCre mutant limbs ( Figure 5F-J , Figure 5—source data 2 ) . The digital image overlap calculator in Zen software showed that 18/50 anterior mesenchymal cilia were Tbx3+ ( 36% , Figure 5—figure supplement 1A , B ) . Of note , no epithelial cilia were Tbx3+ in control limb epithelium , providing an internal negative control for the signal in mesenchymal cilia . This same calculation in Tbx3;PrxCre mutant anterior mesenchyme showed only 2/54 ( <4% ) of mesenchymal cilia had background Tbx3 signal ( Figure 5—figure supplement 1 , C , D ) . These findings were reproduced with a commercially available anti-Tbx3 antibody ( Abcam ab99302 ) which also showed Tbx3 ciliary staining on control limb mesenchymal cilia ( 24/87 , 28% ) that was virtually absent in Tbx3 mutant limbs ( 2/52 , <4%; Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 07897 . 020Figure 5 . Tbx3 localizes to the primary cilia in limb mesenchyme . ( A–D , F–I ) Confocal 100X single Z-plane immunofluorescence images from optically sectioned E10 . 5 control ( top panels A–D ) and Tbx3;PrxCre ( F–I ) anterior limb buds after immunostaining with: Hoechst ( DNA , blue ) , C-terminal anti-Tbx3 antibody ( green , Frank et al . , 2013 ) , anti-Arl13b ( red , cilia ) . Arrowheads demarcate Tbx3 colocalization with cilia marker . Panels E and J are further digital zooms of white boxed cells in D and I . The entire z-stacks containing these planes are in Figure 5—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02010 . 7554/eLife . 07897 . 021Figure 5—source data 1 . Czi file of z-stack through the region of control anterior limb shown in Figure 5A–E . The entire z stack can be viewed using the free download of Zen software http://www . zeiss . com/microscopy/en_de/downloads/zen . html . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02110 . 7554/eLife . 07897 . 022Figure 5—source data 2 . Czi file of z-stack through region of mutant anterior limb shown in Figure 5F–J . Please view as described above . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02210 . 7554/eLife . 07897 . 023Figure 5—figure supplement 1 . Digital image overlap of Tbx3 and Arl13b in limb bud anterior mesenchyme . ( A ) Maximum image projection of Arl13b channel from control limb z-stack shown in Figure 5—source data 1 . Both mesenchymal and epithelial cilia are apparent in the maximum projection . ( B ) Calculated digital image overlap of Arl13b ( cilia ) and Tbx3 positive pixels in control limb bud . Note that all epithelial cilia in the stack are Tbx3 negative and of the 50 mesenchymal cilia , 18 ( 36% ) are Tbx3 positive . Please see Experimental Procedures for use of Zen and Image J software to calculate pixel overlap in separate channels . ( C ) Maximum image projection of Arl13b channel from control limb z-stack shown in Figure 5—source data 2 . Both mesenchymal and epithelial cilia are apparent in the maximum projection . ( D ) Calculated digital image overlap of Arl13b ( cilia ) and Tbx3 positive pixels in Tbx3;PrxCre limb bud . Note that all epithelial cilia in the stack are Tbx3 negative and of the 54 mesenchymal cilia , 2 ( 4% ) are Tbx3 positive consistent with low level of background antibody staining in mutant ( Figure 5 , panels G , I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02310 . 7554/eLife . 07897 . 024Figure 5—figure supplement 2 . Tbx3 immunoreactivity in limb cilia is also detected by a commercial anti-Tbx3 antibody against the N-terminus of Tbx3 . ( A ) Maximum image projection of Arl13b channel from control forelimb z-stack . ( B ) Calculated digital image overlap ( see Methods section ) of Arl13b ( cilia ) and Tbx3 positive pixels in control limb bud shown above using Abcam ( Abcam ab99302 ) anti-Tbx3 antibody to the N-terminus of mouse Tbx3 . 27/97 ( 28% ) of mesenchymal cilia are Tbx3+ . ( C ) Maximum image projection of Arl13b channel from Tbx3;PrxCre mutant forelimb z-stack . ( D ) Calculated digital image overlap of Arl13b ( cilia ) and Tbx3 positive pixels in Tbx3;PrxCre forelimb bud shown above . 2/52 mesenchymal cilia are Tbx3+ . ( E ) Scatter plot obtained using ImageJ comparing Tbx3 and Arl13b intensities from control and mutant anterior forelimb buds stained with Abcam anti-Tbx3 antibody shown in A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 024 Murine embryonic fibroblasts ( MEFs ) are a robust system for studying ciliary proteins ( Chen et al . , 2009; Dorn et al . , 2012; Liem et al . , 2012; Ocbina and Anderson , 2008; Rohatgi et al . , 2007 ) , so we used them to further explore Tbx3 localization and trafficking . In untreated wild type MEFs , our custom C-terminal antibody detected Tbx3 in a subset of cilia ( 30% , N=56; Figure 6A and B , a1-a5 , b1-b5; Figure 6—source data 1 , 2 ) . No Tbx3+ cilia were detected in Tbx3 null MEFS ( Figure 6C–F; Figure 6—source data 3 ) . Treatment of wild type MEFs with the smoothened agonist SAG increased the number of Tbx3+ cilia from 30% to 75% ( Figure 6G-J p<0 . 005 , Figure 6—source data 4 , Figure 6—figure supplement 1A ) . Lysates from control and SAG-treated MEFs showed no detectable difference in Tbx3 protein levels indicating the increased Tbx3 signal in cilia was due to trafficking rather than increased protein levels ( Figure 6—figure supplement 1B ) . The presence of Tbx3 in cilia and response to Hedgehog pathway activity were also detected with a commercially available anti-Tbx3 antibody with both SAG and Shh stimulation ( Figure 6—figure supplement 1C , D and Figure 6—source data 5 , 6 ) . In total , these data show that Tbx3 is present at baseline in cilia , and is trafficked to cilia in response to hedgehog signaling . 10 . 7554/eLife . 07897 . 025Figure 6 . Tbx3 is present in some cilia at baseline in Murine Embryonic Fibroblasts and trafficks to cilia in response to hedgehog pathway activation . ( A , B ) Confocal , 100X single z-plane immunofluorescence images from two different fields of wild type MEFS after immunostaining for: DAPI ( DNA , blue ) , Tbx3 ( green , c-terminal anti-Tbx3 antibody; Frank et al . , 2013 ) , Arl13b ( red , cilia ) . White boxed regions outline single cells that are shown at higher magnification in panels a1–a5 and b1–b5 . Please see Figure 6—source data 1 , 2 for z-stacks . a1–a4 , b1–b4 ) Single cells from white boxed areas in panels A and B . Individual cilia are shown in a5 and b5 . White arrowheads highlight Tbx3+ cilia . ( C–F ) Tbx3 null MEFs show loss of Tbx3 immunoreactivity in cilia and other cellular locations . Please see Figure 6—source data 3 for z-stack . ( G–J ) As in A and B , but MEFs were treated with smoothened agonist ( SAG ) . White arrowheads highlight Tbx3+ cilia . Please see Figure 6—source data 4 for z-stack . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02510 . 7554/eLife . 07897 . 026Figure 6—source data 1 . Czi file showing z-stack of wild type MEFs imaged in Figure 6 panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02610 . 7554/eLife . 07897 . 027Figure 6—source data 2 . Czi file showing z-stack of wild type MEFs imaged in Figure 6 panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02710 . 7554/eLife . 07897 . 028Figure 6—source data 3 . Czi file showing z-stack of Tbx3 null MEFs imaged in Figure 6 panel C–F . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02810 . 7554/eLife . 07897 . 029Figure 6—source data 4 . Czi file showing z-stack of SAG treated MEFs imaged in Figure 6 panel G–J . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 02910 . 7554/eLife . 07897 . 030Figure 6—source data 5 . Czi file showing z-stack of SAG treated MEFs imaged in Figure 6—figure supplement 1 panel C . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03010 . 7554/eLife . 07897 . 031Figure 6—source data 6 . Czi file showing z-stack of SHH treated MEFs imaged in Figure 6—figure supplement 1 panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03110 . 7554/eLife . 07897 . 032Figure 6—figure supplement 1 . Tbx3 immunoreactivity in cilia increases in response to Hedgehog pathway stimulation without an overall increase in Tbx3 protein levels . ( A ) Quantitation of Tbx3+ cilia in wild type MEFS -/+ SAG shows marked increase in Tbx3 immunoreactive cilia in response to SAG . B ) Western blot assaying Tbx3 and btubulin ( loading control ) protein levels in MEFs +/- SAG; the increase in number of Tbx3+ cilia occurs without an increase in amount of total Tbx3 protein . ( C , D ) Immunofluorescence images of SAG-treated ( C ) or SHH ( D ) MEFs assayed with a Santa Cruz commercial anti-Tbx3 antibody ( A20 ) raised against an internal Tbx3 epitope ( green ) confirm colocalization with cilia/Arl13b ( red ) . These merged images include DAPI in blue; white arrowheads highlight ciliary Tbx3 . Please see Figure 6—source data 5 , 6 for z-stacks . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 032 The presence of Tbx3 in cilia and the known association of Gli3 and Kif7 in primary cilia ( Endoh-Yamagami et al . , 2009 ) led us to test for interaction between endogenous Tbx3 and Gli3 . Co-immunoprecipitation of E10 . 5 forelimb lysates showed that Tbx3 co-IPs with endogenous Gli3 ( Figure 7A , and with Kif7 as shown previously ) , and that both Gli3FL and Gli3R are complexed with Tbx3 in the limb bud ( Figure 7B , lane 1 ) . These interactions are also detected in whole embryos ( Figure 7C; specificity/efficiency of Gli3 and Kif7 IPs are shown in Figure 7C’ and C” , respectively . Additional experiments showing this interaction are in Figure 7—figure supplement 1 ) . Overexpression of Flag-tagged Gli3 and Myc-tagged Tbx3 in HEK293 cells followed by co-immunoprecipitation did not indicate direct interaction in this cellular context ( Figure 7—figure supplement 2 ) , suggesting that their association occurs within a larger complex that includes Kif7 ( Figure 4 ) . 10 . 7554/eLife . 07897 . 033Figure 7 . Tbx3 interacts with Gli3 in the limb bud and trafficks with Gli3 in primary cilia . ( A , B ) Immunoprecipitation ( IP ) of E10 . 5 forelimb bud protein lysates with antibodies listed at the top of panel and immunoblotted ( IB ) to detect Tbx3 ( A ) or Gli3 ( B ) . Black arrowhead indicates IgG . Gli3FL and Gli3R ( red arrowheads in B ) both co-immunoprecipitate with Tbx3 . ( C ) Immunoprecipitation ( IP ) of E10 . 5 whole embryo protein lysates with antibodies listed at the top of panel and immunoblotted to detect Tbx3 . Tbx3 co-IPs with Gli3 . Specificity and efficiency of anti-Gi3 and anti-Kif7 antibodies in whole embryo lysates tested are shown in panels C’ and C” . Additional experiments demonstrating Tbx3/Gli3 interactions are in Figure 7—figure supplement 1 . Em , empty lane ( D–I’ ) Confocal 100X single Z-plane immunofluorescence images of vehicle ( DMSO ) treated MEFS after immunostaining for: DAPI ( DNA , blue ) , Tbx3 ( green , Frank et al . , 2013 ) , Gli3 ( red ) , Arl13b ( pink , cilia ) . Panel H is merged image of ( D–G ) . Panel I is 2 . 5X digital zoom of the boxed cell in panel H , and I’ shows the pink ( cilia ) channel pixel shifted to permit visualization of colocalized Tbx3 and Gli3 ( yellow ) within the cilia . White arrowheads highlight Gli3/Tbx3 colocalization . Please see Figure 7—source data 1 for z-stacks . ( J–O’ ) As above , but MEFS were treated with SAG in DMSO . Please see Figure 7—source data 2 for z-stack . ( P ) Quantitation of Tbx3+ and Gli3+ cilia in MEFS -/+ SAG . SAG treatment causes the majority of cilia to become Tbx3+ and these ciliary Tbx3 signals all colocalize with Gli3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03310 . 7554/eLife . 07897 . 034Figure 7—source data 1 . Czi file showing z-stack of wild type MEFs imaged in Figure 7 panel D-I’ . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03410 . 7554/eLife . 07897 . 035Figure 7—source data 2 . Czi file showing z-stack of SAG treated MEFs imaged in Figure 7 panel J-O’ . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03510 . 7554/eLife . 07897 . 036Figure 7—figure supplement 1 . Tbx3 and Gli3 coimmunoprecipitate in whole embryo protein lysates . ( A , B ) Immunoprecipitation ( IP ) of E10 . 5 whole embryo protein lysates with antibodies listed at top of panels and immunoblotted ( IB ) to detect Tbx3 ( A ) or Gli3 ( B ) . Black arrowhead indicates IgG . Gli3FL and Gli3R ( red arrowheads in B ) both co-immunoprecipitate with Tbx3 . Em , empty laneDOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03610 . 7554/eLife . 07897 . 037Figure 7—figure supplement 2 . Tagged Tbx3 does not co-IP with tagged Gli3 in HEK293 cells . Co-IP assay of Myc-tagged Tbx3 and Flag-tagged Gli3 overexpressed in HEK293 cells . IP was performed with antibodies listed at top and immunoblotted for Gli3 . Myc-tagged Tbx3 does not interact with tagged Gli3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03710 . 7554/eLife . 07897 . 038Figure 7—figure supplement 3 . Tbx3 does not co-IP with Sufu or Spop in mouse embryo lysates . ( A , B ) Immunoprecipitations/Immunoblot assaying for interaction between endogenous Tbx3 and Sufu in E10 . 5 mouse embryo lysates . Sufu did not co-IP Tbx3 ( A ) nor did Tbx3 co-IP Sufu ( B ) . ( C ) Immunoprecipitations/Immunoblot assaying for interaction between Tbx3 and Spop in control and Tbx3Δfl/Δfl E10 . 5 mouse embryo lysates . No interaction was detected . Note increased Spop in mutant , consistent with previous result in Figure 4J and data in Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 038 Immunofluorescence assay of untreated MEFs for Tbx3 , Gli3 , and Arl13b by triple immunocytochemistry showed that 90% of cilia were Gli3+; Tbx3 colocalized with 66% of the Gli3 signals ( N=38 total cilia scored , Figure 7D–I’ Figure 7—source data 1 ) . Treatment with SAG increased the fraction of Gli3+ cilia to >95% and all but two of those Gli3 signals colocalized with Tbx3 ( N=34 , Figure 7J–O’ , Figure 7—source data 2 ) . These results are quantified in Figure 7P . To obtain further mechanistic understanding into how loss of Tbx3 results in decreased Gli3 proteins , we assayed levels and interactions of members of the Gli3 processing machinery . Gli3FL stability and partial processing versus complete degradation are regulated by opposing actions of Suppressor of fused ( Sufu ) and speckle-type POZ protein ( Spop ) , respectively ( Wang et al . , 2010; Wen et al . , 2010 ) . In the absence of Smoothened activation , mammalian Sufu recruits GSK3β to phosphorylate Gli3FL downstream of PKA , allowing for its partial processing to Gli3R ( which requires Kif7 and intact cilia ) ( Kise et al . , 2009; Tempe et al . , 2006; Wang et al . , 2000 ) . Spop is an adaptor for Cullin3-based E3 ubiquitin ligase that drives complete degradation of Gli3FL in the absence of Sufu , but facilitates its processing to Gli3R when Sufu is present ( Humke et al . , 2010; Wang et al . , 2010; Zhang et al . , 2006 ) . In contrast to Kif7 and Gli3 , we did not detect interactions between Tbx3 and either Sufu or Spop ( Figure 7—figure supplement 3 ) . The observation that Gli3FL is virtually undetectable in Tbx3;PrxCre anterior mesenchyme ( Figure 4A' ) indicates that despite the increased amount of Sufu ( Figure 4H , Figure 8—figure supplement 1 ) , it fails to prevent degradation of Gli3FL in the absence of Tbx3 . We next tested whether decreased levels of Gli3 proteins in the anterior mesenchyme reflected perturbed interactions between Sufu , Gli3 and Kif7 . Limitations in sample quantity made it unfeasible to perform multiple co-IPs on isolated anterior forelimb buds , so we assayed lysates from control and Tbx3Δfl/Δfl E10 . 5 embryos . The altered levels of protein expression seen in mutant limb buds were recapitulated in whole embryos ( Figure 8—figure supplement 1 ) . As seen in mutant limb buds , the amount of both Gli3FL and Gli3R protein is reduced in Tbx3Δfl/Δfl embryos ( Figure 8A , lanes 1–4; Figure 8—figure supplement 1; Figure 8—figure supplement 2A and B ) . Notably , the interaction between Gli3 and Sufu is reproducibly decreased in excess of the decrement in Gli3 proteins; this is evident by quantitating and comparing the ratio of Gli3 proteins detected in control versus mutant to that complexed with Sufu . For example , in the representative experiment shown in Figure 8A , the Gli3 FL band intensity ratio is ~1 . 6 fold greater in controls ( Figure 8A , lanes 1 and 3 ) than in mutants ( Figure 8A , lanes 2 , 4 ) , while the amount of Gli3FL protein that was immunoprecipitated with Sufu is 4 . 6 fold greater in controls than mutants ( Figure 8A , lane 9 versus 10 ) . Mean band intensity ratios of 3 replicate experiments are shown in Figure 8A’ . The decreased Gli3/Sufu interaction in the absence of Tbx3 was also evident when assayed in the opposite direction , that is , by IP of Gli3 and immunoblotting for Sufu ( Figure 8B , panel B’ shows relative band intensities for the experiment shown ) . Sufu/Gli3R interactions were also affected ( Figure 8A’; Figure 8—figure supplement 2A–B ) . These findings indicate that normal stoichiometry of the interaction of Gli3 proteins with Sufu requires Tbx3 . 10 . 7554/eLife . 07897 . 039Figure 8 . Altered stoichiometry of interactions between Gli3 and members of its processing/degradation complex . ( A ) Anti-Gli3 immunoblot ( IB ) on immunoprecipitates ( IP ) from antibodies listed at top on lysates from E10 . 5 control ( wt ) and Tbx3Δfl/Δfl ( ko ) embryos . Gli3FL and Gli3R are denoted by red arrowheads , IgG heavy chain with black arrowhead . Note decreased levels of IP’d Gli3FL and Gli3R in mutants compared with controls; the IPs in lanes 1–2 and 3–4 are two independent biologic replicates and the band intensity ratio of control to mutant for both GliFL and Gli3R was ~1 . 6 . The interaction between Gli3 and Sufu is decreased more than can be explained by the overall decrement in Gli3 protein levels: in this representative experiment Sufu co-IPs 4 . 6X more Gli3FL in controls than in mutants ( lane 9 versus 10 ) . ( A’ ) Bar graphs show the results of quantitation of band intensities from three3 replicate experiments measured with densitometry and presented as the ratio of signal detected in controls relative to mutants . ( B ) As in A but immunoblot probed for Sufu . Comparison of lanes 1 and 2 confirms decreased interaction between Gli3 and Sufu , despite preserved levels of Sufu in the mutants ( lane 10 ) . ( C ) Anti-Gli3 immunoblot with IPs as listed at top . Note increased interaction between Kif7 and Gli3 in mutants ( lane 4 ) , despite overall decreased level of Gli3 ( lane 8 ) . ( C’ ) Quantitation of band intensities from three replicate experiments measured with densitometry and presented as a ratio of signal detected in control relative to mutant . Even though there is less total Gli3 protein , since there is increased interaction between Gli3 and Kif7 in mutants , the Kif7 co-IP control to mutant ratios are <1 . ( D ) Anti-Kif7 immunoblot with IPs as listed at top . Confirms increased interaction between Gli3 and Kif7 in mutants . ( D’ ) Quantitation of experiment in D . ( E ) Anti-Kif7 immunoblot with IPs as listed at top . The interaction of Kif7 and Sufu is decreased in the absence of Tbx3 ( lane 4 ) despite preserved levels of both proteins ( lane 6; Figure 8 , panels B , D and F; Figure 8—figure supplements 1 and 2 ) . ( E’ ) Quantitation of band intensities from two replicate experiments measured with densitometry and presented as ratio of signal detected in control relative to mutant . There is 2 fold less interaction between Kif7 and Sufu in mutants . ( F ) Anti-Sufu immunoblot with IPs as listed at top . The interaction of Kif7 and Sufu is decreased in the absence of Tbx3 ( lane 3 versus 4 ) . ( F’ ) Quantitation of findings in F . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 03910 . 7554/eLife . 07897 . 040Figure 8—figure supplement 1 . Altered protein levels observed in mutant limb buds are also apparent in whole embryos . ( A , B ) Immunoblots on protein lysates prepared from E10 . 5 forelimb buds ( A ) and whole embryos ( B ) . Actin is loading control . ( C ) Quantitation of protein levels in A and B comparing amount detected in control to mutant . Note increased levels of Sufu and Spop in mutant limbs and embryos that results in ratios of control to mutant <1 . nd , not detectedDOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 04010 . 7554/eLife . 07897 . 041Figure 8—figure supplement 2 . Replicate experiments confirming altered stoichiometry of interactions between Gli3 and members of its processing complex in Tbx3 mutants . ( A ) Anti-Gli3 immunoblot ( IB ) on immunoprecipitates ( IP ) from antibodies listed at top on lysates from E10 . 5 control ( wt ) and Tbx3Δfl/Δfl ( ko ) embryos . Gli3FL and Gli3R are denoted by red arrowheads , IgG heavy chain by black arrowhead . Note decreased levels of IP’d Gli3R in mutants ( lane 4 ) compared with control ( lane 5 ) . ( A’ ) Quantitation of IPd proteins detected in A . The interaction between Gli3R and Sufu is decreased more than can be explained by the overall decrement in Gli3R protein levels: in this experiment , >9 fold more Gli3R co-IPs with Sufu in controls than mutants ( lanes 3 versus 4 ) , whereas the decrement in Gli3R is only 3 . 8 fold ( lanes 5 versus 6 ) . ( B , B’ ) Additional replicate co-IP experiment confirming decreased interaction between Gli3R and Sufu ( lanes 7 and 8 ) in excess of decrement in overall Gli3R level ( lanes 5 and 6 ) . ( C , C’ ) Anti-Gli3 immunoblot ( IB ) and quantitation of IPs from antibodies listed at top on lysates from E10 . 5 control ( wt ) and Tbx3Δfl/Δfl ( ko ) embryos . 3 . 3 fold more Gli3R co-IPs with Kif7 in mutants ( lane 2 ) than in controls ( lane 1 ) , despite an overall 2 . 9 fold decrease in Gli3R levels ( lanes 5 versus 6 ) . ( D , D’ ) Anti-Kif7 immunoblot ( IB ) and quantitation of IPs from antibodies listed at top on lysates from E10 . 5 control ( wt ) and Tbx3Δfl/Δfl ( ko ) embryos . In this example , the amount of Kif7 that co-IPs with Sufu is decreased in 2 . 4 fold in Tbx3 mutants , consistent with previous results shown in Figure 8 panels E and F . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 041 In wild type embryos , we detected only trace interaction between Kif7 and Gli3R assaying by co-IP in either direction ( Figure 8C , lane 3; Figure 8D , lane 7; Figure 8—figure supplement 2C , lane1 ) . However , there was a robust and reproducible interaction in mutants despite the decrease in the total amount of Gli3 proteins . In the representative experiment shown in Figure 8C , in which we IP’d for Kif7 and assayed for Gli3 , the band intensity ratios of control ( lane 3 ) to mutant ( lane 4 ) were 0 . 1 for Gli3FL and 0 . 25 for Gli3R . Quantification of the band intensity ratios from three replicate experiments in which we IP’d for Kif7 and assayed for Gli3 is shown in Figure 8C’ . Furthermore , the increased interaction of Kif7 with Gli3 was confirmed by IP of Gli3 and assay for Kif7 , as shown in Figure 8D: the band intensity ratio of control ( lane 7 ) to mutant ( lane 8 ) was 0 . 1 , indicating a marked increase in interaction in mutants ( Figure 8D’ graphs relative band intensities in experiment 8D ) . Lastly , the interaction between Sufu and Kif7 was reproducibly decreased when assayed by co-IP in either direction ( Figure 8E , lane 3 versus 4; Figure 8F , lane 3 versus 4; Figure 8—figure supplement 2D , lane 7 versus 8 ) . Quantification of the band intensity ratios from replicate experiments in which we IP’d for Sufu and assayed for Kif7 is shown in Figure 8C’ . Figure 8F’ graphs relative band intensities in experiment 8F in which we IP’d for Kif7 and assayed for Sufu . Note that this decrease in Sufu/Kif7 interaction occurs despite increased and normal levels of these proteins , respectively . In total , these data indicate that Tbx3 is required for normal stoichiometry and function of the Sufu/Kif7 complex that stabilizes and processes Gli3 as modeled in Figure 9 . 10 . 7554/eLife . 07897 . 042Figure 9 . Model of compartment specific functions of Tbx3 in forelimb bud mesenchyme and altered interactions and stoichiometry of the Kif7/Sufu Gli3 processing complex in Tbx3;PrxCre mutants . In posterior forelimb mesenchyme , Tbx3 is required for normal levels of Hand2 upstream of Shh . Shh pathway activity and other Tbx3-reponsive factors promote digit 5 formation . In the absence of Tbx3 , there is decreased expression of Hand2 and Shh and other digit 5 promoting pathways . In anterior mesenchyme , Tbx3 is in a complex with Gli3 proteins , Kif7 and Sufu and required for the stability of Gli3 FL and Gli3R . In the absence of Tbx3 , Sufu and Spop protein levels are increased yet there is decreased interaction between Sufu and Kif7 , and Sufu and Gli3 . In mutant anterior mesenchyme , Gli3FL is barely detected: it is either degraded or converted to Gli3R . Levels of Gli3R are abnormally low due to a combination of decreased amount of Gli3FL precursor and its processing to Gli3R , and excess degradation . These findings are consistent with decreased function of cilia and Kif7 ( required for processing of Gli3FL to Gli3R ) , and of Sufu ( required for stability of both Gli3FL and Gli3R ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07897 . 042 Tbx3 expression in the LPM is required for normal Tbx5 expression ( Figure 2A–C ) implicating Tbx3 as one of the earliest limb initiation factors . This is consistent with our finding that Tbx3+ progenitors in the LPM give rise to the majority of limb bud mesenchyme . Determining the fate of these progenitors in the future will help reveal the mechanism for loss of ulna/fibula and digits 2–5 in Tbx3 null embryos . Intrinsic differences in anatomically bilaterally symmetric structures have long been suspected: acetazolamide teratogenizes only the right limb in rats ( Layton and Hallesy , 1965; Wilson et al . , 1968 ) and nitroheterocyclics such as valproate also induce unilateral defects ( Coakley and Brown , 1986; Fantel et al . , 1986 ) . Directional asymmetry in limb size in human fetuses was recently reported ( Van Dongen et al . , 2014 ) . Left/right identity differences in bilaterally symmetric structures such as the limbs may be upstream of secondary patterning events such as dorsal/ventral axis appropriate to body side . Shiratori and colleagues reported asymmetric expression of Pitx2c in the developing mouse limb and postulated functional left/right differences ( Shiratori et al . , 2014 ) . Our finding that the left limb is more sensitive to Tbx3 than the right is consistent with this hypothesis and warrants further investigation in humans with UMS and other mouse models . A model of molecular mechanisms contributing to the different anterior and posterior limb phenotypes of Tbx3;PrxCre mutants is shown in Figure 9 . Digit 5 formation is exquisitely sensitive to Shh activity and Grem1 ( Harfe et al . , 2004; Scherz et al . , 2007; Zhu and Mackem , 2011; Zhu et al . , 2008 ) , so the altered expression of these genes in posterior mutant mesenchyme helps to explain loss of this digit . Shh null heterozygotes ( Shh+/- ) have normal digit number , suggesting that the loss of digit 5 in Tbx3;PrxCre mutants is not solely due to decreased Shh expression and pathway activity in posterior mesenchyme . Importantly , it is not known if compensatory mechanisms preserve normal Shh protein levels or activity in Shh+/- mutants to support normal limb development . Indeed , a 'buffering system' to modulate polarizing activity by Shh was proposed in the chick limb ( Sanz-Ezquerro and Tickle , 2000 ) . We found evidence of dysfunction of other digit 5 promoting pathways in posterior mesenchyme of Tbx3;PrxCre mutants . Compound Hand2/Gli3 null mutants have more severe polydactyly than Gli3 mutants ( Galli et al . , 2010 ) , indicating that the role of Hand2 in digit formation is not limited to regulation of Shh expression . Aberrant expression of other BHLH factors ( Supplementary files 1 , 3: Hand1 , Bhlhe40 , Bhlhe41 , Bhlha15 ) may also contribute to the phenotype because the stoichiometry and interactions of BHLH factors have complex roles in limb development ( Firulli et al . , 2005 ) . Altered levels of BMP responsive targets ( Dkk1 , Noggin , Grem1 , Grem2 ) in Tbx3;PrxCre mutants suggest disrupted BMP signaling . Overexpression of Gata6 decreases Shh and Grem1 expression and causes loss of posterior digits ( Kozhemyakina et al . , 2014 ) ; we detected increased Gata6 expression in the posterior mesenchyme of Tbx3;PrxCre mutants ( Supplementary file 3 ) . Additional studies are needed to determine if Tbx3 transcriptional or post-transcriptional functions directly regulate these pathways . Although Tbx3 is also present in posterior mesenchymal cilia , we did not detect any evidence of altered stability or processing of Gli3 in the posterior mesenchyme ( Figure 4A , representative of three replicates ) . Nonetheless , it is possible that decreased Shh signaling in the posterior of Tbx3;PrxCre mutant limb buds creates changes in Gli3 levels/ratios below our ability to detect . If so , decreased Shh activity would be predicted to increase Gli3R , decreasing digit number . It is notable that decreased expression of Tbx2 observed in Tbx3;PrxCre mutants would be predicted to result in increased Grem1 expression ( Farin et al . , 2013 ) however , Grem1 expression is decreased ( Figure 3 ) . Decreased Shh signaling and Grem1 expression in the posterior mesenchyme would both be predicted to result in decreased Fgf4 and Fgf8 expression in the posterior AER ( Khokha et al . , 2003; Michos et al . , 2004 ) . Rather , the finding that these transcripts are increased in the posterior ( Figure 3—figure supplement 4 ) indicates that loss of Tbx3 in posterior mesenchyme disrupts the Shh-Grem1-FGF signaling loop . In wild type anterior mesenchyme , our data support the model that Tbx3 is part of a Kif7/Sufu complex that drives processing of the majority of Gli3FL to Gli3R and also prevents complete degradation of Gli3FL by Spop ( Wang et al . , 2010 ) . Note that our model and data show a marked decrease in Gli3FL ( virtually undetectable by western blot in anterior mesenchyme ) , but residual Gli3R in mutants . This is consistent with the phenotype of Gli3 deficiency , and digit 1 polysyndactyly is also seen with decreased function of Kif7 or Sufu . Residual Gli3R in the mutants is sufficient to prevent the extreme polydactyly seen in complete absence of Gli3 , Kif7 or Sufu . In Tbx3;PrxCre mutant limb buds , Kif7 and Sufu proteins are present at normal and increased levels , respectively , but their interaction with each other , and that of Sufu with Gli3 , are decreased . Because Spop facilitates processing of Gli3FL to Gli3R in the presence of Sufu , increased levels of Spop in mutants drives complete degradation of any Gli3FL not converted to Gli3R , resulting in undetectable levels of Gli3FL in anterior mesenchyme . The decreased levels of Gli3R in the anterior likely result from a combination of decreased levels of Gli3FL precursor , inefficient processing due to altered Kif7/Sufu function , and excess Gli3R degradation , as evident by the lower molecular weight species present in Figure 3A . This synthesis is consistent with the large body of published data indicating that anterior digit number is tightly regulated by the balance of Gli3A and Gli3R , in turn controlled by Kif7 and Sufu ( Cao et al . , 2013; Wang et al . , 2000; 2007a; 2007b; Zhulyn et al . , 2014 ) . Kif7-/- and other ciliary mutants maintain robust levels of Gli3FL but have a marked decrease in processing it to Gli3R and their PPD phenotypes are consistent with decreased Gli3R ( Cheung et al . , 2009; Endoh-Yamagami et al . , 2009 ) . Furthermore , recent studies from Chi Chung Hui’s group demonstrate that Kif7;PrxCre mutants have anterior PPD and that increasing Gli3R or decreasing Gli activators rescues all but digit 1 duplication ( Zhulyn and Hui , 2015 ) . Tbx3 mutant limbs also display evidence of Sufu dysfunction with markedly decreased levels of Gli3R and absent Gli3FL . Levels of Gli2 , Gli3FL and Gli3R are all drastically reduced in Sufu mutants , even in the absence of cilia , consistent with the fact that Spop is not a ciliary protein and drives degradation of the full length proteins in the absence of Sufu ( Chen et al . , 2009; Jia et al . , 2009; Svard et al . , 2006; Wang et al . , 2010 ) . However , βTrCP and Spop can only mediate processing of Gli3FL to Gli3R in the presence of both Sufu and Kif7/intact cilia/intraflagellar transport ( Endoh-Yamagami et al . , 2009; Law et al . , 2012; Liu et al . , 2005; Wang et al . , 2010; Wen et al . , 2010 ) . This processing is believed to occur at the basal body/centrosome ( Ryan and Chiang , 2012; Wang et al . , 2013; Wen et al . , 2010; Wigley et al . , 1999 ) . Thus , our data strongly support that Tbx3 functions at the cilia or basal body , as a part of the complex that regulates Gli3 processing and stability ( Ryan and Chiang , 2012; Wen et al . , 2010 ) . Loss of Tbx3 could also influence stability and processing of Gli2 . The anterior PPD phenotype of Gli3 heterozygotes is slightly more severe in a Gli2 null background ( Bowers et al . , 2012; Mo et al . , 1997 ) . Gli2 is also expressed in the posterior mesenchyme where it regulates digit patterning but not digit number ( Bowers et al . , 2012 ) . Bowers’ study also demonstrated that the role of Gli activators is in AP patterning of the posterior limb , whereas Gli repressors regulate digit number and anterior limb AP patterning . This is consistent with our findings that Tbx3 affects anterior digit number by regulating Gli3 repressor stability in the anterior mesenchyme . The functional significance of increased interaction between Kif7 and Gli3R that we detect in mutants requires additional study nonetheless , since the anterior phenotype of Tbx3;PrxCre mutants is one of Gli3R deficiency rather than absence , we conclude that either there is a pool of Gli3R unbound by Kif7 and/or Gli3R complexed with Kif7 still has repressor function . In addition to its function as a Gli3R co-repressor , Zic3 also regulates cilia morphogenesis and function ( Sutherland et al . , 2013 ) . Since decreasing Zic3 levels in Gli3+/- mutants rescues their preaxial polydactyly ( Quinn et al . , 2012 ) , Zic3 overexpression in Tbx3;PrxCre mutants may contribute to the PPD phenotype . Our data exclude ectopic Shh pathway activity as a cause of the PPD in Tbx3;PrxCre mutants: using multiple methods at multiple stages , we did not detect anterior Shh , Gli1 , or Ptch1 expression in the anterior mesenchyme . Notably , expression of the BMP antagonist Grem1 is normally increased in PPD associated with ectopic anterior Shh pathway activity ( Lopez-Rios et al . , 2012; Zhang et al . , 2009 ) , but in Tbx3 mutants , Grem1 expression is markedly decreased . This is additional evidence in support of unique Tbx3-dependent pathways regulating digit number . Experiments were conducted in strict compliance with IACUC/AALAC standards . The Tbx3flallele was detailed in Frank et al . ( 2013 ) . Prx1Cre , RARCre and Rosa26LacZ were previously reported ( Soriano , 1999; Moon and Capecchi , 2000; Logan et al . , 2002 ) . Generation of the Fgf8mcm and Tbx3mcm alleles will be described elsewhere . Males bearing Fgf8MCM or Tbx3MCMalleles were crossed with Rosa26LacZ/+ females . Females were gavaged with tamoxifen ( 10mg/gm body weight ) at stages stated in text . β−galactosidase activity was assayed using established protocols ( Park et al . , 2006 ) . E15 . 5 fetuses were fixed in 4% PFA overnight , rinsed in water for 2 days and alcian blue stained for 30 hr , then cleared in BABB . Older specimens were processed as in Moon et al . ( 2000 ) . Digoxigenin-labeled riboprobes were generated according to manufacturer’s instructions ( Roche ) . Embryos were processed using a standard protocol ( Park et al . , 2006 ) . Total RNA preparation , cDNA generation and qPCR were carried out as described in Yu et al . ( 2010 ) . Primer sequences are provided in Supplementary file 5 . All qPCRs were performed on a minimum of three biologic replicates of pooled forelimb buds ( Figure 3 ) or anterior and posterior forelimb bud segments ( Figure 3—figure supplement 4 ) . Total RNA was prepared from three pools of dissected E10 . 25 control ( Tbx3 fl/+ ) and Tbx3;PrxCre mutant forelimbs using the RNAeasy Micro Kit ( Qiagen 74004 ) . The microarray and genomic analysis and bioinformatics core facilities at the University of Utah performed Agilent mouse whole-genome expression arrays and array image data analysis using Agilent Feature Extraction software . Subtle intensity-dependent bias was corrected with LOWESS normalization , with no background subtraction . Statistical analysis of normalized log-transformed data was performed in GeneSifter ( www . genesifter . net ) . Differentially expressed transcripts were defined ( adjusted for multiple testing using the Benjamini and Hochberg method ) as p<0 . 05 . The results presented in Supplementary file 1 show transcripts that were statistically differentially expressed +/-1 . 3 fold in the mutant limb buds; yellow highlighting indicates changes that were replicated by RNA-Seq . Total RNA was isolated from pools of dissected anterior and posterior regions of E11 control ( Tbx3fl/+ ) and Tbx3;PrxCre forelimbs using the RNAeasy Micro Kit ( Qiagen ) . Each pool contained 12 forelimbs and two biologic duplicate pools were assayed . cDNA was generated , sequenced , and raw sequence reads were processed as described in Kumar et al . ( 2014b ) . Supplementary file 2 contains transcripts that are differentially expressed +/-1 . 3 fold ( +/-0 . 38 in log base 2 , column L ) in control anterior ( CA ) compared to control posterior ( CP ) limb segments . The transcripts listed in Supplementary file 3 were the result of mining the data to detect differential expression +/-1 . 3 fold ( +/- 0 . 38 in log base 2 , column L ) in control posterior ( CP ) relative to mutant posterior ( MP ) segments . Supplementary file 4 shows the result of mining the data to detect differential expression +/-1 . 3 fold ( +/-0 . 38 in log base 2 , column L ) in control anterior ( CA ) relative to mutant anterior ( MA ) limb segments . Note that transcripts that were not differentially expressed +/-1 . 3 fold are not listed on the tables . The complete unmined dataset is available on GEO . Protein lysates were prepared from E10 . 5 dissected limb buds or embryos or cultured MEFs using Dignam buffer . 50 ug of total protein were then subjected to SDS-PAGE analysis followed by immunoblotting according to standard protocols . Primary antibodies: Sufu ( #2522 , Cell signaling ) , Spop ( #PA5-28522 ) , Myc ( #SC-789 , Santa Cruz ) , Flag ( #F7425 , Sigma ) , E2F1 ( #137415 , Abcam ) , Ubiquitin ( #7780 , Abcam ) , Gli3 ( AF3690 , R and D systems ) , Tbx3 C-terminal antibody ( Frank et al . , 2013 ) ; Kif7 ( ab 95884 , Abcam ) ; β tubulin ( Santa Cruz ) . Immunoprecipitations were performed as described in Kumar et al . , 2014a and 2014b . Briefly , protein lysates were prepared from limb buds or embryos at E10 . 5 or transfected HEK293 cells using Dignam buffer C . Cleared protein lysates were obtained by centrifugation at 12 , 000g for 10 min . Equal amounts of protein lysates were incubated with 5–10μg of respective antibodies over night at 4°C with gentle shaking . Immune complexes were isolated with Protein-G Dynal beads and washed three times with Dignam buffer C . Precipitates were eluted from the beads by boiling in SDS-loading dye for 10 min , and analysed by western blotting by standard procedures using indicated antibodies at a dilution of 1:1000 . Immunoblot signals were quantified by densitometry using ImageJ64 software as per the procedure described by Luke Miller ( http://www . lukemiller . org/ImageJ_gel_analysis . pdf ) . Sections: Samples were fixed in 4% PFA at 4°C for 2 hr then washed with 0 . 3% Triton-X100 ( in PBST ) . Heat retrieval in citrate buffer ( Vector Laboratories ) was performed for 2 min in a pressure cooker . Slides were washed with PBST and incubated in PBST with 5% serum corresponding to the secondary antibody origin for 1 hr . Slides were incubated with primary antibody in PBST 5% serum overnight at 4°C . Primary antibodies and dilutions: Tbx3 C-terminal antibody 1/200 ( Frank et al . , 2013 ) , Tbx3 N- terminal antibody 1/100 ( Abcam ab99302 ) , Tbx3 internal epitope antibody ( Santa Cruz A-20 ) , Arl13b 1/100 ( USDavis/NIH NeuroMab Clone N259B/66 ) , Gli3 1/100 ( R&D systems AF3690 ) , pHH3 ( Ser10 ) , Millipore 06–570 ( 1:2000 ) ; ( 1:50 ) ; Kif7 ( 1:200 , kind gift from Dr . C-c Hui ) . After washing in PBST for 15 min , slides were incubated with secondary antibody from either Invitrogen or Jackson Immunoresearch diluted 1/1000 in PBST 2% BSA with Hoechst 33 , 342 at 1ug/ml for 1h at room temperature . Final wash was with PBST for 15 min and mounted using Fluoromount-G from Southern Biotech . Secondary antibodies: Donkey anti-rabbit Alexa 488; Donkey anti-goat Alexa 488; Donkey anti-mouse Alexa 647; Donkey anti-goat Alexa 594 . Whole forelimb buds: samples were fixed in 4% PFA for 1 hr at room temperature with PBST . Heat retrieval in citrate buffer ( Vector Laboratories ) was performed for 5 min at 100°C followed by washing with PBST . Limb buds were incubated in PBST with 5% serum that correspond to the secondary antibody origin ( Goat or Donkey ) for 2 hr . Limb buds were incubated with primary antibody in PBST with 5% serum overnight at 4°C , then washed with PBST once and incubated in PBST with 2% BSA for 4 hr . Incubated with secondary antibody and Hoechst as above with overnight at 4°C . Prior to imaging , samples were washed with PBST for 4 hrs and incubated in PBS/50% glycerol overnight . TUNEL: was performed as described in ( Park et al . , 2006 ) Imaging was performed using a Zeiss confocal microscope LSM710 with Zen black imaging software http://www . zeiss . com/microscopy/en_us/downloads/zen . html . 100x objective was used . To generate the Arl13b/Tbx3 colocalization maps , we used Zen to define pixels in each plane of z that exceeded an arbitrary threshold of 0 . 1 relative intensity in both Arl13b and Tbx3 channels using the imaging calculation subtab in the image processing menu . This was followed by calculation of the maximum intensity projection . To quantitate Tbx3 positive cilia , the 3D object counter in ImageJ ( Bolte and Cordelieres , 2006 ) was used employing the “redirection” option to superimpose Arl13b+ objects into the Tbx3 channel . Cilia volume was assayed by quantitating Arl13b signal on 100x images of sectioned forelimbs from four pairs of mutant and control embryos stained for Arl13b and analyzing resulting in ImageJ . Imaged were Gaussian blurred using radius equal to '1' and further 3D object counter was used to measure volumes and surface area with threshold of 900 out of 4000 range . Distributions and parameters of the volume distributions shown in Figure 4—figure supplement 2A and B were calculated in Excel . We calculated that the surface area and volume of both WT and mutant cilia tightly fit the equation: Surface=8 . 01 X Volume0 . 69 ( Figure 4—figure supplement 2D ) , which indicates that the cilia shapes are the same in the mutants and controls , and change size proportionally . In this case , the relative change in volume between mutant and control ( Volume mutant ) / ( Volume control ) =1 . 1747 , which correspond to a change in length of 6% . MEFs from E10 . 5 embryos were plated on fibronectin coated Matteks . MEFs were cultured and processed as in Kumar et al . ( 2014a ) ; cilia outgrowth was stimulated by culture in 0 . 5% FBS for 24 hr followed by incubation with 100 nM SAG or 100 nM SHH ( Phoenix Pharmaceuticals ) for 24 hr . Cells were washed 2x with PBS and fixed in 4% PFA for 20 min on ice , then washed 2X with PBS and permeabilized with 0 . 2% Triton X-100 in PBS for 10 min . After blocking in 5% donkey serum + 3% BSA for 1 hr , cells were incubated with anti-TBX3 ( custom C-terminal or Santa Cruz sc-17871 ) and anti-Arl13b ( 1:50 NeuroMAb ) at room temperature for 2 hr . Cells were washed 5 X 5 min in PBS and then incubated in secondary donkey anti-goat-488 or anti-rabbit 488 and donkey antimouse -594 ( Jackson Immunoresearch ) . After washing 3 x 5min in PBS , cells were incubated in DAPI for 20 min . Cells were then washed 3X with PBS and mounted in Dabco for imaging . Images were captured on a Zeiss LSM-710 confocal microscope and processed using Zen software as described above . HEK-293 cells were grown in DMEM supplemented with 10% fetal bovine serum ( FBS ) and penicillin/streptomycin . Cells were maintained in 5% C02 incubator at 37°C . Myc- tagged full length TBX3 was generated as previously described ( Kumar et al . , 2014b ) . The Flag-tagged Gli3 construct was obtained from Addgene ( 51246 ) and the Flag-tagged Kif7 construct was a kind gift of Kathryn Anderson . Transfections of plasmids were performed with Lipofectamine 2000 reagent ( Invitrogen ) as per the manufacturer’s protocol .	Mutations in the gene that encodes a protein called T-box3 cause serious birth defects , including deformities of the hands and feet , via poorly understood mechanisms . Several other proteins are also important for ensuring that limbs develop correctly . These include the Sonic Hedgehog protein , which controls a signaling pathway that determines whether a protein called Gli3 is converted into its “repressor” form . The hair-like structures called primary cilia that sit on the surface of animal cells also contain Gli3 , and processes within these structures control the production of the Gli3-repressor . Emechebe , Kumar et al . have now studied genetically engineered mice in which the production of the T-box3 protein was stopped at different stages of mouse development . This revealed that turning off T-box3 production early in development causes many parts of the limb not to form . This type of defect appears to be the same as that seen in mice that lack the Sonic Hedgehog protein . If the production of T-box3 is turned off later in mouse development in the rear portion of the developing limb , the limb starts to develop but doesn’t develop enough rear toes . When T-box3 production is turned off in the front portion of the developing limbs , mice are born with too many front toes . This latter problem mimics the effects seen in mice that are unable to produce Gli3-repressor or that have defective primary cilia . Further investigation unexpectedly revealed that T-box3 is found in primary cilia and localizes to the same regions of the cilia as the Gli3-repressor . Furthermore , T-box3 also interacts with a protein complex that controls the stability of Gli3 and processes it into the Gli3-repressor form . In the future , it will be important to determine how T-box3 controls the stability of Gli3 and whether that process occurs in the primary cilia or in other parts of the cell where T-box3 and Gli3 coexist , such as the nucleus . This could help us understand how T-box3 and Sonic Hedgehog signaling contribute to other aspects of development and to certain types of cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2016	T-box3 is a ciliary protein and regulates stability of the Gli3 transcription factor to control digit number
Herd immunity , a process in which resistant individuals limit the spread of a pathogen among susceptible hosts has been extensively studied in eukaryotes . Even though bacteria have evolved multiple immune systems against their phage pathogens , herd immunity in bacteria remains unexplored . Here we experimentally demonstrate that herd immunity arises during phage epidemics in structured and unstructured Escherichia coli populations consisting of differing frequencies of susceptible and resistant cells harboring CRISPR immunity . In addition , we develop a mathematical model that quantifies how herd immunity is affected by spatial population structure , bacterial growth rate , and phage replication rate . Using our model we infer a general epidemiological rule describing the relative speed of an epidemic in partially resistant spatially structured populations . Our experimental and theoretical findings indicate that herd immunity may be important in bacterial communities , allowing for stable coexistence of bacteria and their phages and the maintenance of polymorphism in bacterial immunity . The term ‘herd immunity’ has been used in a variety of ways by different authors ( see Fine et al . , 2011 ) . Here , we define it as a phenomenon where a fraction of resistant individuals in a population reduces the probability of transmission of a pathogen among the susceptible individuals . Furthermore , if the fraction of resistant individuals in a population is sufficiently large the spread of a pathogen is suppressed . Experimental research into the phenomenon has focused mostly on mammals ( Jeltsch et al . , 1997; Mariner et al . , 2012 ) , birds ( van Boven et al . , 2008; Meister et al . , 2008 ) , and invertebrates ( Konrad et al . , 2012; Wang et al . , 2013 ) . In human populations the principles of herd immunity were employed to limit epidemics of pathogens through vaccination programs ( Fine et al . , 2011 ) , which in the case of smallpox lead to its eradication between 1959 and 1977 ( Fenner , 1993 ) . Alongside advances in vaccination programs , the formalization of a general theory of herd immunity was developed . The theory is based on a central parameter , R0 , which describes the fitness of the pathogen , as measured by the number of subsequent cases that arise from one infected individual in a population ( for a historical review of R0 see [Heesterbeek , 2002] ) . Thus , R0 indicates the epidemic spreading potential in a population . Given R0 the herd immunity threshold is defined as , ( 1 ) H=R0-1R0 , which determines the required minimum fraction of resistant individuals needed to halt the spread of an epidemic . R0 and subsequently also H are affected by the specific details of transmission and the contact rate among individuals ( Grassly and Fraser , 2008 ) . Many theoretical studies have addressed the influence of some of these details , in particular maternal immunity ( Anderson and May , 1992-08 ) , age at vaccination ( Anderson and May , 1982; Nokes and Anderson , 1988 ) , age related or seasonal differences in contact rates ( Schenzle , 1984; Anderson and May , 1985; Yorke et al . , 1979 ) , social structure ( Fox et al . , 1971 ) , geographic heterogeneity ( Anderson and May , 1984; Lloyd and May , 1996; Real and Biek , 2007 ) , and the underlying contact network of individuals ( Ferrari et al . , 2006 ) . Interestingly , little work has focused on the potential role of herd immunity in microbial systems which contain a number of immune defense systems and have an abundance of phage pathogens . These defenses vary in their potential to provide herd immunity as they target various stages of the phage life cycle , from adsorption to replication and lysis . Early defense mechanisms include the prevention of phage adsorption by blocking of phage receptors ( Nordström and Forsgren , 1974 ) , production of an extracellular matrix ( Hammad , 1998; Sutherland et al . , 2004 ) , or the excretion of competitive inhibitors ( Destoumieux-Garzón et al . , 2005 ) . Alongside these bacteria have evolved innate immune systems that target phage genomes for destruction . These include host restriction-modification systems ( RMS ) ( Blumenthal and Cheng , 2002 ) , argonaute-based RNAi-like systems ( Swarts et al . , 2014 ) , and bacteriophage-exclusion ( BREX ) systems ( Goldfarb et al . , 2015 ) . In addition to innate systems , bacteria have evolved an adaptive immune system called CRISPR-Cas ( clustered regularly interspaced short palindromic repeat ) ( Sorek et al . , 2013 ) . In order for any of these immune systems to provide herd immunity , they must prevent further spread of the pathogen . Therefore , unless the phage particles degrade in the environment at a timescale comparable to the phage adsorption rate , the immune system must provide a ‘sink’ for the infectious particles reducing the average number of successful additional infections below one . Unlike the early defense mechanisms that may simply prevent an infection but not the further reproduction of infectious particles , the RMS , BREX , argonaute-based RNAi-like , and the CRISPR-Cas systems degrade foreign phage DNA after it is injected into the cell , and thus continue to remove phage particles from the environment , which increases their potential to provide herd immunity . In order for herd immunity to arise , the population must also be polymorphic for immunity , which can be achieved if immunity is plasmid borne . In addition to this , the CRISPR-Cas system is unique in that it is adaptive allowing cells to acquire immunity upon infection ( see Figure 1A , B and C ) , which can lead to polymorphism in immunity even if the system is chromosomal . In addition to immune system-specific factors , the reproductive rate of phage depends strongly on the physiology of the host bacterium ( Hadas et al . , 1997 ) , and the underlying effective contact network which may vary greatly in bacterial populations depending on the details of their habitat . Thus , herd immunity will be influenced by the physiological state of the bacteria and the mobility of the phage in the environment through passive diffusion and movement of infected individuals . Taken together these details call into question the applicability of the traditional models of herd immunity from vertebrates to phage-bacterial systems . Thus , experimental investigation and further development of extended models that take into account the specifics of microbial systems are required . To investigate under which conditions herd immunity may arise in bacterial populations , we constructed an experimental system consisting of T7 phage and bacterial strains susceptible and resistant to it . Our experimental system can be characterized by the following features . First , we used two strains of Escherichia coli , one with an engineered CRISPR-based immunity to the T7 phage , and the other lacking it ( Figure 1D ) . Second , we examined the dynamics of the phage spread in different environments – spatially structured and without structure . Furthermore , we developed and analyzed a spatially explicit model of our experimental system to determine the biologically relevant parameters necessary for bacterial populations to exhibit herd immunity . We engineered a resistant E . coli strain by introducing the CRISPR-Cas Type II system from Streptococcus pyogenes with a spacer targeting the T7 phage genome ( see Material and Methods ) . We further characterized the ability of the system to confer resistance to the phage . We find a significant level of resistance as measured by the probability of cell burst when exposed to T7 ( Figure 2A ) . However , resistance is not fully penetrant as approximately 1 in 1000 resistant cells succumb to infection . In addition , we observe that as phage load increases ( multiplicity of infection , MOI ) the probability that a cell bursts increases ( Figure 2A ) . In order to determine the herd immunity threshold in our experimental system , we constructed the resistant strain such that upon infection the cell growth is halted , yet the cell still adsorbs and degrades phages ( Figure 2B , C ) . This feature is important as it prevents the action of frequency dependent selection which in naturally growing populations will favor the resistant strain until its frequency reaches the herd immunity threshold . Thus , in our system if the frequency of the resistant strain is below the herd immunity threshold , the resistant cells remain below the threshold and are unable to stop the epidemic and the whole population collapses . In contrast , if the frequency of resistant individuals in the population is above the herd immunity threshold , the resistant individuals provide complete herd immunity and the population survives . These properties allow us to quantify the expanding epidemic in both liquid media and on bacterial lawns ( without and with spatial structure , respectively ) using high throughput techniques . Specifically , it allows us to control for the complex dynamics of the system arising from frequency dependent selection and simultaneous changes in the physiological states of the cells ( growth rates depending on the nutrient concentrations ) and phage ( burst size , latent period depending on the cell’s physiology ) . It should be noted that our model does not reflect this artificial property – it assumes that resistant bacteria keep growing after successfully overcoming a phage infection ( see Equation ( 2d ) ) . This discrepancy , however , does not affect the model prediction of the herd immunity threshold in our experimental system for the following reason: time scale of an epidemic spread through a population ( double exponential phage growth ) is substantially shorter than the time scale of bacterial population growth ( exponential growth ) . Therefore , whether or not an epidemic is established does not depend on later dynamics of frequencies of resistant and susceptible individuals in the population , it only depends on the initial conditions . Similarly , the model correctly captures the dynamics of an epidemic in spatially structured populations as the phage spreads radially and in every time-point the epidemic front encounters a naive population with a constant ratio of resistant to susceptible individuals . To understand the influence of spatial population structure , or lack thereof , we first measured the probability of population survival ( i . e . , whether the cultures are cleared or not ) in well mixed liquid environments ( no spatial structure ) consisting of differing proportions of resistant to susceptible individuals and T7 phage . When the percentage of resistant individuals is in excess of 99 . 6% all 16 replicate populations survive a phage epidemic ( i . e . , show no detectable difference in growth profiles to the phage free controls; Figure 3 ) . Populations with 99 . 2% and 98 . 4% resistant individuals show intermediate probabilities of survival – 10 out of 16 replicate populations and 4 out of 16 replicate populations survive , respectively ( Figure 3 ) . The likely explanation as to why some populations survive and others collapse is due to the stochastic nature of phage adsorption after inoculation: If the population composition is close to the herd immunity threshold a stochastic excess of phage particles adsorbing to susceptible cells may trigger an epidemic , whereas if chance increases the number of phages adsorbing to resistant individuals , the epidemic is suppressed . However , when populations have fewer than 96 . 9% resistant individuals all 16 replicate populations fail to survive and collapse under the epidemic ( Figure 3 ) . As mentioned in the introduction , phage and bacterial physiology may affect the herd immunity threshold . To test this we altered bacterial growth by reducing the concentration of nutrients in the medium by mixing LB broth with 1X M9 salts in different ratios ( Figure 4 ) , which concurrently alters the T7 phage’s latent period and burst size ( Figure 5A , B and Table 1 ) . Indeed , we observe as bacterial growth rates decline the fraction of resistant individuals necessary for population survival decreases ( Figure 5C ) . When the populations are grown in a 50% diluted growth medium , the fraction of resistant individuals required for a 100% probability of survival is 99 . 2%; when the fraction of resistant individuals is 75% or less populations go extinct . In a 20% growth medium the fraction of resistant individuals required for survival decreases to 96 . 9% , while the fraction when all replicates collapse to 50% . From the experimental observations of the herd immunity threshold values we infer the phage R0 using Equation 1 . In an undiluted growth medium the phage R0 falls between 32 and 256 and decreases to between 4 and 128 in 50% and between 2 and 32 in 20% nutrient medium . These data indicate that bacterial populations can exhibit herd immunity in homogeneous liquid environments . However , bacteria typically live in spatially structured environments such as surfaces , biofilms or micro-colonies , therefore we extended our experiments to consider the potential impact of spatially structured populations . In order to discern the role , if any , spatial structure plays in herd immunity we conducted a set of experiments in spatially structured bacterial lawns on agar plates . Spatially structured bacterial populations provide a more fine grained measure of herd immunity , compared to the population survival assays done in liquid culture . On bacterial lawns , phages spread radially from a single infectious phage particle and the radius of plaque growth on different proportions of resistant to susceptible individuals can be easily quantified . In addition , these data allow for estimating the speed of the epidemic wave front in these different regimes using real-time imaging ( Figure 6A ) . We observe a decline in the number of plaque forming units ( see Appendix 2—figure 1 ) and a significant decrease in final plaque sizes as the proportion of resistant individuals in the populations increases ( Figure 6B , C ) . A reduction in the final plaque size compared to a fully susceptible population was statistically significant with as few as 10% resistant individuals in a population ( p=0 . 004 , t53 = 2 . 744 ) . In order to determine the effect of resistant individuals during the earlier phase of bacterial growth ( until the bacterial density on the agar plate reaches saturation; Figure 4A ) , we analyze the velocities of plaque growth between 0 and 24 hr post inoculation ( h⁢p⁢i ) . We find that the speed is significantly reduced after 11⁢h⁢p⁢i when the population consists of as few as 10% of resistant individuals ( p=0 . 0317 , t32 = 1 . 923 ) . As the fraction of resistant individuals further increases , the speed declines significantly at earlier and earlier time points: 6⁢h⁢p⁢i with 20% ( p=0 . 0392 , t62 = 1 . 79 ) , and 5 . 67⁢h⁢p⁢i with 30% ( p=0 . 0286 , t53 = 1 . 943 ) . In fact , when the fraction of resistant individuals exceeds 40% , the reduction in the speed of the spread is statistically significant immediately after the plaques are visually detectable ( Figure 7 ) . It should be noted that all populations with such low percentages of resistant individuals in liquid environment collapsed , indicating that spatial structure plays a role in herd immunity . The results presented in this and the previous section would allow us to use Equation 1 to infer a value for R0 from the observed threshold between surviving and collapsing bacterial populations , Figures 3 , 5 . We also observe that herd immunity is strongly influenced by spatial organization of the population , Figure 6 . How the exact value of H ( and subsequently the ‘classical’ epidemiological parameter R0 ) is affected by various factors such as bacterial growth rate , phage burst size and latent period is , however , difficult to resolve experimentally . Similarly , quantification of the effect of spatial structure is hardly achievable solely by experimental investigation . In order to disentangle the roles of all the factors mentioned above , we proceed with development and analysis of a mathematical model of the experimental system . We developed a model of phage growth that takes several physiological processes into account: bacterial growth during the experiment , bacterial mortality due to phage infection , and phage mortality due to bacterial immunity . Furthermore , we use the previously reported observation that phage burst size β and latent period λ depend strongly on the bacterial growth rate α ( see Table 1 ) . The main processes in our model system can be defined by the following set of reactions , ( 2a ) Bs+yN⟶α2Bs , ( 2b ) Br+yN⟶α2Br , ( 2c ) Bs+P⟶A ( BsP ) ⟶1/λβP , ( 2d ) Br+P⟶A ( BrP ) {⟶fastBr , ⟶slowβP . Susceptible ( Bs ) and resistant ( Br ) cells grow at a rate α ( no significant difference in growth rate between strains , α⁢ ( Bs ) =0 . 551±0 . 045⁢h-1 , α⁢ ( Br ) =0 . 535±0 . 023⁢h-1 , t70=1 . 867 , p=0 . 066 ) , Equation 2 , by using an amount y of the nutrients N . Phage infection first involves adsorption to host cells , Equation 2c and Equation 2d , with the adsorption term A specified below . Infected susceptible bacteria produce on average β phage with a rate inversely proportional to the average latency λ . In contrast , resistant bacteria either survive by restricting phage DNA via their CRISPR-Cas immune system or – less likely – succumb to the phage infection . However , when the MOI is large even resistant cells become susceptible to lysis resulting in the release of phage progeny ( see Figure 2 ) ( Westra et al . , 2015; Chabas et al . , 2016 ) . In our system , bacteria eventually deplete the available nutrients , N ( t>Tdepl ) =0 , resulting in the cessation of growth . This decline in bacterial growth affects phage growth – latency increases and burst size decreases , such that phage reproduction declines dramatically ( see Table 2 ) . We define the critical time point at which cells transition from exponential growth to stationary phase as , ( 3 ) Tdepl≈1α⁢log⁡ ( B∞B0 ) . Here , B0 and B∞ are the initial and final bacterial densities , respectively . In the initial exponential growth phase , our estimates from experimental data for growth parameters are α=0 . 63⁢h-1 , β=85 . 6⁢phages/cell and λ=0 . 60⁢h , for bacteria and phages , respectively ( Tables 1 and 2 ) . After time Tdepl , bacterial growth rate is set to zero ( α=0 ) and phage growth is reduced to βdepl=3 . 0⁢phages/cell and λdepl=1 . 69⁢h . Such a two state model – constant growth rate while nutrients are present and no growth after depletion – describes the observed population trajectories in experiments sufficiently well ( see Figure 4 ) . An important parameter for estimating herd immunity is the fraction S of susceptible bacteria in the population . As a first estimate , a phage infection spreads in well mixed bacterial cultures if βS>1 , which leads to a continuous chain of infections: the product of burst size β of phage particles and the probability S of infecting a susceptible host has to be larger than one . As a first approximation , one could identify R0 with the burst size β , which is compatible with the observed herd immunity thresholds when inverting Equation 1 . However , the growing bacterial population could outgrow the phage population if the former reproduces faster ( e . g . , in the case of RNA coliphages , van Duin , 1988 ) , which introduces deviations from the simple relation between R0 and H as shown in Equation 1 . We capture this dynamical effect in a correction to the previous estimate as βS>1+λα ( see Materials and methods ) : more phages have to be produced for the chain of infections to persist in growing populations . The correction λ1/α is the ratio of generation times of phages over bacteria – usually , such a correction is very small for non-microbial hosts and can be neglected . Ultimately , herd immunity is achieved if the threshold defined by H=1-Sc is exceeded , with Sc the critical value in the inequality above . Rearranging , we obtain an expression for the herd immunity threshold ( 4 ) H=β-1-λ⁢αβ . This estimate of H coincides to a very good extent with the population compositions of susceptible and resistant bacteria where we observe the transition from surviving and collapsed populations in experiments ( see Figure 3 ) . Moreover , simulations presented in the Appendix ( section Simulation of recovery rate ) show a range in the bacterial population composition with non-monotonic trajectories for Bs and Br ( see Appendix 1—figure 1B ) , which is comparable to the range in composition we find in both outcomes , that is , some surviving and some collapsing populations in experiments . For such parameter choices , stochastic effects could then decide the observed fates of bacteria . As presented above , the herd immunity threshold changes when the bacterial cultures grow in a diluted growth medium . In a set of independent experiments we measured bacterial growth rate α , phage burst size β and phage latent period λ under such conditions ( see Figure 4B and Table 2 ) . From these data we estimated the dependence of the phage burst size on the bacterial growth rate , β⁢ ( α ) , using a numerical quadratic fit ( Figure 5A ) . Similarly , we estimated the dependence of the phage latent period on the bacterial growth rate , λ⁢ ( α ) ( Figure 5B ) . Using these estimates we calculated the expected growth rate–dependent herd immunity threshold ( 5 ) H⁢ ( α ) =β⁢ ( α ) -1-λ⁢ ( α ) ⁢αβ⁢ ( α ) , which gives a very good prediction of the shift in the herd immunity threshold to lower values for slower growing populations ( green line in Figure 5C ) . This verification of our model shows that it correctly captures the dependence of the herd immunity threshold on bacterial and phage growth parameters . The deviations from the herd immunity threshold depicted by the green area in Figure 3 and green error bars in Figure 5C are derived from uncertainty in measurements in β , λ and α . The inherent stochasticity of the adsorption process thus provides additional uncertainty , which is not captured by the depicted error bars . This additional stochasticity can explain wider transition zone in experiments with slower growing populations ( dilution 0 . 5 and 0 . 2 ) , because the fate of the population is more prone to stochastic effects as the phage replication rate is slower than in a fast growing population . This stochastic effect might be reduced by larger phage inocula . This could , however , also shift the observed transition between collapsing and surviving populations towards higher frequencies of resistant bacteria ( and away from the actual herd immunity threshold ) as protection by the immune system is less effective with increasing number of phages per cell ( see Figure 2A ) . The dynamics of phage spread differ between growth in unstructured ( e . g . , liquid ) and structured ( e . g . , plates ) populations . In order to quantify the effect of spatial structure in a population , we extend our model to include a spatial dimension . In structured populations growth is a radial expansion of phages defined by the plaque radius r and the expansion speed v , for which several authors have previously derived predictions ( Kaplan et al . , 1981; Yin and McCaskill , 1992; You and Yin , 1999; Fort and Méndez , 2002a; Ortega-Cejas et al . , 2004; Abedon and Culler , 2007; Mitarai et al . , 2016 ) . We assume phage movement can be captured by a diffusion process characterized with a diffusion constant D , which we estimate in independent experiments as D=1 . 17 ( ±0 . 26 ) ⋅10−2 mm2/h ( see Materials and methods , Figure 8 ) . However , we assume that only phages disperse and bacteria are immobile as the rate of bacterial diffusion does not influence the expanding plaque on timescales relevant in the experiment . Adsorption of phages on bacteria is modeled with an adsorption constant δ⋆ . Taking these considerations together , allows to write a reaction-diffusion dynamics for growth of phages P on the growing bacterial population as ( 6 ) ∂t⁡P=D⁢∂x2⁡P+δ⋆⁢ ( β⁢S-1-λ⁢α ) ⁢P . The first term accounts for the diffusive spread of phages , while the second term describes phage growth . This second term includes the correction λ⁢α which arises due reproduction of bacteria , derived in the unstructured liquid case . The spreading infection will sweep across the bacterial lawn with the following speed ( 7 ) v=2⁢D⁢δ⋆⁢β⁢S-1-λ⁢α , which is computed in more details in the Materials and methods . This expression Equation 7 indicates that the population composition crucially influences the spreading speed at much lower fractions of resistant bacteria than the herd immunity threshold Equation 4 , where phage expansion stops completely . Consequently , the resulting plaque radius r decays with increasing fractions of resistants and reaches zero at H . A prediction for r can be obtained by integrating Equation 7 over time . In our ( simplified ) model , time-dependence of the speed only enters via the fraction of susceptibles S , which is assumed to stay at the initial S0 value until it encounters the epidemic wave of phages . Furthermore , we use the experimental observation that plaque expansion ceases upon depletion of nutrients , coinciding with a cessation of bacterial growth . This leads to the approximation r≈v⁢Tdepl , with Tdepl given by Equation 3 . Using this expression we estimated the adsorption constant δ⋆ from the growth experiments as it is difficult in practice to measure on plates . The green line in Figure 6B is the best fit , yielding the value δ⋆=4 . 89⁢ ( ±0 . 19 ) ⋅10-2⁢bacteria/phage⁢h for the adsorption constant . Our results for spatially structured populations allows us to speculate on a general epidemiological question: If an infection is not stopped by herd immunity in a partially structured population , by how much is its spread slowed down ? By generalizing Equation 7 we can derive a relative expansion speed , compared to a fully susceptible population , ( 8 ) vrel=1-1-SH . This expression , Equation 8 , is devoid of any ( explicit ) environmental conditions , which are not already contained in the herd immunity threshold H itself . Thus , it could apply to any pathogen-host system . Ultimately , this relative speed approaches zero with a universal exponent of 1/2 , when the fraction of resistant individuals 1-S approaches the herd immunity threshold H . However , a few caveats exist when using Equation 8 , as several conditions have to be fulfilled: Obviously , a pathogen is expected not to spread in a population exhibiting complete herd immunity – the relative speed should only hold for populations below the herd immunity threshold . Moreover , if dispersal cannot be described by diffusion , but rather dominated by large jumps ( Hallatschek and Fisher , 2014 ) , the diffusion approach we used for traveling waves is not applicable , and thus also renders Equation 8 inadequate . An increase in the number of long range jumps of phages can be considered as a transition between the two cases we treated here – spatially explicit dynamics on plates and completely mixed populations in liquid culture , respectively . Potential long range jumps of phages can be mediated by host cells moving distances that the phages cannot achieve on their own . In such cases , dispersal of the phages is a convolution of movement of their hosts with their own ability to spread locally . These long range jumps would therefore increase the overall expansion speed and area of the epidemic . We expect that in our setup bacterial motility does not substantially contribute to phage spread because ( i ) bacteria become motile only in late exponential/early stationary phase ( Amsler et al . , 1993 ) when phage reproduction drops to very low levels , and ( ii ) the soft agar concentration used in our experiments ( ≈0 . 525% ) effectively blocks bacterial motility ( Croze et al . , 2011 ) . However , we would not expect that long range jumps change the herd immunity threshold H⁢ ( α ) itself . Spread of pathogens still stops when the fraction of susceptible hosts S is small such that βS<1+λα , and will continue as long as βS>1+λα is fulfilled . The spread of a pathogen may be halted or slowed by resistant individuals in a population and thus provide protection to susceptible individuals . This process , known as herd immunity , has been extensively studied in a wide diversity of higher organisms ( Jeltsch et al . , 1997; Mariner et al . , 2012; van Boven et al . , 2008; Meister et al . , 2008; Konrad et al . , 2012; Wang et al . , 2013 ) . However , the role of this process has largely been ignored in microbial communities . To delve into this we set out to determine under what conditions , if any , herd immunity might arise during a phage epidemic in bacterial populations as it could have profound implications for the ecology of bacterial communities . We show that herd immunity can occur in phage-bacterial communities and that it strongly depends on bacterial growth rates and spatial population structure . Average growth rates of bacteria in the wild have been estimated as substantially slower than in the laboratory ( generation time is ≈7 . 4 fold longer [Gibson et al . , 2017] ) , a condition that we have shown to facilitate herd immunity . Furthermore , bacterial populations in the wild are also highly structured , as bacteria readily form micro-colonies or biofilms ( Hall-Stoodley et al . , 2004 ) and grow in spatially heterogeneous environments such as soil or the vertebrate gut ( Fierer and Jackson , 2006 ) , a second condition we have shown to facilitate herd immunity . These suggest that herd immunity may be fairly prevalent in low nutrient communities such as soil and oligotrophic marine environments . In an evolutionary context , herd immunity may also impact the efficacy of selection as the selective advantage of a resistance allele will diminish as the frequency of the resistant allele in a population approaches the herd immunity threshold , H . This has two important implications . First , while complete selective sweeps result in the reduction of genetic diversity at linked loci , herd immunity may lead to only partial selective sweeps in which some diversity is maintained . Second , herd immunity has a potential to generate and maintain polymorphism at immunity loci , as has been shown for genes coding for the major histocompatibility complex ( MHC ) ( Wills and Green , 1995 ) . Polymorphism in CRISPR spacer contents have been demonstrated in various bacterial ( Tyson and Banfield , 2008; Sun et al . , 2016; Kuno et al . , 2014 ) and Archaeal ( Held et al . , 2010 ) populations and communities ( Pride et al . , 2011; Zhang et al . , 2013; Andersson and Banfield , 2008 ) . While these studies primarily explain polymorphisms in CRISPR spacer content as a result of rapid simultaneous independent acquisition of new spacers , we suggest that observed polymorphisms may result from frequency-dependent selection on resistance loci arising from herd immunity . In such a case , herd immunity is likely to maintain existing polymorphism in CRISPR spacer content in 1-H fraction of the population , unless the current major variant goes to fixation due to drift . However , considering the large population sizes of bacteria , drift is unlikely to have a strong effect , allowing herd immunity to maintain a large fraction of immunity polymorphism . It has also been suggested that herd immunity might favor coexistence between hosts and their pathogens ( Hamer , 1906 ) , which can lead to cycling in pathogen incidence and proportions of resistant and susceptible individuals over time , e . g . , in measles before the era of vaccination ( Fine , 1993 ) . This cycling is caused by the birth of susceptible individuals , which , once their proportion exceeds the epidemic threshold ( 1-H ) , lead to recurring epidemics . CRISPR-based immunity is , however , heritable meaning that descendants of resistant bacteria remain resistant . One might speculate that analogous cycling in phage epidemics may occur if immunity is costly . In turn , a computer simulation study of coevolution of Streptococcus thermophilus and its phage found both cycling and stable coexistence of different CRISPR spacer mutants and phage strains ( Childs et al . , 2014 ) . The extent to which herd immunity facilitates maintenance of CRISPR spacer polymorphism and coexistence with phage requires further experimental and theoretical investigation . We also developed a mathematical model and show how the herd immunity threshold H ( Equation 4 ) depends on the phage burst size β and latent period λ , and on the bacterial growth rate α . This dependence arises as phages have to outgrow the growing bacterial population , as host and pathogen have similar generation times in our microbial setting . In addition to these parameters , we also describe how the speed v ( Equation 7 ) of a phage epidemic in spatially structured populations depends on phage diffusion constant D , phage adsorption rate δ⋆ , and the fraction of resistant and susceptible individuals in the population . All of which are likely to vary in natural populations . We also derived the relative speed of spread for partially resistant populations , as measured relative to a fully susceptible population , and show that it can be parametrized solely with the herd immunity threshold H ( Equation 8 ) . This relative speed of the spread of an epidemic should be applicable to any spatially structured host population where the spread of the pathogen can be approximated by diffusion . Both our experiments and the modelling show that even when the fraction of resistant individuals in the population is below the herd immunity threshold the expansion of an epidemic can be substantially slowed , relative to a fully susceptible population . In conclusion , we have presented an experimental model system and the connected theory that can be usefully applied to both microbial and non-microbial systems . Our theoretical framework can be useful for identifying critical parameters , such as H ( and to some extent R0 ) , from the relative speed of an epidemic expansion in partially resistant populations so long as the process of pathogen spread can be approximated by diffusion . This approximation has been shown to be useful in such notable cases as rabies in English foxes ( Murray et al . , 1986 ) , potato late blight ( Scherm , 1996 ) , foot and mouth disease in feral pigs ( Pech and McIlroy , 1990 ) , and malaria in humans ( Gaudart et al . , 2010 ) . Spatial modelling of phage expansion has produced several predictions for how plaque radius r and expansion speed v are influenced by experimentally adjustable parameters ( Kaplan et al . , 1981; Yin and McCaskill , 1992; You and Yin , 1999; Fort and Méndez , 2002a; Ortega-Cejas et al . , 2004; Abedon and Culler , 2007; Mitarai et al . , 2016 ) . Here , we try to use a minimal model to estimate these two observables , based on the considerations of previous sections . One of the main complications arises from the fact that all densities in Equation 12 have a spatial dimension in addition to their time dependence , Bi=Bi⁢ ( x→ , t ) , i∈{s , r} . As explained in the main text we only consider phage diffusion , heterogeneities in all other densities are generated only by phage growth . The additional spatial dimension imposes a particular contact network between phages and bacteria , which are not entirely random encounters anymore: One can expect that the size of the bacterial neighborhood B^ phages are able to explore is only slightly determined by the actual density B , and can be assumed constant for most of the experiment , B^⁢ ( B ) ≈c⁢o⁢n⁢s⁢t . Consequently , the adsorption term can be written in the following way , ( 24 ) A[Bi , P\|Bs , Br]=δ⋆BiBs+BrP , i∈{s , r} , which only depends on the relative frequencies of bacterial strains . The adsorption constant δ⋆ is both the rate of adsorption and inter-host transit time as determined by the diffusion constant D . Thus , one can expect the formal dependence δ⋆=δ⋆⁢ ( D , B^⁢ ( B ) ) . For our particular experimental setup , however , δ⋆ will be treated as a constant . This adsorption term Equation 24 leads to the dynamics of phages ( 25 ) ∂t⁡P=D⁢∇2⁡P+G⁢[P , S] , where we collected all contributions to phage growth in G⁢[P , S] and added the spatial diffusion term D⁢∇2⁡P . For simplicity , we consider only expansion in a single dimension ( ∇2≡∂x2 ) , which has been found to coincide well with the dynamics of plaque growth ( Yin and McCaskill , 1992 ) . The growth term for phages is then defined as , ( 26 ) G⁢[P , S]=δ⋆⁢ ( S⁢β-1-λ⁢α ) ⁢P , where we also consider the correction λ⁢α obtained from the analysis in liquid culture . Due to the different absorption dynamics on plates , however , this correction might be a slight overestimate of the actual term that accounts for bacterial growth . Reaction-diffusion equations similar to Equation 25 have been first analyzed about 80 years ago ( Fisher , 1937; Kolmogorov et al . , 1991 ) and since then treated extensively , e . g . ( Murray , 2002; van Saarloos , 2003 ) . They admit a traveling wave solution – here , this corresponds to phages sweeping over an uninfected bacterial lawn . In general , the asymptotic expansion speed for the traveling wave solutions is given by the following expression , ( 27 ) v=2D ( ∂PG ) [0 , S]=2Dδ⋆Sβ−1−λα . Only the linearized growth rate of phages at very low densities is relevant for the expansion speed , ∂PG[P=0 , S] . Thus , the fraction of susceptible individuals S should be unchanged from its initial value S0 . It should be noted , that only for S0β>1+λα does Equation 27 remain valid , otherwise we have v=0 . Such a scenario is relevant when nutrients are depleted and phage growth parameters changes to βdepl and λdepl . The expression for the expansion speed also shows the need for the spatial adsorption model in Equation 24 , in contrast to the liquid case Equation 13 . If adsorption would directly depend on the bacterial density B , the additional linear dependence on B in Equation 26 would lead to an exponentially increasing speed during the experiment . This is in clear contradiction to experimental observations . The density of phages behind the expanding front is large and as previously noted at large MOIs the CRISPR-Cas system fails to provide effective immunity ( see section Materials and methods and appendix Infection load and efficiency of the CRISPR/Cas system ) . However , in comparison to an un-structured environment ( e . g . , liquid ) the structured environment effectively limits transit of phage from within a plaque to the expanding front: The combined effect of growth and diffusion usually generates a much faster expansion of phages during plaque formation , than diffusion alone . Only when nutrients are depleted , can pure diffusion processes explain the slow decrease in speed observed in experiments ( see Figure 7A ) . Our model assumes a sharp drop to v=0 at Tdepl for small S . In order to derive an expression for the plaque radius r , we integrate the expansion speed Equation 7 over time , r⁢ ( t ) =∫0td⁢t′⁢v⁢ ( t′ ) . Employing the simplification that only two values of phage growth are necessary to describe the dynamics – before Tdepl phages grow normally with β and λ , after Tdepl phage growth changes to βdepl and λdepl – we can evaluate the integral for the radius directly , arriving at , ( 28 ) r ( t ) ={2tDδ⋆Sβ−1−λα , 0<t<Tdepl , 2Dδ⋆ ( TdeplSβ−1−λα+ ( t−Tdepl ) Sβdepl−1 ) , Tdepl<t . Using this expression we estimated the adsorption constant δ⋆ from the growth experiments as it difficult to measure in practice . This estimate is done for radii exactly at the time of nutrient depletion Tdepl , and excluding the control experiment with only susceptible cells . Predictions of our model show a discrepancy from experimental results on plates after depletion . We independently estimated βdepl=3 . 0 , which results in Hdepl= ( βdepl-1 ) /βdepl≈0 . 67 . Thus , all experiments with S>0 . 33 should exhibit expanding plaques after nutrients are depleted . In the experimental setup plaques stop expanding in all mixtures of resistant to susceptible cells ( S≤0 . 9 ) , which would correspond to βdepl<1 . 1 . This value is , however , still within experimental accuracy of our estimates of βdepl .	When a disease spreads through a population , it encounters certain individuals it cannot infect . If there are enough of these individuals , the epidemic stops . This phenomenon is known as ‘herd immunity’ , and it occurs in many animals – for example , it plays an important role in human vaccination schemes . While bacteria can cause disease , they are themselves targeted by viruses called ‘phages’ . Bacteria can overcome this threat , and they resist phage attacks in ways that are well understood at the molecular level . However , little is known about the impact of this resistance at the scale of the population . Can herd immunity occur in bacteria ? If so , what factors influence the threshold at which it will occur ? In other words , what affects the minimum percentage of immune bacteria needed to stop the spread of a phage infection ? To answer these questions , Payne et al . used both experimental and mathematical methods . For the experiments , a phage and two strains of bacteria were used , one immune to the virus and one not . The two strains were combined to form several populations with different percentages of resistant bacteria , and the phage was added . How the virus could spread in these different populations was recorded . This confirmed that herd immunity does occur in bacteria and showed how the resistant bacteria influence the speed which an epidemic spreads in a population . Building on the experiments , Payne et al . then produced a mathematical model to explore how different factors affect herd immunity . For example , the model showed that the thresholds for herd immunity can be predicted from how quickly bacteria and phages replicate . The thresholds are lower when bacteria reproduce more quickly , but higher when it is the phages that multiply faster . The model also helps infer a formula that informs on how diseases spread in any species , such as humans . In particular , it becomes possible to predict herd immunity thresholds based on how quickly an epidemic spreads in a population where few people are vaccinated . Future research is needed to adapt the formula to the specific factors that shape disease outbreaks in humans . Ultimately , this could help policymakers design strategies to deal with infectious diseases , such as yearly outbreaks of the flu .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2018	CRISPR-based herd immunity can limit phage epidemics in bacterial populations
Rhomboid proteases reside within cellular membranes , but the advantage of this unusual environment is unclear . We discovered membrane immersion allows substrates to be identified in a fundamentally-different way , based initially upon exposing ‘masked’ conformational dynamics of transmembrane segments rather than sequence-specific binding . EPR and CD spectroscopy revealed that the membrane restrains rhomboid gate and substrate conformation to limit proteolysis . True substrates evolved intrinsically-unstable transmembrane helices that both become unstructured when not supported by the membrane , and facilitate partitioning into the hydrophilic , active-site environment . Accordingly , manipulating substrate and gate dynamics in living cells shifted cleavage sites in a manner incompatible with extended sequence binding , but correlated with a membrane-and-helix-exit propensity scale . Moreover , cleavage of diverse non-substrates was provoked by single-residue changes that destabilize transmembrane helices . Membrane immersion thus bestows rhomboid proteases with the ability to identify substrates primarily based on reading their intrinsic transmembrane dynamics . Signaling networks rely on the specificity of individual components for their targets , avoiding unwanted crosstalk and driving emergent properties of the system . Proteases are ubiquitous in all life , and are particularly well-adapted for serving as regulatory nodes in networks ( López–Otín and Bond , 2008 ) . These enzymes achieve their exquisite specificity by recognizing a short binding motif surrounding the scissile bond to align the substrate along the elongated active site cleft of the protease ( Schechter , 2005 ) . Residues flanking the scissile bond are designated P1 and P1′ on the amino and carboxy sides , respectively , and numbered moving outwards , while the accommodating protease subsites are correspondingly termed S1 and onwards . This geometric pairing confers high target specificity as the number of residues in the interaction increases , and results in a single and invariant substrate cleavage site ( Schechter , 2005; Ng et al . , 2009 ) . While soluble proteases have been attractive topics for study for over a century and are now understood at a sophisticated level ( López–Otín and Bond , 2008 ) , the more-recent discovery of intramembrane proteases uncovered a fundamentally separate path of protease evolution ( Brown et al . , 2000 ) . These polytopic membrane proteins assemble a protease active site within the membrane using catalytic residues contributed by different transmembrane ( TM ) segments ( Erez et al . , 2009 ) . Even more remarkable is the broad array of cellular networks that intramembrane proteases have come to regulate . Intensively studied is γ-secretase , an aspartyl intramembrane protease that releases signaling domains from the membrane , including the intracellular domains of the Notch receptor and the amyloid-β precursor protein ( APP ) implicated in Alzheimer's disease ( De Strooper et al . , 1998 , 1999; Wolfe et al . , 1999 ) . Homologous signal peptide peptidases function in immunity by liberating signaling domains of TNFα and FasL ( Fluhrer et al . , 2006; Friedmann et al . , 2006; Kirkin et al . , 2007 ) . Site-2 proteases are metalloenzymes that release transcription factors from the membrane to regulate membrane biogenesis and stress responses in diverse organisms from bacterial pathogens to man ( Rawson et al . , 1997; Makinoshima and Glickman , 2005 ) . Lastly , rhomboid proteases act as prime regulators of signaling in insects by activating EGF signals through proteolytic shedding ( Urban et al . , 2001 , 2002 ) , and play prominent functions in diverse pathogen signaling and adhesion ( Urban , 2009 ) . The presence of intramembrane proteases in all forms of life indicates that they possess a particularly useful property as regulatory enzymes ( Kinch et al . , 2006; Lemberg and Freeman , 2007 ) . However , comparative approaches have been instructive only in highlighting convergent similarities in catalytic chemistry with soluble serine proteases ( Vinothkumar et al . , 2010 ) ; other properties of intramembrane proteases remain unexplored ( Erez et al . , 2009; Urban , 2010 ) . Since these membrane-immersed proteases evolved within the hydrophobic milieu of the membrane , a fundamentally different environment compared to soluble proteases , could this novel environment confer different enzymatic properties ? Particularly important is target specificity , because intramembrane cleavage is usually the signal-generating step that alone is sufficient for pathway activation ( Brown et al . , 2000 ) . With over a dozen crystal structures and well-developed reconstitution systems for their study , arguably the best understood biochemically are rhomboid proteases ( Urban , 2010 ) . Mutational analyses have identified some sequence determinants in rhomboid substrates , most notably small P1/P1′ residues ( Urban and Freeman , 2003; Akiyama and Maegawa , 2007 ) and large , hydrophobic residues at P4 and P2′ ( Strisovsky et al . , 2009 ) . However , despite the recent wealth of biochemical and structural information ( Bondar et al . , 2009; Urban , 2010 ) , particularly on the E . coli rhomboid GlpG , most current studies have been confined to detergent systems ( Lemberg et al . , 2005; Strisovsky et al . , 2009 ) . Achieving a true understanding of rhomboid function can only be realized by defining its properties in the natural context of the membrane . We therefore used biochemical and spectroscopic methods to determine the contribution of the membrane to proteolysis . These approaches revealed rhomboid proteases rely upon constraints imposed by the membrane on TM segment conformational dynamics to achieve high proteolytic specificity . Further interrogation of proteolysis directly in living cells suggest that rhomboid proteases expose the propensity of TM helices to exit the membrane and unwind as a substrate-discrimination mechanism , rather than relying on recognition-sequence binding like all other known specific proteases . In order to identify any specific contributions of the cell membrane to proteolysis , we compared catalysis in living cells to catalysis in detergent micelles that support high levels of rhomboid activity . Mass spectrometry revealed that rhomboid proteolysis is notably site-specific , in contrast to other intramembrane proteases ( Fraering et al . , 2004; Fluhrer et al . , 2006; Friedmann et al . , 2006; Sato et al . , 2006 ) . In fact , cleavage of the Drosophila EGF ligand Spitz always proceeded between the first two residues ( AS ) of its TM segment even with eight diverse rhomboid proteases and in bacterial , insect and human cells ( and different organelles ) that harbor lipid composition differences ( Fast , 1966 ) ( Figure 1A , also see Figure 1—figure supplement 1A ) . Although Spitz is the most-studied rhomboid substrate , its cleavage site had never been mapped in cells . 10 . 7554/eLife . 00173 . 003Figure 1 . The membrane directs site and substrate specificity by rhomboid proteases . ( A ) Western analysis of GFP-Spitz-Flag expressed in Drosophila S2R+ cells . Denoted throughout are uncleaved ( ∇ ) and cleaved forms ( * ) . In vivo cleavage sites were mapped by mass spectrometry following anti-flag immunocapture of GFP-Spitz-Flag processed in Drosophila S2R+ cells by DmRho1 , as well as other rhomboid proteases in both mammalian and bacterial cells and in different organelles ( also see Figure 1—figure supplement 1A ) . The invariant cleavage site is denoted with a blue arrow in Spitz ( first seven TM residues are shown ) . ( B ) The cleavage site in Spitz generated in vitro shifted also to the second AS when assayed in dodecyl-β-D-maltoside ( DDM ) detergent . Arrows and asterisks are color matched throughout . Cleavage products isolated from N-Flag and C-Flag tagged APP + Spi7 substrates revealed the same cleavage sites with the expected relative proportions . ( C ) Reconstituting substrates and rhomboid proteases from detergent into proteoliposomes in vitro restored cleavage to the natural site . ( D ) Cleavage of APP + Spi7-Flag vs its GA to LM mutant by GlpG in 0 . 25% DDM detergent or reconstituted into proteoliposomes . Note that upon reconstitution , the local concentration of substrate is higher than in detergent solution . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 00310 . 7554/eLife . 00173 . 004Figure 1—figure supplement 1 . Cleavage site of Spitz in animal and bacterial cells , and APP + Spi7 in vitro . ( A ) Cleavage site of GFP-Spitz-Flag processed in human HEK293 cells by DmRho1 ( in the Golgi apparatus ) , DmRho1-KDEL ( in the ER ) , human RHBDL4 ( ER ) and RHBDL2 ( cell surface ) , DmRho4 in COS cells ( cell surface ) , as well as in Escherichia coli cells by Providencia stuartii ( AarA ) , Bacillus subtilis ( YqgP ) , and Aquifex aeolicus ( AqRho ) . Cleaved Spitz was immuno-captured with anti-Flag agarose and subjected to molecular mass analysis by MALDI-TOF mass spectrometry . The shoulder to the right of each cleaved Spitz peak results from doubly-charged immunoglobulin light chain carryover . All rhomboid proteases , irrespective of in which organelle or cell type they were expressed , cleaved Sptiz between the first two residues ( AS ) of its transmembrane segment . ( B ) Cleavage site of APP + Spi7 ( C-terminal Flag tag ) shifted to predominantly between the GA residues when processed by YqgP in mixed DDM:phospholipid ( 0 . 1%:0 . 1% ) micelles ( since pure YqgP is inactive in detergent alone ) and AarA in detergent . Highlighted are uncleaved ( ∇ ) and cleaved forms ( * ) . ( C ) Cleavage site of APP + Spi7 with AarA isolated using an N-terminal Flag tag . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 00410 . 7554/eLife . 00173 . 005Figure 1—figure supplement 2 . Cleavage site of N-Flag-Spitz cleaved in proteoliposomes composed of different lipids in vitro . Sptiz was cleaved between the first two residues ( AS ) of its transmembrane segment irrespective of the lipid composition of the proteoliposomes . Cleaved N-terminal Flag-tagged Spitz was immuno-captured with anti-Flag agarose and subjected to molecular mass analysis by MALDI-TOF mass spectrometry . The asterisks mark the mass peak corresponding to the cleaved Flag-Spitz fragment . The lipid environments denoted are: PC , phosphatidylcholine; PA , phosphatidic acid; PE/LyPE , 1:1 molar mix of phosphatidylethanolamine and lyso-phosphatidylethanolamine; PG , phosphatidylglycerol; PE/PG , 70:30 molar mix of phosphatidylethanolamine and phosphatidylglycerol; SM , sphingomyelin; DPPE/DPPG , 70:30 molar mix of 1 , 2-dipalmitoyl-sn-glycero-3-phosphatidylethanolamine and 1 , 2-dipalmitoyl-sn-glycero-3-phosphatidylglycerol , POPG/DAG , molar 90:10 mix of phosphatidic acid and diacylglycerol; cholesterol was at 10% molar ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 005 Such dramatic site-specificity suggested that sequence binding positions Spitz in the active site , as occurs with soluble proteases . However , when we examined proteolysis in detergent micelles , we found that the cleavage site in Spitz also shifted +3 residues deeper into the TM segment ( Figure 1B ) . The shift was even more dramatic with APP + Spi7 , an engineered substrate that harbors seven TM residues of Spitz within the C-terminal 99 residues of human APP ( Urban and Wolfe , 2005 ) . In fact , some rhomboid enzymes abandoned the natural AS entirely in favor of cleavage +3 and/or +5 residues deeper ( Figure 1B , also see Figure 1—figure supplement 1B ) . Without exception analyzing both N- and C-terminal cleavage products revealed that each substrate is cut only once in vitro without successive trimming , but the cut site is free to shift in position ( Figure 1B , also see Figure 1—figure supplement 1C ) . Notably , small residues flanking the cleavage site ( P1/P1′ ) were strongly preferred . We found that the membrane itself is the basis for the discrepancy in site-specificity in cells vs in detergent micelles; reconstituting pure rhomboid and substrate in vitro from detergent into defined proteoliposomes restored cleavage to the natural site in Spitz , and even in APP + Spi7 ( Figure 1C ) . Reconstitution into proteoliposomes comprised of a wide variety of lipids all restored site-specificity ( Figure 1—figure supplement 2 ) , revealing that the composition of the membrane affects the efficiency of proteolysis , but not its site-specificity . Therefore , the uncompromising site-specificity of rhomboid proteases is not an inherent property of the enzyme itself , but rather results from the membrane somehow directing the position of cleavage . Reconstituting rhomboid and substrate into proteoliposomes from detergent micelles also revealed a second role for the membrane in substrate discrimination . Distal GA residues are a hallmark requirement for Spitz cleavage in cells ( Urban and Freeman , 2003 ) , and these residues were also important for cleavage in vitro when both rhomboid and substrate were reconstituted into proteoliposomes ( Figure 1D ) . In contrast , cleavage of a GA mutant substrate was rescued to nearly wildtype levels in detergent micelles , suggesting that the membrane plays a direct role in restricting substrate specificity in addition to specifying the cleavage site . Since proteolysis in detergent was notably plastic , we investigated the protein dynamics of both the protease and substrates , neither of which have yet been studied for any intramembrane protease . We functionally identified TM5 of GlpG as part of the lateral gate for substrate access to the active site ( Baker et al . , 2007; Urban and Baker , 2008 ) , although other models have also been proposed ( Ha , 2009 ) . Since cleavage sites shifted only deeper into the TM segment , we examined gate dynamics . We introduced a nitroxide spin label onto TM5 at W236 , a position we previously identified to be key for gating , and onto the overlying extramembraneous Cap loop at M247 as a control . The particular W236 and M247 sites were also attractive because neither contribute to GlpG's structural stability ( Baker and Urban , 2012 ) . We then monitored dynamics directly at these sites using electron paramagnetic resonance ( EPR ) spectroscopy . As expected , we observed two spectral components: a dynamic form ( α in Figure 2A ) and an immobilized form ( β in Figure 2A ) . These are consistent with the gate-open and gate-closed conformations , respectively , observed by X-ray crystallography ( reviewed in Urban , 2010 ) . Interestingly , the relative proportion of these two forms changed when GlpG was in different environments . Both TM5 and Cap sites were readily observed in the highly-dynamic form when GlpG was in the detergent-solubilized state . However , while the Cap site retained a high proportion in the mobile form when GlpG was reconstituted into proteoliposomes ( Figure 2B ) , the TM5 position shifted almost completely to the strongly restrained form ( Figure 2A ) . This is consistent with conversion to a predominantly gate-closed form , indicating that the membrane confers site-specificity ( Figure 1C ) by restricting gate dynamics ( Figure 2B ) . 10 . 7554/eLife . 00173 . 006Figure 2 . The membrane restrains rhomboid gate dynamics . Side-view of GlpG ( left ) showing positions ( in spheres ) of nitroxide spin probes . EPR spectroscopy was conducted at 37°C in 0 . 5% DDM detergent or proteoliposomes formed from E . coli lipids . Shown are 100G wide first derivative spectra with the relative signal intensity between samples normalized by quantifying the absolute number of spins . Vertical dashed lines denote magnetic field value positions . ( A ) Dynamics at the W236 gate position in DDM detergent vs proteoliposomes: note the dramatically increased amount of the immobile β component when GlpG was analyzed in proteoliposomes ( red arrow ) relative to in DDM detergent . ( B ) Proteins dynamics at the M247 Cap position showed a major proportion in the dynamic form when GlpG was analyzed both in DDM detergent and in proteoliposomes . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 006 We next probed the structural properties of rhomboid substrates directly by examining long peptides corresponding to the entire TM segments of Providencia TatA , the only known bacterial substrate ( Stevenson et al . , 2007 ) , and APP + Spi7 , as well as their corresponding mutants by circular dichroism ( CD ) spectroscopy . CD is a powerful tool for studying TM structure , but has never been applied to the analysis of intramembrane proteolysis . Interestingly , although distal helix-destabilizing residues were required for proteolysis in proteoliposomes , both wildtype and mutant TM peptides reconstituted into proteoliposomes formed helices of indistinguishable stability as revealed by overlapping ellipticity troughs at 208 and 222 nm ( Figure 3A ) . Moreover , oriented CD analysis revealed comparable spectra for both wildtype and mutant TM peptides ( Figure 3—figure supplement 1 ) , suggesting that the tilt of the substrate and uncleavable TM segments in the membrane is similar . The knowledge-based Ez algorithm ( Schramm et al . , 2012 ) also failed to detect any tilt or structural differences between these wildtype and mutant TM segments . 10 . 7554/eLife . 00173 . 007Figure 3 . The membrane restrains substrate TM dynamics . ( A ) CD spectroscopy of substrate ( red traces ) and non-substrate ( blue traces ) TM peptides revealed both display similarly stable helices when reconstituted into proteoliposomes , which was dramatically higher than in DDM detergent micelles ( green dashed traces ) . ( B ) CD spectroscopy of substrate TM peptides in detergent micelles ( red traces ) revealed them to be strongly reduced in helicity compared to non-substrates ( blue traces ) ; in comparison , non-substrates formed helices in detergent micelles similar in stability to those in the helix-inducing solvent TFE ( black dashed traces ) . All values are mean residue ellipticity , with relative peptide concentrations determined by simultaneously monitoring actual peptide bond absorbance during CD scanning . The TatA GSP-VVL mutant is G11V + S12V + P13L . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 00710 . 7554/eLife . 00173 . 008Figure 3—figure supplement 1 . Oriented CD Spectroscopy of wildtype and mutant APP and TatA TM segments in phospholipid bilayers . TatA ( A ) and APP ( B ) TM peptides in oriented bilayer sheets were analyzed by CD spectroscopy . Oriented CD spectra are shown in diamonds , with the red designating non-substrate TMs and blue designating rhomboid substrate TMs . The observed oriented spectra are consistent with TM peptides adopting a transbilayer orientation with no obvious differences in tilt angle between wildtype and mutant pairs . Although anisotropic , the green spectra ( TM peptides in trifluoroethanol ) are provided to illustrate the expected pattern for TM peptides tilted a full 90° relative to the membrane normal . Oriented bilayers at a 80:1 lipid to peptide molar ratio were assembled as described in Qian S , Wang W , Yang L , Huang HW . 2008 . Biophysical J 94:3512–22 . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 008 However , the key difference was that rhomboid substrates actually rely on the membrane to form these stable helices . TatA was 31% less helical , and APP + Spi7 over threefold less helical , in DDM micelles compared to being reconstituted into proteoliposomes ( Figure 3A ) . This observation explains the relaxed need for helix-destabilizing residues for cleavage in detergent ( Figure 1D ) , because these TM segments are already partly unwound . Moreover , we found that this instability is a defining feature of rhomboid substrates relative to non-substrates; analysis in detergent micelles revealed that the APP + Spi7 peptide was a remarkable 36% less helical than the non-substrate APP , while TatA was 46% less helical than its uncleavable GSP-VVL mutant ( Figure 3B ) . Spectroscopic interrogation revealed for the first time that rhomboid substrates are unable to maintain a stable helix without the membrane , raising the possibility that differences in intrinsic TM dynamics first and foremost is the property that defines substrates , rather than serving a secondary role in exposing the substrate backbone for hydrolysis as currently thought ( Ha , 2009; Strisovsky et al . , 2009 ) . This dynamic nature could allow substrates , but not non-substrates , to enter the catalytic center for proteolysis to ensue . While this model explains our biophysical observations , it's based entirely on in vitro measurements . We therefore sought to test the physiological , as well as functional , relevance of our model in living cells by examining its central predictions . The distinguishing prediction of our model is that substrate position in the active site is dictated primarily by protein dynamics caused by residues of the TM segment that , when not supported by the membrane , disrupt helical stability and promote entry into the hydrophilic active site . Substrate position should therefore be predictably shifted by moving , enhancing , or limiting intrinsic substrate dynamics . In contrast , if rhomboid achieves specificity by regimented binding of a recognition motif like soluble proteases , then the site of cleavage must necessarily be fixed , while altering TM dynamics should only change the efficiency of cleavage . We tested this prediction with Drosophila rhomboid-1 ( DmRho1 ) , the natural Spitz protease , and E . coli GlpG , the best understood rhomboid protease , under physiological conditions . To reveal substrate position in the active site at the time of catalysis we mapped substrate cleavage sites generated in living cells , since a protease:substrate complex structure has never been achieved . First , we reasoned that we could increase substrate TM dynamics but leave unperturbed all residues thought to be important for Spitz ‘binding’ ( P4–P2′ ) by mutating two residues near the middle of the Spitz TM segment ( at P8′ and P9′ ) to glycine . Circular dichroism spectroscopy revealed that the two glycines dramatically decreased helical stability by >60% relative to wildtype Spitz ( Figure 4A ) . Next , we tested Spitz + GG cleavage and found it to be cleaved more efficiently than wildtype Spitz in Drosophila cells , but the cleavage site shifted +3 residues ( Figure 4B ) . The cleavage site also shifted in living E . coli cells by GlpG ( Figure 4—figure supplement 1A ) . This effect was not limited to glycine , since changing a distal cysteine ( at P8′ ) to a helix-destabilizing serine also resulted in very efficient cleavage , but again induced a +3 residue shift in cleavage site ( Figure 4—figure supplement 1B ) . These shifts are particularly instructive because the natural sequence containing any possible binding motif is completely unperturbed , yet is abandoned for an ‘impermissible’ sequence ( Strisovsky et al . , 2009 ) with alanine and glycine at P4 and P2′ , respectively ( Figure 4C ) . Moreover , addition of helix-destabilizing residues at other positions could also induce shifts: mutating the P2′ isoleucine to glycine caused a +1 shift ( Figure 4D ) . 10 . 7554/eLife . 00173 . 009Figure 4 . Substrate dynamics position the cleavage site by Drosophila rhomboid-1 in living cells . ( A ) CD spectroscopy revealed that incorporating two glycines ( +GG ) near the middle of the Spitz TM further reduces its helical stability . ( B ) Western analysis of Spitz mutant processing by DmRho1 in living Drosophila cells assessed by amount of Spitz released into the media ( top left panel ) . In vivo cleavage sites of Spitz + GG revealed a +3 site shift in cleavage . Black triangles denote IgG light chain from the immunoisolation . ( Figure 4—figure supplement 1 shows cleavage of Spitz C146S , and a +5 site shift in cleavage of Spitz + GG by GlpG in E . coli cells ) . ( C ) Alignment of different Spitz mutant cleavage sites generated by DmRho1 in living cells . All mutants were cleaved efficiently ( in B ) , yet note the dramatic redistribution of residues at P4 , P3 , P2 , and P2′ positions as cleavages sites shifted . ( D ) The cleavage site of the Spitz I140G mutant shifted +1 residues with Drosophila rhomboid-1 . Mutating the distal , helix-destabilizing G143 residue to leucine shifted cleavage site −1 residues outwards . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 00910 . 7554/eLife . 00173 . 010Figure 4—figure supplement 1 . TM protein dynamics position the cleavage site . ( A ) Incorporating two glycine residues ( +GG ) deep within the Spitz TM segment resulted in a +5 shift in cleavage site with GlpG in E . coli cells . ( B ) Mutating a single cysteine at position 146 to serine deep within the Spitz TM segment and far away from the cleavage site ( at P8′ ) resulted in a +5 shift in cleavage site with Drosophila rhomboid-1 in living cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 010 Second , since we found site-specificity and gate dynamics were inversely correlated , we next examined whether gate-open mutants cause cleavage site shifts under physiological conditions . Indeed , all gate-open GlpG mutants ( Baker et al . , 2007; Urban and Baker , 2008 ) analyzed in living E . coli shifted the Spitz cleavage site +3 residues deeper into the TM segment ( Figure 5A ) . The effect of these mutations is unlikely to be explained by interfering with any putative recognition-binding site on the protease , because all gate-open mutants , irrespective of position , resulted in cleavage site shifts ( Figure 5—figure supplement 1A ) . In fact , wildtype GlpG in detergent produced the same +3 and +5 cleavage sites as gate-open mutants ( Figure 5C , also see Figure 5—figure supplement 1B ) , suggesting that , in the absence of the membrane , even the wildtype gate opens fully ( which is also independently supported by EPR analysis ) . 10 . 7554/eLife . 00173 . 011Figure 5 . Rhomboid gate and substrate dynamics position the cleavage site by bacterial rhomboid proteases . ( A ) Lateral view of GlpG showing TM5 interfacing sidechains ( boxed ) whose mutation opens the substrate gate ( red ) and increases protease activity . Gate-open mutants of GlpG analyzed in E . coli shifted cleavage site deeper into the Spitz TM segment ( red ) . Cleavage of Spitz with helix-destabilizing S142 mutated to alanine by gate-open GlpG in E . coli cells produced a complete shift towards the top of the TM segment ( blue ) . ( B ) Cleavage of Spitz with G143 , and G143 + A144 , mutated to leucine by gate-open GlpG in E . coli cells produced a gradual shift in cleavage site outwards . ( C ) Wildtype and gate-open ( F153A + W236A ) GlpG proteases produced identical , deeper cleavage sites when assayed in detergent micelles , indicating that the gate is fully open in the absence of the membrane . ( D ) Mutating the distal GA residues to LM also shifted the cleavage of APP + Spi7 in DDM detergent micelles to the natural AS site . Under these conditions the gate of wildtype GlpG is ‘open’ by virtue of the membrane being absent . ( E ) Diagram illustrating the ‘turn propensity’ effect of incorporating a helix-destabilizing/membrane-exiting residue ( asterisk ) into a long TM segment ( Monné et al . , 1999b ) . Lower diagram proposes an analogous effect for intramembrane proteolysis: residues of high ‘turn propensity’ could promote lateral substrate partitioning into the rhomboid active site . Right: change in turn propensity of substrate mutants plotted against the change in cleavage site occurring in natural membranes of living cells . ( F ) A hexapeptide encompassing the entire recognition motif ( P4–P2′ ) of P . stuartii TatA , the most efficient rhomboid substrate , failed to block cleavage of two different substrates by any rhomboid tested in detergent or reconstituted liposomes ( see Figure 5—figure supplement 2 for APP + Spi7 cleavage in liposomes and TatA cleavage in detergent ) . Black and green triangles denote substrate and cleavage product bands , respectively . The highest tested peptide concentration was 1 mM , while substrates were maintained at ≤1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 01110 . 7554/eLife . 00173 . 012Figure 5—figure supplement 1 . Cleavage site shifts with gate-open mutants . ( A ) Cleavage site of Spitz by the gate-open TM5 mutant W236G , TM2 mutant F153A , and triple TM5 mutant L229V + F232V + W236V , in living E . coli cells shifted deeper into the transmembrane segment between the second AS pair for all gate-open mutants irrespective of their position . ( B ) The gate-open W236G mutant of GlpG assayed in 0 . 1% DDM detergent produced identical , deeper cleavage sites as the wildtype enzyme ( see Figure 5C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 01210 . 7554/eLife . 00173 . 013Figure 5—figure supplement 2 . Poor competitive inhibition of proteolysis by a 1000-fold excess of a recognition motif peptide . A hexapeptide ( Ac-IATAAF-amide ) corresponding to the entire recognition motif ( P4–P2′ ) of TatA , the most efficient substrate for any rhomboid protease , did not block substrate proteolysis in vitro even at extreme ( 1 mM ) concentrations . Black and green triangles denote substrate and cleavage product bands , respectively . The highest tested peptide concentration was 1 mM , while substrates were maintained at ≤1 μM . Also see Figure 5F . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 013 Third , we ‘limited’ substrate dynamics , which our model predicts should decrease substrate ‘reach’ into the rhomboid active site , and thereby shift cleavage to the top of the substrate TM . Strikingly , the cleavage site of Spitz by gate-open GlpG in vivo shifted completely −3 residues when helix-destabilizing S142 alone was replaced by alanine ( Figure 5A ) . Moreover , a partial −3 shift also occurred when we replaced G143 with leucine , and almost fully to the outer AS when both GA residues were replaced with leucine ( Figure 5B ) . A shift also occurred with Drosophila rhomboid-1 ( Figure 4D ) . Importantly , we observed the same shifts with mutant substrates and wildtype GlpG in detergent micelles , in which the gate is open by virtue of the membrane being absent ( Figure 5D ) . Therefore , even when the gate is open , helix-destabilizing residues are required for substrates to enter rhomboid's active site . Although we made as conservative mutations as possible with respect to size when altering TM dynamics , mutants should be interpreted with caution because they can also have unintended effects , including altering the TM surface and/or interface with rhomboid . To evaluate further what physical property is most likely responsible for the effects on proteolysis , we searched for a correlation between cleavage site shifts in living cells vs changes that our mutants made in residue properties using several independently generated physical scales . We found the strongest correlation with a ‘TM turn propensity’ scale ( Monné et al . , 1999a , 1999b ) . This scale had been derived by placing guest residues into the center of a long TM segment and measuring whether the residue prefers to stay in the middle of the membrane in the TM helix , or moves outside the membrane in a turn that breaks the long TM helix into two shorter TM helices ( Figure 5E ) . The quantified propensity scale displayed a correlation coefficient of 0 . 86 with the direction and degree of cleavage site shifts that we observed with 12 mutant substrates cleaved by rhomboid proteases in cellular membranes ( Figure 5E ) . Taken together , these observations , conducted in living cells , reveal that gate and TM dynamics , rather than binding of a specific sequence within Spitz , play the predominant yet completely overlooked role in positioning substrates into the active site . The new cleavage sites further revealed dramatic re-alignments of the substrate in the rhomboid active site ( Figure 4C ) , even when the natural putative recognition was unperturbed , such that stereotypical binding of a complementary recognition sequence between rhomboid and substrates is unlikely to be the main driving force for protease specificity . But to evaluate this further , we examined the ability of the proposed P4–P2′ recognition binding region of TatA , the most efficient bacterial substrate , to inhibit proteolysis competitively by a panel of bacterial rhomboid enzymes in vitro both in detergent micelles and reconstituted into proteoliposomes . Even at millimolar concentration ( ∼1000× the substrate concentration ) , the peptide did not block rhomboid proteolysis of TatA or APP + Spi7 substrates ( Figure 5F and Figure 5—figure supplement 2 ) , independently suggesting that sequence-specific binding is not the main feature of rhomboid specificity . Discovering rhomboid specificity is driven by exposing intrinsic TM dynamics raised an independent prediction: non-substrates of various sequence should be converted into substrates simply by increasing their intrinsic TM dynamics . APP , Delta and TGFα have been characterized genetically , cell biologically , and in vitro with pure proteins as non-substrates for rhomboid enzymes ( Peschon et al . , 1998; Urban and Freeman , 2003; Lemberg et al . , 2005; Urban and Wolfe , 2005; Adrain et al . , 2011 ) . We reasoned that incorporating a single proline , which has the highest helix-and-membrane exit propensity of all residues ( Monné et al . , 1999a , 1999b ) , might increase their intrinsic TM dynamics . Indeed , CD analysis revealed that incorporating a single proline into the TM of APP was sufficient to decrease TM helix stability by over twofold ( Figure 6A ) . Remarkably , installing single prolines alone was sufficient to convert APP , Delta and TGFα into efficient rhomboid substrates in vivo and in vitro ( Figure 6B , C ) . In fact , intramembrane cleavage become so efficient that in all cases it outcompeted the natural juxtamembrane cleavage by metalloproteases , and in the case of Delta , little unprocessed full-length protein could be detected . Cleavage was dependent on rhomboid activity , because it required the catalytic serine of both Drosophila and human rhomboid proteases , and γ-secretase inhibition did not affect cleavage ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 00173 . 014Figure 6 . A single proline converts non-substrates into rhomboid substrates . ( A ) CD spectroscopy revealed that incorporating a single proline into the TM of APP ( at position 7 ) dramatically reduces its helical stability . ( B ) Incorporating a single proline into the TM segment of Delta , TGFα , or APP converted each into an efficient substrate for rhomboid proteases ( DmR is Drosophila rhomboid-4 , HsR is human RHBDL2 , SA is the catalytic serine mutant of DmR ) . ( − ) shows non-specific anti-Flag bands in untransfected HEK293 cells . The natural juxtamembrane cleavage product generated by metalloprotease shedding ( denoted by a white triangles ) was outcompeted by intramembrane cleavage . ( C ) Incorporating a single proline also converted APP-Flag into an efficient substrate in vitro with pure bacterial rhomboid enzymes ( cleaved products denoted with asterisks ) . ( D ) Effect of proline position on rhomboid proteolysis ( shown are cleaved product bands , highlighted with asterisks ) . Cleavage sites of Delta + Pro-Flag and GFP-APP + Pro-Flag with different proline positions were mapped from living HEK293 cells ( right panel ) . TM segment residues are underlined and the exogenous proline is boxed . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 01410 . 7554/eLife . 00173 . 015Figure 6—figure supplement 1 . Induction of non-substrate cleavage by rhomboid proteases . Cleavage of non-substrates harboring transmembrane proline depends on rhomboid protease activity not γ-secretase . We examined both the human ( above shown with Delta ) and Drosophila ( above shown with APP ) rhomboid proteases . ‘SA’ denotes co-transfection of HEK293 cells with rhomboid protease constructs harboring the catalytic serine mutated to alanine . These catalytically dead constructs failed to induce cleavage ( highlighted with red asterisks ) . DAPT was added to a final concentration of 10 μM , which is approximately 100-fold higher than its reported IC50 against γ-secretase , yet it did not block proteolysis . ‘UN’ designates control untransfected cells , which serve to identify cross-reactive background bands . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 015 By scanning the proline into each of the first 10 TM positions , we found multiple sites and surprising sequence diversity could accommodate rhomboid cleavage . Although incorporating proline in any TM position between 2 and 10 resulted in Delta and TGFα cleavage , APP displayed a more limited profile of acceptable proline site position . This may reflect influence by neighboring sequence context on overall helical stability and/or propensity to partition into the active site ( Li et al . , 1996; Yang et al . , 1997 ) . Intriguingly , the relative cleavage site preference was dependent on the position of the proline , and moved deeper as the proline position descended into the TM segment ( Figure 6D ) . These observations provide compelling evidence that simple changes in intrinsic TM dynamics under physiological conditions drive substrate specificity for membrane-immersed proteolysis . Immersion of a hydrolytic reaction within the hydrophobic membrane has long been studied as a biochemical conundrum ( Erez et al . , 2009 ) , yet the possible advantage of this arrangement has not been explored . We integrated diverse analytical approaches including spectroscopy , defined reconstitution systems , and cleavage site mapping in living cells to probe the role of the membrane . These new approaches ultimately converged to reveal that the main property conferred by membrane-immersion is the ability to identify substrates through a mechanism centered on exposing intrinsic TM dynamics instead of the protein-protein binding strategy used by other specific proteases ( Figure 7 ) . 10 . 7554/eLife . 00173 . 016Figure 7 . Model of rhomboid proteolysis driven by intramembrane protein dynamics . The membrane imposes two constraints on protein dynamics to ensure high proteolytic specificity; it induces helix formation of TM segments ( left red arrows ) and limits rhomboid gate ( light blue ) opening ( right red arrow ) . Substrates form a stable helix only in the membrane; partial exposure to the aqueous environment within rhomboid triggers an entropy-driven conformational switch , facilitated by helix-destabilizing residues , allowing substrates to reach the catalytic residues ( in orange ) . Bottom panel depicts a non-substrate:rhomboid complex , in which the TM segment maintains a stable helix and therefore cannot reach the catalytic residues . The non-substrate TM segment eventually dissociates without being cleaved ( far right ) . Induction of efficient non-substrate cleavage suggests that the initial docking interaction between rhomboid and TMs is non-specific . The exact order of events , and what triggers each step , remain speculative . Membrane thinning surrounding GlpG as observed in molecular dynamics simulations is illustrated ( Bondar et al . , 2009; Zhou et al . , 2012 ) , although its functional consequence remains unclear . Structures 2IC8 ( closed GlpG ) , 2NRF ( open GlpG ) , 1MOX ( Spitz-EGF ) , and 2TGF ( TGFα-EGF ) were used to diagram the model . DOI: http://dx . doi . org/10 . 7554/eLife . 00173 . 016 Applying CD spectroscopy provided the first opportunity to measure the structural properties of rhomboid substrate TMs relative to non-substrates , and in different environments . Coupling this information with examining proteolysis in living cells suggests that the main factor keeping many type I TM segments from becoming rhomboid substrates is their ability to maintain a stable TM helix . Differences between substrate and non-substrate TM helices are minimal while they reside in the membrane . However , a defining property of rhomboid substrates is a meta-stable TM helix that actually relies on the membrane for stability , and unravels without it . Mutations that stabilize TM helices compromise proteolysis , while simply introducing helical instability into non-substrate TMs , as measured directly by CD , converted three of three unrelated type I membrane proteins into substrates for seven different rhomboid proteases despite diversity in the cleaved sequences . It is worth emphasizing that mutational analysis has played an important role in our study of rhomboid proteolysis , and while this is a proven strategy for elucidating enzyme mechanisms , all of our mutants that change helicity also necessarily change sequence . To limit this inherent drawback , we were careful , whenever possible , to make conservative mutations , measure effects on helicity directly by CD , and ultimately base our analysis on >30 mutations . Nevertheless , we further evaluated what underlying physical property drives the cleavage site shifts and ultimately found the strongest correlation with a ‘turn propensity’ scale . The informative feature of this novel scale is that it integrates both the helical propensity of a residue , as well as its preference to be inside the membrane versus seeking a more hydrophilic environment ( Monné et al . , 1999a , 1999b ) . As such , the intrinsic dynamics in rhomboid substrate TMs thus likely derives from a combination of residues that destabilize the TM helix structure directly , as well as those of limited hydrophobicity that increase the likelihood of this polypeptide region escaping the helix-inducing environment of the membrane ( in favor of the hydrophilic active site of rhomboid , see below ) . This observation independently suggests that the key differences in TM dynamics are evaluated in a different , non-membranous environment . This may explain why alanines are known to be important for rhomboid proteolysis yet themselves are not helix-destabilizing directly , and require mutation to large hydrophobic residues to compromise proteolysis . The strong correlation with the turn propensity scale could also have predictive value for finding new rhomboid substrates by sequence analysis . However , it should be noted that our analysis had the benefit of quantifying ‘changes’ in turn propensity ( ‘∆ turn propensity’ in Figure 5E ) between two sequences that differ at only one or two residues . In practice , the turn propensity of a residue has been shown to be altered in non-linear ways by differences in TM segment length , the local sequence context , and the exact position of the residue within the TM segment ( Monné et al . , 1999b ) . If these confounding parameters also apply to rhomboid proteolysis , accurately predicting absolute propensities of natural TM segments from sequence alone may present challenges . Nevertheless , cautious optimism is warranted , since these challenges could be overcome by additional scale refinement with rhomboid proteases directly . While our analyses consistently indicate TM dynamic state is the defining feature of rhomboid substrates , they do not neglect that sequence also contributes a secondary role . The shifts that we observed reveal sequence requirements for proteolysis with positive data , because they mark substrate position in the active site at the time that catalysis was proceeding efficiently . Preference for cleavage to shift to between small residues re-affirmed prior studies ( Urban and Freeman , 2003; Akiyama and Maegawa , 2007 ) . However , little appears to be essential for cleavage beyond small P1/P1′ residues , since we found a great diversity of residues at other positions allowed efficient proteolysis to proceed by multiple rhomboid proteases . This is particularly informative , because prior analyses necessarily focused on mutations that block cleavage , but could not rule out the possibility that they interfere with proteolysis indirectly , for example , by promoting TM oligomerization ( which is required for TatA function ) . By incorporating a single proline at various positions we were able to convert three unrelated proteins into rhomboid substrates . Nevertheless it is important to note that this too does not mean that any TM sequence can become a rhomboid substrate . Rather , it highlights that the natural sequence diversity in many TM segments provides ample opportunities for finding acceptable cleavage sites . True substrates would nevertheless be expected to have further sequence optimization that would be specific to a particular rhomboid protease , because greater proteolytic efficiency would be favored as substrate and protease co-evolve . One current example might be TatA proteolysis , because TatA cleavage is thought to proceed rapidly and automatically for quorum sensing to operate ( Stevenson et al . , 2007 ) . However , this is unlikely to be representative of most rhomboid functions . Even so , a peptide comprised of all P4–P2′ residues requires millimolar concentrations to affect proteolysis of TatA by AarA , indicating that even in this context recognition sequence binding is not the main determinant for specificity . Considered together , our observations suggest a new working model in which canonical rhomboid proteases patrol the membrane for meta-stable TM helices for cleavage . A gate-open rhomboid may be sufficient to provide a microenvironment in which ‘intrinsic’ TM segment differences can be unmasked , and recent molecular dynamics simulations and biophysical measurements indicate that this is a stable rhomboid conformation in the membrane ( Baker and Urban , 2012; Zhou et al . , 2012 ) . Because of both helix-destabilizing residues and limited hydrophobicity , substrates are poised to exit from the membrane and partition into the hydrophilic rhomboid active site . Proteolysis then ensues because such extended TMs are susceptible to proteolysis , and/or are able to ‘reach’ the internal catalytic apparatus . Such substrate ‘partitioning’ is consistent with a strong correlation with a ‘turn propensity scale’ , and may further explain the unexpected observation that scanning a single proline along the TM segment moves the cleavage site deeper , at times positioning it at unacceptable residues . Importantly , the central yet overlooked component of this specificity system is the membrane itself , which limits proteolysis both by inducing TMs to form helices and restricting gate-opening ( Figure 7 ) . Ultimately our analyses have both coalesced into a general framework for how rhomboid intramembrane proteolysis functions , and emphasize that further work is required to define the specific details of how this complex system operates . Differences between substrates and non-substrates in the membrane beyond the resolution of our experimental approaches are possible , although direct evidence suggests that the most dramatic difference occurs when TMs leave the influence of the membrane . Moreover , structural analysis of the rhomboid-substrate complex is required to define the fine-detail interactions that mediate proteolysis , although this might prove particularly challenging if true substrates are indeed intrinsically dynamic . Finally , the precise order of events , and what triggers each step in the cleavage reaction , also remain unclear . It's well recognized that proteolytic release of factors from the membrane regulates many signaling networks . Yet dozens of examples reveal that simply anchoring a protease domain to the membrane through a TM segment satisfies these needs ( Blobel et al . , 2009; Antalis et al . , 2010 ) . So why did complicated , membrane-immersed enzymes evolve and become so wide-spread across all forms of life ? Our analyses indicate that rhomboid proteases achieve substrate specificity first through reading TM dynamics , which endows them with different properties relative to soluble proteases . Intramembrane proteases might therefore be a distinct group of enzymes in the cell , not because they are proteases that release proteins from the membrane , but because they are membrane-immersed; they live in a world with different rules , giving the cell a set of enzymes with unique properties that it can harness for evolving new functions . Rhomboid proteases and substrates were expressed in E . coli , purified , and cleavage reactions were conducted at 37°C for 1–2 hr in 50 mM Tris–HCl pH 7 . 4 , 150 mM NaCl , and either reconstituted into proteoliposomes or in 0 . 1% DDM as described previously ( Urban and Wolfe , 2005 ) . Reaction products were resolved and quantified by infrared fluorescence ( LiCor Biosciences , Lincoln , NE ) using western analysis ( Baker et al . , 2007 ) . Substrates from in vitro proteolysis assays or transformed/transfected cells ( lysed in RIPA buffer ) were subjected to anti-Flag immunoaffinity purification with the M2 resin ( Sigma , St Louis , MO ) , and analyzed by MALDI-TOF mass spectrometry using sinapinic acid as the matrix as described previously ( Baker et al . , 2007 ) . Changes in turn propensity of substrate mutants was calculated by subtracting the average turn propensity value ( from table 1 in Monné et al . , 1999b ) of the wildtype residue from the mutant residue . This value was plotted against the change in cleavage site of the mutant substrate whereby a value of 1 represents a complete shift in cleavage site while values <1 signify the proportion of cleavage occurring at the new site ( s ) . Positive values denote a C-terminal shift ( deeper into the TM ) while negative values indicate an N-terminal shift ( outward shift ) . Cysteines were introduced at positions 236 ( TM5 ) or 247 ( L5 loop ) of GlpG in which the endogenous C104 was mutated to alanine . Proteins were expressed and purified as described previously ( Wu et al . , 2006 ) , and labeled with 250 μM MTSL ( Toronto Research Chemicals , Canada ) . Free probe was removed by gel filtration chromatography , followed by NiNTA affinity chromatography and washing for 2–3 days at room temperature . X-band EPR spectra of samples in 0 . 9-mm quartz capillaries were recorded at 37°C ( 310 K ) on a Bruker EMXmicro spectrometer equipped with a PremiumX ultra low noise microwave bridge , a high-sensitivity ER4119HS resonator , and an ER4141VT temperature control unit ( Bruker Biospin , Billerica , MA ) . Spectra were background corrected and subtracted , and normalized by the absolute number of spins in each sample as quantified by double integration using Xenon software ( Eaton et al . , 2010 ) . Thirty-two residue long peptides containing the entire TM sequence ( with flanking lysines to increase solubility ) were dissolved at 10–105 μM in 95% trifluoroethanol ( Sigma , St Louis , MO ) 1 mM DTT , or 10 mM Hepes pH 7 10 mM NaCl 1 mM DTT containing either 0 . 25% DDM or 1% SDS . Peptides were electrophoresed on 16% tricine gels to verify concentration and lack of aggregation , while peptides reconstituted into proteoliposomes were further examined by ultracentrifugation . Ellipticity and UV absorbance at 205 nm ( to quantify peptide concentration accurately during scanning ) were measured simultaneously at 25°C through a 0 . 2 mm path length cuvette in a Jasco J-810 spectropolarimeter ( Jasco Inc . , Easton , MD ) . Analyses were conducted by averaging 10 scans at 50 nm/min , background was determined and subtracted , and mean residue ellipticity was calculated . Drosophila S2R+ and human HEK293 cells were transiently transfected with plasmids for the expression of GFP-tagged and/or Flag-tagged substrates and 3× HA-tagged rhomboid proteases ( Baker et al . , 2006 ) . For non-substrates , we used GFP fused to the C-terminal most 99 residues from APP , GFP fused to full-length TGFα , and full-length Drosophila Delta ( all constructs had a single Flag tag at their C-terminal ends ) . 24 hr after transfection , serum-free media was conditioned for an addition 18–24 hr . Media and cell samples were analyzed by quantitative westerns . E . coli cells were transformed with two plasmids for the inducible expression of bacterial rhomboid proteases and Flag-tagged substrates , grown under double antibiotic selection , and induced with IPTG as described ( Urban and Baker , 2008 ) .	Proteases are enzymes that break the peptide bonds that hold proteins together , and have a central role in many physiological processes , including digestion , blood clotting and programmed cell death . An important characteristic of proteases is that they are highly selective , only cutting proteins that contain well-defined sequences of amino acids in accessible regions . Proteases that are soluble in water have been studied for over a century and are now well understood , as are proteases that need to be tethered to the membrane of a cell to work properly . In 1997 researchers discovered a protease that was immersed in the cell membrane , and it soon became clear that these intramembrane proteases were widespread and involved in a wide range of processes in cells . Examples of intramembrane proteases include γ-secretase , which is implicated in Alzheimer's disease , and various site-2 proteases that regulate pathogenic circuits in bacteria . There are many similarities between soluble and intramembrane proteases . However , given that intramembrane proteases evolved within the hydrophobic environment of the membrane , whereas soluble proteases evolved in an aqueous environment , there should there should also be significant differences between them . The best understood intramembrane proteases in terms of their biochemistry are probably the rhomboid proteases . However , most studies of their function have been performed in detergent systems rather than in real membranes . Moin and Urban now report that the main strategy used by rhomboid proteases to identity the proteins that they selectively cut is completely different from that used by soluble proteases . Through a combination of biochemical and spectroscopic methods , they have discovered that rhomboid proteases identify the proteins they act on mainly by detecting changes in dynamic behavior: only those proteins that lose a stable helical structure when they exit the lipid phase to interact with the rhomboid protease will be cut by the rhomboid protease . Soluble proteases , on the other hand , achieve specificity by looking for proteins with a particular sequence of amino acids . The novel strategy used by rhomboid proteases allows them to patrol the membrane for unstable helices and selectively cut them . This discovery provides the first explanation of why these complicated enzymes evolved to have active sites immersed within the cell membrane .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2012	Membrane immersion allows rhomboid proteases to achieve specificity by reading transmembrane segment dynamics
Carboxysomes are protein-based bacterial organelles encapsulating key enzymes of the Calvin-Benson-Bassham cycle . Previous work has implicated a ParA-like protein ( hereafter McdA ) as important for spatially organizing carboxysomes along the longitudinal axis of the model cyanobacterium Synechococcus elongatus PCC 7942 . Yet , how self-organization of McdA emerges and contributes to carboxysome positioning is unknown . Here , we identify a small protein , termed McdB that localizes to carboxysomes and drives emergent oscillatory patterning of McdA on the nucleoid . Our results demonstrate that McdB directly stimulates McdA ATPase activity and its release from DNA , driving carboxysome-dependent depletion of McdA locally on the nucleoid and promoting directed motion of carboxysomes towards increased concentrations of McdA . We propose that McdA and McdB are a previously unknown class of self-organizing proteins that utilize a Brownian-ratchet mechanism to position carboxysomes in cyanobacteria , rather than a cytoskeletal system . These results have broader implications for understanding spatial organization of protein mega-complexes and organelles in bacteria . Bacterial microcompartments ( BMCs ) are protein-based organelles that encapsulate specialized metabolic processes in ~20% of all sequenced bacteria ( Axen et al . , 2014 ) . Over 23 different classes of BMC are broadly distributed across bacterial phyla , but perhaps the best characterized class are the large ( ~175 nm ) carbon-fixing carboxysomes of cyanobacteria ( reviewed here: Rae et al . , 2013; Kerfeld and Melnicki , 2016 ) . Carboxysomes encapsulate the key Calvin-Benson-Bassham cycle enzyme , ribulose-1 , 5-bisphosphate carboxylase/oxygenase ( RuBisCO ) , and enhance the carbon fixation efficiency of this enzyme by creating a microenvironment that is enriched for CO2 . Specifically , carboxysomes consist of an outer ‘shell’ layer that encapsulates both RuBisCO and carbonic anhydrase: the localized conversion of bicarbonate to CO2 near RuBisCO is thought to greatly reduce its oxygenase activity , thereby suppressing the energetically-costly process of photorespiration ( Rae et al . , 2013 ) . In the model rod-shaped cyanobacterium Synechococcus elongatus PCC 7942 ( hereafter S . elongatus ) , it has been shown that carboxysomes align along the longitudinal axis of the cell ( Savage et al . , 2010 ) . This equidistant positioning is thought to promote equal inheritance of carboxysomes to both daughters of a dividing cell , which has been shown to impact cyanobacterial fitness under ambient CO2 ( Savage et al . , 2010; Rae et al . , 2013; Kerfeld and Melnicki , 2016 ) . Since cyanobacteria contribute to greater than 25% of global carbon-fixation , how carboxysome organization is maintained is of considerable ecological importance . However , the mechanisms used to position any BMC , including carboxysomes , has remained elusive . The most comprehensive study of BMC positioning to date showed that a ParA-type ATPase ( hereafter McdA - Maintenance of carboxysome distribution A ) displayed oscillatory dynamics in S . elongatus and was required for proper positioning of carboxysomes ( Savage et al . , 2010 ) . ParA family members have been most comprehensively studied as factors important for the segregation of genetic material; bacterial chromosomes and low-copy plasmids ( reviewed here: Baxter and Funnell , 2014 ) . Early hypotheses of ParA function favored a cytoskeletal model in which ParA formed filaments that self-assemble into a larger scaffold capable of segregating genetic cargo , analogous to a primitive mitotic spindle ( Ringgaard et al . , 2009; Ptacin et al . , 2010 ) . These models were supported in part by in vitro observations of fibrous , long bundled filaments of purified members of the ParA/MinD family of proteins ( Reviewed here: Vecchiarelli et al . , 2012 ) . Such filament-based models , combined with the requirement of McdA for carboxysome positioning , led to the prevailing theory that carboxysomes ( Savage et al . , 2010 ) and other BMCs ( Parsons et al . , 2010 ) could be tethered to a cryptic cytoskeletal element that traverses the cell length and/or exerts positioning forces . More recently , filament-based models of ParA have been challenged by experiments utilizing reconstituted cell-free systems ( Hwang et al . , 2013; Vecchiarelli et al . , 2013; Vecchiarelli et al . , 2014 ) , super-resolution microscopy ( Le Gall et al . , 2016; Lim et al . , 2014 ) , crystallography ( Zhang and Schumacher , 2017 ) , and mathematical modelling approaches ( Hu et al . , 2015; Hu et al . , 2017; Le Gall et al . , 2016; Lim et al . , 2014; Surovtsev et al . , 2016 ) . These new data are consistent with a model whereby asymmetric distributions of ParA dimers on the nucleoid drive directed and persistent movement of DNA cargos ( e . g . plasmids ) towards increased concentrations of ParA via a Brownian-ratchet mechanism that does not require a cytoskeletal element . ParA-mediated DNA segregation via this proposed mechanism requires only three factors: ( i ) a ParA-type ATPase that dimerizes and non-specifically binds the nucleoid in the presence of ATP ( Leonard et al . , 2005; Hester and Lutkenhaus , 2007; Castaing et al . , 2008; Vecchiarelli et al . , 2010 ) , ( ii ) a partner protein , ParB , that site-specifically binds DNA and stimulates the ATPase activity of ParA to displace it from the nucleoid ( Davis et al . , 1992; Bouet and Funnell , 1999; Bouet et al . , 2000 ) , and ( iii ) a centromere-like site on the DNA cargo ( parS ) that ParB loads onto ( Davis and Austin , 1988; Funnell , 1988 ) . In this model , multiple dimers of ParB form a large protein-DNA complex around the parS site , which leads to a break in ParA symmetry across the nucleoid due to the formation of local ParA depletion zones around individual ParB-bound cargos ( Adachi et al . , 2006; Hatano et al . , 2007; Hwang et al . , 2013; Vecchiarelli et al . , 2014 ) . In turn , transient ParA-ParB interactions could translate the asymmetrical distribution of ParA across the nucleoid into a directional cue for processive motion of cargo towards the highest local concentration of ParA . Despite extensive research into carboxysome biogenesis and organization , as well as efforts to identify the full complement of proteins associated with many BMC classes , the only evidence for an underlying cytoskeletal-mode of BMC organization is indirect . Brownian ratchet-based models are gaining favor within the ParA field , but there is not yet a broad consensus and there is substantial uncertainty as to whether ParA-type ATPases form functional filaments in vivo ( Wagstaff and Löwe , 2018 ) . Additionally , no factors suitable to play a role analogous to ParB or parS have been identified for McdA , preventing any direct assessment of a putative Brownian ratchet-based mechanism for carboxysomal organization . Therefore , whether McdA acts as part of a ParA-like system or utilizes a unique mechanism has remained an open question . Indeed , several fundamental questions remain unanswered in relation to carboxysome positioning , including: ( i ) Does McdA form a cytoskeletal structure or follow a Brownian-ratchet mechanism ? ( ii ) Upon what cellular surface does McdA bind in order to processively oscillate from pole-to-pole ? ( iii ) What factors contribute to the emergence of higher-order McdA patterning ? ( iv ) How do oscillations of McdA contribute to carboxysome distribution ? Here , we identify a novel factor , we term McdB ( Maintenance of carboxysome distribution protein B ) that localizes to carboxysomes via interaction with outer shell proteins and regulates carboxysome positioning within the cytosol . While McdB has no identifiable similarities with any known ParB-family members , we find that McdB can directly interact with McdA to stimulate its ATPase activity and release it from DNA in vitro , and promote its pole-to-pole oscillation in vivo . Changes in McdB expression resulted in loss of McdA oscillatory dynamics , loss of equidistant carboxysome positioning and alteration of carboxysome ultrastructure . Although several features of McdAB differ significantly from those of classic ParAB family members , we find that a Brownian ratchet model of localized concentration gradients of McdA on the nucleoid is consistent with our results and may also reconcile past observations of carboxysome positioning . We discuss our results in light of their implications for BMC positioning and biogenesis , as well as the insights that this carboxysome positioning system can provide for a broader class of self-organizing proteins including the ParAB family . In the DNA partition process , ParA-type ATPases successively bind ATP , dimerize , and bind non-specifically to DNA ( Leonard et al . , 2005; Hester and Lutkenhaus , 2007; Castaing et al . , 2008; Vecchiarelli et al . , 2010 ) . In vivo , this mechanism establishes the nucleoid as the biological surface upon which directed DNA cargo motion occurs ( Hatano and Niki , 2010; Sengupta et al . , 2010; Castaing et al . , 2008; Le Gall et al . , 2016 ) . In the model rod-shaped cyanobacterium S . elongatus , the ParA-like protein we call McdA ( Synpcc7942_1833; Maintenance of carboxysome distribution protein A ) is required for positioning carboxysomes via an unknown oscillatory mechanism ( Savage et al . , 2010 ) . However , the ParA family of ATPases contains MinD proteins , which bind membranes , ParA proteins , which bind DNA , and McdA has been broadly hypothesized to form filaments . Therefore , it remained unclear if McdA uses a biological surface to self-organize in the cell or if it forms a free-standing cytoskeletal network as previously proposed ( Savage et al . , 2010 ) . Since a C-terminal GFP fusion of McdA was previously used to observe McdA oscillation and its involvement in carboxysome positioning in vivo ( Savage et al . , 2010 ) , we purified McdA-GFP-His and examined its capacity to bind DNA via an Electrophoretic Mobility Shift Assay ( EMSA ) . We observed that McdA-GFP-His significantly shifted non-specific DNA ( nsDNA ) in the presence of ATP ( Figure 1A ) . To more directly examine the interaction of McdA with DNA , we used Total Internal Reflection Fluorescence Microscopy ( TIRFM ) to visualize McdA-GFP-His dynamics upon a DNA-coated surface; a technique that also has sufficient resolution to resolve oligomeric McdA filaments proposed to be involved in carboxysome positioning ( Savage et al . , 2010; Yokoo et al . , 2015 ) . A flowcell unit was decorated with nsDNA fragments ( ~500 bp sonicated salmon sperm DNA ) at high density ( ~1000 fragments/µm2 ) ( Vecchiarelli et al . , 2013 ) to create a visualizable DNA-coated surface . Consistent with EMSA analysis , McdA-GFP-His uniformly bound the DNA carpet when infused into the flowcell with ATP , ( Figure 1B and Video 1 ) , but showed no appreciable DNA binding when ADP or ATP-γ-S were added , or when nucleotides were omitted ( Figure 1C ) . No McdA filaments were observed forming under any of these conditions ( Video 1 ) . We then sought to examine endogenous localization and dynamics of McdA , therefore we generated N- and C-terminal fluorescent fusions of mNeonGreen ( mNG ) to McdA using the native mcdA promoter and chromosomal location . Interestingly , our native C-terminally tagged reporter ( McdA-mNG ) did not show dynamic oscillations and instead formed a uniform distribution of signal along the longitudinal axis ( ≥99% of cells; n = 950 cells ) ( Figure 1D , Figure 1—figure supplement 1A and Table 1 . ) . Alternatively , an N-terminally tagged reporter ( mNG-McdA ) displayed robust oscillations ( ≥99% of cells; n = 442 cells ) ( periodicity of 15 . 3 min per 3 . 3 µm , ~5–6 x faster than previously reported using McdA-GFP [Savage et al . , 2010] ) that formed a bimodal distribution of signal intensity ( Figure 1EG , Figure 1—figure supplement 1B and Video 2 ) . A carboxysome reporter was then generated by insertion of an additional copy of the small subunit of RuBisCO ( RbcS ) fused at the C-terminus to mTurquoise2 ( mTQ ) and expressed using a second copy of the native rbcS promoter . In this line , an average of 2 carboxysomes per micron of cell length was observed ( Figure 1—figure supplement 1C ) . Cells bearing the mNG-McdA construct maintained carboxysome positioning along the longitudinal axis ( ≥99% of cells; n = 374 cells ) ( Figure 1F , Figure 1—figure supplement 1D and Video 3 ) , indicating the N-terminal fusion fully complemented McdA’s known functions . To assay in vivo whether mNG-McdA could be binding the nucleoid , we stained the cyanobacterial nucleoid with 4′ , 6-Diamidine-2′-phenylindole dihydrochloride ( DAPI ) and recorded the mNG-McdA signal as it traversed the length of the cell . We found that the topology of the mNG-McdA signal closely resembled that of the DAPI-stained nucleoid ( Figure 1H ) , providing additional evidence that the nucleoid is the surface upon which McdA dynamics are occurring . Since an N-terminal fusion of McdA was more functional in vivo , we performed an EMSA with an N-terminal fusion of McdA . While we found that wild-type McdA and His-mNG-McdA were insoluble and prone to degradation , an N-terminal fusion of McdA to Maltose Binding Protein ( MBP ) ( His-MBP-McdA ) was highly soluble . Consistent with our McdA-GFP-His EMSA , His-MBP-McdA shifted nsDNA in an ATP-dependent manner ( Figure 1I ) . Together , our results demonstrate that ATP-bound N- or C-terminally tagged McdA binds DNA and McdA-GFP-His does not display indications of polymer formation at the resolution limits of our microscope . Traditional ParA-family members require cognate ParB proteins to stimulate their ATPase activity and promote oscillatory dynamics . Yet no ParB-like ortholog has been identified for McdA . Although no obvious chromosomally-encoded homolog of parB could be detected in S . elongatus , we identified a parB-like gene ( Synpcc7942_B2626 ) on the large plasmid ( pANL ) ( Figure 2A ) . However , deletion of pANL parB did not disrupt oscillation of mNG-McdA ( ≥99% of cells; n = 554 cells ) ( Figure 2B and Figure 1—figure supplement 1E ) . Two additional hypothetical genes were then selected due to their proximity to the mcdA gene , Synpcc7942_1834 and Synpcc7942_1835 ( Figure 2A ) . While deletion of Synpcc7942_1835 had no observable effect on mNG-McdA oscillation ( ≥99% of cells; n = 834 cells ) ( Figure 2C and Figure 1—figure supplement 1F ) , deletion of Synpcc7942_1834 resulted in complete loss of mNG-McdA dynamics ( ≥99% of cells; n = 373 cells ) . In the ΔSynpcc7942_1834 background , mNG-McdA concentrated in the center of the cell in the vicinity of the nucleoid , but without any consistent asymmetrical patterning ( Figure 2D and Figure 1—figure supplement 1G ) . To more descriptively designate the activities we observe for Synpcc7942_1834 in this work , we will hereafter refer to this gene as maintenance of carboxysome distribution B ( mcdB ) . Bioinformatic analysis of the McdB protein by BlastP ( protein-protein blast ) revealed that McdB lacked homology to any known ParB family member , nor any identifiable conserved regions with known ParB proteins . We therefore used the jPred4 platform ( Drozdetskiy et al . , 2015 ) to analyze this 17 kDa protein , which predicted that McdB possesses a secondary structure consisting mainly of alpha-helices , a highly charged N-terminal a1-helix , and a coiled-coil C-terminal a8-helix ( Figure 2E ) . Because McdB appears to be a novel protein and there are no characterized proteins with comparable sequence , Phyre2 ( Kelley et al . , 2015 ) was unable to generate a reliable protein homology model . We sought to determine if McdA and McdB directly interact by performing a bacterial two-hybrid assay ( B2H ) between N- and C-terminally tagged McdA and McdB ( Figure 2F ) . McdA and McdB were able to self-associate in the B2H analysis . Self-association of C-terminally tagged McdA proteins was faint , but confirmed on X-gal plates ( Figure 2—figure supplement 1A ) . We also observed a reciprocal interaction between N-terminally tagged McdA and N-terminally tagged McdB ( Figure 2F ) . However , C-terminally tagged McdB failed to show an interaction with McdA , while C-terminally tagged McdA association with McdB was dependent upon expression conditions . These results suggest that N-terminally tagged McdA only interacts with N-terminally tagged McdB , while C-terminal fusions of either protein partially disrupts function , consistent with our in vivo observation of mNG-McdA dynamics in comparison to McdA-mNG ( Figure 1DE ) . The ParA/MinD family of ATPases are defined by the presence of two lysine residues within their deviant Walker-A motif ( KGGXXGKT ) required for dimerization , ATP-binding , and ATP hydrolysis ( Lutkenhaus , 2012 ) . Interestingly , S . elongatus McdA lacks the signature amino terminal lysine residue ( Figure 2G ) , suggesting that McdA might have an activity uncharacteristic of proteins from this family . Therefore , we set out to determine if McdA displayed ATPase activity . His-MBP-McdA displayed strong ATPase activity alone , significantly higher ( >200 fold ) than that of traditional ParA family ATPases SopA of F plasmid and ParA of P1 plasmid ( Figure 2G and Figure 2—figure supplement 1B ) . Because the ATPase activity was uncharacteristically high , we confirmed that the measured ATPase activity co-eluted with His-MBP-McdA from a size exclusion chromatography column and could not be attributed to a contaminating protein ( Figure 2—figure supplement 1C ) . Relative to the constant specific activity of F SopA-His and P1 ParA with increasing protein concentrations ( Figure 2—figure supplement 1D ) , the specific ATPase activity of His-MBP-McdA declined at higher protein concentrations ( Figure 2—figure supplement 1E ) . This decrease in ATPase activity was not due to substrate limitation during the course of the in vitro assay , as ATP was provided in excess , but could be indicative of a regulatory mechanism or product inhibition that is not characteristic of traditional ParA family members . ParB partners stimulate the ATPase activity of their cognate ParA synergistically with nsDNA ( Davis et al . , 1992; Bouet and Funnell , 1999; Bouet et al . , 2000 ) . ParB stimulation is suggested to be coupled to ParA depletion on the nucleoid in the vicinity of ParB-bound cargo , as ADP does not support ParA binding to nsDNA ( Leonard et al . , 2005; Hester and Lutkenhaus , 2007; Castaing et al . , 2008; Vecchiarelli et al . , 2010 ) . We tested whether McdB-His could stimulate McdA ATPase activity , analogously to traditional ParB members . When adding only nsDNA or McdB-His to the reactions , we observed very mild stimulation of His-MBP-McdA ATPase activity ( Figure 2H ) . When both nsDNA and McdB-His were added simultaneously , the ATPase activity of His-MBP-McdA was further stimulated ( Figure 2H and Figure 2—figure supplement 1G ) . An alternative preparation of McdA which was tagged at the C-terminus with GFP also exhibited intrinsically high ATPase activity that could be stimulated with the addition of nsDNA ( Figure 2—figure supplement 1F ) , but addition of McdB-His did not further increase ATPase activity , in agreement with our prior data ( Figures 1D and 2F , Figure 1—figure supplement 1A ) that suggests C-terminal fusions of McdA may prevent interactions with McdB . In comparison to classic examples within the ParA/ParB family , McdB’s stimulation of McdA ATPase activity is relatively mild ( 2–3 fold; Figure 2G ) . We therefore assayed if this stimulatory effect was sufficient to influence McdA’s binding to DNA substrates using a gel shift assay . As shown previously ( Figure 1I ) , with ATP present , non-specific DNA fragments exhibit slowed mobility in the presence of increasing concentrations of purified His-MBP-McdA ( Figure 2I ) . Conversely , when the experiment was conducted with a constant concentration of His-MBP-McdA ( 2 . 5 µM ) , the shift in DNA mobility was reversed by addition of increasing amounts of McdB-His ( Figure 2J ) . This demonstrates that McdB favors the release of McdA from DNA . McdB-His alone did not exhibit any DNA binding activity in our gel shift assay ( Figure 2K ) . Taken together , these results further support a direct interaction between McdA and McdB , and suggest that McdB drives the removal of McdA from its DNA substrate; contributing to the emergent oscillatory dynamics of McdA we observe in vivo . We next sought to elucidate McdB’s localization in vivo . Similar to our native McdA reporters , we generated N- and C-terminal mNG fusions of McdB in its native genomic locus downstream of mcdA . C-terminal fusions of McdB displayed a diffuse localization with random punctate-like patterns ( ≥99% of cells; n = 371 cells ) ( Figure 3A and Figure 3—figure supplement 1A ) . In contrast , N-terminal mNG-McdB was observed as multiple discrete fluorescent foci near the central longitudinal axis of the cell ( ≥99% of cells; n = 699 cells ) , a result that strongly resembles the localization pattern of native carboxysomes ( Figure 3B and Figure 3—figure supplement 1B ) . We confirmed co-localization of McdB with carboxysomes by co-expression of the carboxysome fluorescent reporter in the mNG-McdB strain . Both the mNG-McdB and RbcS-mTQ signals strongly colocalized ( ≥99% of cells; Pearson’s Correlation Coefficient [PCC]=0 . 92; n = 316 cells ) as fluorescent foci near the long central axis of the cell ( Figure 3C and Figure 3—figure supplement 1C ) . Next , we investigated if McdB’s interaction with the carboxysome is direct , and if so , what carboxysome components bind McdB . During biogenesis , carboxysomes first form a core structure containing RuBisCO and carbonic anhydrase , which are coordinated into an ordered array through interactions with CcmM ( Figure 3D ) ( Long et al . , 2007; Cot et al . , 2008; Long et al . , 2010; Cameron et al . , 2013 ) . This core is thought to recruit outer shell proteins through the mediating protein CcmN , thereby forming the carboxysome coat ( Fan et al . , 2012; Kinney et al . , 2012 ) ( Figure 3D ) . CcmK2 is the dominant shell protein that composes the facets of the shell , and which has been shown to directly interact with CcmN ( Kinney et al . , 2012 ) . Along with CcmK2 , proteins CcmO , CcmL , CcmK3 , CcmK4 , and CcmP are also recruited to complete compartmentalization ( Figure 3E ) ( Tanaka et al . , 2008; Tanaka et al . , 2009; Rae et al . , 2012; Cai et al . , 2013 ) , although the relative arrangement of these structural components of the shell remains uncertain . We explored if the outer shell proteins of the carboxysome could be involved in recruiting McdB through a bacterial two-hybrid screen . Using N- or C-terminally tagged McdA or McdB as bait , the assay suggested only N-terminally tagged McdB interacts with the shell proteins CcmK2 , CcmK3 , CcmK4 , CcmL , and CcmO , but not CcmP ( Figure 3F and Figure 3—figure supplement 1D ) . In contrast , we did not find evidence for direct interaction between McdA and carboxysome shell proteins ( Figure 3F and Figure 3—figure supplement 1D ) . To further investigate the association of McdB with the carboxysome , we examined mNG-McdB dynamics in a S . elongatus background that lacks functional carboxysomes . While carboxysomes are essential for growth under an ambient atmosphere , mutants deleted for the Carbon Concentrating Mechanism ( ccm ) operon ( ΔccmK2LMNO; Figure 3G ) can be recovered in high CO2 ( Price et al . , 1993; Cameron et al . , 2013 ) . We therefore examined the localization of mNG-McdB in a ΔccmK2LMNO background and found that mNG-McdB signal was diffuse ( ≥99% of cells; n = 389 cells ) in the absence of carboxysomes ( Figure 3H and Figure 3—figure supplement 1E ) , further indicating a recruitment of McdB to assembled carboxysomes . As expected , RbcS-mTQ signal was also diffuse in ΔccmK2LMNO cells , confirming the absence of carboxysomes ( Figure 3H and Figure 3—figure supplement 1E ) . Interestingly , in the absence of carboxysomes , mNG-McdA did not oscillate and formed a homogenous distribution along the nucleoid similar to that of our mNG-McdA∆mcdB strain ( ≥99% of cells; n = 227 cells ) ( Figure 3I , Figure 3—figure supplement 1F and Figure 2D ) . This result suggests that the interaction and/or concentration of McdB onto carboxysomes is required for self-organization of McdA and could be an important prerequisite for strong stimulation of McdA ATPase activity . Since we found that both McdA and McdB are implicated in the regulation of carboxysome positioning , we next investigated how carboxysomes were distributed in strains lacking these proteins . To reduce potential off-target effects in the ∆mcdA and ∆mcdB lines , we generated knockouts in a manner designed to minimize alterations in expression of neighboring genes . This included insertion of the kanamycin resistance cassette outside of the McdA operon , and duplication of the mcdA promoter upstream of the mcdB gene in constructs where interruption of mcdA might be expected to disrupt downstream gene expression ( see Figure 4AB ) . Simultaneously , we also inserted RbcS fused to mOrange2 ( mO ) expressed from the native rbcS promoter to visualize carboxysomes ( Figure 4ABC ) . As before , mNG-McdA distributed along the nucleoid and did not oscillate in ∆mcdB lines ( ≥99% of cells; n = 373 cells ) ( Figure 4A ) , and carboxysomes were observed as large irregularly shaped polar fluorescent foci with smaller randomly distributed signals within the cell ( Figure 4A ) . Consistent with our proposed role for McdB in removing McdA from the nucleoid , we did not observe depleted McdA signal on the nucleoid in the vicinity of carboxysomes within a ∆mcdB background . In ∆mcdA lines , mTQ-McdB still localized to carboxysomes , indicating McdA is not required for the association ( ≥99% of cells; PCC = 0 . 85; 416 cells ) . Carboxysomes also formed large fluorescent foci in ∆mcdA lines and , where multiple foci could be resolved , they frequently clustered in close proximity rather than distributing throughout the cell ( Figure 4B ) , similarly to ∆mcdB lines . In the absence of both mcdA and mcdB , carboxysomes appeared as irregular foci of varying sizes randomly distributed through the cell ( ≥99% of cells; n = 399 cells ) ( Figure 4C ) . The fluorescence intensity from the RbcS-mO reporter was also unexpectedly ~4 fold weaker in these lines , likely due to unintended read-through effects from the mcdA promoter . We next investigated carboxysome positioning in McdA and McdB overexpression lines . For these experiments , we inserted the RbcS-mO fluorescent reporter in neutral site 1 , a genomically neutral locus ( Clerico et al . , 2007 ) , and overexpressed either mNG-McdA or mTQ-McdB from a synthetic riboswitch ( Nakahira et al . , 2013 ) inserted into neutral site 2 ( Clerico et al . , 2007 ) ( Figure 4DE ) . Upon the overexpression of mNG-McdA , we observed loss of McdA oscillation and the formation of large irregularly-shaped RbcS-mO fluorescent foci reminiscent of ∆mcdA , ∆mcdB and ∆mcdAB lines ( ≥99% of cells; n = 340 cells ) ( Figure 4D ) . With endogenous levels of McdB present , the McdA signal was generally depleted in the vicinity of these carboxysome aggregates . Upon overexpression of mTQ-McdB , we observed mTQ-McdB signal colocalized at irregularly-shaped RbcS-mO fluorescent foci as well as diffuse within the cell ( ≥99% of cells; PCC = 0 . 80; n = 345 cells ) ( Figure 4E ) . Moreover , we occasionally observed ( <10% of cells ) much larger RbcS-mO bar-shaped structures within the cell ( Figure 4E; yellow arrow ) , indicating that McdB levels might influence carboxysome size and ultrastructure . Carboxysome distribution displayed some variation even in wildtype strains , therefore we quantified carboxysome distributions in hundreds of cells utilizing MicrobeJ to automatically detect and characterize fluorescent carboxysome foci ( Ducret et al . , 2016 ) . In our RbcS-mO only reporter line , carboxysomes were observed predominately along the central axis and mean focus diameter was 140 nm ± 100 nm ( Figure 5A , G ) . In comparison to this strain , carboxysomes in ∆mcdA ( 510 nm ± 270 nm ) , ∆mcdB ( 350 nm ± 190 nm ) , and ∆mcdAB ( 370 nm ± 250 nm ) lines had a much broader distribution off the central axis with larger mean foci diameters and deviations ( Figure 5BCDG ) . Likewise , overproduction of McdA produced carboxysome distributions off the central axis and a mean foci diameter ( 320 nm ± 300 nm ) similar to those of deletion lines ( Figure 5BCDEG ) . Lastly , McdB overexpression produced carboxysomes that were distributed mostly along the central longitudinal axis , but exhibited the largest mean foci diameter and deviation ( 530 nm ± 470 nm ) ( Figure 5FG ) . This data suggests that in addition to spatially regulating carboxysome distributions , McdA and McdB could influence carboxysome size and/or ultrastructure . To differentiate whether the changes in size of RbcS-mO foci were due to clustering of multiple carboxysomes or changes in carboxysome ultrastructure , we used Transmission Electron Microscopy ( TEM ) . In contrast to faithfully distributed carboxysomes in our wildtype TEM images ( Figure 5H and Figure 5—figure supplement 1B ) , we observed multiple carboxysomes tightly clustered in our ∆mcdA , ∆mcdB , and ∆mcdAB strains ( Figure 5IJK and Figure 5—figure supplement 1ACDE ) . This observation suggested McdA and McdB are required to separate neighboring carboxysomes , or that newly synthesized carboxysomes may be incompletely detached from one another during biogenesis ( Cameron et al . , 2013; Chen et al . , 2013 ) in the absence of McdA or McdB . Upon overproduction of McdA , we observed irregularly-shaped carboxysomes with ‘rounded’ edges that tightly clustered ( Figure 5L and Figure 5-figure supplement 1AF ) . It should be noted however , that this cell line displayed severe growth arrest and possessed an unusually abundant number of granules which we suggest could be polyphosphate bodies based on their structure and dense staining . Consistent with this observation , overproduction of ParA family proteins in many organisms is lethal ( Lasocki et al . , 2007 ) . Most strikingly , carboxysomes did not cluster in McdB overproduction strains either , instead , carboxysome ultrastructure was dramatically altered . Unlike the classic icosahedral-shape , carboxysomes were observed as large irregular bar-like structures ( Figure 5M and Figure 5—figure supplement 1AG ) . In some instances , these ‘bar-carboxysomes’ extended hundreds of nanometers ( Figure 4E ) , resembling previous reports of improperly assembled carboxysomes in cells lacking ccmL ( Price and Badger , 1989; Cameron et al . , 2013 ) . Together , these results highlight the importance of the McdAB system for spatially separating carboxysomes and regulating the underlying size and ultrastructure . In ParA-based plasmid partitioning , the low-copy number of plasmid cargo allows for the direct observation of how one or two plasmid copies influence the distribution of ParA on the nucleoid , and vice versa . Through in vivo and in vitro experimentation , it has been demonstrated that ParB-bound plasmids and chromosomes form ParA depletion zones by stimulating the release of ParA from the nucleoid in their vicinity ( Hatano et al . , 2007; Ringgaard et al . , 2009; Schofield et al . , 2010; Hwang et al . , 2013; Vecchiarelli et al . , 2013; Vecchiarelli et al . , 2014 ) . But in S . elongatus , carboxysome copy-number correlates with cell length and ranges between 3 to 15 copies ( Figure 1—figure supplement 1C ) . How multiple carboxysomes are equally spaced along the cell-length by a global oscillation of McdA protein is not intuitively obvious . Therefore , we wished to determine if McdB-bound carboxysomes also cause similar local depletions of McdA on the nucleoid and whether oscillation of McdA was a requirement for carboxysome motion and equidistant positioning , even at low-copy numbers . In time-lapse experiments of mNG-McdA and RbcS-mTQ strains , we frequently observed that carboxysomes move towards the highest local concentration of McdA on the nucleoid; this was especially apparent as the wavefront of the oscillating McdA pool approaches carboxysomes ( Videos 3–5 ) . The most rapid directed motions of carboxysomes were visible as the wavefront approaches a carboxysome , and then as the wavefront passes . To explore this observation in more depth , we developed the ability to regulate the initiation of carboxysome formation as well as to modulate the number of carboxysomes per cell . To accomplish this , we replaced the native ccmK2 promoter with a Ptrc promoter lacking the lacI repressor and attached a 5’ synthetic riboswitch preceding ccmK2 ( RS::ccmK2LMNO; Figure 6A ) . In the absence of inducer , genes regulated by this riboswitch are tightly off , and expression is highly tunable with increasing concentrations of theophylline ( Nakahira et al . , 2013 ) . In the absence of theophylline , we observed that RbcS-mTQ signal was diffuse and mNG-McdA was distributed homogenously along the nucleoid ( ≥99% of cells; n = 204 cells ) ( Figure 5—figure supplement 1H ) , consistent with our prior results from ∆ccmK2LMNO mutants ( Figure 3I ) . When these strains were induced with either 400 µM or 600 µM theophylline , we were able to generate on average one or two carboxysomes per cell , respectively ( Figure 6BC ) . In the presence of 1 carboxysome , mNG-McdA signal remained evenly distributed except for a depletion zone that correlated with the nucleoid region in the vicinity of the carboxysome ( Figure 6B ) . Likewise , with two carboxysomes , mNG-McdA signal again distributed along the nucleoid but was depleted in areas correlating to carboxysomes ( Figure 6C ) . In either case , mNG-McdA signal was highly reduced in areas of RbcS-mTQ signal ( ≥99% of cells; PCC = 0 . 20; n = 391 cells ) , indicating McdB-bound carboxysomes have mNG-McdA depleted in their vicinity . In this strain , we performed real-time imaging of mNG-McdA dynamics and RbcS-mTQ motion . In instances where cells contained two , closely spaced carboxysomes , one carboxysome could be clearly observed to move in the direction of the higher McdA concentration ( Figure 6D ) . When a sufficient distance was reached between the two carboxysomes , mNG-McdA was re-recruited to the depleted nucleoid region between the two carboxysomes ( Figure 6D ) . As mNG-McdA rebound the nucleoid , movement of the centralized carboxysome halted and slightly regressed back in the opposite direction ( Figure 6D ) . This result is consistent with the Brownian-ratchet mechanism for genetic cargo movement towards the highest local concentration of ParA ( Vecchiarelli et al . , 2010; Vecchiarelli et al . , 2014; Hu et al . , 2017 ) . We next sought to determine if we could reconstitute carboxysome-dependent oscillation of McdA . Even at relatively high concentrations of theophylline inducer , our synthetic riboswitch was unable to generate wildtype quantities of carboxysomes . Therefore , we used a variant of a previously published approach ( Cameron et al . , 2013 ) by replacing the ccmK2 promoter with the Ptrc promoter and inserted an upstream lacI repressor ( Figure 6E ) . This promoter is generally capable of driving higher expression levels of gene targets . In the absence of Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) , some carboxysome formation was observed due to leaky expression of the Ptrc promoter in cyanobacteria ( Figure 6F ) . Similar to prior results , mNG-McdA was depleted in the vicinity of carboxysomes ( ≥99% of cells; n = 453 cells ) ( Figure 6F ) . Following induction with IPTG , multiple carboxysomes formed throughout cells and mNG-McdA oscillations emerged ( ≥99% of cells; n = 439 cells; Figure 6F ) . Altogether , these experiments strongly indicate that McdB is concentrated upon carboxysomes and that this localized pool of McdB changes the dynamics of McdA bound to neighboring regions of the nucleoid . Under conditions where there are relatively few ( 1-3 ) carboxysomes , McdB appears to continuously stimulate the release of nearby McdA . It is only at higher numbers of carboxysomes ( 4+ ) when a self-organized oscillation of McdA from end-to-end of the nucleoid emerges . Moreover , we observe multiple instances of directed motion of carboxysomes towards increased McdA concentrations on the nucleoid , consistent with the Brownian-ratchet model of cargo movement . Carboxysomes are frequently described as being linearly arranged along the longitudinal axis of S . elongatus . In addition to linear distributions , we also routinely observed carboxysomes that were equidistant to each other , but no longer linearly arranged ( Figure 7A ) . Instead , carboxysomes displayed a hexagonal packing phenomenon where the linear arrangement along the longitudinal axis looked kinked or displayed a zig-zag pattern . The hexagonal packing arrangement is non-intuitive assuming carboxysome distribution were to be based solely on either an underlying cytoskeleton or McdA oscillations , as these features are oriented longitudinally to the cell . Hexagonal packing was typically correlated with cells that had a high number of carboxysomes relative to the cell’s length . However , it is difficult to ascertain if this different packing arrangement is due solely to carboxysome number , or any number of other factors that could be differentially regulated between distinct cells . To better understand this hexagonal packing phenomenon , we examined carboxysome positioning in hyperelongated cells because these cells have unusual nucleoid features useful to supplement our observations in cells of wildtype length ( 3–6 µm ) . Importantly , such hyperelongated S . elongatus cells are known to contain nucleoid clusters which are separated by intermittent cytoplasmic gaps that physically separate the nucleoid clusters from one another ( Miyagishima et al . , 2005 ) ; Figure 7—figure supplement 1 ) . One targeted method to increase cell length is to overexpress the FtsZ regulatory protein Cdv3 , which we have previously reported to cause division arrest and subsequent cell elongation up to 2 mm ( Jordan et al . , 2017; MacCready et al . , 2017 ) . In cells elongated by this method , we observed both linear and hexagonal carboxysome packing ( Figure 7B ) , sometimes observing different packing arrangements on neighboring nucleoid clusters within the same cell ( Figure 7—figure supplement 1D ) . Carboxysomes in hyperelongated cells always co-localized with a nucleoid cluster , as visualized by DAPI staining ( Figure 7—figure supplement 1A ) , and were never observed in the gaps between clusters , consistent with a role for McdAB in tethering carboxysomes to DNA . By contrast , in a ΔmcdA or ΔmcdB background , when we induced hyperelongation by expressing Cdv3 , carboxysomes were frequently observed in these cytoplasmic gap regions ( Figure 7-figure supplement 1BC ) . In individual hyperelongated cells , we frequently observed carboxysomes both in linear and hexagonal-packing arrangements within the same cell but on different nucleoid clusters ( Figure 7—figure supplement 1D ) . Because cells containing both linear and hexagonal packing arrangements share the same cytosol , it is unlikely that the carboxysome packing is regulated by a global change within a cell ( such as a diffusible factor ) . Instead , we found once again that the hexagonal packing was typically observed when the number of carboxysomes were higher on a given nucleoid cluster . These results suggest that carboxysome packing arrangement may be a self-emergent property related to the density of carboxysomes on a given nucleoid surface area . To assess if the Brownian-ratchet model of carboxysome positioning could account for both carboxysome spacing and patterning ( i . e . linear vs . hexagonal ) , we turned to an established in silico mathematical model that has successfully described several aspects of the Brownian-ratchet mechanism for ParA-mediated partitioning of plasmids ( Hu et al . , 2017 ) . Since we have yet to determine which biochemical parameters of the Mcd system differ from that of traditional Par systems , in this treatment , we simply increased the number of cargo copies on the nucleoid matrix while keeping all other biochemical parameters as previously described so as to determine if increasing cargo copy number is enough to convert linear positioning into hexagonal packing . We programmed the geometry of the nucleoid surface area ( 2 . 5 µm by 0 . 6 µm rounded rectangle ) and carboxysome cargo ( 175 nm ) based off of previously measured values of wildtype S . elongatus ( Rae et al . , 2012; Murata et al . , 2016 ) . All simulations are initiated with tightly clustered carboxysomes near the center of the nucleoid ( Figure 7C; left images ) , but carboxysomes are allowed to travel towards the highest gradient of McdA using the previously-established parameter values . With the Brownian-ratchet model , five or less carboxysomes will linearly distribute on a rectangular surface representative of S . elongatus’ nucleoid ( Figure 7C ) . As cargo number increases the linear arrangement is maintained , but with tighter spacing ( Figure 7CD ) . However , above a certain density threshold ( six or more cargos on the same nucleoid under our simulation parameters ) , cargo positioning switches from a linear arrangement to hexagonal packing , reminiscent of the in vivo distributions ( Figure 7ABCD ) . This change in packing arrangement can be understood if each carboxysome is independently seeking the highest local concentration of McdA on the nucleoid: as carboxysome density increases , a staggered conformation maintains the maximal nearest-neighbor distance . As many other cyanobacterial species exhibit spherical morphology , including the model Synechocystis sp . PCC 6803 , we also examined the predicted distribution of carboxysomes upon a 1 . 7 µm circular nucleoid ( Figure 7EF ) . We suggest that the linear arrangement of carboxysomes in rod-shaped cells is largely a byproduct of nucleoid geometry , and thus , cell morphology . In support of this proposition , many spherical ( e . g . , see Figure 1A in Kerfeld et al . , 2005 ) and filamentous ( e . g . , see Figure 1 in Montgomery , 2015 ) cyanobacterial cells also show a hexagonal carboxysome arrangement . Furthermore , S . elongatus cells grown under environmental conditions that increase carboxysome synthesis also dominantly display hexagonal packing ( e . g . , see Figure 2 in Sun et al . , 2016 ) . Homologs of ParA-type ATPases have been identified within extended carboxysome operons of cyanobacteria ( Axen et al . , 2014 ) . Therefore , we examined other distant cyanobacterial species for possible McdAB homologs . One such case is the primitive thylakoid-less cyanobacterium Gloeobacter kilaueensis JS1 , which drives expression of an mcdA-like gene from the rbcL promoter . Interestingly , upon further examination , we found a small coding sequence following this mcdA-like gene with weak similarity to mcdB . BlastP determined that S . elongatus McdA had 22 . 5% pairwise sequence identity to the G . kilaueensis JS1 McdA-like protein , while S . elongatus McdB had only 18 . 4% pairwise identity to the McdB-like protein ( Figure 8A ) . To investigate the possibility that the McdB-like protein of G . kilaueensis JS1 functions similarly to S . elongatus McdB , we expressed a fluorescent fusion of the G . kilaueensis JS1 mcdB-like gene , mTQ-McdB ( Gk ) , in our S . elongatus ∆mcdB strain . Despite the low primary sequence identity of McdB ( Gk ) , we found that mTQ-McdB ( Gk ) colocalized with RbcS-mO ( PCC = 0 . 79; n = 248 ) , indicating that mTQ-McdB ( Gk ) can interact with S . elongatus carboxysomes ( Figure 8B ) . These results suggest that carboxysome positioning by McdA and McdB may be widespread among cyanobacteria . Carboxysomes are essential components of the photosynthetic metabolism of cyanobacteria , yet the mechanisms underlying their positioning within the cell has remained an outstanding question . Prior work in S . elongatus showed that a ParA-like protein ( McdA ) was required for maintaining carboxysome positioning ( Savage et al . , 2010 ) . Largely influenced by models for DNA segregation by ParA-type ATPases at the time , the observation that C-terminally tagged McdA ( McdA-GFP ) oscillated in vivo and that carboxysomes were mispositioned following the disruption of McdA or MreB ( an actin-related component of the cytoskeleton ) led to a widely-adopted hypothesis that carboxysomes were positioned by a cytoskeletal mechanism ( Savage et al . , 2010; Murat et al . , 2010; Rae et al . , 2013; Yokoo et al . , 2015 ) . This model proposed that adjacent carboxysomes were connected via McdA filaments that continually polymerize and depolymerize , exerting physical force upon carboxysomes in order to maintain their equidistant positioning – although no direct evidence for this model has been demonstrated to date . Here we show that McdA-GFP homogeneously binds non-specifically to DNA , but exhibits no signs of filament formation ( Figure 1A–C and Video 1 ) . We show multiple lines of evidence indicating that McdA is capable of binding DNA in an ATP-dependent manner , and that the bulk of McdA in S . elongatus is concentrated upon the nucleoid . Furthermore , we identify a novel protein , McdB , that interacts with McdA , stimulating its intrinsic ATPase activity and release from the nucleoid . Taken together , these results strongly suggest that McdA does not form independently-standing filaments , but instead attaches to the nucleoid body at the center of the cell , using it as a scaffolding surface to support an oscillating wave from one end of the cell to the other . We provide evidence that carboxysomes are in turn tethered to the nucleoid through interactions with McdB , that in turn can bind McdA . Our results provide insight into the molecular mechanisms of McdA , and extend upon the limited characterization in the literature of this unusual ParA-family member . While we confirm that McdA can oscillate in vivo as was previously shown ( Savage et al . , 2010 ) , we observe oscillatory waves that traverse the cell within ~10 min , which is significantly faster than that reported for McdA-GFP . The discrepancy may be related to the use of C-terminally tagged McdA reporters , which we find are unable to interact effectively with McdB and was originally over-expressed in a background with an additional endogenous copy of McdA ( Savage et al . , 2010 ) . We observe activities for McdA ( tagged on either terminus ) that are consistent with other ParA family members , including ATP-dependent DNA binding ( Figure 1A ) , self-association ( Figure 2F and Figure 2—figure supplement 1A ) , and DNA stimulated ATPase activity ( Figure 2H and Figure 2—figure supplement 1F ) . However , we also observe key distinctions between McdA and more canonical family members ( see below ) , which may be important in the adaption of this system for the segregation of large protein cargos . Importantly , we do not find any direct evidence supporting filament formation by McdA . Using high-resolution imaging techniques , we observe McdA-GFP to coat evenly across DNA in a carpeted flowcell , without any indication of large oligomer formation . While we cannot exclude the possibility that McdA-GFP fusions are disrupted in their capacity to form filaments , we note that this fusion retains many of its characteristics ( Figure 1A , Figure 2—figure supplement 1A , and Figure 2—figure supplement 1F ) . Furthermore , our data suggests that the morphology of the cyanobacterial nucleoid is important for carboxysome positioning ( Figures 1 , 2JK , 3C , 6 , 7 ) , which could also explain why ΔmreB mutants exhibit disorganized carboxysomes ( Savage et al . , 2010 ) . Both cell morphology and nucleoid topology are grossly altered in ΔmreB strains ( Hu et al . , 2007 ) , suggesting that the influence of MreB in carboxysome positioning is likely indirect . Instead , our results support an alternative model for McdA in carboxysome positioning that does not require an underlying cytoskeleton and which utilizes a Brownian-ratchet based mechanism ( see below ) . Our model of self-organized carboxysome positioning is both informed by the ParA-based mechanisms used to segregate low-copy number plasmids , but also provides a novel platform to study the dynamics of self-organized protein segregation systems . Low-copy plasmids often contain DNA regions ( e . g . parS ) that bind ParB , which drives the directed and persistent movement of plasmids towards increased concentrations of ParA on the nucleoid ( Vecchiarelli et al . , 2010; Le Gall et al . , 2016 ) . In this way , it is proposed that ParA can provide a positional cue allowing plasmid cargo to ‘surf’ along the larger bacterial chromosome without a separate cytoskeletal system ( Vecchiarelli et al . , 2012 ) . While conceptual similarities exist between the better-established systems for plasmid positioning and results we report here for carboxysome positioning , a number of key distinctions separate McdAB from ParAB models . First , S . elongatus’ McdA lacks the signature lysine residue in the Walker A box that defines the ParA family of ATPases ( KGGXXGT; Figure 2G ) . The serine substitution in McdA at a position universally conserved in ParA members may underlie the unusually high ATPase activity of McdA ( Figure 2G ) , which displays a maximum specific activity that is roughly two-orders of magnitude greater than that of other well-studied ParA systems ( Vecchiarelli et al . , 2010; Ah-Seng et al . , 2009 ) . McdB is an even more divergent protein , bearing no identifiable sequence similarity to any known ParB proteins; indeed , no homologous proteins have been characterized in other species . This novel protein also recognizes and binds a large protein-based cargo ( carboxysomes; Figure 3C , F , H ) , further distinguishing it from all characterized ParB-like proteins that recognize genetic cargo . Even though McdB and ParB share no similarity , we find that McdB: ( i ) interacts with McdA ( Figure 2F ) , ( ii ) stimulates McdA ATPase activity ( Figure 2H ) ( iii ) removes McdA from DNA ( Figure 2J ) , and ( iv ) is responsible for emergent dynamics of McdA along the nucleoid ( Figure 2D ) ; analogous to the roles played by ParB in well-characterized plasmid partitioning systems . Furthermore , we observe that a pool of McdB enriched at the carboxysome is necessary to locally deplete McdA ( Figures 4 and 6B–D ) , suggesting that prolonged McdB activity may stimulate the local release of McdA from the nucleoid . We propose that McdB is therefore acting to interface carboxysomes with nucleoid-bound McdA , processively pulling this protein cargo towards the highest local McdA concentration , and thereupon stimulating McdA ATPase activity and release ( Figure 6D and Videos 3–5 ) . The parallels between features of the McdAB system and Brownian ratchet ParAB models make it tempting to speculate that McdB has a distinct evolutionary origin from ParB-family members , but that these independent protein families convergently evolved to use nucleoid gradients of ParA-like proteins to segregate entirely different classes of macromolecular structures . The colocalization of signal between mNG-McdB and carboxysomes ( RbcS-mTQ ) ( Figure 3C ) , coupled with the carboxysome requirement for providing site-specificity to mNG-McdB in vivo ( Figure 3H ) , provide strong evidence that McdB is associating with carboxysomes and that this interaction is needed for emergent dynamics of McdA ( Figure 3I ) . It is curious that McdB is able to associate with a number of different shell proteins in our B2H assay ( Figure 3F ) . Taken together with the evidence that a McdB homolog from G . kilaueensis JS1 with low sequence-identity still concentrated upon S . elongatus carboxysomes ( Figure 8B ) , the most parsimonious hypothesis is that McdB-carboxysome shell interactions are mediated by structural and/or charge features common to many distinct shell proteins . Indeed , evolutionarily distant hexameric shell proteins of the bacterial microcompartment ( BMC-H ) family share a number of similarities in structural features and key residues at hexamer interfaces that are largely conserved ( Cai et al . , 2015; Sommer et al . , 2017; Young et al . , 2017 ) . This suggests some of these common structural features could be important in mediating interactions with McdB , which might explain why McdB displays an interaction with different shell protein paralogs . Our B2H analysis also indicates that McdB may have a higher affinity to some shell proteins ( CcmK2 and CcmK3 ) than others ( CcmK4 , CcmL and CcmO ) ( Figure 3F ) . This may be related to the observation of clustered carboxysomes in ∆ccmK3-4 mutants ( Rae et al . , 2012 ) , as this may reduce the amount of McdB recruited to the carboxysome surface . We note however , that given McdB’s poor sequence conservation and without further knowledge of the structure and interaction domains of McdB , we cannot rule out that McdB is a 'sticky' protein by the B2H assay and is instead recruited through an alternative adaptor protein to the vicinity of carboxysomes . Moreover , deleting individual shell components , such as CcmK2 , CcmL or CcmO , prevents mature carboxysomes formation and subsequent biogenesis ( Cameron et al . , 2013 ) , preventing in vivo testing of McdB/carboxysomes interaction . Additional experiments will be required to identify the domain ( s ) that mediate McdB-shell interaction , and without a more detailed analysis , it remains possible that McdB can directly integrate within the shell of mature carboxysomes . Some indirect evidence would argue against the possibility that McdB is an integral shell protein , including both our observation that ∆mcdB strains did not possess a high CO2-requiring phenotype and McdB has not been identified in previously-published carboxysome purification studies ( Faulkner et al . , 2017 ) . One surprising result of our study was that we observed that both McdB and carboxysomes themselves were required for the emergence of McdA oscillations along the nucleoid ( Figure 2D , 3HI ) . Furthermore , a critical threshold number of carboxysomes were required to be localized on the same nucleoid in order for McdA oscillation to ensue ( generally >3; Figure 6F ) . This suggests that it is not sufficient for McdB to be merely present , it must be specifically localized , concentrated , and/or activated to promote the McdA oscillations . Furthermore , we note that McdA oscillation per se is not required to segregate one carboxysome from another . In cells where we induced the formation of only two carboxysomes , or in cells were the Ptrc promoter was leaky , these carboxysomes reliably separated from one another despite the fact that no McdA oscillations were present ( Figure 6DF ) . Likewise , our Brownian-Rachet simulations were able to recapitulate the separation between carboxysomes in silico without any requirement for an oscillating pool of McdA ( Figure 7CDEF ) . These results could suggest that ‘global’ McdA oscillations have a secondary role in regulating carboxysome positioning relative to the influence of local gradients of McdA on the nucleoid , and also raise other questions related to how McdA oscillations emerge . One possibility is that McdA oscillation itself might be a byproduct of the motion of multiple carboxysomes removing McdA along the nucleoid . Indeed , such a model has been proposed for plasmid segregation , termed ‘DNA relay’ ( Surovtsev et al . , 2016 ) , where the plasmids recruit ParB to form cargo complexes that themselves oscillate from pole-to-pole in the cell , removing nucleoid-bound ParA in their wake . In the DNA relay model , the long-range motion of the cargo itself drives the emergent oscillation of ParA . Our simultaneous imaging of carboxysomes and McdA dynamics precludes such a model that would require long-range cargo movement ( carboxysomes move much shorter distances and over longer time scales than the McdA oscillatory wave ) , but we cannot rule out more subtle carboxysome motions being involved in the emergence of McdA oscillations . The dynamics of carboxysome motion are complex at rapid time scales and are responsive to McdA wavefronts . During McdA oscillation , we observed that some carboxysomes at the wave front paused , and in some cases , were observed to get sucked into the approaching wave ( Videos 3–5 ) . While in the wave , carboxysome diffusion was suppressed . As the wave passed , carboxysome diffusion was anisotropic , drifting in the direction of the wave . Away from the wave , carboxysome diffusion was isotropic . An alternative hypothesis is that the global dynamics of McdA oscillation are dependent upon a balanced level of activities between McdA and McdB . In this case , recruitment of soluble McdB to a defined location ( the carboxysome ) may concurrently act to remove it from the bulk cytosol , reducing the concentration that McdA perceives when not near a carboxysome . There is precedence for this interpretation in the ParA-like family of proteins , including the oscillatory behaviors of MinD and MinE . MinD binds to the plasma membrane when bound to ATP and exhibits an emergent pole-to-pole localization that is driven by the ATPase-stimulating activities of the partner protein MinE ( Lutkenhaus , 2007 ) . The ratio of these activities is important for their higher-order behaviors and when MinD:MinE ratios become severely unbalanced , oscillatory patterns collapse ( Fange and Elf , 2006; Loose et al . , 2008; Loose et al . , 2011; Zieske and Schwille , 2014; Vecchiarelli et al . , 2016; MacCready et al . , 2017 ) . It is intriguing to speculate whether McdA would oscillate in cyanobacteria displaying different morphologies , such as the spherical Synechocystis sp . 6803 or the filamentous Fremyella diplosiphon . While carboxysomes in these organisms are equidistantly spaced , they display a packing more reminiscent of the hexagonal arrangement rather than a linear distribution ( e . g . , see Figure 1A in Kerfeld et al . , 2005; e . g . , see Figure 1 in Montgomery , 2015 ) . Our modeling suggests that this could be a natural outcome of the McdAB system operating on a nucleoid topology that is more spherical , rather than rod-shaped . While McdA oscillation could still be possible , it is unclear what patterns would be expected . Further analysis of McdAB dynamics in other cyanobacteria is required to elucidate the effects of nucleoid morphology on McdA pattern formation . Our analysis provides a number of lines of evidence to suggest that McdA and McdB activities can also influence the ultrastructure of carboxysomes in S . elongatus . Cyanobacterial strains that are genetic knockouts of mcdB display carboxysomes that are significantly enlarged ( Figure 5—figure supplement 1AD ) . Furthermore , overexpression of McdB resulted in massive carboxysome globules that sometimes spanned the entire short axis of the cell ( Figure 5—figure supplement 1AG ) , while overexpression of McdA resulted in irregularly-shaped carboxysomes with rounded edges ( Figure 5—figure supplement 1AF ) . While we cannot rule out the possibility of indirect effects , it is intriguing to speculate that McdA and McdB may act to directly regulate the size or shape of microcompartments as they are formed . In S . elongatus , it has been suggested that new carboxysomes bud off from existing carboxysomes ( Cameron et al . , 2013; Chen et al . , 2013 ) . Our model for carboxysome positioning requires that the interaction of McdB with McdA provide a pulling force exerted on the carboxysome shell that acts to processively move the protein compartment up an McdA gradient . It is therefore possible that these same molecular forces act during the synthesis of a new carboxysome . The relative ratio between McdA and McdB activities may play a role in the differences in carboxysome sizes observed under different environmental conditions or within different species . Future research will be required to confirm the role of McdAB systems in regulating the size of protein-based organelles . Carboxysomes exist in two distinct forms , α and β , depending on the form of RuBisCO they encapsulate . While both are found in cyanobacteria , α-carboxysomes also exist in many actinobacteria and proteobacteria . In these organisms , the vast majority of carboxysome-related genes tend to be found at genomic loci near the respective enzymes they encapsulate ( Axen et al . , 2014 ) . We find that mcdA/B-like sequences frequently fall in regions near α- and β-carboxysome operons ( Figure 8C ) . We propose that the mcdA/B-like sequences near the α-carboxysome operon could also function to equidistantly space α-carboxysomes to ensure equal inheritance following cell division . Further study is now needed to determine how widespread the McdAB system is across evolutionary space . Indeed , many BMC classes exist , are widespread in bacteria , and encapsulate a wide array of enzymatic activities beyond Calvin-Benson-Bassham factors ( Axen et al . , 2014; Kerfeld and Erbilgin , 2015 ) . While putative McdB homologs are widespread in cyanobacteria and can be identified in many α-carboxysome-containing proteobacteria ( Figure 8C ) , it is possible that a more comprehensive bioinformatic approach could identify similar factors associated with other classes of BMC . More broadly , these findings aid in understanding the spatial organization of other protein-based mesoscale assemblies that encode ParA family members and are associated with diverse biological processes , including secretion ( Perez-Cheeks et al . , 2012; Viollier et al . , 2002 ) , conjugation ( Atmakuri et al . , 2007 ) , chemotaxis ( Thompson et al . , 2006; Ringgaard et al . , 2011; Alvarado et al . , 2017 ) , and cell motility ( Youderian et al . , 2003; Kusumoto et al . , 2008 ) . All constructs in this study were generated using Gibson Assembly ( Gibson et al . , 2009 ) from synthetized dsDNA and verified by sequencing . Constructs contained flanking DNA that ranged from 500 to 1500 bp in length upstream and downstream of the targeted insertion site to promote homologous recombination into target genomic loci ( Clerico et al . , 2007 ) . All S . elongatus cultures were grown in 125 mL baffled flasks ( Corning ) containing 50 ml BG-11 medium ( SIGMA ) buffered with 1 g L−1 HEPES to pH 8 . 3 . Flasks were cultured in a Multitron II ( atrbiotech . com ) incubation system with settings: 80 µmol m−2 s−1 light intensity , 32°C , 2% CO2 , shaking at 130 RPM . Cloning of plasmids was performed in E . coli DH5α chemically competent cells ( Invitrogen ) . All S . elongatus transformations were performed as previously described ( Clerico et al . , 2007 ) . Cells were plated on BG-11 agar with either 12 . 5 mg ml−1 kanamycin or 25 mg ml−1 spectinomycin . Single colonies were picked into 96-well plates containing 300 μl of BG-11 with identical antibiotic concentrations . Cultures were verified for complete insertion via PCR and removed from antibiotics . N- and C-terminal T18 and T25 fusions of McdA , McdB and shell proteins CcmK2 , CcmK3 , CcmK4 , CcmL , CcmO and CcmP were constructed using plasmid pKT25 , pKNT25 , pUT18C and pUT18 , sequence-verified and co-transformed into E . coli BTH101 in all pairwise combinations ( Karimova et al . , 1998 ) . Several colonies of T18/T25 cotransformants were isolated and grown in LB medium with 100 μg/ml ampicillin , 50 μg/ml kanamycin and 0 . 5 mM IPTG overnight at 30°C with 225 rpm shaking . Due to the self-assembling nature of carboxysome shell proteins , overnight IPTG induction for cotransformants bearing T18/T25 shell protein fusions was carried out at 0 . 1 mM IPTG . Overnight cultures were spotted on indicator MacConkey plates supplemented with 100 μg/ml ampicillin , 50 μg/ml kanamycin and 0 . 5 mM IPTG . Plates were incubated at 30°C up to 48 hr before imaging . Overproduction of mNG-McdA and mTQ-McdB were accomplished by inducing strains with 1500 µM theophylline for 48 hr . For carboxysome induction under the riboswitch , strains were incubated in 400 µM theophylline ( one carboxysome ) or 600 µM theophylline ( two carboxysomes ) for 24 hr prior to imaging . Alternatively , for carboxysome induction under the Ptrc promoter and LacI repressor , cells were incubated with 1000 µM IPTG for 16 hr prior to imaging . To increase cell lengths of the carboxysome reporter only strain or the ∆mcdA , ∆mcdB and ∆mcdAB with the carboxysome reporter strains , RS::Cdv3 was overexpressed for 48 hours using 1500 µM theophylline . All live-cell microscopy was performed using exponentially growing cells . Two mL of culture was spun down at 5000xg for 30 s , resuspended in 200 µl of BG-11 and 2 µl transferred to a square 1 . 5% agarose +BG-11 pad on glass slides . All images were captured using a Zeiss Axio Observer A1 microscope ( 100x , 1 . 46NA ) with an Axiocam 503 mono camera except the carboxysome induction experiments . Carboxysome induction experiments were performed using a Nikon Ti2-E motorized inverted microscope with LED-based light sources ( 100x , 1 . 45NA ) with a Photometrics Prime 95B Back-illuminated sCMOS Camera . Image analysis was performed using Fiji v 1 . 0 . Cultures were grown to OD750 = 0 . 7 in BG-11 . Cells were pelleted and fixed overnight at 4 ˚C with 2 . 5% formaldehyde/2 . 5% glutaraldehyde in 0 . 1 M sodium cacodylate buffer ( pH 7 . 4 ) , suspended into a 2% agarose bead and cut into ~1 mm cubes . Following three washes with 0 . 1 M sodium cacodylate buffer , cells were suspended in 1% osmium tetroxide/1 . 5% potassium ferrocyanide and incubated overnight at 4 ˚C . After incubation , cells were washed with HPLC-quality H2O until clear . Cells were then suspended in 1% uranyl acetate and microwaved for 2 min using a MS-9000 Laboratory Microwave Oven ( Electron Microscopy Science ) , decanted , and washed until clear . Cells were dehydrated in increasing acetone series ( microwave 2 min ) and then embedded in Spurr’s resin ( 25% increments for 10 min each at 25°C ) . A final overnight incubation at room temperature in Spurr’s resin was done , then cells were embedded in blocks which were polymerized by incubation at 60 ˚C for three days . Thin sections of approximately 50 nm were obtained using an MYX ultramicrotome ( RMC Products ) , post-stained with 1% uranyl acetate and Reynolds lead citrate , and visualized on a JEM 100CX II transmission electron microscope ( JEOL ) equipped with an Orius SC200-830 CCD camera ( Gatan ) . Cultures were grown to OD750 = 0 . 7 in BG-11 . Multiple individual images of the fluorescent reporters and chlorophyll autofluorescence were obtained for each strain and analyzed using MicrobeJ 5 . 11 n . In each line , cell perimeter detection was performed using the rod-shaped descriptor and default thresholding algorithm . Carboxysome detection was performed using the foci function with a tolerance of 15 and Z-score of 50 . For fluorescent McdA lines , localization was quantified via detection of the single brightest point ( tolerance = 2000 ) . For fluorescent McdB lines , localization was quantified via multiple smoothed foci detections ( tolerance = 15 and Z-score = 50 ) . Associations , shape descriptors , profiles and distances were recorded for each strain . Heatmaps were automatically generated with counts , contour and a spot size of 5 . Mean foci area and standard deviation for each maxima was automatically calculated . The gene sequence , mcdA–GFP–6xHis , was codon optimized for E . coli and synthesized by Genscript . The fragment was inserted into the NcoI/BamHI cloning sites of the expression vector pET15b to create the pAV30 plasmid . pAV30 was transformed into BL21 ( AI ) cells ( Invitrogen ) and a 100 mL overnight culture containing 100 µg/mL of carbenicillin was grown at 20°C with shaking at 225 rpm . LB supplemented with 100 µg/mL of carbenicillin and a drop of Antifoam Emulsion ( 1 L per 2 . 5 L Fernbach flask ×4 ) was pre-warmed to 37°C and inoculated with 10 mL of overnight culture per flask . The cells were grown at 37°C with shaking at 225 rpm to an O . D . of 0 . 4 . The flasks were then plunged in an ice bath until the temperature of the culture dropped to 16°C . Protein expression was then induced at O . D . 0 . 6 by the addition of 10 mL of a 0 . 1 M IPTG/20% Arabinose solution to each flask . Cells were then grown overnight at 16°C with shaking at 225 rpm ( ∼15 hr induction ) . The cells were transferred to 1 L Beckmann bags and bottles , which were spun in a JLA 8 . 1 rotor at 4 , 500 rpm for 1 hr . The supernatant was poured out , and the cell pellets were frozen in the bags with liquid nitrogen and stored at −80°C . Frozen cell pellets were combined in a beaker with 10 mL of cold Lysis Buffer per gram of cell pellet ( ∼150 mL total ) , three Protease Inhibitor Mixture Tablets ( Sigma ) and 0 . 1 mg/mL lysozyme ( Sigma ) . A homogenizer was used to ensure that the cell pellets were thoroughly dispersed , and two passes through a Microfluidizer lysed the cells . The lysate was cleared with a 30 min ultracentrifugation at 35 , 000 rpm and 4°C using a 45Ti rotor . The lysate was then passed through a 0 . 45 µm syringe filter . Using a peristaltic pump , the cleared lysate ( ∼200 mL ) was loaded at a flow rate of 2 mL/min onto a 5 mL HisTRAP HP cassette ( GE ) and equilibrated with Lysis Buffer ( 50 mM HEPES–KOH ( pH 7 . 6 ) , 1 M KCl , 10% Glycerol , 20 mM Imidazole ( pH 7 . 4 ) , 2 mM β-mercaptoethanol ) . The protein was eluted with a 20 mM to 1 M imidazole gradient ( total volume = 60 mL ) . Peak protein fractions were pooled and concentrated using an Amicon Ultra Centrifugal Device ( 10 KD MWCO ) . The sample was passed through a 26/10 salt-exchange column equilibrated in Q-Buffer ( 50 mM HEPES–KOH ( pH 7 . 5 ) , 200 mM KCl , 10% Glycerol , 0 . 1 mM EDTA , 2 mM DTT ) . The sample was then immediately loaded onto a 1 mL Mono Q 5/50 anion exchange column ( GE ) equilibrated in Q-Buffer . The protein was eluted with a 200 mM to 1 M KCl gradient . Peak fractions were pooled and concentrated to a no more than 100 µM . The sample was then separated over a 10/300 GL Superdex200 gel-filtration column equilibrated in Q Buffer ( but with 600 mM KCl ) . Peak fractions were pooled , concentrated to no more than 100 µM , frozen with liquid nitrogen , and stored at −80°C . Due to insolubility issues encountered when expressing McdA-6xHis , a construct was designed where a 6xHis-MBP-tag was encoded upstream of a Tobacco Etch Virus ( TEV ) cleavage site and fused to the N-terminus of the mcdA gene in a pET15b expression backbone to create pAH2 plasmid . pAH2 was transformed into ArcticExpress ( DE3 ) competent cells ( Agilent ) and protein expression was carried out by growing transformants at 37°C and 225 rpm until an OD600 of 0 . 6–0 . 8 was reached . Following an ice bath plunge to lower the culture temperature to 15°C , protein expression was induced with the addition of 0 . 5 mM IPTG . Induction was allowed to continue overnight at 15°C . The cells were pelleted , flash frozen with liquid nitrogen , and stored at −80°C . Cells were then lysed in Buffer A ( 50 mM HEPES pH 7 . 6 , 50 mM KCl , 10% glycerol , 20 mM imidazole pH 7 . 4 , 5 mM BME , 50 µg/ml lysozyme , 1 . 25 kU benzonase , 2 Protease Inhibitor Cocktail tablets ) using a probe sonicator with 15 s on , 15 s off pulsation for 8 min . Cell debris was removed by centrifugation at 14 , 000 rpm for 40 min in a FiberliteTM F15−8 × 50 cy Fixed Angle Rotor ( ThermoFisher Scientific ) and the resulting lysate was filtered through a 0 . 45 µm syringe filter prior to being loaded onto a HiTrapTM Q HP 5 ml cassette ( GE ) connected in tandem to a 5 ml HiTrapTM TALON Crude cassette ( GE ) . The protein was eluted from the Q cassette with a 50 mM – 1 M KCl gradient in an anion exchange chromatography step . The His-tagged protein was then eluted from the TALON column with a 20 mM – 1M imidazole gradient . Peak fractions were pooled , concentrated and further separated by gel filtration on a Superdex200 10/300 GL column ( GE ) pre-equilibrated with 50 mM HEPES pH 7 . 6 , 50 mM KCl , 10% glycerol , 5 mM DTT . Individual peak fractions were concentrated to no higher than 20 µM and frozen aliquots were kept at −80°C . A codon-optimized gene sequence of mcdB-6xHis was inserted into the NcoI/BamHI cloning sites of the expression vector pET15b , yielding pAV42 plasmid . The construct was transformed into BL21 ( AI ) and protein expression was carried out in the same manner as McdA-GFP-His . Frozen cell pellets were thawed and resuspended in lysis buffer containing 50 mM HEPES pH 7 . 6 , 500 mM KCl , 10% glycerol , 20 mM imidazole pH 7 . 4 , 5 mM MgCl2 , 2 mM BME , 50 µg/ml lysozyme , 1 . 25 kU benzonase , 2 Protease Inhibitor Cocktail tablets . Resulting cell lysate was centrifuged , filtered and loaded onto a 5 ml His-TrapTM Ni-NTA cassette ( GE ) . Following protein elution with a 20 mM – 1M imidazole gradient , peak fractions were pooled , concentrated and loaded onto a Superdex200 HiLoad 16/600 PG column pre-equilibrated with 50 mM HEPES pH 7 . 6 , 500 mM KCl , 10% glycerol , 5 mM MgCl2 , 2 mM DTT for final separation . Peak fractions were concentrated to no more than 70 µM and flash frozen aliquots were kept at −80°C . ATPase assays were performed in a buffer containing 50 mM HEPES ( pH 7 . 6 ) , 10 mM MgCl2 , 100 mM KCl , 0 . 1 mg/ml BSA , 2 mM DTT , and 0 . 1 mg/ml sonicated salmon sperm DNA ( when present ) . Unlabeled ATP was spiked with [γ-32P]-ATP and purified from contaminating 32Pi prior to use with a 1 ml gel filtration ( P-2 fine resin , Bio-Rad ) column . The radiolabeled ATP mix was added to reactions at 1 mM . Reactions were assembled on ice at the protein concentrations indicated , with His-MBP-McdA , McdA-GFP-His , F SopA-His or P1 ParA being added last . The 20 µl reactions were incubated for 1 hr at 30°C and immediately quenched by adding 10 µl of a 1% SDS , 20 mM EDTA solution . Two microliters of the quenched reactions were spotted and analyzed by thin-layer chromatography as previously described ( Fung et al . , 2001 ) . Due to the feeble ATPase activities of SopA-His and P1 ParA , specific activities were determined from experiments carried out as shown above , but the 30°C incubation period was carried out for 3 hr . Electrophoretic mobility shift assays ( EMSAs ) were performed in a final reaction volume of 10 µl in a buffer containing 50 mM HEPES ( pH 7 . 6 ) , 5 mM MgCl2 , and 100 mM KCl with 10 nM pUC19 plamsid ( 2 . 8 kb ) as the DNA substrate . At the concentrations indicated , McdA-GFP-His or His-MBP-McdA was incubated for 30 min at 23°C with ADP , ATP or ATPγS ( 1 mM ) . When used , McdB-His was added at the concentrations specified . Reactions were then mixed with 1 μl 80% glycerol , run on 1% agarose gel in 1X TAE at 110V for 45 min and stained with ethidium bromide for imaging . The peak fractions representing the dimer form of His-MBP-McdA from Superdex200 size exclusion chromatography were used . Quartz flowcell construction and DNA-carpeting of the flowcell surface were performed as previously described ( Vecchiarelli et al . , 2014 ) . For the imaging of McdA-GFP-His binding to the DNA carpet , prism-type TIRFM was performed using an Eclipse TE2000E microscope ( Nikon ) with a PlanApo 60 × NA = 1 . 40 oil-immersed objective and magnifier setting at 1 . 5× . Movies were acquired using an Andor DU-897E camera ( Andor Technology ) with integrated shutter . The camera settings were digitizer , 3 MHz ( 14-bit gray scale ) ; preamplifier gain , 5 . 2; vertical shift speed , 2 MHz; vertical clock range: normal , electron-multiplying gain 40 , EM CCD temperature set at –98°C , baseline clamp ON , exposure time 100 ms , frame rate 0 . 5 Hz . The baseline of ∼100 camera units was subtracted from the intensity data . The excitation for McdA-GFP-His was provided by 488 nm diode-pumped solid-state ( Sapphire , Coherent ) laser . Total internal reflection fluorescence illumination had a Gaussian shape in the field of view with measured horizontal and vertical half maximum widths of ∼65 μm × 172 μm at 488 nm . Intensity data for the DNA carpet-bound populations of McdA-GFP were taken from the middle of the illumination profile . The laser power of 488 nm illumination was 15 μW . Metamorph seven software ( Molecular Devices ) was used for camera control and image acquisition . ImageJ was used for analysis . The display brightness and contrast were set to the same levels for all TIRFM movies . ImageJ was used for conversion of Metamorph movies ( . stk ) into . avi format and Adobe Premiere was used for added text . Movie accelerations are indicated in the movie and figure legends . McdA-GFP ( 0 . 5 μM ) was preincubated in McdA Buffer [50 mM Hepes ( pH 7 . 6 ) , 100 mM KCl , 10% ( vol/vol ) glycerol , 5 mM MgCl2 , 2 mM DTT , 0 . 1 mg/mL α-casein , 0 . 6 mg/mL ascorbic acid] with 1 mM of the indicated nucleotide ( or no nucleotide ) . The sample was incubated for 15 min in a 1 ml syringe connected to one of the two inlets of a Y-shaped flowcell . The sample was infused onto the DNA carpet at a rate of 20 μL/min . The fluorescence intensity of McdA-GFP that bound the DNA carpet was measured over time . At t = 3 min , flow from the sample inlet was stopped and immediately switched to the second inlet that was connected to a wash buffer ( McdA Buffer without McdA-GFP or nucleotide ) . Wash buffer was flowed at a rate of 20 μL/min , and the decrease in fluorescence intensity was monitored over time . The two-inlet flowcell had a Y-patterned configuration and imaging took place at the point of flow convergence to minimize the effect of protein rebinding to the DNA carpet during dissociation when flow was switched to the wash buffer . We find that McdA , McdB , and the carboxysome cargo show in vivo dynamics strikingly similar to that found for ParA-mediated DNA partition systems . Therefore , we leveraged our established Brownian ratchet model of ParA/ParB-mediated partition ( Hu et al . , 2015; Hu et al . , 2017 ) to theoretically interrogate the carboxysome positioning process in cyanobacteria . Briefly , the model describes the mechanochemical interplay between nucleoid-bound McdA and carboxysome-bound McdB . McdA and McdB in the current model fulfil exactly the same roles of ParA and ParB as in the low-copy plasmid partition case , respectively . While carboxysome alone diffuses randomly , its motility can be greatly modulated when carboxysome-bound McdB interacts with nucleoid-bound McdA . Specifically , carboxysome-bound McdB stimulates the ATPase activity of nucleoid-bound McdA , which triggers the dissociation of McdA from the nucleoid substrate surface . The slow rate of dissociated McdA resetting its DNA-binding capability generates an McdA-depleted zone behind the moving cargo . The resulting asymmetric McdA distribution perpetuates the directed movement of the carboxysome cargo . Transient tethering arising from the McdA-McdB contacts collectively drives forward movement of the cargo and also quenches diffusive motion in orthogonal directions . This way , McdA/McdB interaction – when at proper mechanochemical coupling – drives directed and persistent movement of carboxysomes ( Hu et al . , 2015; Hu et al . , 2017 ) . The model treats carboxysomes as circular disks that move on the nucleoid surface , which is modeled as a 2D simulation domain . The simulation domain is bounded by the reflective boundary condition . To study the effects of nucleoid geometry on carboxysome positioning , we constructed the simulation domains to mimic I . ) a circular nucleoid and II . ) a more elongated rounded rectangle nucleoid , which consists of a rectangle with two spherical caps at the ends of its long axis . We simulated the model by the same kinetic Monte Carlo technique as in ( Hu et al . , 2015; Hu et al . , 2017 ) , which describes the coupling between the stochastic reaction-diffusion processes involving McdA and McdB , and their mechanochemical interplay . Specifically , we investigated the effects of carboxysome number and nucleoid geometry on carboxysome positioning . In each case defined by different carboxysome number and nucleoid geometry , the simulation starts with the initial positions of the carboxysomes that cluster around the center of the simulation domain ( see Figure 7CE in the main text ) . For each case , we identify the parameter regime in which the carboxysomes undergo ‘directed segregation’ — a motility mode in which the cargoes move away from each other and then become relatively stationary ( e . g . , see Figure 3 in Hu et al . , 2017 ) . That is , the carboxysomes are segregated and then stably positioned with a large inter-spacing . We then tracked the time evolutions of each of the simulated trajectories that are 10 min long , from which the final positions of , and separation distances between , carboxysomes were then calculated and reported in each case ( average ± standard deviation , n = 36 trajectories ) .	Cyanobacteria are tiny organisms that can harness the energy of the sun to power their cells . Many of the tools required for this complex photosynthetic process are packaged into small compartments inside the cell , the carboxysomes . In Synechococcus elongatus , a cyanobacterium that is shaped like a rod , the carboxysomes are positioned at regular intervals along the length of the cell . This ensures that , when the bacterium splits itself in half to reproduce , both daughter cells have the same number of carboxysomes . Researchers know that , in S . elongatus , a protein called McdA can oscillate from one end of the cell to the other . This protein is responsible for the carboxysomes being in the right place , and some scientists believe that it helps to create an internal skeleton that anchors and drags the compartments into position . Here , MacCready et al . propose another mechanism and , by combining various approaches , identify a new partner for McdA . This protein , called McdB , is present on the carboxysomes . McdB also binds to McdA , which itself attaches to the nucleoid – the region in the cell that contains the DNA . McdB forces McdA to release itself from DNA , causing the protein to reposition itself along the nucleoid . Because McdB attaches to McdA , the carboxysomes then follow suit , constantly seeking the highest concentrations of McdA bound to nearby DNA . Instead of relying on a cellular skeleton , these two proteins can organize themselves on their own using the nucleoid as a scaffold; in turn , they distribute carboxysomes evenly along the length of a cell . Plants also obtain their energy from the sun via photosynthesis , but they do not carry carboxysomes . Scientists have tried to introduce these compartments inside plant cells , hoping that it could generate crops with higher yields . Knowing how carboxysomes are organized so they can be passed down from one generation to the next could be important for these experiments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "microbiology", "and", "infectious", "disease" ]	2018	Protein gradients on the nucleoid position the carbon-fixing organelles of cyanobacteria
The pervasive occurrence of sexual dimorphism demonstrates different adaptive strategies of males and females . While different reproductive strategies of the two sexes are well-characterized , very little is known about differential functional requirements of males and females in their natural habitats . Here , we study the impact environmental change on the selection response in both sexes . Exposing replicated Drosophila populations to a novel temperature regime , we demonstrate sex-specific changes in gene expression , metabolic and behavioral phenotypes in less than 100 generations . This indicates not only different functional requirements of both sexes in the new environment but also rapid sex-specific adaptation . Supported by computer simulations we propose that altered sex-biased gene regulation from standing genetic variation , rather than new mutations , is the driver of rapid sex-specific adaptation . Our discovery of environmentally driven divergent functional requirements of males and females has important implications-possibly even for gender aware medical treatments . The ubiquity of sexual dimorphism in dioecious organisms reflects the discordant selection pressure driven by divergent reproductive roles of males and females ( Chapman , 2006 ) . For instance , males typically evolve to increase their mating frequency and success of fertilization , while females benefit from better resource allocation to their offspring ( Brengdahl et al . , 2018; Civetta and Clark , 2000; Friberg and Arnqvist , 2003 ) . Often , such differential requirements of males and females results in sexual conflict , preventing males and females to reach sex-specific trait optima ( Bonduriansky and Chenoweth , 2009; Lande , 1980; Mank , 2017a; Rice , 1992 ) . Based on the widespread sexual dimorphism , several models for the evolution of sexual dimorphism from a largely shared genome have been proposed ( Barson et al . , 2015; Day and Bonduriansky , 2004; Mank , 2017b; Parsch and Ellegren , 2013; Pennell and Morrow , 2013; Rice , 1984; Telonis-Scott et al . , 2009 ) . One implicit assumption of these studies is that stable sex-specific fitness landscapes are persisting over long evolutionary time scales . However , ecological changes , such as environmental fluctuations , occur at high rates ( Reznick and Ghalambor , 2001 ) . If such environmental factors affect the sex-specific fitness landscapes , sudden ecological changes may impose selection for novel/altered sexual dimorphism in a population ( Camus et al . , 2019 ) . To date , limited attention has been given to the evolutionary dynamics of sex differences in response to changing environments . The clinal variation of sexual dimorphism for a small number of phenotypes ( Blanckenhorn et al . , 2006; Chenoweth et al . , 2008 ) and gene expression ( Allen et al . , 2017; Hutter et al . , 2008 ) in Drosophila suggests that sex-specific adaptation in response to environmental heterogeneity is not uncommon . When the requirements of males and females differ in an environment-specific manner , the adaptive response is contingent on the availability of segregating variants with sex-specific or sex-biased effects . Without the corresponding variants , sex-specific adaptation requires new mutations , resulting in slow evolutionary responses . Here , we use experimental evolution for direct experimental evidence that sex-specific adaptation can be triggered by a rapid environmental shift within a few generations . We explored the phenotypic evolution of males and females by studying gene expression because many of these molecular phenotypes can be scored with a high precision at moderate costs . Furthermore , in contrast to high-level phenotypes , which are typically selected on a priori criteria , the analysis of gene expression is unbiased . We measured gene expression of 10 replicate populations which evolved independently for more than 100 generations in a simple and well-controlled high-temperature selection regime ( Barghi et al . , 2019 ) . In each sex , we screened for genes with parallel changes in expression across the replicated evolved populations compared to their same-sex ancestors . After accounting for allometric changes during evolution ( see Materials and methods ) , we identified 2366 and 4151 genes ( 25% and 44% of all expressed genes , N = 9 , 457 ) showing evolutionary responses in males and females respectively ( FDR < 0 . 05; Supplementary file 1 and Figure 1—figure supplement 1 ) . The evolution in gene expression was vastly different between the sexes , resulting in almost uncorrelated gene expression changes ( Figure 1a ) . Only 760 genes ( 14%; 469 up-regulated and 291 down-regulated ) evolved concordantly in both sexes . 1295 genes ( 24% ) changed expression specifically in males ( 657 up-regulated and 638 down-regulated ) and 3080 genes ( 57% ) evolved in females only ( 1877 up-regulated and 1203 down-regulated ) . Interestingly , 311 genes ( 6% ) displayed divergent responses to selection in the two sexes ( Figure 1b ) . The discordant gene expression evolution of males and females indicates different functional requirements in the novel environment . To determine the diverged functional requirements of males and females in the new environment , we tested for enrichment of gene ontology ( GO ) terms and tissue-specific expression ( Figure 1c and d , Supplementary files 2 and 3 ) . We found a striking pattern of enrichment that suggested sex-specific evolution of fatty acid metabolism in both the GO term and tissue-specific enrichment analyses . Genes highly expressed in fat body tissue were over-represented among the 1280 genes with upregulation in males , but over-represented among the 1648 genes with downregulation in females ( FET , FDR < 0 . 01 in both tests , Figure 1d and Supplementary file 3 ) . GO enrichment analysis of genes with male-specific upregulation further highlighted biological processes like ‘lipid metabolic process’ , ‘acyl-CoA biosynthetic process’ , ‘fatty acid elongation’ and ‘triglyceride catabolic process’ ( Supplementary file 2 ) . Similar GO categories were enriched among the 154 antagonistically evolving genes that were upregulated in males but downregulated in females ( Supplementary file 2 ) . Interestingly , two apparently counteracting processes , fatty acid synthesis and degradation , were both upregulated in males ( Figure 2a ) whereas in females , only genes involved in fatty acid synthesis were significantly downregulated ( Figure 2a ) . A link between these changes in gene expression and a higher-level phenotype is suggested by the observation that these laboratory populations experienced a significant decrease of fat content only in females but not in males ( Barghi et al . , 2019; Figure 2b ) . In addition , sex-specific responses to selection in gene expression were also related to neuronal signaling . The evolution of dopamine signaling during temperature adaptation has previously been reported in male flies of the same population ( Jakšić et al . , 2019 ) . The 1086 genes that evolved decreased expression in males were enriched in brain and ganglion tissues ( FET , FDR < 0 . 001 in both tests; Figure 1d and Supplementary file 1 ) whereas there was no enrichment in these tissues for females . Likewise , gene expression of dopaminergic processes ( e . g . : Ddc , DAT and Dop1R2 ) evolved downregulation in males but did not evolve in females ( Figure 2c ) . In contrast , only females evolved increased expression of genes involved in octopamine biosynthesis and signaling ( e . g . : Tdc1 , Tdc2 and Octα2R ) ( Figure 2c ) . The sex-specific modulation of transcriptional activity in different neuronal circuits may trigger changes in sex-specific fitness-related behaviors such as male courtship and female oviposition . In support of this hypothesis , the GO terms ‘copulation’ and ‘male courtship behavior’ were enriched among the 154 antagonistic genes up-regulated in males , as was ‘oviposition’ among the 1877 genes with female-specific up-regulation ( Supplementary file 2 ) . The increased fecundity of evolved females ( Barghi et al . , 2019 ) fits the expectations for increased octopamine synthesis ( Cole et al . , 2005; Monastirioti , 2003 ) . Female fecundity is , however , a complex trait which may be affected by many factors other than increased octopamine level . We tested therefore another octopamine-related phenotype that was not selected in the experiment , ovarian dormancy in response to cold temperatures ( Andreatta et al . , 2018 ) . Confirming the increased octopamine level in the evolved females , dormancy incidence was lower at two different dormancy-inducing temperatures ( 10°C and 12°C ) ( Figure 2d and Figure 2—figure supplement 1 ) . Further , we also observed changes in male-specific behavior after 100 generations of adaptation; evolved males spent more time chasing females and made more copulation attempts than ancestral ones ( Figure 2e and Figure 2—figure supplement 2 ) . The sexually discordant evolution of several phenotypes , including gene expression , metabolism and behavior , provides evidence that sex-specific adaptive processes occurred in experimental populations exposed to a novel temperature regime . This raises the important question of how potentially conflicting selection pressures on the shared genome have been decoupled during 100 generations of evolution . Sexually dimorphic gene expression is abundant in Drosophila ( Parsch and Ellegren , 2013 ) and 95% of the genes in the ancestral population of this study are also sex-biased ( Supplementary file 1 ) . This implies the decoupling of selection on the gene expression in males and females ( Mank , 2017a ) as well as the presence of a sex-biased regulatory architecture of the transcriptome ( Mank , 2017b; Parsch and Ellegren , 2013; Pennell and Morrow , 2013 ) in the ancestral population . Transcription factors ( TF ) with sex bias in expression or splicing are the key factor underlying this sex-biased regulatory architecture ( Mank , 2017a ) . It has been hypothesized that relatively fast sex-specific responses to discordant selection may be driven by fixation of novel mutations resulting in sex-biased gene expression ( Stewart et al . , 2010; van Doorn , 2009 ) . However , we observe sex-specific expression changes across replicates after only 100 generations , in which case de novo mutations in individual replicates are unlikely to be the driver ( Burke et al . , 2010 ) . Rather , selection on standing genetic variation in existing sex-specific genetic architecture seems the most likely mechanism allowing replicated populations to approach different functional requirements of the two sexes in the new environment over such a short timescale ( Figure 3 ) . Candidate TFs supporting this hypothesis would regulate both genes with sex-biased expression ( criterion 1 ) and genes with a significant evolution of sex bias in expression ( criterion 2 ) . Furthermore , the sex bias of these TFs must have evolved in a direction compatible with the changes of their target genes ( criterion 3 ) . Of 656 annotated TFs expressed in our populations , 300 TFs evolved a change in sex-biased expression ( i . e . either evolve a new sex bias or the ancestral sex bias changes ) ; 210 and 80 evolved either in females or males , respectively , and 10 changed in opposite direction in the two sexes ( Supplementary file 4 ) . Based on cis-regulatory element enrichment , we identified 69 TFs which regulate genes with sex-biased expression and a total of 198 TFs that target genes with sex bias evolving in opposite direction ( Supplementary file 5 ) . In the end , 19 TFs satisfied all our three criteria for the most likely candidates targeted by the discordant selection in the two sexes ( Supplementary file 6 ) . Despite genomic time series data being available for these populations ( Barghi et al . , 2019 ) , extensive linkage structure in the populations preclude an unambiguous identification of selected TF alleles . Future functional studies will show which of these candidate TFs are accomplishing the decoupling of male and female requirements and which molecular processes contribute to adaptation of the two sexes in a novel temperature regime . Nevertheless , we caution that the evolution of gene expression is most likely polygenic , with several-or even many loci contributing to the evolution of sex bias . In this case , both genomic responses and functional tests may be complicated due to the expected small effects of individual loci . Using computer simulations , we further corroborated the hypothesis that sex-specific adaptation can be achieved rapidly in the presence of segregating regulatory variants which alter the sex bias of a trait . Based on the haplotype information of the founder lines initiating the experiment ( Barghi et al . , 2019 ) , we simulated traits ( expression value ) each controlled by 50 additive loci ( TFs ) of which 0 , 1 , 2 , 5 , 10 or 20 are sex-specific ( effect size = 0 in one sex ) /sex-biased ( 2-fold difference in effect size ) . The simulated populations were exposed to a selection regime where males and females of the same population have different fitness optima for the focal trait and we monitored the phenotypic change in each sex during 100 generations . 100 simulations were performed under each scenario . Without sex bias in the effect size ( rmf = 1 ) , neither males nor females could respond to the discordant selection ( Figure 4 ) . With 40% of the loci contributing to the trait being sex-specific ( rmf = 0 . 49 ± 0 . 2 ) or sex-biased ( rmf = 0 . 87 ± 0 . 05 ) , both males and females can evolve toward the opposing optima ( Figure 4 and Figure 4—figure supplement 1 ) . Nevertheless , sex-specific or sex-biased expression is not required for many contributing loci . Already two sex-specific ( rmf = 0 . 94 ± 0 . 08 ) loci significantly decouple the response of the two sexes ( Figure 4b ) under opposing selection pressures . As discussed above , the rapid sex-specific responses , which are highly parallel across replicates , in combination with the gain and loss of sexual dimorphism ( Figure 1—figure supplement 2 ) highlight the importance of standing genetic variation in sex-biased regulatory architecture . This raises the interesting question of how genetic variation with sex-biased effects is maintained . Assuming a simple genetic basis and a stable fitness landscape with pronounced differences between the two sexes , alleles with dimorphic expression are expected to become fixed . We propose two , not mutually exclusive hypotheses to explain the discrepancy to our observation . First , the fitness landscape of some sex-specific phenotypes could vary in response to environmental fluctuation . In this case , alleles controlling the sex difference of a trait could be segregating and maintained in a population . As natural Drosophila populations regularly encounter seasonal temperature fluctuations , candidate alleles regulating sex-specific temperature adaptation can be maintained at sufficiently high frequencies to facilitate rapid responses . The impact of seasonal variation on oscillating allele frequency changes has been recently described experimentally and theoretically ( Bergland et al . , 2014; Wittmann et al . , 2017 ) . The second hypothesis assumes a polygenic basis . We note that unambiguous sex-limited modifiers ( e . g . male and female isoforms of doublesex; Kopp et al . , 2000 ) do not preclude polygenic adaptation — these sex-limited modifiers may regulate many down-stream regulators that respond to the environmental change . Thus , already minor frequency shifts of these down-stream regulators could mediate the observed evolution of sex-specific gene expression changes . Importantly , under polygenic adaptation segregating variation is maintained for rather long time-scales ( Barton and Keightley , 2002; Gillespie , 1984; Gillespie and Turelli , 1989 ) . Indirect support for this hypothesis comes from the observation that no significant SNPs explaining the sex difference for multiple human traits can be identified ( Randall et al . , 2013 ) . Under this scenario , rapid evolution of the sex difference may be achieved by the heterogeneous genotypic changes across replicated populations ( Barghi et al . , 2019 ) . This study demonstrates the power of experimental evolution to study sex-specific adaptation after an environmental shift . A substantial fraction of the transcriptome and related high-level phenotypes rapidly developed discordant changes in the two sexes upon exposure to a new environment . We propose that variation segregating in the ancestral population has facilitated the evolution of sex-biased gene expression , which in turn provides the basis for the sex-specific adaptation evidenced by the broad range of phenotypes evolving in different directions in males and females . While we provided robust evidence for sex-specific adaptation , it is important to keep in mind that the identification of the selected traits in both sexes is an extremely challenging task . While 60% of genes changed expression in a sex-specific manner , it is unlikely that each of them is independently selected . We can anticipate many ways how the sex specific phenotypic changes have been achieved , ranging from allometric changes during adaptation to selection acting on cis-regulatory variation of highly pleiotropic transcription factors . Further characterization of the adaptive changes needs to distinguish between two goals . One goal , which is pursued in many studies , is the identification of the gene ( s ) that experienced a frequency change of a favored variant as contribution to the adaptive phenotype . The other goal is the identification of the selected phenotype . Given the pleiotropic effects of many genes and the polygenicity of most adaptive phenotypes ( Barghi et al . , 2019; Pritchard et al . , 2010 ) , it is apparent that the characterization of individual selected alleles has clear limitations in reaching the later goal . In fact , the justification of studies aiming to characterize adaptive allelic variants has been challenged ( Rockman , 2012 ) . More rewarding would be the characterization of the adaptive trait , which is selected in a sex-specific manner . Our enrichment analysis and characterization of high-level phenotypes aimed towards this direction , but we cannot distinguish between correlated phenotypic changes and the actual selected phenotypes . While most of this report focused on the rapid evolution of sex-specific adaptation , the driving forces behind this have not been discussed to the same extent , largely because they will require further functional characterization . Nevertheless , in line with sex-dependent dietary effects on fitness ( Camus et al . , 2019 ) , the fact that males and females have vastly different functional requirements after being exposed to a novel environment has far reaching consequences-well beyond Drosophila . We anticipate that our results will have profound influence on biomedical research and medical treatments which need to account for the overwhelming differences of the two sexes in particular with respect to new environmental stressors , reaching from diet to climatic conditions . The set-up of the experimental evolution populations is described in Barghi et al . ( 2019 ) . In brief , 10 replicated outbred populations were constituted from 202 isofemale lines derived from a natural Drosophila simulans population collected in Tallahassee , Florida , USA in 2010 . Replicated populations have been independently adapting to a laboratory environment at 18/28°C with 12 hr dark/12 hr light photoperiod for more than 160 generations with a census population size of 1000–1250 adults per population per generation . The collection of samples for RNA-Seq and all other phenotypic assays , was preceded by two generations of common garden rearing . Two different RNA-Seq data sets were generated for this study: The first one , in which highly replicated whole body samples were collected , represents the main dataset that we used to contrast gene expression levels of females and males from ancestral and hot evolved populations . The second one with gonads and carcass being analyzed separately was generated to correct for allometric differences between evolved and ancestral populations . The first data set comes from a common garden experiment ( CGE ) performed after 103 generations of evolution in the hot environment and this CGE has been described in Barghi et al . ( 2019 ) ; Hsu et al . ( 2019 ) ; Jakšić et al . ( 2019 ) . We reconstituted five replicates of the ancestral population from 184 founder isofemale lines by generating five pools with five mated females from each isofemale line . No significant allele frequency differences are expected between the reconstituted ancestral populations and the original ancestral populations initiating the experiment ( Nouhaud et al . , 2016 ) . Because we evaluated phenotypes on the population level , even deleterious mutations will have a very limited impact , because they occur only in a single isofemale line , which represents a very small fraction of the total population . For each of the 10 hot evolved replicates , we generated three sub-replicates by multiple egg lays . The five ancestral replicates and all hot evolved sub-replicates were reared in common garden for two generations with controlled low egg density ( 400 eggs/bottle ) in the same temperature regime as during the evolution experiment . After two generations under CGE conditions , flies were collected from each replicate/sub-replicate a few hours after eclosion and maintained on fresh food under the 18/28°C temperature regime to allow for mating . On the third day after eclosion , sexes were separated under CO2 anesthesia and allowed to recover for two days . At the age of five days , 50 flies of each sex were snap frozen in liquid nitrogen at 2pm and stored at −80°C until RNA extraction . We sequenced the transcriptomes of 50 females and males from each of the five ancestral replicates and from each of the 10 hot evolved replicates with three sub-replicates each for males and two sub-replicates for females . The third sub-replicate of the hot evolved female samples was frozen at a different age which prevented the joint analysis in the context of this study ( Hsu et al . , 2019 ) . The second RNA-Seq data set was generated at generation 140 of the hot evolving populations to correct for potential differences in the relative size of gonadal and carcass tissue between ancestral and evolved populations . CGE set-up and maintenance were repeated as described above , without sub-replication of the hot evolved replicates: 50 whole body samples for females and males were collected from five reconstituted ancestral and all 10 hot evolved replicates and snap-frozen at the age of five days at 2pm . Gonadal and carcass tissue was sampled from six reconstituted ancestral and six randomly chosen hot evolved replicates ( replicates no . 1 , 4 , 5 , 6 , 8 , 9 ) . For each replicate , 50 female and 50 male flies were dissected in PBS at the age of 5 days and dissected gonadal tissues and remaining carcasses were immediately preserved in Qiazol and stored at −80°C . Total RNA was extracted using the same procedure for all samples: homogenized in Qiazol with a pestle . Total RNA was extracted from the homogenate using the Qiagen RNeasy Universal Plus Mini kit ( Qiagen , Hilden , Germany ) with DNase treatment to remove traces of genomic DNA . Libraries were prepared on the Neoprep Library Prep System ( Illumina , San Diego , USA ) starting from 100 ng total RNA and following the manufacturer’s recommended protocol for the TruSeq stranded mRNA Library Prep Kit for Neoprep . Neoprep runs were performed using software version 1 . 1 . 0 . 8 and protocol version 1 . 1 . 7 . 6 with default settings for 15 PCR cycles and an insert size of 200 bp . Libraries were arranged in randomized order on library cards . To avoid batch effects , we used library cards with the same lot number for all samples for which direct comparisons of expression levels were planned ( lot no . 20123465: CGE at generation 103 , males , whole body , all ancestral and hot evolved samples; lot no . 20173962: CGE at generation 103 , females , whole body , all ancestral and hot evolved samples; lot no . 20182049: CGE at generation 140 , females and males , whole body and gonadal tissue ) . 50 bp single-end reads were sequenced on an Illumina HiSeq 2500 . All sequencing reads were trimmed with ReadTools ( Version: 1 . 5 . 2 ) ( Gómez-Sánchez and Schlötterer , 2018 ) based on a quality score of 20 , and mapped with GSNAP ( Version: 2018-03-25; Parameters: -k 15 N 1 m 0 . 08 ) ( Wu and Nacu , 2010 ) to Drosophila simulans reference genome ( Palmieri et al . , 2015 Supplementary file 7 ) . Exon-aligned reads were counted with Rsubread ( Version: 1 . 30 . 9 ) ( Liao et al . , 2013 ) based on the annotation ( Palmieri et al . , 2015 ) and the expression level of each gene was quantified after normalizing the exon-aligned read counts by TMM method implemented in edgeR ( Version: 3 . 22 . 5 ) ( Robinson et al . , 2010 ) . Only genes with more than 0 . 1 count per million base pairs in each sample of the main dataset ( 1st CGE ) were retained for the analysis to avoid biased analyses . Using an independent CGE that consisted of dissected samples ( 2nd CGE , correcting dataset ) , we corrected for potential differences in the relative size of gonadal and remaining carcass tissues in ancestral and hot evolved populations for each gene . For each gene , we formulated its average expression across whole-body samples ( yWb , i¯ ) with the average expression across gonad samples ( yg , i¯ ) and carcass samples ( yc , i¯ ) as: yWb , i¯=αiyg , i¯+ ( 1−αi ) yc , i¯ , where α is the coefficient measuring the relative portion of gonadal expression of a gene in whole body expression , ranging from 0 to 1 ( Supplementary file 10 ) . If a gene is expressed at similar level in both gonadal and somatic tissues , it would not be affected by differences in tissue scaling . We excluded these genes in the comparison of tissue-scaling and applied no correction for them in the subsequent analysis . Leave-one-out cross validation was performed to evaluate the accuracy and robustness of the method . The estimation of the scaling coefficients for each gene was robust ( Supplementary file 8 ) . In addition , the prediction was nearly perfect ( Supplementary file 9 ) . Comparing the distribution of gene-wise estimates of scaling coefficients , we found significant difference between ancestral and evolved populations for both sexes ( Kolmogorov-Smirnov test D = 0 . 18 and 0 . 12 for females and males , respectively; p<0 . 001 in both tests; Supplementary file 11 ) . This suggested that the gonad-carcass size ratio may have significantly changed during the adaptation to the new environment . A proper correction is necessary for unbiased inference . Hence , we normalized the tissue-scales of each ancestral sample to the scale of evolved samples . We reconstructed pseudo whole-body samples using the expression data of dissected samples of the ancestral populations and scaling coefficients estimated from the evolved samples as: yWb , ipseudo=αi^evoyg , i+ ( 1-αi^evo ) yc , i . Finally , the ratio of expression levels between the reconstructed pseudo whole-body samples and the original ones ( yWb , ipseudoyWb , i ) for each gene were calculated as the correcting factors ( γi^ ) . Gene-wise correction was applied to ancestral whole-body samples from the 1st CGE by multiplying the expression value of each gene to corresponding γi^ . The corrected samples were used in all subsequent analyses . After correction , we modeled the effects of sex and evolution on gene expression variation as: Y=group+ε , where Y is the normalized expression values; group indicates the combination of evolution and sex difference with four levels ( ancestral females , ancestral males , evolved females and evolved males ) and ε is the random error . Likelihood ratio tests implemented in edgeR were used to perform differential expression analysis on three contrasts: ( 1 ) female evolution: evolved females vs . ancestral females , ( 2 ) male evolution: evolved males vs . ancestral males and ( 3 ) sex bias: females vs . males . Benjamini and Horchberg’s FDR correction ( Benjamini and Hochberg , 1995 ) was applied with the significance threshold of FDR < 0 . 05 . Genes showing distinct evolutionary patterns were classified based on criteria in ( Supplementary file 2 ) . Gene ontology ( GO ) enrichment was performed using the default ‘weight01’ algorithm implemented in topGO ( version 2 . 32 . 0 ) ( Alexa et al . , 2006 ) . Genes highly expressed in each tissue were identified based on the FlyAtlas expression dataset ( Chintapalli et al . , 2007 ) ( required >2 fold higher expression in a certain tissue than whole-body , Supplementary file 3 ) . Fisher’s exact test was applied for the enrichment of tissue expression . Except for the GO enrichment analysis of which the method already accounts for multiple testing ( Alexa and Rahnenführer , 2018 ) , Benjamini and Horchberg’s FDR correction ( Benjamini and Hochberg , 1995 ) was applied to account for multiple testing . Enrichment of cis-regulatory elements ( CREs ) 5 kb upstream and intronic sequences of the genes of interest ( Supplementary file 5 ) was identified with RcisTarget ( version 1 . 0 . 2 ) ( Aibar et al . , 2017 ) . We searched for enriched motifs using the latest motif ranking file of Drosophila species ( ‘dm6-5kb-upstream-full-tx-11species . mc8nr . feather’ , accessed on 2019-04-08 ) with parameters , nesThreshold = 3 and aucMaxRank = 1% . Transcription factors ( TFs ) annotated to bind on the enriched CREs were considered as candidate master TFs regulating the genes of interest . We performed cis-regulatory element enrichment analysis on female-biased , male-biased , female-specifically up-regulated , down-regulated , male-specifically up-regulated , down-regulated , and two sets of antagonistically evolving genes separately ( Supplementary file 5 ) . We measured the reproductive activity of five reconstituted ancestral populations and five randomly selected hot evolved replicates at generation 140 . After two generations reared in a common garden condition ( 18/28°C cycling ) , 10 five-day-old mated males and females from each population were placed together in an agar-based arena ( 4% agar , 4% sugar ) and filmed for 15 min at 20 FPS ( frame-per-second ) at 28°C using the FlyCapture2 system ( PointGrey , Version 2 . 13 . 3 . 31 ) . In total , 10 video each for reconstituted ancestral and evolved populations were filmed . The movement and behavior of each fly was tracked using flytracker ( Version 1 . 0 . 5 ) ( Eyjolfsdottir et al . , 2014 ) . Videos that failed the tracking process were not used for subsequent analysis . Janelia Automatic Animal Behavior Annotator ( JAABA , Version 0 . 6 . 0_2014a ) was used to annotate and recognize the chasing and attempted copulation behavior ( Kabra et al . , 2013 ) . We imported the output files of JAABA into R for data processing and statistical analysis . The time a male fly spent on chasing and copulation attempt females was quantified . Wilcoxon’s rank sum test was applied to test the difference in reproductive activity of male flies in evolved and ancestral populations . We screened three replicates of the reconstituted ancestral and 10 replicated evolved populations for dormancy incidence at generation 167 . Ancestral and evolved populations were kept at the same temperature regime for four generations before freshly eclosed female flies were collected within two hours post-eclosion and kept under dormancy-inducing conditions ( 10°C and 12°C , LD 10:14 ) for three weeks before dissection . 90 flies from each population and temperature regime were dissected and their oogenesis progression was examined . Each fly was classified as dormant or non-dormant ( Lirakis et al . , 2018 ) . Wilcoxon’s rank sum test was applied to test the difference in dormancy level of female flies in evolved and ancestral populations . We performed forward simulations using qff function implemented in MimicrEE2 ( v208 ) ( Vlachos and Kofler , 2018 ) . Starting with 189 founder haplotypes ( Barghi et al . , 2019 ) , in each sex , we simulated a trait controlled by a varied number of loci ( 0 , 1 , 2 , 5 , 10 , 20 ) conferring sex-specific or sex-biased effects while the total number of contributing loci in each sex was constantly 50 . For each trait , we assumed an additive model ( a~Γ0 . 5 , 2 . 5 ) and relatively high heritability ( h2 = 0 . 8 ) . A sex-specific locus confers additive effect on a trait in one sex but no effect in the other sex while a sex-biased locus is assumed to contribute to the trait in both sexes but there is a 2-fold difference in its additive effect between the two sexes . Sexually discordant selection , where the trait optimum is shifted three units ( i . e . on average , 1 . 9 phenotypic standard deviations ) to the left and to the right for males and females respectively , was imposed to the simulated traits for 100 generations assuming balanced sex-ratio . In total , we performed 100 independent simulations for each of the six scenarios in this study . Then , we measured the normalized phenotypic responses to the selection as Δp¯100−0σ02 , where Δp¯100−0 is the mean phenotypic difference between F100 and F0 of the populations and σ02 is the phenotypic variance when the experiment starts . We calculated the fractions of simulations in which expected phenotypic responses in the two sexes ( increase in males but decrease in females ) were observed . One-sample proportion test was performed to test for significant difference between each scenario to the control group . Bonferroni’s correction was applied to account for multiple testing .	Male and female animals of the same species sometimes differ in appearance and sexual behavior , a phenomenon known as sexual dimorphism . Both sexes share most of the same genes , but differences can emerge because of the way these are read by cells to create proteins – a process called gene expression . For instance , certain genes can be more expressed in males than in females , and vice-versa . Most studies into the emergence of sexual dimorphism have taken place in stable environments with few changes in climate or other factors . Therefore , the potential impact of environmental changes on sexual dimorphism has been largely overlooked . Here , Hsu et al . used genetic and computational approaches to investigate whether male and female fruit flies adapt differently to a new , hotter environment over several generations . The experiment showed that , after only 100 generations , the way that 60% of all genes were expressed evolved in a different direction in the two sexes . This led to differences in how the males and females made and broke down fat molecules , and in how their neurons operated . These expression changes also translated in differences for high-level biological processes . For instance , animals in the new settings ended up behaving differently , with the males at the end of the experiment spending more time chasing females than the ancestral flies . These findings demonstrate that male and female fruit flies adapt many biological processes ( including metabolism and behaviors ) differently to cope with changes in their environment , and that many different genes support these sex-specific adaptations . Ultimately , the work by Hsu et al . may inform medical strategies that take into account interactions between the patient’s sex and their environment .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2020	Rapid sex-specific adaptation to high temperature in Drosophila
The motor cortex ( M1 ) is classically considered an agranular area , lacking a distinct layer 4 ( L4 ) . Here , we tested the idea that M1 , despite lacking a cytoarchitecturally visible L4 , nevertheless possesses its equivalent in the form of excitatory neurons with input–output circuits like those of the L4 neurons in sensory areas . Consistent with this idea , we found that neurons located in a thin laminar zone at the L3/5A border in the forelimb area of mouse M1 have multiple L4-like synaptic connections: excitatory input from thalamus , largely unidirectional excitatory outputs to L2/3 pyramidal neurons , and relatively weak long-range corticocortical inputs and outputs . M1-L4 neurons were electrophysiologically diverse but morphologically uniform , with pyramidal-type dendritic arbors and locally ramifying axons , including branches extending into L2/3 . Our findings therefore identify pyramidal neurons in M1 with the expected prototypical circuit properties of excitatory L4 neurons , and question the traditional assumption that motor cortex lacks this layer . ‘Agranular’ cortical regions such as the primary motor cortex ( M1; area 4 ) are so named as they are commonly held to lack layer 4 ( L4 ) ( Brodmann , 1909 ) . The apparent absence of L4 has strongly influenced theories of cortical organization ( Shipp , 2005; Bastos et al . , 2012; Shipp et al . , 2013 ) . Nevertheless , various observations—such as subtle changes in cell density , expression patterns of various molecular markers , branching patterns of thalamocortical axons , and retrograde labeling termination—suggest that motor cortex might contain some sort of L4 homolog ( Krieg , 1946; von Bonin , 1949; Caviness , 1975; Deschênes et al . , 1979; Skoglund et al . , 1997; Cho et al . , 2004; Kuramoto et al . , 2009; Rowell et al . , 2010; Mao et al . , 2011; Kaneko , 2013; García-Cabezas and Barbas , 2014 ) . For example , Rorb expression in mouse S1 is highest in L4 ( Schaeren-Wiemers et al . , 1997 ) ( Figure 1A ) , and a similar if weaker and thinner pattern is seen in M1 ( Figure 1B ) , coincident with the L3/5A border ( Schaeren-Wiemers et al . , 1997; Shepherd , 2009; Rowell et al . , 2010 ) . In primate M1 , Rorb is also expressed but at lower levels than in sensory cortices ( Bernard et al . , 2012 ) , and a recent report presented evidence for the existence of L4 based on cytoarchitecture and SMI-32 labeling patterns ( García-Cabezas and Barbas , 2014 ) . 10 . 7554/eLife . 05422 . 003Figure 1 . L4 in M1 as a zone of Rorb expression . Images are from the Allen Mouse Brain Atlas ( http://mouse . brain-map . org/ ) ( Lein et al . , 2007 ) showing coronal sections for Rorb in situ hybridization ( probe Rorb-RP_071018_01_H03 ) and corresponding Nissl stains . ( A ) S1 labeling . In a Nissl-stained section ( left ) , L4 is readily identifiable due to cell density differences across layers . In situ hybridization labeling of Rorb ( center , with corresponding expression intensity image shown on the right ) is the strongest in L4 ( long arrow , with borders indicated by lines ) , with weaker labeling present in L5A/B ( short arrow ) . ( B ) M1 labeling . Nissl stain ( left ) showing a region of the lateral agranular cortex corresponding to the forelimb representation area of M1 ( same section as in panel A ) . L4 is not readily identifiable based on cell density differences alone . Nevertheless , in situ hybridization against Rorb ( center and right ) shows the strongest labeling in a laminar zone corresponding to L4 in S1 ( long arrow , with borders indicated by lines ) , with weaker labeling present in L5A/B ( short arrow ) . Scale on the far right shows the normalized cortical distance from pia to white matter ( WM ) . The approximate location of the cortical layers is indicated , based on prior quantitative analysis of the bright-field optical appearance of M1 layers ( Weiler et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 003 Although evidence based on markers is useful and highly suggestive , establishing that M1 truly possesses a functional L4 requires showing that neurons in this band have the same input–output connectivity as their counterparts in sensory areas ( Alfano and Studer , 2012; Feldmeyer et al . , 2013 ) . In barrel and other sensory cortices in rodents , L4 is characterized by strong input from primary thalamus , an extensive and largely unidirectional projection to the superficial cortical layers , a paucity of inputs from other cortical areas , and a paucity of long-range cortical outputs ( Petersen , 2007; Svoboda et al . , 2010; Feldmeyer , 2012 ) . We found that mouse M1 contains pyramidal neurons in a thin laminar zone at the L3/5A border with all these properties . In sensory cortical areas , L4 neurons receive strong thalamocortical ( TC ) excitation from primary sensory thalamic nuclei ( Douglas and Martin , 2004; Feldmeyer , 2012; Harris and Mrsic-Flogel , 2013 ) . If the Rorb-expressing zone in M1 is similarly organized , then neurons in that laminar location should receive strong TC input from the primary motor thalamic nuclei , particularly the ventrolateral nucleus ( VL ) . This is suggested by previous anatomical work ( Strick and Sterling , 1974; Jones , 1975; Cho et al . , 2004; Kuramoto et al . , 2009; Kaneko , 2013 ) ; however , while monosynaptic VL input to pyramidal neurons in the upper layers of vibrissal M1 was recently demonstrated using an optogenetic-electrophysiological approach ( Hooks et al . , 2013 ) , it was not possible to determine if this input terminated in a putative L4 or L2/3 , as vibrissal M1 is highly compressed due to its location at a cortical flexure ( von Economo , 1929; Hooks et al . , 2011 ) . In this study , we focused on the forelimb area of M1 ( Weiler et al . , 2008; Tennant et al . , 2010 ) , located in the lateral agranular cortex ( area 4 ) ( Caviness , 1975 ) where the upper layers are not compressed in this manner , and putative L4 can be more easily distinguished from more superficial layers . ( For convenience , we henceforth refer to this forelimb region simply as ‘M1’ . ) To map input connections , we used an optogenetic strategy ( Hooks et al . , 2013 ) , injecting AAV-ChR2-Venus in VL and subsequently preparing coronal slices containing M1; recording conditions were set to isolate monosynaptic inputs ( Petreanu et al . , 2009 ) . Laminar profiles of the fluorescence intensity of labeled VL axons showed three peaks , in L1 , the L3/5A border , and the L5B/6 border , similar to vibrissal M1 ( Hooks et al . , 2013 ) ( Figure 2A ) . In each slice , we recorded from neurons at the L3/5A border ( i . e . , putative L4 neurons ) and from additional neurons across other layers , thereby obtaining a laminar profile of the excitatory TC input from VL ( Figure 2B , C ) . This analysis revealed two distinct peaks of TC input , the uppermost of which coincided with the L3/5A border ( normalized cortical depth , ∼1/3 ) ( Figure 2C , D , E ) . These data thus indicate that M1 contains neurons in a laminar zone corresponding to L4 that receives strong monosynaptic excitatory TC input from a primary thalamic nucleus associated with this cortical area , thereby fulfilling one important circuit-level criterion for the identification of L4 in M1 . For convenience , we henceforth refer to these as M1-L4 neurons . 10 . 7554/eLife . 05422 . 004Figure 2 . Thalamocortical ( TC ) input to M1-L4 neurons from VL . ( A ) Epifluorescence image of coronal slice containing M1 , showing laminar pattern of labeled thalamocortical axons following injection of AAV carrying ChR2 and GFP in the ventrolateral ( VL ) region of the thalamus . S1 cortex is located laterally ( to the left , as indicated ) . Scale shows normalized cortical distance . Yellow arrow indicates laminar zone of labeling where strong photostimulation-evoked electrophysiological responses were also detected . Plot to the right shows laminar profile of fluorescence intensity , in arbitrary units ( A . U . ) , across layers ( normalized distance from pia ) . ( B ) Responses recorded ( sequentially ) in vitro in multiple M1 neurons in different layers ( as indicated ) to photostimulation of ChR2-labeled axons originating from motor thalamus ( VL ) neurons ( Hooks et al . , 2013 ) . ( C ) Laminar profile of VL input to M1 neurons . The profile exhibits two peaks , one in the upper ∼1/3 of the cortex ( corresponding to L4 ) and the other in the lower part ( corresponding to L5B ) . ( D ) Laminar profiles obtained from multiple slices ( n = 6 ) . Most profiles show a clear peak at a normalized depth of ∼1/3 ( black arrow ) . ( E ) Average laminar profile ( black; bars: s . e . m . ) , calculated by binning the data for each profile ( bin width: 1/10 of the normalized cortical depth ) , averaging within each bin , and then averaging across all profiles . The individual profiles are also shown ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 004 Prior to investigating the intracortical circuits of M1-L4 neurons ( next section ) , we extended this analysis of TC inputs to address whether M1-L4 neurons also receive inputs from the posterior nucleus ( PO ) of the thalamus . The fluorescence intensity of the labeled PO axons showed peaks corresponding to L1 and the L3/5A border ( but , unlike the VL profile , not the L5B/6 border ) , similar to the pattern in vibrissal M1 ( Hooks et al . , 2013 ) ( Figure 3A ) . Laminar profiles of the relative amount of monosynaptic TC input from PO axons to M1 neurons ( Figure 3B , C ) showed a broad peak in the upper layers that included L4 and adjacent layers ( Figure 3D , E ) . Thus , PO's laminar input pattern in M1 resembled its laminar innervation of secondary somatosensory cortex ( S2 ) ( Pouchelon et al . , 2014 ) and the septum-related columns of rat barrel cortex ( Lu and Lin , 1993; Feldmeyer , 2012 ) , rather than its innervation of S1 barrel-related columns themselves ( Feldmeyer , 2012 ) . 10 . 7554/eLife . 05422 . 005Figure 3 . Thalamocortical ( TC ) input to M1-L4 neurons from PO . ( A ) Epifluorescence image of coronal slice containing M1 , showing laminar pattern of labeled thalamocortical axons following injection of AAV carrying ChR2 and GFP in the PO region of the thalamus . S1 cortex is located laterally ( to the left , as indicated ) . Scale shows normalized cortical distance . Yellow arrow indicates laminar zone of labeling where strong photostimulation-evoked electrophysiological responses were also detected . Plot to the right shows laminar profile of fluorescence intensity , in arbitrary units ( A . U . ) , across layers ( normalized distance from pia ) . ( B ) Responses recorded ( sequentially ) in vitro in multiple M1 neurons in different layers ( as indicated ) to photostimulation of ChR2-labeled axons originating from sensory thalamus ( posterior nucleus; PO ) neurons ( Hooks et al . , 2013 ) . ( C ) Response amplitudes of the same neurons plotted as a function of laminar location , providing a laminar profile of VL input to M1 neurons . The profile exhibits one peak , situated in the upper ∼1/3 of the cortex , somewhat wider ( vertically ) compared with the peak of VL input , spanning the laminar zone corresponding to L4 . ( D ) The laminar profiles obtained from multiple slices ( n = 4 ) . Profiles show a broad peak at a normalized depth of ∼0 . 2–0 . 5 ( black arrow ) . ( E ) Average laminar profile ( black; bars: s . e . m . ) , calculated by binning the data for each profile ( bin width: 1/10 of the normalized cortical depth ) , averaging within each bin , and then averaging across all profiles . The individual profiles are also shown ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 005 Next , we tested whether M1-L4 neurons project to L2/3 , as L4 neurons do in sensory cortex ( Feldmeyer , 2012 ) . Previous studies in mouse forelimb M1 using glutamate uncaging and laser scanning photostimulation ( glu-LSPS ) to map local circuits have suggested that neurons around L3/5A border zone can excite L2/3 neurons ( Weiler et al . , 2008; Wood et al . , 2009; Wood and Shepherd , 2010 ) but lacked the spatial specificity to isolate a putative L4 . We examined this pathway by mapping local input to L2/3 neurons at high spatial sampling density ( 75 µm grid spacing ) both in M1 and , for comparison , in the adjacent S1 ( Figure 4A ) . To facilitate this side-by-side comparison , in these experiments ( only ) , we used sagittal instead of coronal slices . Synaptic input maps for M1-L2/3 neurons showed a local peak of excitatory input strength arising at the location of the hypothesized L4 ( Figure 4B ) . Synaptic input maps for S1-L2/3 neurons were generally similar , but with stronger and spatially more focused excitation from L4 , roughly the size and shape of a L4 barrel ( Figure 4C ) . The stronger input may partly reflect the higher cell density in barrels ( Figure 1A , B ) ( Hooks et al . , 2011 ) . The laminar profile of L4 input to L2/3 neurons was topographically similar in M1 and S1 ( Figure 4D , E ) ; that is , the M1 profile was a scaled version of the S1 profile , with a distinct locus of input from L4 ( plot in Figure 4D ) . Thus , in M1 , the strongest ascending input to the L2/3 neurons arose from the L3/5A border , at a normalized cortical depth of ∼1/3 , confirming the presence of a L4→2/3 excitatory projection in M1 . 10 . 7554/eLife . 05422 . 006Figure 4 . Excitatory output from M1-L4 neurons to L2/3 . ( A ) Top: bright-field image of a parasagittal slice containing motor ( M1 ) and somatosensory ( S1 ) cortex . In S1 , L4 barrels are easily discernable , but are absent from M1 , where L5A ( lighter-appearing laminar zone ) appears wider than in S1 . Graduated scale indicates cortical depth in normalized units , from the pia ( 0 ) to the white matter ( 1 ) . Yellow boxes indicate placement of photostimulation grid for mapping inputs to L2/3 neurons in both cortical areas . Bottom: Schematic indicating the major areas and layers of interest in the image . ( B ) Example of a synaptic input map recorded in a L2/3 neuron in M1 . Grid spacing was set to 75 µm , the top of the grid was flush with the pial surface , and the grid was horizontally centered over the soma ( triangle ) . Cortical layers indicated to the left , with the location of the L3/5A border ( as observed under bright-field ) marked by a horizontal line . Inputs arise from both this L4-like laminar zone and the lateral sites in L2/3 . Photosimulation sites where the postsynaptic neuron's dendrites were directly stimulated were excluded from analysis and are shown as black pixels . ( C ) Example of a synaptic input map recorded in a L2/3 neuron in S1 . Same mapping parameters as in B . The input pattern is similar to that of the M1 example shown in B , but with weaker L2/3 and stronger ascending input from the subjacent region corresponding to the L4 barrel layer . ( D ) : Mean input map for M1 neurons ( n = 14 ) . The laminar profile ( plotted to the right of the map; red , M1; gray , S1 ) shows a peak at the level of the L3/5A border ( black arrow ) , ∼0 . 5 mm deep , corresponding to ∼1/3 of the normalized cortical depth ( DN ) in both M1 and S1 . The bottom plot shows the L4 region of the same plot , with the input profiles normalized ( IN ) to their peak values in L4 ( arrow ) ; the scaled M1 profile closely resembles the S1 profile . ( E ) : Mean input map for S1 neurons ( n = 13 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 006 A characteristic feature of L4 in sensory cortex is that the projection from L4 to L2/3 is primarily unidirectional , as L4 neurons receive primarily intralaminar excitatory input ( Lorente de Nó , 1949; Feldmeyer et al . , 2002; Schubert et al . , 2003; Lefort et al . , 2009; Hooks et al . , 2011; Feldmeyer , 2012 ) . To assess whether this also applies to M1 , we drew on a previously acquired data set of glu-LSPS input maps ( Weiler et al . , 2008 ) to further analyze the subset of maps from neurons located in the L4-like zone in M1 . The input maps of these neurons typically showed an overall paucity of input , which arose mainly from nearby intralaminar sites ( Figure 5A ) . Plotting the median input map of these 10 neurons confirmed that inputs to these L4 neurons arose mostly from L4 , with relatively weak input from L2/3 and other layers ( Figure 5B ) . Only 2 of 10 neurons received distinct loci of input from other layers: one neuron received relatively weak inputs from L5B/6 ( Figure 5C ) and a second received strong input from L2/3 ( Figure 5D ) . These two neurons were at similar laminar locations as the others , suggesting some heterogeneity among M1-L4 neurons' local circuits . Nevertheless , input maps of L4 neurons were generally distinct from those of neurons in mid-L2/3 ( see above ) , and from those of neurons in low-L5A , which typically , and in sharp contrast to the L4 neurons studied here , receive strong L2/3 input ( Weiler et al . , 2008; Anderson et al . , 2010 ) . Statistical analysis confirmed that L4 input to L2/3 neurons was greater than L2/3 input to L4 neurons by a factor of nearly 4 ( L4→2/3: −3 . 9 pA median amplitude , n = 17; L2/3→4: −1 . 0 pA , n = 10; p = 0 . 0062 , rank-sum test ) ( Figure 5E ) . These analyses thus confirm that L4 neurons in M1 receive mostly intralaminar , rather than interlaminar , excitatory input , and that L4→2/3 projections are predominantly unidirectional . 10 . 7554/eLife . 05422 . 007Figure 5 . Paucity of local input to M1-L4 neurons from L2/3 and other layers . ( A ) Example of a typical synaptic input map recorded from a M1-L4 neuron , showing mostly intralaminar excitatory input ( arrow ) , and little L2/3 input . The triangle marks the location of the soma , and the black pixels represent photostimulation sites resulting in direct dendritic responses . ( B ) Median input map for M1-L4 neurons ( n = 10 ) . Excitatory input arose mostly from intralaminar sources , with notably little from L2/3 sites . The laminar profile ( plotted to the right of the map; red , median ± median absolute deviation; gray , individual cells ) shows a peak at the level of the L3/5A border , ∼0 . 5 mm deep , corresponding to ∼1/3 of the normalized cortical depth ( DN ) . ( C ) A M1-L4 neuron showing both intralaminar sources of excitation and ascending excitation from L5B/6 sites ( arrow ) . ( D ) An exceptional M1-L4 neuron . ( E ) Comparison of the individual ( gray circles ) and median ( black lines ) amplitudes of L4 input to M1-L2/3 neurons vs L2/3 input to M1-L4 neurons ( p < 0 . 05 , rank-sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 007 L4 neurons in primary sensory areas receive long-range inputs mostly from thalamus but not other cortical areas . In barrel cortex , for example , interhemispheric ( callosal ) axons from contralateral S1 excite neurons in all layers except L4 ( Petreanu et al . , 2007 ) . We therefore assessed whether M1-L4 neurons similarly receive relatively little long-range cortical input from contralateral M1 . We used the same optogenetic-electrophysiological paradigm employed in the TC experiments described above , with AAV-ChR2-Venus injections targeted to contralateral M1 . The fluorescence intensity of the labeled corticocallosal axons showed a dip at the location of putative L4 ( Figure 6A ) , and laminar profiles of electrophysiologically measured callosal input ( Figure 6B , C ) indicated a nadir at the level of the L3/5A border ( Figure 6C , D , E ) . This pattern was complementary to that of thalamic input from VL , that is , callosal input was strong to neurons in L2/3 and weak to those in L4 , and vice versa for VL input ( Figure 6F ) . Thus , L4 neurons in M1 , similar to S1 , receive relatively little long-range corticocallosal input from contralateral M1 . 10 . 7554/eLife . 05422 . 008Figure 6 . M1-L4 neurons receive and send relatively little long-range corticocortical input . ( A ) Epifluorescence image of coronal slice containing M1 , showing laminar pattern of labeled interhemispheric corticocortical axons following injection of AAV carrying ChR2 and GFP in the contralateral M1 . Scale shows normalized cortical distance . Arrow indicates laminar zone of labeling where weakest photostimulation-evoked electrophysiological responses were also detected . Plot to the right shows laminar profile of fluorescence intensity , in arbitrary units ( A . U . ) , across layers ( normalized distance from pia ) . ( B ) Responses recorded ( sequentially ) in vitro in multiple M1 neurons in different layers ( as indicated ) to photostimulation of ChR2-labeled axons of contralateral M1 neurons infected with AAV-ChR2 . ( C ) Response amplitudes of the same neurons plotted as a function of laminar location , providing a laminar profile of contralateral M1 input to M1 neurons . The profile exhibits a dip in the upper ∼1/3 of the cortex ( corresponding to L4 ) ( arrow ) . ( D ) The laminar profiles were obtained from multiple slices ( n = 4 ) . Profiles show a dip ∼1/3 deep in the cortex ( in normalized coordinates ) . ( E ) Average laminar profile ( black; bars: s . e . m . ) , calculated by binning the data for each profile ( bin width: 1/10 of the normalized cortical depth ) , averaging within each bin , and then averaging across all profiles . The individual profiles are also shown ( gray ) . ( F ) Comparison of input from callosal axons ( from contralateral M1; n = 4 slices ) vs thalamic axons ( from VL; n = 6 slices ) , recorded in postsynaptic L2/3 and L4 neurons in M1 ( p < 0 . 01 , rank-sum test ) . For each slice , values within each laminar zone ( in units of normalized cortical depth: L2/3 , 0 . 1 to 0 . 25; L4 , 0 . 29 to 0 . 37 ) were averaged to obtain a single value per profile; these were averaged and plotted with error bars representing the s . e . m . ( G ) Representative epifluorescence image ( left ) showing gap ( marked by yellow arrows ) at the level of L4 ( ∼1/3 deep in the cortex , in normalized distance units ) in the retrograde labeling of M1 neurons following injection of retrograde tracers in contralateral M1 . Two-photon microscopic image ( right ) showing the same labeling pattern at a higher resolution ( different animal ) . Some neurons within the ‘gap’ are labeled with the retrograde tracer ( green arrow ) . ( H ) Laminar profiles of fluorescence intensity of M1 neurons projecting to contralateral M1 . Each trace represents the average profile for one animal , obtained by averaging several M1-containing slices . For display , the profiles were normalized to the value in L4 ( specifically , the value at a normalized cortical depth of 1/3 ) . The bold line is the average of three animals . There is a reduction in labeling intensity in the L4 region ( arrow ) . ( I ) Representative epifluorescence image showing gap at the level of L4 in the retrograde labeling of M1 neurons following injection of retrograde tracers into ipsilateral M2 , S1 , and S2 . ( J ) Laminar profiles of retrograde labeling pattern for ipsilateral injections . Average traces were calculated as in Panel G . The average ( bold line ) of three animals shows a zone of reduced labeling observed ∼1/3 deep in the cortex , corresponding to L4 ( arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 008 As well as receiving little long-range input , L4 neurons of sensory cortex typically send only weak long-range corticocortical output . Although exceptions to this rule exist , for example , in primate V1 ( Nassi and Callaway , 2009 ) , this pattern appears to hold in mouse S1 , where a ‘gap’ is evident in L4 following retrograde tracer injection into ipsilateral vibrissal M1 ( e . g . , Sato and Svoboda , 2010 ) . Furthermore , a similar gap has been noted in the upper layers of vibrissal M1 following retrograde tracer injections in S1 , potentially corresponding to L4 ( Mao et al . , 2011 ) . Consistent with this pattern , we observed relatively sparse labeling in M1-L4 following injection of retrograde tracers into contralateral M1 ( Figure 6G ) . This laminar gap in labeling was not absolute; higher resolution imaging showed that some of the M1-L4 neurons were clearly labeled with the retrograde tracer but at lower intensity compared with cells in adjacent layers ( Figure 6G ) . This gap was centered ∼1/3 deep in the cortex ( n = 3 mice ) ( Figure 6H ) . Similarly , following injection of retrograde tracers into multiple ipsilateral cortical areas ( M2 , S1 , and S2 , in the same animals ) , we observed a laminar zone with reduced labeling intensity , centered ∼1/3 deep in the cortex ( n = 3 mice ) ( Figure 6I , J ) . The overall average depth of the local minimum in L4 was 0 . 35 ( normalized cortical depth; ipsi- and contralateral profiles pooled , n = 6 ) . These labeling patterns indicate a relative paucity of long-range corticocortical projections originating from L4 of M1 , as from L4 of sensory cortices . The collective evidence from the preceding experiments indicated that M1 contains neurons having the expected input–output circuits of L4 neurons . Having established that M1 does contain a L4 in the form of these hodologically defined L4 neurons , we next sought to characterize their cellular properties . First , we assessed the electrophysiological properties of these M1-L4 neurons . We recorded 56 neurons located across the upper layers of M1 , from upper L2 through L5A . Recordings were targeted to any neurons appearing more likely to be excitatory ( pyramidal/stellate ) rather than inhibitory based on familiar soma features ( shape and size ) as observed under bright-field visualization at high magnification ( Hooks et al . , 2011; Apicella et al . , 2012 ) ; all recorded cells had soma features typical of pyramidal neurons , as we did not observe any with stellate-like somata . For each neuron , we analyzed various passive and active membrane properties . Plotting these parameters vs soma depth showed diverse depth-dependent trends and patterns ( circles in plots in Figure 7 ) . To compare neurons in L4 to those in adjacent layers , we binned the data on the basis of the soma depths into three laminar groups , corresponding to L4 , L2/3 , and L5A ( see ‘Materials and methods’ ) . Statistical comparisons ( Figure 7 , Table 1 ) indicated that the electrophysiological properties of L4 neurons were not simply intermediate between those of L2/3 and L5A neurons . For example , L4 neurons were similar to L2/3 neurons but different from L5A neurons in Rinput , Ithresh , and SFA ratio . Conversely , L4 neurons resembled L5A neurons but differed from L2/3 neurons in Vr , AP width , and Vthresh − Vr . In the case of AP amplitude , the average was greater for L4 neurons than for L2/3 and L5A neurons . L4 neurons showed considerable variability in firing patterns , which ranged from regular and non-adapting to moderately and even strongly adapting , and in other cases highly irregular . From this analysis , we conclude that the properties of L4 excitatory neurons tend to differ from those of either L2/3 or L5A but with considerable cell-to-cell variability , particularly in spiking patterns . Firing pattern diversity has also previously been noted for S1-L4 neurons in rat barrel cortex ( Staiger et al . , 2004 ) . 10 . 7554/eLife . 05422 . 009Figure 7 . Electrophysiological properties of M1-L4 cells . ( A ) Resting membrane potential ( Vr ) , plotted as a function of the cortical depth of the soma ( normalized distance from pia ) . A total of 56 neurons were sampled across layers 2/3 through 5A . For analysis , neurons were binned into three main laminar groups corresponding to L2/3 , L4 , and L5A . The blue lines indicate the laminar range of each group , and also represent their mean ± s . e . m . values . Significant differences between groups are marked ( * , rank-sum test ) . See ‘Materials and methods’ for additional details . ( B ) Input resistance ( Rinput ) vs soma depth . ( C ) Action potential ( AP ) width vs soma depth . ( D ) AP amplitude vs soma depth . ( E ) ( Vthresh − Vr ) vs soma depth . ( F ) Current ( I ) threshold vs soma depth . ( G ) Rinput vs Ithresh . ( H ) Slope of the frequency–current ( F–I ) relationship . ( I ) Spike-frequency adaptation ( SFA ) ratio . ( J ) Example traces , representing the various types of repetitive firing patterns observed among L4 neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 00910 . 7554/eLife . 05422 . 010Table 1 . Electrophysiological properties of L2/3 , L4 , and L5A neurons in M1DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 010ParameterL2/3 neurons ( n = 15 ) L4 neurons ( n = 18 ) L5A neurons ( n = 8 ) L2/3 vs L4L4 vs L5AL2/3 vs L5AVr ( mV ) −78 ± 1−72 ± 1−69 ± 20 . 000790 . 0890 . 00095Rinput ( Mohm ) 65 ± 471 ± 7148 ± 170 . 990 . 000950 . 00034Cm ( pF ) 146 ± 12109 ± 1190 ± 70 . 0380 . 420 . 0061Sag ( % ) 2 . 8 ± 0 . 34 . 3 ± 0 . 66 . 0 ± 1 . 00 . 0570 . 130 . 0088f-I slope ( Hz/nA ) 63 ± 480 ± 8103 ± 160 . 120 . 190 . 0045Ithresh ( pA ) 267 ± 19295 ± 26163 ± 230 . 450 . 00300 . 0038SFA ratio0 . 89 ± 0 . 020 . 89 ± 0 . 020 . 74 ± 0 . 050 . 910 . 00540 . 0080Vthresh ( mV ) −33 ± 1−35 ± 1−32 ± 10 . 190 . 210 . 85AP amplitude ( mV ) 72 ± 279 ± 266 ± 20 . 0150 . 000780 . 028AP width ( msec ) 1 . 21 ± 0 . 080 . 98 ± 0 . 011 . 03 ± 0 . 020 . 00600 . 230 . 19Vthresh − Vr ( mV ) 46 ± 238 ± 237 ± 2* 0 . 000280 . 32* 0 . 0041Values under each cell group are mean ± s . e . m . Numbers in the last three columns are p-values for comparisons between the indicated groups ( rank-sum test; asterisks indicate significant differences ) . Vr , resting membrane potential; Rinput , input resistance; Cm , cell capacitance; Ithresh , current threshold for evoking action potential ( s ) ; SFA ratio , spike-frequency accommodation ratio; Vthresh , voltage threshold for action potential; AP amplitude , action potential peak minus threshold; AP width , action potential duration . See ‘Materials and methods’ for additional details . Lastly , we assessed the morphological properties of M1-L4 neurons . Neurons in slices were filled with biocytin during whole-cell recordings , processed , and imaged with two-photon microscopy ( Figure 8A ) . The fluorescently labeled neurons were then digitally reconstructed as three-dimensional tracings ( n = 6 ) ( Figure 8B , C ) . The morphology of M1-L4 neurons consistently had pyramidal-like dendritic morphology , including a basal arbor with multiple dendrites emerging from the soma , and an apical dendrite extending towards the pia and branching into a small apical tuft in L1 ( Figure 8B , C ) . These impressions were borne out by quantitative analysis of dendritic length density across layers ( Figure 8D ) . The axonal morphology of these neurons was variable but typically included branches in multiple layers , especially L2/3 , L4 , and L5A , a pattern evident from inspection of the reconstructions ( Figure 8B , C ) and borne out by length density analysis ( Figure 8E ) . This laminar profile of axonal density peaked in L4 and steadily declined in the ascending direction across L2/3 towards zero in L1 and also in the descending direction across L5A towards a low baseline level in L5B and L6 . The main trunk of each neuron's axon descended towards the white matter ( Figure 8A–C ) , a feature that can also be seen in reconstructed L4 axons from the sensory cortex ( e . g . , Lübke et al . , 2000; Staiger et al . , 2004 ) . In most cases , this descending axon could be traced into the white matter , typically coursing laterally , and sometimes but not always sending a branch medially towards the corpus callosum ( Figure 8B ) . Overall , this morphological analysis thus indicates that M1-L4 neurons are generally pyramidal neurons , with intracortical axons that ramify mainly in the upper half of the cortex . 10 . 7554/eLife . 05422 . 011Figure 8 . Morphological properties of M1-L4 neurons . ( A ) Example fluorescence image of M1-L4 neurons . The neurons were filled with biocytin during whole-cell recordings , fixed and fluorescently labeled , and imaged with a two-photon microscope . The image is a maximum-intensity projection of multiple aligned image stacks . ( B ) Three-dimensional reconstructions of the two L4 neurons in the center and right of the image are shown in panel A . Dendrites are blue , axons are red , and the pia is drawn across the top . ( C ) Three more examples . ( D ) Quantitative analysis of dendritic morphology . The three-dimensional digital reconstructions ( n = 6 neurons ) of L4 neurons' dendrites were converted to two-dimensional length–density maps and averaged . Plot to the right shows the same data as a vertical profile ( black: group mean ± s . e . m . ; colored lines , individual neurons ) . ( E ) Same analysis , for the axons of the same neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 05422 . 011 We tested the hypothesis that M1 , despite lacking a cytoarchitecturally distinct granular layer , nevertheless contains the circuit-level equivalent of L4 in the form of a layer of excitatory neurons at the layer 3/5A border having the same basic synaptic circuit organization as L4 neurons in sensory cortex . Our findings support this hypothesis and additionally reveal area-specific features of these M1-L4 neurons . The familiar hallmarks of L4 neurons' circuits in sensory areas include ( 1 ) input from primary thalamocortical ( TC ) axons; ( 2 ) output to excitatory neurons in other layers , especially L2/3; ( 3 ) largely unidirectional L4→2/3 projections ( i . e . , little input in return from L2/3 ) , and often , although not as a strict rule , ( 4 ) a paucity of long-range corticocortical inputs and outputs . Our results provide evidence for each of these features in M1-L4 neurons . Our results also revealed features that appear distinct from their S1 barrel counterparts . For one , these neurons received TC input not only from VL but also from PO . This contrasts with S1 barrels , where VPM and PO axons target L4 and L5A , respectively ( Feldmeyer , 2012 ) , but is consistent with previous findings of PO input to neurons in L4 of S2 ( Herkenham , 1980; Theyel et al . , 2010; Pouchelon et al . , 2014 ) and inter-barrel septa in L4 of rat S1 . Another difference from S1-L4 neurons was that the M1-L4 neurons were all pyramidal neurons; we did not detect star pyramids or spiny stellate cells , as are found in rodent S1 ( Staiger et al . , 2004; Feldmeyer , 2012 ) . In this respect , M1-L4 is more similar to the primary visual and auditory cortices of rodents , which also do not contain spiny stellate cells ( Peters and Kara , 1985; Smith and Populin , 2001 ) . Indeed , as shown in ferret visual cortex , spiny stellate cells first develop as pyramidal neurons and subsequently lose their apical dendrites through developmental sculpting , indicating a spectrum of L4 morphological subtypes , with ‘pyramidal’ as the default or prototypical structure ( Callaway and Borrell , 2011 ) . The finding that all M1-L4 neurons in our sample extended an axon towards and often into the subcortical white matter does not however represent a difference , as this is also commonly observed for S1-L4 neurons ( e . g . , Lübke et al . , 2000; Staiger et al . , 2004; Shepherd et al . , 2005 ) , and L4 neurons ( including stellates ) with callosal projections have been described in cat V1 ( e . g . , Vercelli et al . , 1992 ) . In several ways , M1-L4 neurons displayed properties that more closely resembled pyramidal than spiny stellate neurons in rodent S1-L4 . For example , the axonal projections of S1-L4 stellate neurons tend to be more focused in a dense beam to L2/3; in contrast , those of S1-L4 pyramidal neurons tend to show more horizontal spread , lower density in L2/3 , and more branching in L5A ( Brecht and Sakmann , 2002; Bender et al . , 2003; Lübke et al . , 2003; Staiger et al . , 2004; Feldmeyer , 2012 ) . Such a pattern is also observed for S1-L4 neurons located in the inter-barrel septa ( in rats ) , which are pyramidal neurons ( Brecht and Sakmann , 2002; Bureau et al . , 2004; Shepherd et al . , 2005 ) . In rat S1 , thalamocortical axons from PO branch in L4 in septum-related columns ( similar to our finding of PO innervation of M1-L4 neurons ) but not in barrel-related columns in rat S1 , where they instead branch in L5A ( and L1 ) ( Lu and Lin , 1993; Wimmer et al . , 2010; Feldmeyer , 2012 ) . The septal region of rat S1 has been proposed to be hodologically organized as a higher order rather than primary sensory cortical area ( Killackey and Sherman , 2003 ) . Consistent with this , PO projects to L4 in S2 in addition to septal-S1 ( Theyel et al . , 2010; Pouchelon et al . , 2014 ) . Thus , our findings are generally suggestive that , at least in terms of its L4-related organization , the ‘primary’ motor cortex more closely resembles higher order than primary sensory cortex . The apparent absence of L4 in M1 and other agranular cortical areas has long been of interest for its implication that the circuit organization of these areas differs fundamentally from that of sensory areas ( Shipp , 2005; Feldmeyer et al . , 2013; Shipp et al . , 2013; García-Cabezas and Barbas , 2014 ) . Our results suggest that L4 in M1 has been ‘lost’ only at the level of cytoarchitecture but not of cellular connectivity , as it is present in the form of a layer of pyramidal neurons with the expected input–output connections of prototypical L4 neurons . This accords with the general notion that cortical circuit organization tends to be conserved rather than reinvented across areas , with variations arising mostly through quantitative differences in a core set of existing circuits ( Harris and Shepherd , 2015 ) . For example , rodent M1 possesses not only a thin L4 but an expanded L5B ( Brecht et al . , 2004; Weiler et al . , 2008; Yu et al . , 2008; Anderson et al . , 2010; Hooks et al . , 2013 ) . In rodent S1 barrel cortex , it is L4 that is instead expanded and shows the most overtly specialized connectivity ( Feldmeyer , 2012 ) ; similarly , L4 in V1 of highly visual mammals is often elaborately differentiated ( Fitzpatrick , 1996; Nassi and Callaway , 2009 ) . Thus , L4 appears to be most elaborate in the primary sensory cortices of modalities that are particularly ethologically relevant to an animal . Motor cortex contains a L4 circuit that is smaller and simpler but retains the same prototypical connectivity patterns . We speculate that like their sensory cortical counterparts , L4 neurons in M1 are specialized for processing of thalamic input before this information is integrated with the activity of other cell classes ( which may also be thalamorecipient ) downstream in the local M1 network . Our results should facilitate further studies of M1-L4 by enabling a shift of focus away from the question of whether L4 neurons are present in M1 to questions of what types of information they process , how they do so , and how this relates to motor behavior . Animal studies were approved by the Northwestern University Animal Care and Use Committee . In vivo stereotaxic injections of retrograde tracers ( fluorescent microspheres , Lumafluor , Durham , NC ) or AAV viruses encoding channelrhodopsin-2 ( AAV-ChR2-Venus ) were performed as described ( Anderson et al . , 2010; Hooks et al . , 2013 ) . Retrograde tracer injections were made in either the contralateral M1 or ipsilateral M2 , S1 , and S2 of 6- to 7-week-old mice , to label corticocortical projection neurons in M1; brain slices were prepared 3–7 days later and imaged as described below . Viral injections were made in the motor thalamus of 3- to 4-week-old mice , targeting the ventrolateral nucleus , and optogenetic experiments in brain slices were performed ∼3 weeks later . Brain slice preparation and electrophysiology were performed as previously described ( Weiler et al . , 2008; Anderson et al . , 2010 ) . Whole-cell patch-electrode recordings were made from neurons in 0 . 3-mm-thick brain slices containing M1 . Data were sampled at 10 kHz ( most experiments ) or 40 kHz ( for intrinsic electrophysiology measurements ) and filtered at 4 kHz . For optogenetic experiments , recordings in L4 ( and other layers ) were generally targeted to neurons with ‘pyramidal’ somata . For experiments aimed at characterizing intrinsic and morphological properties , recordings were targeted to L4 neurons with any soma shape or size except those suggestive of common types of interneurons , particularly basket cells ( Apicella et al . , 2012 ) . Data acquisition was controlled by Ephus software ( www . ephus . org ) ( Suter et al . , 2010 ) . Standard electrophysiological stimulus protocols were delivered to assess intrinsic properties , as previously described ( Suter et al . , 2013 ) . For each cell , after measuring the resting membrane potential , current was injected as needed to set the membrane potential to −70 mV , and then stimulus protocols were delivered to measure electrophysiological properties . Spike-frequency accommodation ( SFA ) ratio was calculated as the ratio of the third to fifth inter-spike interval in the first trace containing ≥6 spikes . Current threshold was defined as the amplitude of the current step that was sufficient to evoke one or more action potentials . For group analyses of electrophysiological properties ( Figure 7 , Table 1 ) , statistical comparisons were performed by pooling neurons into laminar groups corresponding to L4 , L2/3 , and L5A . Based on the results of the circuit analyses ( Figures 1–6 ) , we defined ‘L4’ as a thin zone centered on 0 . 33 ( in units of normalized cortical depth ) and spanning 0 . 05 the cortical thickness ( i . e . , depth range 0 . 305–0 . 355 ) . We defined ‘L2/3’ as the laminar zone spanning 0 . 14–0 . 26 , and ‘L5A’ as the zone 0 . 37–0 . 42 . These laminar zones were separated by small gaps ( 0 . 045 between L2/3 and L4 , and 0 . 015 between L4 and L5A ) , which reduced ( but did not necessarily eliminate ) the likelihood that some neurons were wrongly classified due to slice-to-slice variability in layer thicknesses . A small number of neurons thus fell outside these groups and were excluded from group analyses ( but not from the plots; all data are plotted as circles in Figure 7 ) . Glutamate uncaging and laser scanning photostimulation ( glu-LSPS ) were performed as previously described ( Weiler et al . , 2008; Wood et al . , 2009; Wood and Shepherd , 2010; Shepherd , 2012 ) , using 3- to 5-week-old mice . As described in ‘Results’ , in one set of experiments , we acquired sets of input maps for L2/3 neurons in M1 or S1; in another set , we further analyzed a subset of glu-LSPS mapping data from a previous study ( Weiler et al . , 2008 ) . Temporal windowing was used to detect photostimulation sites where the postsynaptic neuron's dendrites were directly stimulated ( defined as excitatory events arriving within 7 msec post-stimulus ) ( Schubert et al . , 2001 ) , and these sites were excluded from analysis ( shown in the figures as black pixels ) . Optogenetic photostimulation in brain slices was performed as previously described ( Kiritani et al . , 2012; Hooks et al . , 2013 ) , exploiting the retained photoexcitability of ChR2-expressing long-range axons in slices ( Petreanu et al . , 2007 ) and using conditions ( in particular , tetrodotoxin and 4-aminopyridine in the bath solution ) that isolate monosynaptic inputs ( Petreanu et al . , 2009 ) . Responses to blue-LED photostimulation were sampled in voltage-clamp mode for each neuron , and multiple neurons were recorded per slice . Traces were analyzed to determine the average response amplitude in a 50-msec post-stimulus window . For the set of neurons recorded in the same slice , responses were normalized to the strongest response , resulting in a normalized laminar profile for each slice . Profiles from different slices and animals were pooled for group analyses . Imaging and morphological reconstructions were performed as previously described ( Suter et al . , 2013 ) , by acquiring two-photon image stacks of neurons that had been biocytin filled during slice recordings , fixed , and processed for fluorescent labeling . Three-dimensional reconstructions of axons and dendrites were manually traced ( Neurolucida , MBF Bioscience , Williston , VT ) and further analyzed using custom Matlab routines ( Source code 1 ) to quantify dendritic and axonal length density , as previously described ( Shepherd et al . , 2005 ) . Images of expression patterns of molecular markers were obtained from the Allen Mouse Brain Atlas ( http://mouse . brain-map . org ) ( Lein et al . , 2007 ) . Many analyses involved plotting a parameter of interest as a function of cortical depth , providing a laminar profile of that parameter . To facilitate comparisons across slices , we converted from absolute cortical depth ( distance from pia ) to a normalized scale , with pia defined as zero and the cortex–white matter border defined as one . To the extent that the thicknesses of individual cortical layers vary as a constant fraction of the total cortical thickness , this normalization procedure is assumed to reduce some of the slice-to-slice variability; for example , due to small differences in slice angle . Unless noted otherwise , statistical comparisons were performed using non-parametric tests ( rank-sum or signed-rank tests , as appropriate ) with significance defined as p < 0 . 05 . For the group analyses shown in Figure 7 and Table 1 , significance was defined as p < 0 . 05/3 ( multiple-comparison correction ) .	In 1909 , a German scientist called Korbinian Brodmann published the first map of the outer layer of the human brain . After staining neurons with a dye and studying the structures of the cells and how they were organized , he realized that he could divide the cortex into 43 numbered regions . Most Brodmann areas can be divided into a number of horizontal layers , with layer 1 being closest to the surface of the brain . Neurons in the different layers form distinct sets of connections , and the relative thickness of the layers has implications for the function carried out by that area . It is thought , for example , that the motor cortex does not have a layer 4 , which suggests that the neural circuitry that controls movement differs from that in charge of vision , hearing , and other functions . Yamawaki et al . now challenge this view by providing multiple lines of evidence for the existence of layer 4 in the motor cortex in mice . Neurons at the border between layer 3 and layer 5A in the motor cortex possess many of the same properties as the neurons in layer 4 in sensory cortex . In particular , they receive inputs from a brain region called the thalamus , and send outputs to neurons in layers 2 and 3 . Yamawaki et al . go on to characterize some of the properties of the neurons in the putative layer 4 of the motor cortex , finding that they do not look like the specialized ‘stellate’ cells that are found in some other areas of the cortex . Instead , they resemble the ‘pyramidal’ type of neuron that is found in all layers and areas of the cortex . The discovery that the motor cortex is more similar in its circuit connections to other area of the cortex than previously thought has important implications for our understanding of this region of the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	A genuine layer 4 in motor cortex with prototypical synaptic circuit connectivity
Methylation of histone H3 at lysine 36 ( H3K36me ) , a widely-distributed chromatin mark , largely results from association of the lysine methyltransferase ( KMT ) SET-2 with RNA polymerase II ( RNAPII ) , but most eukaryotes also have additional H3K36me KMTs that act independently of RNAPII . These include the orthologs of ASH1 , which are conserved in animals , plants , and fungi but whose function and control are poorly understood . We found that Neurospora crassa has just two H3K36 KMTs , ASH1 and SET-2 , and were able to explore the function and distribution of each enzyme independently . While H3K36me deposited by SET-2 marks active genes , inactive genes are modified by ASH1 and its activity is critical for their repression . ASH1-marked chromatin can be further modified by methylation of H3K27 , and ASH1 catalytic activity modulates the accumulation of H3K27me2/3 both positively and negatively . These findings provide new insight into ASH1 function , H3K27me2/3 establishment , and repression in facultative heterochromatin . Methylation of histone H3 at lysine 36 ( H3K36me ) is largely associated with euchromatic regions of eukaryotic genomes ( Ho et al . , 2014 ) . It serves as a link to transcription , as a H3K36 lysine methyltransferase ( KMT; for example , yeast Set2 ) is directly associated with RNA polymerase II ( RNAPII ) elongation , and the mark is enriched along actively transcribed genes ( Kizer et al . , 2005; Li et al . , 2003; Morris et al . , 2005 ) . As a result , H3K36me is commonly cited as an indicator of ‘active’ chromatin and is thought to exist in an antagonistic relationship with heterochromatin ( Gaydos et al . , 2012 ) . Cohabitation of H3K36me3 with either H3K27me2/3 or H3K9me2/3 on the same histone tail is rare ( Jamieson et al . , 2016; Voigt et al . , 2012; Young et al . , 2009 ) , and deposition of one mark can inhibit deposition of the second ( Schmitges et al . , 2011; Voigt et al . , 2012; Yuan et al . , 2011 ) . Paradoxically , studies of H3K36me have shown that this modification can recruit chromatin remodelers and modifiers that organize and deacetylate nucleosomes , stabilize histones by inhibiting exchange , and restrict access to DNA – effectively conferring features of heterochromatin ( Carrozza et al . , 2005; Fazzio et al . , 2001; Lee et al . , 2013; Li et al . , 2007a; Smolle et al . , 2012 ) . In metazoans , these seemingly dissonant functions are resolved by a division of labor within the H3K36me pathway that: 1 ) links conversion of H3K36me2 to –me3 with transcription elongation by physically tethering the Set2-ortholog to RNAPII , and 2 ) employs specialized RNAPII-independent KMTs to catalyze H3K36me1/2 . The consequence is a complex and poorly understood regulatory network controlling access to and modification of the H3K36 substrate . Though much has been learned about H3K36me3 as a signal , there is little mechanistic understanding of how the RNAPII-independent KMTs are targeted and how their products function . The complexity and significance of the H3K36me regulatory pathway is illustrated both in the range of fundamental genomic processes it underlies ( e . g . transcription initiation and repression , alternative splicing , and DNA replication , recombination and repair ) ( Wagner and Carpenter , 2012 ) , and the frequency with which it is disturbed during oncogenesis . The direct or indirect disruption of H3K36me by mutation of histone H3 genes defines distinct subtypes of pediatric chondroblastoma ( H3 . 3K36M ) and glioblastoma ( H3 . 3G34R/V ) ( Fang et al . , 2016; Lu et al . , 2016; Schwartzentruber et al . , 2012 ) . In addition , recurrent mutation or overexpression of genes that methylate ( Ash1L , Nsd1/2/3 , and Setd2 ) or demethylate ( Kdm2b and Kdm4a ) H3K36 have been implicated as drivers of malignant transformation ( Black et al . , 2013; He et al . , 2011; Jaju et al . , 2001; Kovac et al . , 2015; Liu et al . , 2012; Mar et al . , 2014; Cancer Genome Atlas Network , 2015 ) . The prevalence of aberrant H3K36me regulation in cancer underscores the value of identifying therapeutic options for targeting this pathway . Unfortunately , the complexity and essential nature of the H3K36me pathway in higher organisms has restricted lines of inquiry and has left fundamental aspects of its function largely unexplored . Instead , much of the functional characterization of H3K36me has been performed in the yeasts S . cerevisiae and S . pombe , where H3K36me is nonessential and performed by a single RNAPII-associated KMT ( Strahl et al . , 2002 ) . The simplicity of the H3K36me pathway in yeasts has proven valuable but has limited our understanding of the situation in eukaryotes that possess RNAPII-independent H3K36 KMTs , including filamentous fungi , plants , and animals ( Janevska et al . , 2018; Schuettengruber et al . , 2017 ) . We present the filamentous fungus Neurospora crassa as an experimental bridge between yeasts and higher organisms , and use it to address unresolved questions concerning H3K36me . As in S . cerevisiae and S . pombe , H3K36me is not essential in N . crassa but unlike the case in the yeasts , we found that H3K36 methylation results from a division of labor between the RNAPII-associated SET-2 enzyme which can catalyze mono- , di- , and tri-methylation ( Adhvaryu et al . , 2005 ) and ASH1 ( NCU01932 ) . Notably , like higher organisms , Neurospora possesses both facultative heterochromatin – characterized by Polycomb Repressive Complex 2 ( PRC2 ) -catalyzed H3K27me2/3 ( Jamieson et al . , 2013 ) – and constitutive heterochromatin – characterized by H3K9me3 , HP1 , DNA methylation and HDAC recruitment ( Freitag et al . , 2004; Tamaru and Selker , 2001; Tamaru et al . , 2003 ) . Both of these forms of heterochromatin are nonessential in Neurospora , facilitating studies of their interplay in vivo ( Jamieson et al . , 2016 ) . The study presented here reveals a novel function for ASH1 and elucidates relationships between H3K36me , RNAPII , and facultative heterochromatin . Our examination of the H3K36me pathway in Neurospora began with analyses of set-2 and ash1 mutant strains . In preliminary work , we found that , unlike set-2 , which is dispensable for viability , ash1 appears to be essential , as evidenced by the inability to generate a pure Δash1 strain ( Colot et al . , 2006 ) . Nevertheless , we found that we could build an ash1 strain that should be catalytically inactive by mutation of Y888 , which is required for coordinating the target lysine in SET protein superfamily members ( Figure 1A , B ) ( Dillon et al . , 2005 ) . Strains harboring the ash1 ( Y888F ) mutation displayed severely compromised growth but only minor reductions in global H3K36me2 and –me3 ( Figure 1C , D , E ) . Δset-2 strains showed a dramatic loss of H3K36me2 but only minor impairment of growth ( Figure 1C , D , E ) . We found that deletion of the ‘Set2 Rpb1 Interacting’ ( SRI ) domain of SET-2 , which should decouple the enzyme from RNAPII ( Youdell et al . , 2008 ) , resulted in a loss of H3K36me3 comparable to that seen with a set-2 deletion , suggesting that RNAPII-associated SET-2 is responsible for nearly all H3K36me3 ( Figure 1G ) . Weak H3K36me3 signals remain in each of these backgrounds , raising the possibility that ASH1 is responsible for some H3K36me3 . Consistent with this possibility , set-2; ash1 ( Y888F ) double mutants showed additive loss of the H3K36me2 observed in the single mutants and loss of the residual H3K36me3 signal ( Figure 1E , F , G ) . This suggested ASH1 has weak H3K36me3 catalytic activity in vivo – a surprise given the in vitro activity of its orthologs ( An et al . , 2011; Yuan et al . , 2011 ) and that the protein has a tyrosine at amino acid position 886 ( Figure 1B ) , the predicted site of the ‘Y/F-switch , ’ which is characteristic of SET domains in mono/di-KMTs ( Collins et al . , 2005 ) . As a step to identify the functions of ash1 and set-2 , we investigated the distribution of their activities across the genome . Our observation that the catalytic activity of ASH1 is not essential provided an opportunity to analyze separately the H3K36me2 and –me3 catalyzed by ASH1 and SET-2 by ChIP-seq in set-2 knockout and ash1 ( Y888F ) strains , respectively . Overall , we found that H3K36me2 and –me3 is associated with gene-rich DNA and excluded from constitutive heterochromatin , which is marked by DNA methylation and H3K9me3 ( Figure 2A , Figure 2—figure supplement 1A ) . H3K36me2 catalyzed by ASH1 or SET-2 was found in distinct domains that apparently together produce the overall pattern of wildtype ( WT ) H3K36me2 ( Figure 2A , C ) . We found ASH1-catalyzed H3K36me2 was prominent across the promoter and body of the genes that are silent or poorly transcribed in WT ( Figure 2B , C , Figure 2—figure supplement 1B ) . Conversely , SET-2-catalyzed H3K36me2 was found associated predominantly with moderately- and highly-transcribed genes and was depleted from transcriptional start-sites ( TSS ) but enriched over gene bodies ( Figure 2B , C ) . By this assay , SET-2 was found to mark most ( >80% ) genes , while ASH1 marked the minority ( ~20% ) of genes that lacked SET-2-catalyzed H3K36me2 ( p-value<10−4 ) ( Figure 2B , Figure 2—figure supplement 1C ) . In the ash1 mutant , H3K36me3 was found restricted to sites of SET-2-catalyzed H3K36me2 ( Figure 2C ) , and the intergenic H3K36me3 seen in WT was absent . Similarly , H3K36me3 at domains of ASH1-catalyzed H3K36me2 was lost in the ash1 mutant . Consistent with the results of the western blot ( Figure 1E ) , H3K36me3 was not entirely lost when set-2 was deleted , and signal remained at regions with intense ASH1-catalyzed H3K36me2 ( Figure 2C , Figure 2—figure supplement 1D ) . We carried out RNAseq analyses to assess the effect of ASH1 and SET-2 activity on gene expression . We found that both ash1 and set-2 mutants have substantial , but distinct , changes in gene expression relative to WT . The set-2 deletion showed a relatively symmetrical distribution of gene expression changes with 916 genes up-regulated and 1222 genes down-regulated ( Figure 3A ) . In contrast , the ash1 mutant predominantly resulted in up-regulation ( 1261 genes up-regulated; 228 genes down-regulated; Figure 3A ) . When we limited our analysis to ASH1-marked genes , they were almost exclusively up-regulated in the ash1 ( Y888F ) background , while ASH1-unmarked genes showed no pattern of altered regulation ( Figure 3B , Figure 3—figure supplement 1 ) . When ASH1-marked genes were separated into ‘SET-2-unmarked’ and ‘SET-2-comarked’ categories , we found that co-marked genes were significantly up-regulated , while SET-2-unmarked genes showed little or no change in expression ( Figure 3C ) . Collectively , these results imply that ASH1 and SET-2 independently catalyze H3K36me2 in a manner that differentiates the genome into regions of poorly- or robustly-transcribed genes . The repressed state of poorly transcribed genes is largely dependent upon ASH1 catalytic activity . Upon inactivation of ASH1 , genes that are subject to derepression become co-marked by transcription-coupled SET-2 . The presence of ASH1-catalyzed H3K36me at silent and poorly transcribed genes prompted us to investigate its relation to PRC2-catalyzed H3K27me2/3 , which is also in domains of silent genes ( Jamieson et al . , 2013 ) . Interestingly , we found nearly all ( 220/232 ) annotated domains of H3K27me2/3-marked chromatin ( Klocko et al . , 2018 ) are also marked with ASH1-catalyzed H3K36me2 ( Figure 4A ) . When we looked at where the 12 absent domains were located , we saw they were all found in sub-telomere regions characterized by the presence of H3K27me2/3 , H3K9me3 , and DNA methylation ( Jamieson et al . , 2018 ) , a finding consistent with ASH1-catalyzed H3K36me2 being excluded from constitutive heterochromatin ( Figure 2A ) . When we examined the promoter region of individual genes , we again saw that the distribution of H3K27me2/3 overlapped with that of ASH1-catalyzed H3K36me2 ( Figure 4B ) . Previous mass spectrometry analyses of N . crassa histone H3 suggested K27 and K36 methylation do not typically occur on the same molecule ( Jamieson et al . , 2016 ) , implying that these marks exist as ‘asymmetric’ modifications on the same nucleosome and/or on adjacent nucleosomes ( Voigt et al . , 2012; Yuan et al . , 2011 ) . In all , we found 30% of ASH1-marked genes were co-marked by H3K27me2/3 , and this co-marking was predominantly found at domains of ASH1-catalyzed H3K36me2 that lacked appreciable SET-2-catalyzed H3K36me2 ( Figure 4C , D ) . The consistent overlap of ASH1-catalyzed H3K36me with native H3K27me made us question whether the pattern would hold true in mutant backgrounds in which we had observed new domains of H3K27me . To test this , we first re-examined the H3K27me2/3-defects caused by deletion of the Drosophila Nurf55/Caf1 ortholog , Neurospora p55 ( NPF ) ( Jamieson et al . , 2013 ) . Though the predominant effect of npf deletion is loss of sub-telomeric H3K27me2/3 ( Jamieson et al . , 2013 ) , we also found new domains of H3K27me3 and , interestingly , these were limited to regions of ASH1-catalyzed H3K36me2 ( Figure 4E ) . Next , we took advantage of a situation in which H3K27me2/3 was induced in a normally euchromatic region by insertion of telomere repeats in the vicinity ( Jamieson et al . , 2018 ) . Using an insertion at the csr-1 locus , we found that the discontinuous spread of H3K27me2/3 from the repeats correlated perfectly with the presence of ASH1-catalyzed H3K36me2 ( Figure 4F ) . Altogether , these analyses show that a fraction of ASH1-marked chromatin is asymmetrically modified by PRC2 to generate overlapping profiles of H3K27me and H3K36me at genes that are most refractory to derepression , and that H3K27me-competency is a distinguishing characteristic of ASH1-marked chromatin . Drosophila Ash1 has previously been reported to inhibit PRC2-mediated repression by preventing H3K27me2/3 accumulation ( Papp and Müller , 2006 ) . Similarly , H3K36me3 has been shown to inhibit catalysis of H3K27me2/3 in vitro ( Yuan et al . , 2011 ) . We therefore asked if ASH1-catalyzed H3K36me is influenced by loss of H3K27me , and if H3K27me is influenced by loss of ASH1 activity . Immunoblotting and ChIPseq in a double-mutant strain lacking SET-2 and the H3K27 KMT , SET-7 , revealed ASH1-catalyzed H3K36me2 was unchanged by loss of H3K27me2/3 across the genome ( Figure 5—figure supplement 1 ) , indicating that ASH1-catalyzed H3K36me2 is not dependent on PRC2-catalyzed H3K27me2/3 . We next asked whether some fraction of normal H3K27me regions depend on H3K36 methylation directed by ASH1 . To test this possibility , we performed H3K27me2/3 ChIP in the ash1 ( Y888F ) background . Inactivation of ASH1 showed a striking effect on H3K27me2/3 , resulting in reduction or complete loss of the mark across roughly one-third of ASH1/PRC2-comarked genes ( Figure 5A , B ) . The loss of ASH1-catalyzed H3K36me and resultant loss of H3K27me2/3 was accompanied by accumulation of H3K27 acetylation ( ac ) and derepression of affected genes ( Figure 5C , D , Figure 5—figure supplement 2 ) . In addition to identifying ASH1-dependent H3K27me2/3 , we also found domains of H3K27me2/3-competent chromatin where the ASH1 mark prevented H3K27me2/3 . In the ash1 mutant , 128 genes gain H3K27me2/3 ( defined as >2 fold increase over background ) ( Figure 5E ) , whereas 180 genes lost the H3K27me2/3 mark ( Figure 5B ) . Importantly , these new domains of PRC2-marked chromatin in the ash1 ( Y888F ) strain are delineated by regions normally marked with ASH1-catalyzed H3K36me2 ( Figure 5F ) . Thus , ASH1 catalyzed H3K36me can both positively and negatively influence H3K27me2/3 accumulation . With the notable exception of yeasts , the H3K36me pathway of eukaryotes is divided between Set2 orthologs ( SET2D in humans ) , which can catalyze mono- , di- , and tri-methylation , and a group of specialized KMTs that largely catalyze mono/di-methylation ( Wagner and Carpenter , 2012 ) . Study of the functional relationships between Set2-orthologs and the mono/di-KMTs has been limited , in part because numerous dedicated H3K36 mono/di-KMTs are found in higher organisms ( e . g . seven in mammals ) and these enzymes – as well as the Set2-ortholog – are typically essential , making determination of their independent actions difficult or impossible ( Wagner and Carpenter , 2012 ) . To address unresolved questions regarding the functional relationship between H3K36 KMTs , we took advantage of the simplified H3K36me pathway of N . crassa , which we showed consists of two H3K36 KMTs , SET-2 and ASH1 . Unlike SET-2 ( Adhvaryu et al . , 2005 ) , ASH1 appears to be essential for viability but we found that the organism tolerates a catalytic null mutation of ash1 , allowing us to assess the relative contribution of the two KMTs . By dissecting the roles of these enzymes , we uncovered a previously undescribed pathway that connects ASH1-catalyzed H3K36me to repression of poorly transcribed genes . Curiously , we found that much of ASH1-marked chromatin is characterized by H3K27me2/3-competency . Not only did native domains of H3K27me overlap with domains of ASH1-catalyzed H3K36me , but experimentally-induced domains of H3K27me2/3 selectively spread to ASH1-marked chromatin . RNAPII-associated SET-2 is generally considered to be the only KMT capable of catalyzing H3K36me3 ( Adhvaryu et al . , 2005 ) , and based upon biochemical studies with Drosophila and human orthologs ( An et al . , 2011; Yuan et al . , 2011 ) we expected ASH1 to act as a dedicated H3K36 mono/di-KMT . Conservation of a tyrosine residue at the ‘F/Y-switch’ of its SET domain ( Figure S1A ) supported this expectation ( Collins et al . , 2005 ) . Results of western blotting suggested , however , that ASH1 is responsible for ~5% of global H3K36me3 ( Figure 1 ) in the absence of SET-2 , and the ASH1 homologs of Fusarium fugikuroi and Plasmodium falciparum have also been reported to have H3K36me3 activity ( Janevska et al . , 2018; Jiang et al . , 2013 ) . Although we detected ASH1-catalyzed H3K36me3 in vivo with different , independently validated , antibodies and techniques , it remains possible these antibodies recognized residual H3K36me2 or that ASH1 can convert H3K36me2 to –me3 but only in the absence of SET-2 . Neurospora SET-2 catalyzes H3K36me2/3 across the bodies of active genes , much as in yeast ( Krogan et al . , 2003; Li et al . , 2003 ) , whereas ASH1 is responsible H3K36me2/3 across large domains that encompass multiple genes and intergenic regions . Interestingly , genes marked by ASH1 are silent or poorly transcribed and are largely reliant upon the mark for their repressed state . Genes that were derepressed by inactivation of ASH1 were mostly ‘SET-2-comarked , ’ that is , they only lost H3K36me in the absence of both KMTs . It will be interesting to learn how ASH1 is directed to where it acts , that is , in domains of lowly transcribed genes . Neurospora ASH1 does not display telling conserved protein domains , but does have an AT-hook that might interact with the minor-groove of A/T-rich DNA . Constitutive heterochromatin in Neurospora is characterized by A/T-rich DNA ( Cambareri et al . , 1989 ) , but we found no indication that ASH1 functions at constitutive heterochromatin; in fact , H3K36me appears to be normally excluded from such regions . Our finding that ASH1 has a function in repression is not entirely surprising given prior evidence of H3K36me3 in recruiting repressive chromatin machinery ( Fazzio et al . , 2001; Keogh et al . , 2005; Strahl et al . , 2002 ) but it was striking to see the extent of its repressive influence . When compared to PRC2-catalyzed H3K27me2/3 ( Jamieson et al . , 2013 ) , ASH1-catalyzed H3K36me appears to be the predominant repressive modification of poorly transcribed genes ( H3K27me covers only ~30% of silent , ASH1-marked , genes ) . Given the collaborative relationship between H3K36 KMTs , chromatin remodelers , and histone deacetylases ( HDACs ) described in other organisms ( Lee et al . , 2013 ) , we predict a role for nucleosome positioning and histone deacetylation in ASH1-mediated repression . Here , we observed an accumulation of H3K27ac following loss of ASH1-dependent H3K27me2/3 , but it is unclear if this is a passive product of H3K27me2/3 loss or if there is an active role for H3K27me2/3 in exclusion . Future studies should examine the Neurospora counterpart of the yeast HDAC complex , RPD3 Small ( RPD3S ) , which includes the H3K36me3 reader , EAF-3 ( Joshi and Struhl , 2005; Keogh et al . , 2005 ) . RPD3S activity appears to be dependent upon proper nucleosome spacing established by Isw2 and Chd1 , which together apparently organize and stabilize nucleosomes to restrict internal initiation by RNAPII in the wake of transcription ( Carrozza et al . , 2005; Fazzio et al . , 2001; Lee et al . , 2013; Li et al . , 2007a; Smolle et al . , 2012 ) . Importantly , this mechanism is dependent upon transcription , as H3K36me3 deposition by SET-2 is strictly tied to elongating RNAPII ( Youdell et al . , 2008 ) . Our results support a related transcription-independent mechanism that maintains gene repression at facultative heterochromatin . ASH1 would establish the H3K36me mark required to recruit RPD3S , while chromatin remodelers would establish proper nucleosome positions to facilitate RPD3S deacetylase activity . To test this hypothesis , further study of the RPD3S HDAC will be required , but it will be challenging as the N . crassa ortholog of Rpd3 ( HDA-3 ) is essential , and other units of the complex – EAF-3 , SIN3 , and NPF – are components of various other chromatin modifying complexes ( Jamieson et al . , 2013; Sathianathan et al . , 2016 ) . Notably , orthologs of EAF-3 and NPF ( Mrg15 and Nurf55 , respectively ) have recently been identified as Ash-1 complex members in Drosophila ( Huang et al . , 2017; Schmähling et al . , 2018 ) , further supporting a connection to RPD3 . Perhaps the most surprising observation from our study was the substantial overlap of H3K27me2/3 at ASH1-marked chromatin . Neurospora H3K27me2/3 is catalyzed by a PRC2 complex that is highly similar to those found in metazoans ( Jamieson et al . , 2013 ) but N . crassa has no apparent PRC1 components . Even in higher organisms , the mechanism of repression mediated by PRC2 and H3K27me2/3 is far from clear , necessitating additional studies . Interestingly , we observed in Neurospora that loss of ASH1-dependent H3K36 methylation was associated with both losses and gains of H3K27me2/3 . Early work with Drosophila gave evidence that ASH1 opposes the action of PRC2 function ( Klymenko and Müller , 2004; Shearn , 1989 ) , consistent with the observation that the presence of H3K36me on a histone tail can inhibit PRC2 activity in cis ( Yuan et al . , 2011 ) , but our findings suggest the situation is more complicated . We found new domains of H3K27me2/3 at ASH1-regulated regions when ASH1 was inactivated , suggesting the presence of the ASH1 mark prevents H3K27me2/3 . In addition , we found that ASH1 drives repression , and derepression associated with ASH1 inactivation is frequently accompanied by H3K27me2/3 loss . These seemingly opposing activities may reflect differential histone modifications and accompanying effector proteins found at those regions . Or perhaps , similar to plants , different forms of PRC2 may exist that respond differently to the presence of H3K36me ( Schmitges et al . , 2011 ) . These possibilities will be interesting to investigate in the future . Finally , it is important to note that though it was initially surprising to find genome-wide colocalization of H3K27me2/3 with ASH1-catalyzed H3K36me2 , this does not appear to be unique to Neurospora and other fungi , as recent work with embryonic stem cells revealed apparent cross-talk of these marks ( Streubel et al . , 2018 ) . Our work supports a model in which the genes of N . crassa are separated into two compartments depending upon their source of H3K36me ( Figure 6 ) . Actively transcribed genes possess SET-2-catalyzed H3K36me2/3 specific to the gene body , while silent and infrequently transcribed genes are covered in large domains of ASH1-catalyzed H3K36me2/3 . In both cases , H3K36me appears to act as a repressive mark , protecting active genes against internal cryptic-transcription ( Li et al . , 2007b ) and blocking general transcription at inactive genes . The repressed state of ASH1-modified chromatin is largely contingent upon the presence of ASH1-catalyzed H3K36me , but can be further modified with H3K27me2/3 catalyzed by the PRC2 complex to support repression . All Neurospora strains used in this study are listed in Key resources table . Strains were grown , crossed , and maintained according to standard procedures ( Davis , 2000 ) . Knockout and mutant strains were either taken from the Fungal Genetic Stock Center knockout collection ( Colot et al . , 2006; McCluskey et al . , 2010 ) or generated as previously described ( Honda and Selker , 2009 ) . ASH1 mutations were made with a QuickChange site-directed mutagenesis kit ( Stratagene ) and PCR-based mutagenesis with the In-Fusion HD cloning system ( Takara ) . The follow primer sets were used for quantitative real-time ( qRT ) PCR: 8:G3 ( CGTAGAGAAGGGAAGTAGTAG; GCACAATACGAAGTCACTTTTCGCC ) , NCU07152 ( GGCAACAGAGGCTGTGCTGC , CGCAAAGATGCCGCACCTGTC ) , hH4 ( CATCAAGGGGTCATTCAC , TTTGGAATCACCCTCCAG ) . Immunoblotting was performed as previously described ( Honda and Selker , 2008 ) . Briefly , Neurospora extracts were produced by sonication in extraction buffer ( 50 mM Hepes pH7 . 5 , 1 mM EDTA , 150 mM NaCl , 10% Glycerol , 0 . 02% NP40 ) supplemented with cOmplete ULTRA protease inhibitor cocktail tablets ( Roche , 05892970001 ) . The following antibodies were used for immunoblotting: H3K36me3 ( Cell Signaling , Cat#4909S , Clone ( D5A7 ) ) , H3K36me2 ( Abcam , Cat#ab9049 ) , Histone H3 ( Abcam , Cat#ab1791 ) , IRDye 680RD Goat-anti-Rabbit secondary antibody ( Licor , Cat#926 – 68071 ) . ChIP was performed as previously described ( Jamieson et al . , 2016 ) . qPCR was performed using the Quanta Biosciences PerfeCTa Sybr Green FastMix and an Applied Biosystems Step One Plus Real-Time PCR System . ChIP-libraries were prepared as previously described ( Jamieson et al . , 2016 ) and sequencing was performed using an Illumina NextSeq 500 or HiSeq 4000 sequencer with 75- or 100-nt single-end reads , respectively . All sequencing reads were mapped to the corrected N . crassa OR74A ( NC12 genome ) ( Galazka et al . , 2016 ) using Bowtie2 ( Langmead and Salzberg , 2012 ) . ChIP-seq read coverage was averaged , normalized , and analyzed using tools available from deepTools2 ( Ramírez et al . , 2016 ) and SAMtools ( Li et al . , 2009 ) on the open-source platform Galaxy ( Afgan et al . , 2016 ) . Sequencing tracks are displayed as 25-nt-window TDF or bigWig files with the Integrative Genomics Viewer ( IGV ) ( Robinson et al . , 2011 ) . The following antibodies were used for ChIP: H3K27me3 ( Millipore , Cat#07 – 449 ) , H3K36me3 ( Abcam , Cat#ab9050 ) , H3K36me2 ( Abcam , Cat#ab9049 ) , H3K27ac ( ActiveMotif , Cat#39133 ) , H3K27me2/3 ( ActiveMotif , Cat#39535 ) . RNA was isolated ( Jamieson et al . , 2013 ) , DNase treated ( Thermo Fisher Scientific ) , cleaned ( Agencourt RNAClean XP beads; Beckman Coulter ) , and Poly-A +RNA seq libraries prepared ( KAPA Stranded mRNA-seq kit; KAPA Biosystems ) and sequenced on a Illumina NextSeq 500 or HiSeq 4000 sequencer with 75- or 100-nt single-end reads , respectively . High-quality ( Kmer filtering ) adapter-trimmed reads were identified ( Stacks ) ( Catchen et al . , 2013 ) , mapped ( TopHat2 ) ( Kim et al . , 2013 ) , sorted ( SAMTools ) ( Li et al . , 2009 ) , and directionality-preserved read numbers for genes were calculated ( HTSeq ) ( Anders et al . , 2015 ) . Differential gene expression ( DESeq2 ) ( Love et al . , 2014 ) analysis examined pair-wise differences between WT and mutants or within replicates . Sequencing analysis was performed with previously described software using the open-source platform Galaxy ( Afgan et al . , 2016 ) . Tools available from DeepTools ( Ramírez et al . , 2016 ) were used for the following: 1 ) bamCoverage was used to generate coverage bigWig files; 2 ) bamCompare was used to normalize and obtain log2ratios from two BAM files; 3 ) computeMatrix was used to prepare data for heatmaps or profiles; 4 ) plotHeatmap was used to create heatmaps for score distributions; 5 ) plotProfile was used to create meta-plots of score distributions . Tools available from SAMtools ( Li et al . , 2009 ) were used for the following: 1 ) BedCov was used to calculate read depth over given intervals; 2 ) Merge BAM Files was used to combine replicates . GraphPad Prism was used to analyze frequencies and prepare histograms . Complete ChIP-seq and RNA-seq files , gene expression values , ChIP-seq intensity values have been deposited in NCBI’s Gene Expression Omnibus ( GEO; http://ncbi . nlm . nih . gov/geo ) and are accessible through GEO Series accession number GSE118495 and , as part of a previously reported series GSE82222 ( Klocko et al . , 2016 ) and GSE104019 ( Jamieson et al . , 2018 ) . Requests for materials should be addressed to VTB and EUS . All Neurospora crassa strains are available at the Fungal Genetic Stock Center ( McCluskey et al . , 2010 ) .	Not all genes in a cell’s DNA are active all the time . There are several ways to control this activity . One is by altering how the DNA is packaged into cells . DNA strands are wrapped around proteins called histones to form nucleosomes . Nucleosomes can then be packed together tightly , to restrict access to the DNA at genes that are not active , or loosely to allow access to the DNA of active genes . Chemical marks , such as methyl groups , can be attached to particular sites on histones to influence how they pack together . One important site for such marks is known as position 36 on histone H3 , or H3K36 for short . Correctly adding methyl groups to this site is critical for normal development , and when this process goes wrong it can lead to diseases like cancer . An enzyme called SET-2 oversees the methylation of H3K36 in fungi , plants and animals . However , many species have several other enzymes that can also add methyl groups to H3K36 , and their roles are less clear . A type of fungus called Neurospora crassa contains just two enzymes that can add methyl groups to H3K36: SET-2 , and another enzyme called ASH1 . By performing experiments that inactivated SET-2 and ASH1 in this fungus , Bicocca et al . found that each enzyme works on a different set of genes . Genes in regions marked by SET-2 were accessible for the cell to use , while genes marked by ASH1 were inaccessible . ASH1 also affects whether a methyl group is added to another site on histone H3 . This mark is important for controlling the activity of genes that are critical for development . ASH1 is found in many other organisms , including humans . The results presented by Bicocca et al . could therefore be built upon to understand the more complicated systems for regulating H3K36 methylation in other species . From there , we can investigate how to intervene when things go wrong during developmental disorders and cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2018	ASH1-catalyzed H3K36 methylation drives gene repression and marks H3K27me2/3-competent chromatin
The mechanosensitive channel of large conductance , which serves as a model system for mechanosensitive channels , has previously been crystallized in the closed form , but not in the open form . Ensemble measurements and electrophysiological sieving experiments show that the open-diameter of the channel pore is >25 Å , but the exact size and whether the conformational change follows a helix-tilt or barrel-stave model are unclear . Here we report measurements of the distance changes on liposome-reconstituted MscL transmembrane α-helices , using a ‘virtual sorting’ single-molecule fluorescence energy transfer . We observed directly that the channel opens via the helix-tilt model and the open pore reaches 2 . 8 nm in diameter . In addition , based on the measurements , we developed a molecular dynamics model of the channel structure in the open state which confirms our direct observations . Mechanosensitive ( MS ) channels are essential in both eukaryotes and prokaryotes ( Perozo , 2006; Árnadóttir and Chalfie , 2010; Haswell et al . , 2011 ) . In eukaryotes , they are involved in diverse processes such as embryonic development , touch , pain , hearing , lung growth , and muscle homeostasis ( Hamill and Martinac , 2001; Chalfie , 2009; Árnadóttir and Chalfie , 2010 ) . In bacteria , they are ‘safety valves’ , opening their pores to release the pressure to protect cells from hypo-osmotic shock ( Booth and Blount , 2012 ) . The rise in antibiotic resistance , and the crucial role MS channels play in bacterial adaptation , makes it important to understand the MS channels as potentially new drug targets ( Booth and Blount , 2012 ) . When high pressure ( ∼10 mN/m ) causes the bacterial mechanosensitive channel of large conductance ( MscL ) to open , it forms a large , nonselective pore with a very high conductance ( ∼3 nS ) that is permeable to various ions and small organic osmolytes . In 1998 , MscL from Mycobacterium tuberculosis in the closed state was crystallized by Rees and co-workers ( Chang et al . , 1998 ) . They showed that MscL is a pentamer made up of five identical subunits ( Figure 1A , B ) . Each subunit consists of one cytoplasmic α-helix ( the CP domain ) and two transmembrane α-helices ( the TM1 and TM2 helices ) , which extend through the cell membrane and are joined by a periplasmic loop ( Figure 1B ) . TM1 and TM2 are primarily responsible for gating; it has been shown that complete deletion of the CP domain does not change the gating parameters substantially ( Anishkin et al . , 2003 ) . 10 . 7554/eLife . 01834 . 003Figure 1 . Cartoon representation of the structure of MscL in the closed conformation in the ( A ) top view and ( B ) side view ( PDB ID: 2OAR [Chang et al . , 1998; Steinbacher et al . , 2007] ) , and scheme of single molecule FRET setup . MscL is a homo-pentamer consisting of five identical subunits . Each subunit consists of one cytoplasmic α-helix ( CP ) and two transmembrane α-helices ( TM1 and TM2 ) , which extend through the cell membrane and are joined by a periplasmic loop ( Chang et al . , 1998 ) . ( C ) Residues measured using smFRET . Three residues on each of the transmembrane helices ( M42C , A27C and I25C on TM1; Y75C , Q80C and V82C on TM2 ) were chosen . Note that no residues on the CP were chosen because the complete deletion of the CP does not change the gating parameters substantially ( Anishkin et al . , 2003 ) . ( D ) Labeled MscL proteins were reconstituted into liposomes , which were then immobilized on a coverslip and used for smFRET experiments . ( E ) The addition of LPC traps the protein in the open conformation ( Perozo et al . , 2002b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 003 Despite this progress , the open form of MscL has not been crystallized . This leaves two questions unanswered: what is the exact size of the open pore of MscL , and how does the channel open ? Several techniques , for example , permeation of organic ions ( Cruickshank et al . , 1997 ) , Electron paramagnetic resonance ( EPR ) ( Perozo et al . , 2002a , 2002b ) and ensemble fluorescence resonance energy transfer ( FRET ) ( Corry et al . , 2005b , 2010 ) have attempted to measure the pore size . However , systematic errors likely result in an overestimation of ( Cruickshank et al . , 1997 ) , an underestimation of ( Corry et al . , 2005b , 2010 ) , or an insensitivity to the requisite distances ( Perozo et al . , 2002a ) . For example , EPR was only able to establish that the open pore is >25 Å ( 11 ) . Ensemble FRET , which yielded some insightful results , is potentially sensitive to larger distances ( ∼80–100 Å ) ( Roy et al . , 2008 ) . However , due to multiple labeling , problems with protein clustering , and the need for Monte-Carlo simulations to extract distance information , there was much variability and uncertainty in the results ( Corry et al . , 2005a , 2005b , 2010 ) . Another important question is how the MscL channels open , that is how the helices rearrange upon channel activation ( i . e . , from the closed state to the open state ) . Currently , there exist two predominant models for the opening of MscL: the barrel-stave model and the helix-tilt model ( Figure 2; Perozo , 2006 ) . The barrel-stave model ( Figure 2C , D ) involves motion of the transmembrane helix 1 ( TM1 ) with the transmembrane helix 2 ( TM2 ) remaining stationary; the open pore is lined by both TM1 and TM2 and the helices are fairly vertical ( where the membrane is horizontal ) . This model derives primarily from the number of transmembrane helices and the large size of the open pore . In contrast , the helix-tilt model ( Figure 2E , F ) , which has been proposed more recently ( Sukharev et al . , 2001a , 2001b; Betanzos et al . , 2002 ) , involves motion of TM1 and TM2 , with both swinging away from the pore upon channel opening and both helices tilting toward the plane of membrane . Recent evidence from cysteine-crosslinking experiments , EPR experiments , and ensemble FRET experiments , argue in favor of the helix-tilt model ( Betanzos et al . , 2002; Perozo et al . , 2002a; Corry et al . , 2010 ) . 10 . 7554/eLife . 01834 . 004Figure 2 . The opening models for MscL . The MscL opens from ( A and B ) the closed state , to ( C and D ) the open state via the barrel-stave model or ( E and F ) the open state via the helix-tilt model . The top figures ( A , C , E ) are top views and the bottom figures ( B , D , F ) are the side views . TM1 helices are shown in red while TM2 in blue . In the barrel-stave model ( C and D ) , TM1 swings away from the pore center but TM2 remains stationary upon channel activation , generating an open pore lined by both TM1 and TM2 and the helices are more parallel to the membrane normal than the membrane plane . In the helix-tilt model ( E and F ) , both TM1 and TM2 swing away from the symmetry axis and both helices tilt toward the plane of membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 004 In the present work , we focused on the transmembrane helices involved in the opening of MscL from Escherichia coli ( EcoMscL ) , using a single-molecule fluorescence resonance energy transfer ( smFRET ) . MscL channels were reconstituted in liposomes during smFRET measurements and thus the channels were in their quasi-native environment . In addition , although MscL is a pentamer , we utilized photobleaching to virtually sort out the population of molecules with a single donor and a single acceptor , allowing us to make accurate smFRET measurements . It is the first time that smFRET has been applied to liposome-reconstituted membrane proteins with more than three monomers . We measured movement of three residues on TM1 ( M42C , A27C , and I25C; Figure 1C ) and three residues on TM2 ( Y75C , Q80C and V82C; Figure 1C ) , from which we determined not only the translational movements but also the tilting of each helix . We observed the tilting of the helices in a model-free fashion , arguing strongly in favor of the helix-tilt model . In addition , from the movement of the residue ( I25C ) right at the gating region , we determined directly that the open pore reaches 2 . 8 nm in diameter . Lastly , we developed a molecular dynamics model of the channel structure in the open state based on the smFRET results , while using the crystal structure of the protein in the closed state as a reference . The model of the open structure satisfies all the distance constraints measured from smFRET experiments . The developed open structure confirmed that the pore size of the fully open channel is 2 . 8 nm in diameter , achieved via the helix-tilt opening model . Purified MscL mutants ( Figure 3—figure supplement 1 ) were labeled with Alexa Fluor 488 ( AF488 ) and Alexa Fluor 568 ( AF568 ) and reconstituted into ∼50 nm liposomes made of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) with 2% 1 , 2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-biotinyl ( BPE ) ( Figure 1D ) . The liposomes were then immobilized on a glass coverslip , via biotin-avidin interaction , for smFRET measurements ( Figure 1D ) . To access the open state of the channels , 1-oleoyl-2-hydroxy-sn-glycero-3-phosphocholine or lysophosphatidylcholine ( LPC ) of 25% molar ratio was added ( Perozo et al . , 2002b , 2002a; Corry et al . , 2005b , 2010; Figure 1E ) and incubated for >10 min before immobilization . Just before performing smFRET experiments , the fluorescence spectra of the samples ( ±LPC ) were recorded with excitation at 488 nm to confirm that the channels open up with LPC by observing the shift in the FRET peaks . ( The channel activity is also determined by observing the opening of the channels upon application of negative pressure [suction] to the patch pipette . The labeled proteins for patch-clamp experiments are from a different aliquot , although the same batch , of the labeled proteins for smFRET experiments ) . We emphasize that , although smFRET has been applied to study the conformational changes of channels and transporters ( Choi et al . , 2010; Zhao et al . , 2010 , 2011; Akyuz et al . , 2013 ) , to our knowledge , it is the first time that smFRET has been used with channels reconstituted to liposomes . Via smFRET measurements , we observed fluorescence intensity traces with one or two photobleaching steps ( Figure 3—figure supplement 2A , B ) . This is the expected result because MscL is a homo-pentamer and the labeling of fluorophores is stochastic . The number of photobleaching steps tells the number of fluorophores attached to a channel . Only the traces showing a single photobleaching step in both the donor and acceptor channels , ensuring that only a single donor and/or acceptor fluorophore , were included in the analysis ( Figure 3—figure supplement 2A ) . Donors were , in most cases , photobleached first , resulting in simultaneous dropping of the fluorescence intensities in both donor and acceptor channels ( Figure 3—figure supplement 2A , B ) . Subtraction of the intensities before and after photobleaching gives the intensities of donor ( ID ) and acceptor ( IA ) , which are used for the calculation of FRET efficiency . Note that the intensities , ID and IA , automatically remove the direct excitation of acceptor ( i . e . , the leakage of acceptor emission in the donor channel ) . However , the leakage of donor emission in the acceptor channel is still present . To measure this leakage , MscL channels were labeled with donors-only and the leakage coefficient ( ℓ ) was measured experimentally: ℓ = IDA/IDD ≈ 0 . 09 , where IDA is the intensity of donor emission in the acceptor channel and IDD is the intensity of donor emission in the donor channel . Furthermore , to determine the actual FRET efficiency , another instrumental correction was made through the correction factor γ , which accounts for the differences in quantum yield and detection efficiency between the donor and the acceptor . It was calculated as the ratio of change in the acceptor intensity , ΔIA , to the change in the donor intensity , ΔID , upon acceptor photobleaching: from the traces where the acceptor photobleached first ( Roy et al . , 2008 ) , we estimated the value γ = ΔIA/ΔID ≈ 0 . 89 ± 0 . 06 ( Figure 3—figure supplement 2C ) . We analyzed a few hundred traces ( varying between 134 and 577 traces ) with single photobleaching steps in the absence and presence of LPC for each mutant ( Figure 1C , Figure 3 ) . Here we focus on the mutant M42C for the sake of illustration . For the single photobleaching steps of M42C , 428 and 577 traces , in the absence and presence of LPC , respectively , were analyzed . The corrected FRET efficiencies were calculated and their distribution was then plotted and fitted with Gaussians via maximum likelihood estimates , shown in Figure 3A , B , while the number of Gaussians was determined according to the corrected Akaike information criterion ( AICc ) and the Bayesian information criterion ( BIC ) ( Table 1; Akaike , 1974; Schwarz , 1978; Sugiura , 1978 ) . In the absence of LPC , we observed three peaks at E = 0 . 1 , 0 . 28 and 0 . 63 , respectively ( Figure 3A ) . In the presence of LPC , the third peak showing the highest FRET efficiency diminishes , leaving mainly two Gaussians ( E = 0 . 1 and 0 . 23 , Figure 3B ) . This transition ( i . e . , the highest peak decreases and the lowest peak increases ) is more obvious when we plotted the difference between the normalized FRET distributions ( ∑PX=1 , where X = + for in the presence of LPC and X = − for in the absence of LPC ) under the two conditions , as shown in Figure 3C: after adding 25% LPC , the peak at E ∼0 . 6 went away but the fraction of the peak at E ∼0 . 1 built up . Note that the highest peak at E ∼0 . 6 does not completely disappear in the presence of 25% LPC , which is consistent with ( Perozo et al . , 2002b ) . 10 . 7554/eLife . 01834 . 005Figure 3 . Single molecule FRET results for MscL M42C . The distribution of FRET efficiency of M42C in the ( A ) absence and ( B ) presence of LPC were plotted and fitted with Gaussians . ( C ) The difference between the normalized FRET distributions under the two conditions ( ±LPC ) , ΔP , emphasizes the diminishing of the third peak at E ∼0 . 6 after adding LPC . ( D ) The variance between the normalized FRET distributions under the two conditions ( ±LPC ) , ΔP2 , decreases as the fraction of BPE in the liposomes is increased from 2% to 16% . ( E ) Histograms of FRET efficiencies in the absence ( top row , −LPC ) and presence ( bottom row , +LPC ) of LPC for the other five residues ( I25C , A27C , Y75C , Q80C , and V82C ) measured in the current study . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 00510 . 7554/eLife . 01834 . 006Figure 3—figure supplement 1 . Examples of FPLC traces and SDS-PAGE gel for purification of MscL . ( A ) Absorbance at 280 nm ( black ) and 260 nm ( red ) of MscL proteins are monitored during the running of FPLC . Fractions of 0 . 5 ml are collected and the fractions corresponding to MscL ( green area , 2 ml ) is used in the later experiments . ( B ) The purity of MscL after FPLC purification is estimated and verified by SDS-PAGE . The purity is in general >95% ( in terms of mass ) . For this specific sample , the purity is ∼97% ( 2 . 3% for very faint bands at high molecular weight ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 00610 . 7554/eLife . 01834 . 007Figure 3—figure supplement 2 . Examples of fluorescent intensity traces showing ( A ) a single photobleaching step , ( B ) multiple photobleaching steps , and ( C ) a single photobleaching step but the acceptor photobleached first . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 00710 . 7554/eLife . 01834 . 008Figure 3—figure supplement 3 . FRET efficiencies between non-neighboring subunits ( Ef , o ) . Ef , o is consistent and compatible with the final model . The final model satisfies Rn , o , from which Rf , o can be calculated based on geometry . The corresponding Ef , o are shown as blue bars . As a comparison , the measured Ef , o values ( as well as errors ) are shown as magenta and gray bars . Note that both samples before and after adding LPC give Ef , o ( as–LPC sample is a mixture of closed and open states ) . The values from samples without LPC are shown in magenta while the values from samples with LPC are shown in gray . We found that the measured Ef , o values are consistent ( within error ) from samples without and with LPC . We also found that all mutants ( except that M42C is slightly off ) give Ef , o measurements compatible ( within error ) with the final model . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 00810 . 7554/eLife . 01834 . 009Table 1 . The best model of the fittings of FRET efficiency distribution were determined by calculating the corrected Akaike information criterion and Bayesian information criterion: AICc=−2⁡ln⁡LM +2k+2k ( k+1 ) N−k−1 , BIC=−2⁡ln⁡LM+k⁡ln⁡N where LM is the maximum likelihood by the model , k the number of parameters of the model , N the number of datapoints used in the fit ( Akaike , 1974; Schwarz , 1978; Sugiura , 1978 ) DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 009# of fitting peaks− ln LMAICcBICM42C; No LPC 1−1 . 90 . 38 . 4 2−43 . 4−76 . 6−56 . 4 3−56 . 4−96 . 4−64 . 3 4−58 . 1−93 . 5−49 . 5M42C; With LPC 1−190 . 9−377 . 9−369 . 2 2−279 . 5−549 . 0−527 . 3 3−290 . 7−565 . 1−530 . 5 4−292 . 7−563 . 0−515 . 5The lowest AICc and BIC values give the best fitting model: three peaks ( bold values ) . In the absence of LPC , the existence of three peaks , rather than two peaks , can be explained by considering the effect of tethering on the liposome . As the MscL channel is a homo-pentamer , we initially expected two distances between donor and acceptor in each state ( Rn and Rf in Figure 4A ) and thus two peaks for the distribution of FRET efficiency , assuming that all the channels are closed in the absence of LPC . However , this assumption is not necessarily true , especially in our situation where liposomes are immobilized and the proteins are responsive to membrane tension . It had been predicted by theories and observed in experiments that immobilization of liposomes ( or vesicles ) results in significant membrane tension and possibly rupture ( Zhdanov et al . , 2006; Chung et al . , 2009; Serro et al . , 2012 ) . In our experiments , the membrane tension is expected to be high , ∼30–40 kBT , due to the strong interaction between BPE and the surface via biotin-neutravidin ( Miyamoto and Kollman , 1993; Rico and Moy , 2007 ) . With such strong interaction , giant unilamellar vesicles ruptured spontaneously , as has been observed experimentally ( Chung et al . , 2009 ) . The consequence is that some of the MscL channels switch to the open conformation upon the immobilization of the liposomes . ( However , the fraction of open channels might be different for different mutants even if the membrane tension is similar ) . Therefore , the FRET histogram for the no-LPC sample includes a mixture of closed and open MscL channels . To test this hypothesis , control experiments were performed by increasing the fraction of BPE in the liposomes , guided by a theoretical prediction ( Zhdanov et al . , 2006 ) : if the hypothesis was true , the membrane tension in the liposomes due to immobilization would be higher , more channels would open in the absence of LPC , and therefore the difference between the FRET histograms with and without LPC would be smaller . We varied the fractions of BPE in the liposomes from 2% to 16% and , indeed , observed that the difference between the samples with and without LPC decreases ( Figure 3D ) . We quantified the difference by the ( unscaled ) variance , ΔP2=∑i=1N ( Pi+−Pi− ) 2 , where PiX is the probability distribution of FRET efficiency , ∑i=1NPiX=1 . We observed that the variance ΔP2 decreased by 98% , from 0 . 084 to 0 . 002 , when the fraction of BPE in the liposomes increased from 2% to 16% , supporting the hypothesis that the sample without LPC is a mixture of closed and open channels . 10 . 7554/eLife . 01834 . 010Figure 4 . Movement of residues . ( A ) Each residue ( highlighted in green ) defines a circumcircle ( dashed red circle ) of radius r ( or diameter D , where D , as shown , is Dclosed , although upon opening would be Dopen ) , centered at the pore center ( O ) . Upon channel activation , the protein expands ( radius changes from rclose to ropen ) , or equivalently , the residue moves by Δr = ropen−rclose , measured from the pore center ( O ) . ( B ) Sketch of MscL from closed state ( blue pentagon ) to open state ( purple pentagon ) . The residue of interest ( vertices of the pentagons ) moves Δr from the pore center . ( C and D ) Translational movements ( Δr ) of residues on TM1 and TM2 measured via smFRET . All the residues move away from the pore center , arguing in favor of the helix-tilt model . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01010 . 7554/eLife . 01834 . 011Figure 4—figure supplement 1 . Molecular structures of fluorophores used in the experiments . ( A and B ) 5′ isomer of Alexa Fluor 488 ( donor ) ; ( C and D ) 6′ isomer of Alexa Fluor 488 ( donor ) ; ( E and F ) 5′ isomer of Alexa Fluor 568 ( acceptor ) ; ( G , H ) 6′ isomer of Alexa Fluor 568 ( acceptor ) . The donor is 17 . 1 Å ( 5′-isomer ) or 16 . 3 Å ( 6′-isomer ) while the acceptor is 17 . 4 Å ( 5′-isomer ) or 17 . 4 Å ( 6′-isomer ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01110 . 7554/eLife . 01834 . 012Figure 4—figure supplement 2 . Geometric analysis of the distances of interest while taking into account the finite size of fluorescent probes and the breaking of fivefold symmetry of the protein due to attachment of probes . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01210 . 7554/eLife . 01834 . 013Figure 4—figure supplement 3 . Effect of hexameric MscL in protein preparation . ( A ) Effect of impurities on FRET peak ( averaged over 100 simulations for each fraction of hexamers ) . ( B ) The peak broadening in a simulation for ∼30% hexamers as observed in Gandhi et al . ( 2011 ) . ( C ) The conductance of wild-type EcoMscL ( WT ) and a mutant ( I25C ) , agreeing with that of an EcoMscL pentamer . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 013 A simple estimation based on the crystal structure of MscL in the closed state ( Chang et al . , 1998 ) and previously predicted/estimated open pore-size ( Cruickshank et al . , 1997; Perozo et al . , 2002a; Corry et al . , 2010 ) indicates that it is likely that , due to limited resolution of FRET , the middle peak ( E = 0 . 28 ) is an overlap of two peaks corresponding to Rn of the open state ( Rno ) and Rf of the closed state ( Rfc ) ( for Rn and Rf , Figure 4A , B ) . The geometry of the protein ( i . e . , fivefold symmetry ) gives Rn=D⋅sin ( π5 ) and Rf=D⋅sin ( 2π5 ) where D is the protein diameter . Take M42C as an example: the crystal structure gives the diameter of the protein in the closed state ( using the Cα atoms ) , Dc = 4 . 4 nm , resulting in Rfc=Dc⋅sin ( 2π5 ) =4 . 2 nm . The expected change in the protein size ( ΔD ) is from 2 . 5 to 4 nm ( Cruickshank et al . , 1997; Perozo et al . , 2002a; Corry et al . , 2010 ) , with the most recent report of 2 . 8 nm from ensemble FRET measurements ( Corry et al . , 2010 ) . Then the expected Do = Dc + ΔD is 7 . 2 nm , which gives Rno=Do⋅sin ( π5 ) =4 . 2 nm , exactly the same as Rfc ( The expected Rno is 4 . 1–4 . 9 nm when taking into account the expected range from the literature ) . In the simple calculation above , the positions of Cα atoms for the estimations were used . However , the side chains of the residues and the attached fluorescent probes will add an additional length on the order of 2 nm , resulting in that the chance for Rfc and Rno to overlap is even higher . Furthermore , it has been observed that ≤30% of MscL are hexamers , instead of pentamers , in detergents such as n-Dodecyl-β-D-maltopyranoside ( DDM ) used in the current study ( Gandhi et al . , 2011 ) . This would tend to ‘smear’ the middle peaks of FRET in the absence of LPC . Therefore , to be consistent and accurate , we always use the highest FRET efficiency ( Enc and Eno ) for the calculation of distance changes ( ΔRn ) . On the other hand , we did find that all mutants give Efo measurements compatible ( i . e . , within error ) with the final model ( except that M42C is slightly off ) , as shown in Figure 3—figure supplement 3 . FRET between neighboring MscL channels on the same liposome had been a problem in previous ensemble FRET experiments . To decrease the likelihood of its happening , and to effectively solve the problem , we applied two strategies . First , we used 5% labeled channels together with 95% unlabeled ones for reconstitution in liposomes . As a result , we had 16x lower molar ratio of labeled proteins ( pentamers ) to lipids than that in the ensemble FRET experiments: 1:4000 vs 1:250 ( Corry et al . , 2005b , 2010 ) , greatly reducing the likelihood of inter-molecular FRET . We found that adding a mixture of labeled and unlabeled protein helps to obtain no more than one fluorescent channel per liposome while ensuring efficient incorporation of channels into liposomes . In addition , only traces showing a single photobleaching step in both donor and acceptor channels were included in analysis , which helps further removing the FRET between neighboring MscL channels in the analysis . These strategies reduce significantly the likelihood that energy transfer happens between two adjacent channels even in the presence of MscL clustering , simplifying the interpretation of FRET results . Another note is that we used maximum-likelihood estimation ( MLE ) ( In Jae , 2003 ) for peak fitting . This method was chosen particularly because it does not require binning the data before fitting . Although there are mathematical ways for selection of ‘good’ bin sizes ( Shimazaki and Shinomoto , 2007 ) , the selection of bin size is , in practice , subjective , and the peaks derived can be affected with different bin sizes . After MLE fitting , we then bin the data and plot the histograms for the sake of presentation purpose . How the data is binned does not change the fitting parameters . The Förster radius ( R0 ) for AF488 and AF568 is calculated by means of R0 ∝ ( κ2 QD ) 1/6 ( Förster , 1948; Iqbal et al . , 2008 ) . Because κ2 and QD , can be environmentally sensitive , we measured the quantum yield and orientation factor for the fluorophores conjugated to each and every channel mutant ( Fery-Forgues and Lavabre , 1999; Lakowicz , 1999; Figure 5 ) . The quantum yields of AF488 conjugated to various MscL mutants are summarized in Table 2; Figure 5A , corrected for polarization effects ( Fery-Forgues and Lavabre , 1999; Lakowicz , 1999 ) . It is noted that the fluorophores used in the current study are mixtures of 5′ and 6′ isomers . However , it was expected that this will not affect the results because ( 1 ) they have successfully been used in many smFRET studies ( Marras et al . , 2002; Yin et al . , 2005; Jäger et al . , 2006; Granier et al . , 2007; Majumdar et al . , 2007 ) ; ( 2 ) the chromophores of the isomers are exactly the same while the only difference between the isomers lies in where the linker of carbon-chain [ ( CH2 ) 5NHCO] is attached; ( 3 ) we examined the molecular structures of the probe-isomers and confirmed that the difference in molecular size is <5% between isomers ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 01834 . 014Figure 5 . Measurement of R0 . ( A ) Absorbance and fluorescence spectra of AF488-MscL and fluorescein ( as a standard ) , used to determine the quantum yield of AF488 conjugated to MscL mutants . ( B ) Anisotropy of AF488 and AF568 conjugated to MscL mutant ( M42C ) , corrected for the intrinsic polarization properties of the microscope , and for the high numerical aperture of the objective . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01410 . 7554/eLife . 01834 . 015Figure 5—figure supplement 1 . Anisotropy of AF488 and AF568 conjugated to MscL mutants , corrected for the intrinsic polarization properties of the microscope , and for the high numerical aperture of the objective . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01510 . 7554/eLife . 01834 . 016Table 2 . Measurements of smFRET experimentsDOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 016ResidueHelixQdAdAaR0 ( nm ) EncEnoΔRn ( nm ) Δr ( nm ) ΔD ( nm ) 42TM10 . 330 . 220 . 015 . 5+0 . 4−0 . 30 . 630 . 231 . 7+0 . 7−0 . 51 . 4+0 . 6−0 . 42 . 8+1 . 1−0 . 827TM10 . 280 . 120 . 095 . 3+0 . 4−0 . 30 . 720 . 331 . 5+0 . 6−0 . 41 . 3+0 . 5−0 . 32 . 5+0 . 9−0 . 625TM10 . 420 . 190 . 065 . 7+0 . 5−0 . 40 . 780 . 421 . 4+0 . 6−0 . 51 . 2+0 . 6−0 . 42 . 4+1 . 1−0 . 875TM20 . 620 . 190 . 115 . 6+0 . 6−0 . 50 . 600 . 162 . 4+1 . 0−0 . 72 . 0+0 . 8−0 . 64 . 0+1 . 6−1 . 280TM20 . 220 . 180 . 095 . 1+0 . 5−0 . 30 . 720 . 291 . 6+0 . 7−0 . 41 . 4+0 . 6−0 . 42 . 7+1 . 1−0 . 882TM20 . 390 . 240 . 136 . 1+0 . 7−0 . 60 . 760 . 351 . 6+0 . 9−0 . 81 . 4+0 . 8−0 . 62 . 7+1 . 5−1 . 3Qd is the quantum yield of donor ( AF488 ) after conjugation to each MscL mutant . Ad and Aa are the anisotropy of donor ( AF488 ) and acceptor ( AF568 ) after conjugation , respectively . R0 is the Förster radius . Enc and Eno are the FRET efficiencies in the closed and open states , respectively . ΔR is the change in the distances between donor and acceptor ( ΔRn = Rno−Rnc , Figure 4A , B ) . ΔD is the change of the protein diameter ( ΔD = Do−Dc ) . Δr is the translational movement of the residue , measured from the pore center , Δr = ΔD/2 . Note that the errors shown in the table are the maximum possible errors due to anisotropic orientation of the dyes . The actual errors are expected to be much smaller . The orientation factor κ2 , was determined by measuring the anisotropy of the conjugated fluorophores ( Table 2; Figure 5B , Figure 5—figure supplement 1 ) . The anisotropy of both donor and acceptor for most residues is <0 . 2 and therefore κ2 , in fact , is close to 2/3 ( Clegg , 1992; Andrews and Demidov , 1999; Roy et al . , 2008 ) . Nonetheless , we calculated the maximum possible errors in R0 due to anisotropic orientation of the dyes ( Table 2; Figure 5B , Figure 5—figure supplement 1 ) ; the actual errors in R0 should be much smaller . Another source of error in R0 lies in the measurements of QD , which were performed for AF488-MscL in detergent ( PBS + 1 mM DDM ) , which was not exactly the same environment for fluorophore-MscL conjugates in smFRET experiments ( incorporated in liposomes in PBS ) , although the buffer was kept the same . Furthermore , it is possible that the addition of LPC and the conformational change of MscL changes QD as well , resulting in additional errors in R0 and in the distances calculated below . It is noticed that the anisotropy of the acceptor is consistently lower than that of the donor although the acceptor is larger . This could be attributed to many factors , as stated by the Perrin equation ( Perrin , 1926 ) : A0/A = 1 + 6Dτ , including the rotational diffusion coefficient ( D ) , fluorescence lifetime ( τ ) and , more significantly , the fundamental anisotropy ( A0 ) which varies according to wavelength ( Weber and Shinitzky , 1970; Lakowicz , 1999 ) . The anisotropies given in Table 2 were measured at the wavelengths where smFRET experiments were performed . We note that the finite size and length of the fluorescent probes brings additional difficulty to converting FRET measurements to the estimation of distances and therefore to structural modeling . To overcome this difficulty , our strategy is to focus on the movement of residues , instead of the absolute distances , and then to develop the structural model based on the changes of distances . For illustration purpose , we set up the general scene in Figure 4—figure supplement 2 . The pentameric MscL channel ( light blue ) , centered at point O , is labeled with two probes ( dark blue ) on two cysteine residues ( specified by residue # ) , while the other three residues ( with the same residue # but on different subunits ) remain empty/unlabeled ( dashed green ) . ( Here the breaking of symmetry is taken into account in this general case ) . To emphasize the size and length of the probes , filled green and red circles are used to indicate the actual chromophores . Then the actual chromophores of the labeled probes ( green and red circles ) and the center of the protein ( O ) define a pentagon ( dotted orange pentagon , with a side length of R ) and a circumcircle ( black circle , with a diameter of D , or a radius of r = D/2 ) . We call R , D and r apparent distance/diameter/radius because they could be measured/calculated from FRET experiments ( with corrections ) and they gave the apparent size of the protein ( i . e . , the protein appears to have a radius of r based on FRET experiments ) . There is another set of distances , which are more relevant to the protein and to the MD simulations for the structural model . We call this set of distances true values . For example , the true radius of the protein ( rt ) ( referring to a specific residue ) is defined as the distance between the Ca atom of the residue and the center of protein ( O ) . We note that the exact atom chosen for the definition of the true radius ( rt ) does not matter . Generally r = rt + rp ≠ rt due to the finite size of the probes ( rp ≠ 0 ) . As a result , converting the FRET measurements into a structural model is not straightforward . However , if the size of the probes does not change ( i . e . , Δrp = 0 ) upon channel activation , we then have Δr = Δrp + Δrt = Δrt ( similarly ΔD = ΔDt ) . We believe Δrp = 0 is a reasonable assumption for the following reasons: ( 1 ) no chemical reactions are happening for the probes and thus the structures of the probes do not change before and after channel opening; ( 2 ) anisotropy measurements show that the orientation of probes are not constrained significantly . As a result , the change in the apparent distances is the same as the change in the true distances . In other words , the movement of residues ( in the radical direction , Δrt ) can be obtained from the FRET measurements , even if the sizes and lengths of probes are nonzero . A note to make is that we have assumed that donors and acceptors have similar sizes in the argument above . This assumption could be justified by the molecular structures of the probes ( shown in Figure 4—figure supplement 1 ) , which shows that the difference between the donor-size and acceptor-size is <2% . On the other hand , a caveat is that , although not likely , there are possible situations where the size of the probe can change ( i . e . , Δrp ≠ 0 ) upon channel opening , due to , for example , steric hindrance . To conclude , the finite size of probes ( rp ∼1 . 7 nm ) brings additional difficulties to converting FRET measurements to estimation of distances: FRET results gave the distances between the chromophores of donors and acceptors , which is different from the distances between the Cα atoms of residues of interest . However , on the other hand , the movement of the residues ( or the movement of the Cα atoms of the residues ) in the radial direction is the same as the movement of the chromophores assuming that the size of the probes does not change ( i . e . , Δrp = 0 ) upon channel opening . We also note that , although the fivefold symmetry is broken due to the binding of one donor and one acceptor per pentamer , the geometric construction will not be affected . From here on , we use Δr ( and ΔD ) for the movement of residues ( and the change of the protein diameter ) without any subscript . We measured the change of FRET efficiency of MscL before and after channel activation using smFRET . For example , for M42C , the FRET efficiency changed from 0 . 63 ( closed state ) to 0 . 23 ( open state ) . We also determined experimentally the Förster radii ( R0 = 5 . 5+0 . 4−0 . 3 nm for M42C ) . This permitted us to estimate the change in the distance between donors and acceptors from the closed to open states ( Figure 4 ) , ΔRn = Rno–Rnc = R0 ( Eno−1–1 ) −1–R0 ( Enc−1–1 ) −1 ( ≈1 . 7 nm for M42C ) . We emphasize that some of the distances between fluorophores ( Rno and Rnc in Table 3 ) are indeed out of the sensitive range of EPR measurements , making FRET a more suitable technique in this context . 10 . 7554/eLife . 01834 . 017Table 3 . Measurements of smFRET experimentsDOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 017ResidueHelixEncEnoRnc ( nm ) Rno ( nm ) ΔRn ( nm ) Δr ( nm ) ΔD ( nm ) 42TM10 . 630 . 235 . 0+0 . 4−0 . 36 . 7+0 . 5−0 . 41 . 7+0 . 7−0 . 51 . 4+0 . 6−0 . 42 . 8+1 . 1−0 . 827TM10 . 720 . 334 . 6+0 . 3−0 . 26 . 0+0 . 4−0 . 31 . 5+0 . 6−0 . 41 . 3+0 . 5−0 . 32 . 5+0 . 9−0 . 625TM10 . 780 . 424 . 6+0 . 4−0 . 36 . 0+0 . 5−0 . 41 . 4+0 . 6−0 . 51 . 2+0 . 6−0 . 42 . 4+1 . 1−0 . 875TM20 . 600 . 165 . 7+0 . 5−0 . 48 . 1+0 . 8−0 . 62 . 4+1 . 0−0 . 72 . 0+0 . 8−0 . 64 . 0+1 . 6−1 . 280TM20 . 720 . 294 . 4+0 . 4−0 . 35 . 9+0 . 5−0 . 41 . 6+0 . 7−0 . 41 . 4+0 . 6−0 . 42 . 7+1 . 1−0 . 882TM20 . 760 . 354 . 7+0 . 5−0 . 56 . 2+0 . 7−0 . 61 . 6+0 . 9−0 . 81 . 4+0 . 8−0 . 62 . 7+1 . 5−1 . 3Enc and Eno are the FRET efficiencies in the closed and open states , respectively . Rnc and Rno are the distances between donor and acceptor . ΔR is the change in the distances between donor and acceptor ( ΔRn = Rno−Rnc ) . ΔD is the change of the protein diameter ( ΔD = Do−Dc ) . Δr is the translational movement of the residue , measured from the pore center , Δr = ΔD/2 . Note that the errors shown in the table are the maximum possible errors due to anisotropic orientation of the dyes . The actual errors are expected to be much smaller . As illustrated in the previous section , we focus on the more relevant distance of interest: the movement of the residue away from the pore center , Δr ( Figure 4B ) , or the change of protein diameter measured from the residue , ΔD . Because of the fivefold symmetry of the MscL channel , ΔD and Δr can be calculated readily according to ΔD = ΔRn/sin ( π /5 ) ≈ 2 . 8 nm , which yields Δr = ΔD/2 ≈ 1 . 4 nm ( for M42C ) . The Δr values of the residues are summarized in Table 2 . This value is above 2 . 5 nm , a lower bound predicted by EPR experiments ( Perozo et al . , 2002a ) , but larger than ΔD obtained from the previous ensemble FRET measurement: ΔDM42C = 2 . 8 nm ( smFRET ) vs ΔDM42C = 1 . 6 nm ( ensemble FRET ) ( Corry et al . , 2005b , 2010 ) . We emphasize that the measurements of two more residues ( I25C and A27C ) in the current study were also reported previously ( Corry et al . , 2010 ) . Our results are close to the values in their simulations ( ΔDI25C = 2 . 4 vs 2 . 5 nm; ΔDA27C = 2 . 5 vs 2 . 6 nm ) but differ significantly from the values measured directly from ensemble FRET experiments ( ΔDI25C = 2 . 4 vs 0 . 2 nm ) . It should be noted that ensemble experiments gave inconsistent measurements for ΔDI25C = 0 . 2 nm and ΔDA27C = 2 . 9 nm , although the two residues are close . In contrast , smFRET results show that ΔDI25C = 2 . 4 nm is similar to ΔDA27C = 2 . 5 nm . This clearly demonstrates the advantage of smFRET . We note that fluorophores/linkers at different residues are likely to be constrained differently . Furthermore , how they are constrained differently is not clear , partly due to the unavailability of the crystal structure of EcoMscL . However , certain residues are in agreement between the EcoMscL and the MtMscL ( Perozo et al . , 2001 ) . Nevertheless , the distances between donors and acceptors are not good to compare for different residues of EcoMscL . A more reasonable way is to compare the changes of distances , that is , the movements of residues . The calculations above were performed with the assumption that EcoMscL are pentamers . However , it is noted that a mixture of hexamers and pentamers were observed in certain detergents for EcoMscL ( Gandhi et al . , 2011 ) ( ≤30% hexamers in DDM ) . In our experiments , we ran FPLC ( Superdex 200 10/300 GL column ) for our MscL proteins and used the proteins from a single peak and checked with SDS-PAGE ( Figure 3—figure supplement 1 ) . In addition , in the smFRET experiments , we reconstituted the channels into liposomes , different from proteins in detergents where a mixture of hexameric channels and pentameric channels were observed ( Gandhi et al . , 2011 ) . Third , as the conductance of the mutants ( from electrophysiological recordings ) used in this study agrees with that of MscL pentamers ( Figure 4—figure supplement 3C ) , it is likely that we are reconstituting MscL pentamers into liposomes . Nonetheless , the possibility of having a ( small ) portion of hexamers in the sample in smFRET experiments could not be excluded completely . To estimate the introduced uncertainties , we investigated quantitatively the effect of the presence of hexamers via numerical simulations and found that the main effect of the presence of hexamers is not the shift of the peak center but the broadening of peak width ( Figure 4—figure supplement 3A-B ) . If the sample was 100% hexamers , the diameters of the protein , as well as the pore size , will be greater by a factor of sin ( π/5 ) /sin ( π/6 ) −1 ≈ 17 . 6% . In the presence of 30% hexamers , as observed in ( Gandhi et al . , 2011 ) , the calculations would be off by about 7 . 5% . Because the size of both Alexa fluorophores is significant ( ∼1 . 7 nm ) , it is possible that the attachment of the fluorophores to MscL channel results in various effects on the protein and on the FRET measurements . For example , the presence of the fluorophores might sterically hinder the conformational change of the proteins and prevent them from opening or closing . On the other hand , the steric hindrance might constrain the orientation of fluorophores , affect the relative orientation between the fluorophores and therefore add more errors on the distances converted from FRET efficiencies . In addition , the insertion of fluorophores to the protein might force the channel to be in a state different from the fully closed state , resulting in the distance change measurement is underestimated . However , we would like to emphasize that the expected effect is insignificant for the following reasons . First , if the insertion of fluorophore would result in significant steric hindrance on the protein , it is expected that the labeling is difficult ( i . e . , it takes much more effort for the fluorophores to be attached due to the steric hindrance ) . In other words , it is expected that steric hindrance is not significant on the mutants that are labeled well . More importantly , the channels after being labeled with AF488 and AF568 were confirmed to be functional by both ensemble FRET experiments ( by observing the shift in the FRET peak ) and patch-clamp measurements ( by observing the opening of the channels upon application of negative pressure to the patch pipette ) as shown in Figure 6 and previous publications with the same fluorophores ( Corry et al . , 2010 ) . 10 . 7554/eLife . 01834 . 018Figure 6 . Activation thresholds , Pa , of MscL mutants at the proximity of the narrowest pore constriction . The activation thresholds were determined by electro-physiological recordings by patch-clamping without and with 10 mM DTT . Three recordings in the presence of DTT are shown as examples: ( A ) G22C , ( B ) I24C , and ( C ) I25C . ( D ) Comparison of the mutants with the wild type ( WT ) shows that the thresholds for mutants G22C and I24C are more than twice higher than the wild type , indicating the function of the channel was affected by the mutations . This was also observed via ensemble and single molecule FRET experiments . However , the mutation I25C does not affect the gating parameter substantially . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 01810 . 7554/eLife . 01834 . 019Figure 6—figure supplement 1 . Positions of MtMscL mutants at the proximity of the narrowest pore constriction . ( A ) Top view and ( B ) side view of the close state as seen in the crystal structure . ( C ) Top view and ( D ) side view of the open state modeled from smFRET measurement . Three residue are shown: A20 ( red ) , V22 ( blue ) , and I23 ( green ) , which are equivalent to G22 , I24 , and I25 , respectively , in E coli MscL . A20 and V22 are closer to the pore than I23 . I23 is the only one among then that is facing outward from the channel axis and is accessible from the periphery of the protein . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 019 With smFRET , we measured the movements of three residues on TM1 ( M42C , A27C , and I25C ) and three on TM2 ( Y75C , Q80C and V82C ) summarized in Table 2 and Figure 4C , D . We observed directly and reliably for the first time , that both TM1 and TM2 swing away from the pore , supporting the helix-tilt model . Note that , among the three residues on each helix , two sites were very close to each other ( A27C and I25C on TM1 , Q80C and V82C on TM2 ) . They were chosen purposefully to be close; they served as consistency checks and confirmed that our smFRET measurements are accurate ( Table 2 ) . In addition , the top of both helices ( periplasmic side , Figure 1B , C; residues 42 on TM1 and 75 on TM2 ) moves further than the bottom ( 1 . 4 nm vs 1 . 2 nm for TM1 and 2 . 0 nm vs 1 . 4 nm for TM2 ) , indicating that rotational tilting of the helices ( toward the membrane plane ) is involved . We emphasize that it is the first direct ( model-free ) observation of both TM1 and TM2 swinging away from the pore center and of the tilting of the transmembrane helices . Therefore it is the first direct observation in favor of the helix-tilt model . To quantitatively investigate in detail how the MscL channel opens ( i . e . , how the helices move and rotate upon opening ) , we developed a computational model for the open structure of the MscL , starting from the crystal structure of MscL in the closed state ( PDB: 2OAR ) ( Chang et al . , 1998; Steinbacher et al . , 2007 ) and employing the measured residue movements . For this purpose , we performed MD simulation with distance constraints ( Brünger et al . , 1986; Trabuco et al . , 2009 ) ( i . e . , a virtual spring , Figure 7—figure supplement 1 ) using NAMD 2 . 9 ( Phillips et al . , 2005 ) . Although similar modeling attempts have been made by Corry et al . ( 2010 ) and Deplazes et al . ( 2012 ) by using distance changes measured from ensemble FRET , we would like to emphasize that all smFRET measurements were used for the simulation while previously only a selected subset of ensemble data were used ( as other data were not consistent with the resultant model ) ( Corry et al . , 2010 ) . For each measured residue , ten virtual springs were placed , five springs between the central carbon atom Cα of identical residues ( highlighted green in Figure 7—figure supplement 1 ) from adjacent monomers ( red springs in Figure 7—figure supplement 1 ) and five springs between the Cα of identical residues from non-adjacent monomers ( yellow springs in Figure 7—figure supplement 1 ) . The virtual springs were not applied to side chains because the flexibility of side chains likely introduces errors under large forces in the modeling process . The equilibrium lengths of the springs were chosen by adding the distance changes measured from smFRET to the equilibrium distances seen in the closed state , thereby , opening the crystal structure of M . tuberculosis MscL ( PDB: 2OAR ) ( Chang et al . , 1998; Perozo et al . , 2001; Steinbacher et al . , 2007 ) . In the simulation , the virtual springs pushed corresponding residues from the distance in the closed state to the equilibrium length in the open state . We note that the uncertainty due to the size of the FRET probes was minimized by focusing on the change of the distances between the closed and open state , rather than absolute distances as discussed in previous section . We note several limitations in the modeling: as the spring constant was kept constant through the simulations , resulting in a large force at beginning of the simulation , we applied both secondary structure restraints ( Trabuco et al . , 2009 ) and symmetry restraints ( Chan et al . , 2011 ) to prevent structural distortion . The secondary structure restraints prevents some subtle changes in the structure , such as kinks observed previously in the upper part of TM1 in the open model of MscL ( Deplazes et al . , 2012 ) . Therefore , we limit our discussion of the open model to pore size and helix tilting . The membrane tension , which causes membrane thinning , plays an important role in the MscL opening process ( Corry et al . , 2010; Louhivuori et al . , 2010; Deplazes et al . , 2012 ) . However , the restraint MD simulation cannot address the question of how the channel is activated . For the simplicity of the modeling , membrane tension is not considered here . We did observe that the membrane near the MscL becomes thinner during the channel opening process to match with the flattening MscL ( Figure 7—figure supplement 2 ) , confirming that a thinning membrane , likely caused by tension , matches the open channel better . The resulting open state structure of MscL is shown in Figure 7B , D , and compared with the crystal structure of MscL in the closed state ( Figure 7A , C ) . The open structure satisfies all the distance constraints measured in our smFRET experiments . In contrast , previous models based on ensemble FRET measurements failed to be consistent with all experimental measurements ( Corry et al . , 2010 ) . In the open conformation , the pore is mainly lined by helices TM1 ( indicated by blue arrows ) , consistent with the helix-tilt model . In addition , it is observed that both TM1 and TM2 indeed tilt toward the membrane plane ( horizontal ) upon channel activation . For example , the orientation of TM1 tilts from the green arrow orientation ( Figure 7C , closed state ) to the yellow arrow orientation ( Figure 7D , open state ) . The change in tiling angle of the TM1 and TM2 helices is ∆θ1≈27° and ∆θ2≈19° , respectively , where θ is the angle between helix and the fivefold symmetry axis . The all-atom model and backbone model of the open state resulting from the current study are provided in PDB format in SI . 10 . 7554/eLife . 01834 . 020Figure 7 . Model of the MscL structure in the open conformation . ( A and C ) The crystal structure of MscL in the closed state is shown for comparison ( PDB: 2OAR [Chang et al . , 1998; Steinbacher et al . , 2007] ) . ( B and D ) The structure of MscL in the open state ( Source Code 1 and 2 ) was developed based on the smFRET measurements , satisfying all the distance constraints measured from smFRET experiments . In the open conformation , the pore is mainly lined by TM1 ( indicated by blue arrows ) , consistent with the helix-tilt model . In addition , both TM1 and TM2 tilt toward the membrane plane ( horizontal ) upon channel activation , which is emphasized by the green and yellow arrows in the side views . The green arrows show the orientation of TM1 in the closed state while the yellow arrow indicated the orientation of TM1 in the open state . The angle between the two arrows is 27° . ( E ) A sphere with a diameter of 2 . 7 nm ( blue ) is shown in the MscL channel in the top view . ( F ) The surfaces of water molecules ( green ) inside the tunnel of MscL ( magenta ) are drawn and the narrowest constriction is ∼2 . 7–2 . 8 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 02010 . 7554/eLife . 01834 . 021Figure 7—figure supplement 1 . Developing a model for the open structure of MscL by inserting virtual springs . For each measured residue ( highlighted in green ) , 10 springs were inserted: five springs between identical residues from adjacent monomers ( red ) and five between residues from non-adjacent monomers ( yellow ) . The equilibrium lengths of the springs are based on the smFRET measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 02110 . 7554/eLife . 01834 . 022Figure 7—figure supplement 2 . Side views of MscL open model ( orange ) in the POPC lipid bilayers . For clarity purpose , the POPC bilayer is described only by phosphorus atoms displayed as glossy gray van der Waals spheres . A thinning membrane was observed near the channel when lipids try to match the flatting MscL in the open state . DOI: http://dx . doi . org/10 . 7554/eLife . 01834 . 022 We used two independent methods to measure the pore size of MscL in the open state . The first method is to measure the movements of the residues forming the narrowest pore constriction of the channel , that is residues around I25 for E . coli MscL ( Chang et al . , 1998; Perozo et al . , 2001 , 2002a; Corry et al . , 2010 ) . However , this method , although straightforward , has its limitations . It is likely that the function of the channel is affected by mutation and labeling of ( some of ) the residues at the pore region . For example , the activation thresholds ( Pa , defined as the pressure at which the first channel opening was observed [Nomura et al . , 2012] ) of mutants G22C and I24C are more than double the wild-type thresholds ( Figure 6 ) and both ensemble and single molecule FRET measurements of these mutants showed no change in the FRET efficiency after adding 25% LPC . The effect of the point mutations near the pore on the electro-physiological properties of the channel can be quantitatively explained by the closed and open structure of MscL as shown in Figure 6—figure supplement 1 , the residue G22 ( A20 in M . tuberculosis MscL ) is very close to the pore and is facing the pore . The residue V22 ( V22 in M . tuberculosis MscL ) is also close to the pore and sandwiched between helix 1 and neighboring helix 1 . Mutating these two residues is likely to perturb the channel function . On the other hand , the residue I25 is further from the pore than G22 and I24 . The mutation I25C is less likely effect the channel properties . Indeed the I25C mutation does not affect the channel’s gating parameters ( Figure 6C , D ) . I25 is still close enough to the pore , making it a perfect candidate for measuring pore size . Furthermore , among the three mutated residues shown in Figures 6 , I25 ( green ) is the only one facing outward from the channel axis and accessible from the periphery of the protein ( Figure 6—figure supplement 1B , D ) . We were able to determine the movement of residue I25C ( Corry et al . , 2010 ) ; and measured that the residue I25 moves away from the pore center by Δr = 1 . 2 nm , indicating that the pore opens up by ΔD = 2 . 4 nm in diameter . Taking into account that the pore diameter in the closed state ( Φclose ) is 0 . 4 nm ( Chang et al . , 1998 ) , we conclude that the pore size in the open state ( Φopen ) is Φopen = Φclose + ΔD = 2 . 8 nm , which agrees with previously reported values ( Perozo et al . , 2002a; Corry et al . , 2010 ) . The second method is based on the open state model of MscL constructed by means of molecular dynamics . The surfaces of water molecules inside the channel were rendered ( Figure 7E , F ) using VMD ( Humphrey et al . , 1996 ) and the narrowest constriction seen provided an estimate of the pore size . This estimate accounts for all residues of the transmembrane domain and therefore is expected to be more accurate than the estimate of the first method . Using this method we estimate that the pore size of the MscL channel in the fully open state is 2 . 7–2 . 8 nm , which is consistent with the value from the first method , 2 . 8 nm . We used a combination of experimental smFRET and computational modeling to study the conformational change of MscL upon channel activation . It is the first time that single molecule FRET has been applied to liposome-reconstituted membrane proteins with more than three monomers . We measured the distance changes of multiple residues from the MscL transmembrane α-helices ( TM1 and TM2 ) during gating of the channel . For the first time , it is observed directly that both transmembrane helices swing away from the pore center , with rotational tilting involved . The results argue clearly in favor of the helix-tilt model . In addition , we developed by means of computational modeling a model of the channel structure in the open state based on the smFRET results and the crystal structure of the protein in the closed state as a reference . This model also confirms the helix-tilt model and yields a pore diameter of 2 . 8 nm . The smFRET experiments carried out in the present study observe MscL channels dynamics in lipid bilayers ( liposomes ) and not in detergents , which is a great advantage over crystallography that can result in different oligomeric states like those seen in the tetrameric structure of S . aureus MscL ( Liu et al . , 2009 ) . It is possible that the detergent used in purification caused some portion ( ≤30% ) of the MscL as hexamers , instead of the assumed pentamers . Nevertheless , our conclusion of the helix-tilt opening model is independent of the percentage of hexameric structure . However , the exact value of the open pore diameter would be slightly greater , 3 . 0 nm ( 30% hexamers ) , up to 3 . 3 nm ( 100% hexamers ) , still agreeing with previously reported values ( Sukharev et al . , 2001b; Perozo et al . , 2002a; Corry et al . , 2010 ) . The current study focused on the closed and fully open state of MscL . The fully open state was achieved by adding LPC to the liposomes ( Perozo et al . , 2002a , 2002b ) . However , the technique introduced is not limited to these two states only . Single molecule FRET together with other techniques–for example , with patch-clamping done simultaneously–can answer many more questions than a crystal structure . For instance , it could probe the conformation of the channel during sub-conducting levels that involve partial MscL openings , or probe sequence of movements of the individual channel domains during opening of the channel . The E . coli MscL gene ( EcoMscL ) was cloned into plasmid pQE-32 ( Qiagen , Hilden , Germany ) as the BamHI-SalI fragment , which also added a hexa-histidine tag ( his-tag ) to the protein at the N-terminus . The protein was expressed in E . coli ( M15 strain ) ( Qiagen ) that were lysed by sonication and purified from DDM solubilized membranes using TALON Metal affinity chromatography ( Clontech Laboratories , Inc , Mountain View , CA ) , followed by a further purification step using fast protein liquid chromatography ( FPLC; Superdex 200 10/300 GL column , GE Healthcare , Pittsburgh , PA ) . Purification was performed in the presence of 1 mM DDM . The wild type of MscL protein does not contain any cysteine . To label the proteins with fluorescent probes , MscL was mutated using site-directed mutagenesis such that a residue at the desired position was replaced by a cysteine . Because the MscL protein is a homo-pentamer ( Chang et al . , 1998 ) , this mutation introduced five identical cysteine sites . The protein with his-tag was then labeled with Alexa Fluor 488 ( AF488 ) and/or Alexa Fluor 568 ( AF568 ) maleimide , which specifically reacted with the introduced cysteines ( Kim et al . , 2008 ) . Right before labeling , proteins were reduced with 10 mM DTT for 30 min , followed by purification using PD-10 desalting columns ( GE Healthcare ) . We titrated the pentameric protein-to-fluorophore molar ratio from 1:1 to 1:5 and used the molar ratio of 1:5 for labeling in all the experiments . Under our labeling conditions , this ratio gave satisfying results such that most of the proteins are labeled ( averagely ∼1 . 7 donors and ∼1 . 3 acceptors per pentamer ) and that many of proteins are attached by a single donor and a single acceptor ( ∼30% of good traces show multiple donors and/or acceptors ) . Excess fluorophores were then removed using PD-10 desalting columns . The sample was reduced with 10 mM DTT before this purification step . A note to make is that the fluorophores ( Alexa Fluor 488 maleimide and Alexa Fluor 568 maleimide ) come as mixtures of 5′ and 6′ isomers , which would potentially complicate interpretation of smFRET data . However , we expect that the results would not be affected because the exactly same fluorophores have been successfully used in many single molecule FRET studies ( Marras et al . , 2002; Yin et al . , 2005; Jäger et al . , 2006; Granier et al . , 2007; Majumdar et al . , 2007 ) . MscL channels were reconstituted into artificial liposomes ( ∼50 nm diameter ) , following the protocol described in Perozo et al . ( 2002a , 2002b ) . Liposomes were prepared by drying , rehydrating and extruding lipids through filters with ∼50 nm pores . The lipids used in all the measurements were a mixture of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC , Avanti Polar Lipids , Inc , Alabaster , AL ) and 1 , 2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-biotinyl ( BPE , Avanti Polar Lipids , Inc . ) dissolved in chloroform at a molar ratio of POPC:BPE = 1000:20 . BPE was used for immobilization . To incorporate MscL channels into the liposomes , a mixture of unlabeled and labeled MscL proteins ( 5% labeled ) was then reconstituted into the liposomes , at a final volume of 1 ml , with a protein/lipid ( molar ) ratio of 1:200 , resulting in a molar ratio of 1:4000 for the labeled proteins to lipids . The liposomes were immobilized onto a glass coverslip . This immobilization was achieved by biotin-avidin linkages between biotinylated-PEG molecules on the surface to a neutravidin molecule , and then biotinylated lipids ( BPE ) in the liposomes ( Roy et al . , 2008 ) . To open the MscL channels in the liposomes , a conical lipid , 1-oleoyl-2-hydroxy-sn-glycero-3-phosphocholine or lysophosphatidylcholine ( LPC , Avanti Polar Lipids , Inc ) , was added to the liposomes , at a molar fraction of 25% . As LPC incorporates itself into the outer leaflet of a lipid bilayer , it introduces membrane tension , changes the lipid pressure profile , and triggers the MscL to open ( Perozo et al . , 2002a , 2002b ) . MscL protein purification and reconstitution into soybean azolectin liposomes were described previously ( Nomura et al . , 2012 ) . All results were obtained with proteoliposomes at the protein: lipid ratio of 1:200 ( wt/wt ) . Channel activities of the wild-type and mutant MscL were examined in inside-out liposome patches using patch-clamp technique . Borosilicate glass pipettes ( Drammond Scientific Co , Broomall , PA ) were pulled using a Narishige micropipette puller ( PP-83; Narishige , Tokyo , Japan ) . Pipettes with resistance of 2 . 5–4 . 9 MΩ were used for the patch-clamp experiments . Pipette and bath solution contained 200 mM KCl , 40 mM MgCl2 , and 5 mM HEPES ( pH 7 . 2 adjusted with KOH ) . The current was amplified with an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA ) , filtered at 2 kHz and data acquired at 5 kHz with a Digidata 1440A interface using pCLAMP 10 acquisition software ( Molecular Devices , Sunnyvale , CA ) and stored for analysis . Negative pressure ( suction ) was applied to the patch pipettes using a syringe and was monitored with a pressure gauge ( PM 015R , World Precision Instruments , Sarasota , FL ) . Since the MscL channel is a homo-pentamer ( Chang et al . , 1998 ) ( or possibly homo-hexamer [Gandhi et al . , 2011] ) , there is always a distribution of various donor/acceptor combinations . To exclude signal from those channels having multiple donors or multiple acceptors , the fluorescence intensity of single channels ( and hence the step-wise photobleaching ) was monitored . Because multiple donors or acceptors have multiple ‘staircase’ photobleaching , these channels were simply not used . Only the traces with a clear single-step photobleaching in both donor and acceptor channels were included in the analysis . Subtraction of the intensities ( averaged ) before and after photobleaching gives the intensities of donor ( ID ) and acceptor ( IA ) , which are then used for FRET efficiency calculation as described below . Single molecule FRET experiments were performed using total internal reflection fluorescence microscopy ( TIRFM ) with a 1 . 45 NA 100X oil immersion objective ( Selvin and Taekjip , 2007; Roy et al . , 2008 ) . The fluorescence intensities were used to calculate the energy transfer efficiency by the corrected FRET equation: E = ( IA − ℓID ) / ( IA + γID ) : where E is the FRET efficiency , ℓ represents leakage of donor signals in the acceptor channel , γ is the correction factor which accounts for the differences in quantum yield and detection efficiency between the donor and the acceptor , IA and ID represent the acceptor and donor intensities , respectively ( Roy et al . , 2008 ) . Note that the direct excitation of the acceptor by the donor excitation has been corrected automatically when getting the acceptor intensity from the fluorescence traces . The distance between the donor and acceptor is given by R = R0 ( E−1−1 ) 1/6 , where R0 is the Förster radius ( Förster , 1948 ) . The Förster radius , R0 , given by R0= ( 0 . 529 κ2 QD J ( λ ) NA n4 ) 1/6∝ ( κ2QD ) 1/6 , and its error were measured experimentally by measuring the absorbance and fluorescence spectra , quantum yield of the donor , AF488 , ( QD = QAF488 ) and anisotropy ( Aa and Ad which give the maximum possible error in κ2 ) of the fluorescent probes conjugated to proteins . The quantum yield of AF488 conjugated to MscL was measured using fluorescein in 0 . 1 M NaOH as a standard ( Fery-Forgues and Lavabre , 1999; Lakowicz , 1999 ) using the equation QX=ASAX×FXFS× ( nXnS ) 2×QS , where Q is the quantum yield , A is the absorbance at the excitation wavelength ( 470 nm ) ; F is the area under the corrected emission curve , and n is the refractive index of the solvent . Subscripts S and X refer to the standard ( fluorescein ) and to the unknown ( AF488 ) , respectively . The spectra of absorbance and fluorescence of AF488-MscL in PBS + DDM ( 1 mM DDM ) were measured using Agilent 8453 UV-Vis absorbance spectrophotometer ( Agilent Technologies , Santa Clara , CA ) and PC1 spectrofluorimeter ( ISS , Inc . , Champaign , IL ) , respectively . In order to determine the maximum error in the orientation factor , κ2 , and therefore the error in R0 , the anisotropy of the fluorophores conjugated to MscL was measured . The fluorophores-protein conjugates were immobilized on a glass coverslip which was covered with PEG ( 5% biotinylated ) , then a layer of neutravidin ( Thermo Scientific , Waltham , MA ) , followed by a layer of penta-his biotin conjugate ( Qiagen ) . The emission of the fluorophores-protein conjugates were split into two channels of polarization and used to calculate the anisotropy , A=I∥−I⊥I∥+2 I⊥ , where I∥ is the fluorescence emission with polarization parallel to the excitation polarization and I⊥ is the fluorescence emission with polarization perpendicular to the excitation polarization ( Lakowicz , 1999 ) . Anisotropies were corrected for the intrinsic polarization properties of the microscope by calibrating to known freely diffusing fluorophores . Anisotropies were also corrected for the high numerical aperture of the objective . Then the maximum range of κ2 was given by κ2max = 2/3 ( 1 + 2 . 5Ad+2 . 5Aa ) and κ2min = 2/3 ( 1−1 . 25Ad−1 . 25Aa ) where Ad and Aa are the anisotropy of AF488 ( donor ) and AF568 ( acceptor ) , respectively ( Dale et al . , 1979; Cha et al . , 1999 ) . To evaluate directly the sizes of the fluorescent probes used in our FRET experiments , the molecular structures of the AF488-C5-Maleimide and AF568-C5-Maleimide were constructed using Avogadro ( Hanwell et al . , 2012 ) . Both the 5′- and 6′-isomers were constructed . These structures were then optimized in Avogadro with molecular dynamics using the universal force field ( UFF ) ( Rappe et al . , 1992 ) . From the optimized molecular structures ( shown in Figure 4—figure supplement 1 ) , we estimated the probe sizes which were defined as the distance between the oxygen atom of the fluorophore ( indicated by the magenta arrows in Figure 4—figure supplement 1B , F ) and the nitrogen atom of the maleimide group ( indicated by the cyan arrows in Figure 4—figure supplement 1A–E ) . We found that the donor is 17 . 1 Å ( 5′-isomer ) or 16 . 3 Å ( 6′-isomer ) while the acceptor is 17 . 4 Å ( 5′-isomer ) or 17 . 4 Å ( 6′-isomer ) . The difference in the molecular size between donor-isomers or between acceptor-isomers is small , <5% . Due to lack of an E . coli MscL ( EcoMscL ) crystal structure , the simulation were performed using the structure of MscL from M . tuberculosis ( MtMscL , PDB: 2OAR ) ( Chang et al . , 1998; Steinbacher et al . , 2007 ) . The CP domain was truncated in the simulation because the complete deletion of the CP does not change the gating parameters substantially ( Anishkin et al . , 2003 ) . The residues to which the distance constraints were applied , were shifted according to the sequence alignment in Chang et al . ( 1998 ) . A spring constant of 0 . 2 kcal mol−1Å−2 was used for the virtual spring in the distance constrained simulation . Both secondary structure restraints ( Trabuco et al . , 2009 ) and symmetry restraints ( Chan et al . , 2011 ) were applied to prevent structural distortion under large force in the distance constrained simulation . Total simulation time is 5 ns . A model of MscL in the open state was obtained at the end of the distance constrained simulation , when the simulation satisfied all the distance constraints measured by means of smFRET experiment . The restraint MD simulation procedure is similar to the one used previously ( Corry et al . , 2010; Deplazes et al . , 2012 ) . The simulation system was prepared by first imbedding the crystal structure of MscL ( PDB: 2OAR ) ( Chang et al . , 1998; Steinbacher et al . , 2007 ) into a membrane patch with 1727 POPC lipids . Solvent was then added to both sides of the membrane , and the system was neutralized with 200 mM NaCl using VMD ( Humphrey et al . , 1996 ) . The final simulation system contained 1 , 137 , 413 atoms . The all-atom MD simulations were performed using NAMD 2 . 9 ( Phillips et al . , 2005 ) with the TIP3P model ( Jorgensen et al . , 1983 ) for explicit water and the CHARMM36 force field ( Best et al . , 2012 ) . The simulation was conducted in the NPT ensemble ( constant pressure and temperature ) with periodic boundary condition . Constant temperature of 300 K was maintained using a Langevin thermostat with a damping coefficient of 1 ps−1 . A Nosé–Hoover Langevin piston barostat was used to maintain a constant pressure of 1 atm with a period of 200 . 0 fs and damping timescale of 100 . 0 fs . The multiple time-stepping algorithm was employed , with an integration time step of 2 fs , the short-range force being evaluated every time step , and the long-range electrostatics every second time step . Non-bonded energies were calculated using particle mesh Ewald full electrostatics and a smooth ( 10–12 Å ) cutoff of the van der Waals energy .	Bacterial cells are full of fluid , and they will burst if they are not able to respond to a build up of pressure . Fortunately , the membrane of a bacterial cell contains channels that can detect the increased mechanical stress on the cell membrane and then open to relieve the pressure . In many bacterial cells , the last defence against the cell exploding is called the mechanosensitive channel of large conductance ( MscL ) . This is made of five proteins , each of which consists of TM1 and TM2 helixes , which are responsible for opening and closing the channel . Two models have been proposed to explain how the channels are opened . In the barrel-stave model , the TM1 helix moves , while the TM2 helix remains stationary . This results in an open pore that is lined with TM1 and TM2 helixes in the same way that wooden staves line a barrel . In the helix-tilt model , both helixes tilt towards the membrane to open the channel . Wang et al . have now used a technique called single-molecule fluorescence resonance energy transfer ( FRET ) to explore the structure of the open channel in E . coli in order to determine which model is correct . In this technique an individual channel is labeled with two different fluorescent molecules . By illuminating the channel with light of a wavelength that excites the first fluorescent molecule , and measuring the strength of the fluorescence from the second molecule , it is possible to work out the distance between the two molecules . From this , the structure of the channel and how it opens and closes can be explored . Previous attempts to measure the diameters of open channels using fluorescence techniques have suffered from issues caused by the use of large numbers of fluorescent molecules . This has made it necessary to use computational modeling to extract the required data . By looking at a series of individual proteins , Wang et al . overcame these problems and found that the diameter of the fully open pore is 2 . 8 nm . The result provides strong support for the helix-tilt model .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2014	Single molecule FRET reveals pore size and opening mechanism of a mechano-sensitive ion channel
Spontaneous fluctuations in neuronal activity emerge at many spatial and temporal scales in cortex . Population measures found these fluctuations to organize as scale-invariant neuronal avalanches , suggesting cortical dynamics to be critical . Macroscopic dynamics , though , depend on physiological states and are ambiguous as to their cellular composition , spatiotemporal origin , and contributions from synaptic input or action potential ( AP ) output . Here , we study spontaneous firing in pyramidal neurons ( PNs ) from rat superficial cortical layers in vivo and in vitro using 2-photon imaging . As the animal transitions from the anesthetized to awake state , spontaneous single neuron firing increases in irregularity and assembles into scale-invariant avalanches at the group level . In vitro spike avalanches emerged naturally yet required balanced excitation and inhibition . This demonstrates that neuronal avalanches are linked to the global physiological state of wakefulness and that cortical resting activity organizes as avalanches from firing of local PN groups to global population activity . When the brain is not engaged in any particular sensory , cognitive , or motor task , cortical neurons nevertheless give rise to coordinated group activity . This so-called resting activity delineates functional networks ( Fox and Raichle , 2007; Haimovici et al . , 2013 ) , modulates responses to sensory input ( Arieli et al . , 1996; Womelsdorf et al . , 2012 ) , predicts cortical responses ( Luczak et al . , 2009 ) , and changes in disease or sleep deprived states ( Greicius et al . , 2004; Meisel et al . , 2013 ) . It is , therefore , crucial to identify and understand the dynamical constraints that determine coordinated neuronal group activity at rest in the awake ( AW ) state . Resting activity , recorded from neuronal populations in vivo using the local field potential ( LFP; Gireesh and Plenz , 2008; Petermann et al . , 2009; Yu et al . , 2011; Priesemann et al . , 2013 ) , functional magnetic resonance imaging ( fMRI; Fraiman and Chialvo , 2012; Haimovici et al . , 2013 ) , the magnetoencephalogram ( MEG; Palva et al . , 2013; Shriki et al . , 2013 ) , or electrocorticogram ( ECoG; Solovey et al . , 2012 ) , has been shown in rodents , non-human primates , and humans to be composed of activity cascades called neuronal avalanches . Neuronal avalanches identify a specific organization of activity patterns in which avalanche sizes distribute according to a power law with slope of −1 . 5 , that is , the relative occurrences of avalanche sizes are constant ( Beggs and Plenz , 2003 ) . This scale-invariant organization , which transcends many spatial and temporal scales , is an indication that cortical dynamics reside at or close to a critical state in which interactions between local elements give rise to long-range spatial and long-term temporal fluctuations ( Plenz and Thiagarajan , 2007; Chialvo , 2010; Beggs and Timme , 2012; Plenz , 2012; Marković and Gros , 2014; Plenz and Niebur , 2014 ) . Predictions from the theory of criticality ( Bertschinger and Natschlager , 2004; Nykter et al . , 2008; Chialvo , 2010; Plenz and Niebur , 2014 ) and neural modeling ( Beggs and Plenz , 2003; Kinouchi and Copelli , 2006; Rämö et al . , 2007; Shew et al . , 2009; Tanaka et al . , 2009; de Arcangelis and Herrmann , 2010 ) suggest that cortical networks that reside in such a fluctuation-dominated regime can improve various aspects of information processing , as was demonstrated experimentally in vitro ( Shew et al . , 2009 , 2011; Yang et al . , 2012; Shew and Plenz , 2012 ) . Yet , further exploration of the origin and potential functional advantages of avalanche dynamics are limited by the ambiguity in the composition of the LFP , ECoG , MEG , and BOLD fMRI signals , with respect to their spatiotemporal and cellular origin as well as synaptic input and action potential ( AP ) output . It is now well established that neuronal avalanches can emerge within cortex as demonstrated in vitro ( Beggs and Plenz , 2003; Stewart and Plenz , 2006; Pasquale et al . , 2008 ) , and sensibly depend on the balance of fast synaptic excitation/inhibition ( E/I ) ( Beggs and Plenz , 2003; Pasquale et al . , 2008; Shew et al . , 2009 ) and neuromodulators ( Stewart and Plenz , 2006; Pasquale et al . , 2008 ) . However , it is currently not known how the scale-invariant , macroscopic organization of avalanches measured at the population level relates to the output of the principal cells of cortex , that is , AP firing in pyramidal neurons ( PNs ) and how this organization relates to the global physiological state of the animal . AP firing in PNs is commonly reported as spontaneous and irregular ( Softky and Koch , 1993; Shadlen and Newsome , 1998 ) with low average correlation in firing between neurons during spontaneous activity and a high variability in evoked AP responses ( Gawne and Richmond , 1993; Kerr et al . , 2007; Sato et al . , 2007; Poulet and Petersen , 2008; Ecker et al . , 2010; Komiyama et al . , 2010; Renart et al . , 2010 ) . Here , we show experimentally in vivo that ongoing fluctuations in AP firing in single cortical neurons amount to scale-invariant AP avalanches at the neuronal group level . The emergence of spike avalanches marks the AW state and is absent under anesthesia . Similarly , spike avalanches spontaneously organize in layer 2/3 PN groups from organotypic slice cultures and yet are sensitive to the E/I balance . We propose that critical dynamics govern the organization of resting activity in the awake animal from AP firing in individual PNs to the activity in large neuronal groups across cortex . To identify the relationship between AP firing and neuronal avalanches , which are primarily found in superficial layers of cortex ( Stewart and Plenz , 2006; Petermann et al . , 2009 ) , we expressed the genetically encoded calcium indicator ( GECI ) YC2 . 60 ( Yamada et al . , 2011 ) in layer 2/3 ( L2/3 ) PNs of rats using in utero electroporation at embryonic day 15 . 5 ± 0 . 5 ( Saito , 2006 ) . Labeled mature neurons distributed throughout dorsolateral frontal and sensorimotor cortex . They exhibited PN morphology ( Figure 1A ) , and their synaptic transmission was blocked by glutamate receptor antagonists ( data not shown ) . To record ongoing AP activity in local PN groups , we performed 2-photon imaging ( 2-PI ) of YC2 . 60-expressing PNs in head-restrained rats ( Figure 1A , B; depth = 270 ± 50 µm; cortical area = 0 . 15 ± 0 . 05 mm2; 10–15 min per recording ) . Recordings were done under anesthesia ( AN; 1–2% isoflurane ) , during wakening ( WK; 5–20 min at 0% isoflurane ) , and in the awake state ( AW; after >20 min at 0% isoflurane ) . Intracellular calcium transients produced fluorescence changes in visually identified somatic ROIs ( Figure 1B ) , which were converted into ratiometric time courses ( ΔR/R ) and then deconvolved to obtain an instantaneous firing rate estimate , λ , for each neuron ( Vogelstein et al . , 2010 ) ( see ‘Materials and methods’; Figure 1C , D ) . In control experiments , we showed ( 1 ) YC2 . 60 reliably and linearly reported AP activity at physiological temperature from single APs to AP bursts up to 28 Hz ( Figure 1—figure supplement 1A , B ) and ( 2 ) λ linearly recovered spike trains at different temporal resolutions ( Figure 1—figure supplement 2 ) . We first recorded at a temporal resolution of Δt = 250 ms ( n = 6 rats; 38 recordings; 15–30 active PNs/recording; >1 AP/min ) . Neuronal activity was stationary in λ and in the average crosscorrelation , R , between ROIs ( Figure 1E , F , respectively ) . Neurons fired on average more during AW compared to WK and AN ( ANOVA , F ( 2 , 35 ) = 23 . 05 , p < 0 . 001; probability density function ( PDF ) shown in Figure 1G ) . Under all three conditions , though , neurons fired irregular APs interspaced by relatively long periods of quiescence . This was quantified by three measures . First , λ distributed exponentially for single neurons [log-likelihood ratio ( LLR ) comparison between power law vs exponential , >98% of ROIs in favor of exponential distribution , p < 0 . 05; Figure 2A , single distributions and average for one PN group; Figure 2B , averages over all PN groups] . Second , neurons tended to not fire at all within Δt ( Figure 2A , B , left; arrow ) . The corresponding probability of quiescence , Pq ( λ < minimal λ threshold , λthr , set to 0 . 5 ) , was highest for AN and WK ( Figure 2B , inset; ANOVA , F ( 2 , 35 ) = 23 . 05 , p = 0 . 002 ) . Both of these characteristics remained true for higher temporal resolutions despite the expected increase in λ fluctuations ( Figure 2B , right; additional n = 6 rats; n = 19 recordings; Δt = 167 and 88 ms during AW; LLR in favor of exponential , p < 0 . 05 ) and Pq ( ANOVA , F ( 2 , 28 ) = 31 . 46 , p < 0 . 001 ) . Third , the normalized duration of quiescent times , IBInorm , between firing ( i . e . , λ < λthr = 0 . 5 ) also distributed exponentially for all conditions ( Figure 2A , right; Figure 2C; LLR: >98% of ROIs with p < 0 . 05 ) . The corresponding CV was larger than 1 for all conditions , was significantly higher for AW than WK and AN ( Figure 2C , left; inset , AW: 1 . 5 ± 0 . 2; WK: 1 . 2 ± 0 . 1; AN: 1 . 2 ± 0 . 1; mean ± SD , F ( 2 , 35 ) = 25 . 66 , p < 0 . 001 ) , and increased further with temporal resolution ( AW , Figure 2C , right; Δt = 167 ms , 1 . 9 ± 0 . 4; Δt = 88 ms , 2 . 1 ± 0 . 3 , mean ± SD; F ( 2 , 28 ) = 15 . 48 , p < 0 . 001 ) . This irregularity was also robust to minimal AP activity: increasing λthr smoothly reduced the average firing rate λavg ( data not shown ) , yet maintained a CV larger than 1 for all conditions and Δt ( Figure 2—figure supplement 1 ) . CV values for single units ( n = 26; average firing rate = 1 . 3 Hz; range: 0 . 1–6 . 2 Hz ) recorded with chronic microelectrode arrays from superficial layers in the AW rat ( Figure 1—figure supplement 2 ) compared favorably with λ results from our imaging analysis and ranged between 1 . 6 ± 0 . 4 ( Δt = 0 . 033 ms ) and 1 . 3 ± 0 . 3 ( resampled at Δt = 250 ms ) , respectively . 10 . 7554/eLife . 07224 . 003Figure 1 . Imaging of ongoing spiking activity in groups of L2/3 PNs in the awake ( AW ) rat . ( A ) Z-stack side projection of YC2 . 60-expressing PNs in L2/3 in vivo . ( B ) Single imaging plane ( dotted line in A ) containing a group of PNs with significant changes in fluorescent intensity ∆R/R over time ( colored ROIs ) . ( C ) Time course of ΔR/R for individual ROIs ( from B ) . ( D ) Top: Binary raster display of instantaneous spike rate estimate λ ( λthr = 0 . 1 ) . Middle: Expanded period showing color coded λ amplitude . Bottom: Overplot of λ time course for individual , color-coded ROIs . ( E , F ) Stationary firing rate estimate λ and pairwise crosscorrelation Rnorm ( normalized by the correlation during the first 30 s ) over the course of acquisition . Firing rate increased from anesthetized to AW conditions but remained stable throughout the recording , suggesting that our measures were not affected by slow modulations of activity ( Ecker et al . , 2010 ) . Shown are averages over all PN groups . ( G ) Distributions of the average firing rate estimate , λavg , for the three different states . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 00310 . 7554/eLife . 07224 . 004Figure 1—figure supplement 1 . Single AP detection in YC2 . 60-expressing neurons at physiological temperature and performance of the OOPSI deconvolution algorithm . ( A ) Using whole-cell patch recording of YC2 . 60-expressing PNs in cortical slice cultures , we confirmed that YC2 . 60 reliably resolved spontaneous single AP firing at physiological temperature ( ∼32°C ) , in line with previous reports ( Yamada et al . , 2011 ) . Gray: Individual trials: Black: Average . Inset: Zoomed view of bar graph from 1 AP subpanel . Note the decay in ΔR/R by ∼2/3 within 1–2 s . Responses from single PN . ( B ) Peak and integral of instantaneous rate λ as well as peak ΔR/R linearly increase with the number of spontaneous APs/250 ms ( n = 7 neurons ) . ( C ) Single movie data showing that the minimum reconstruction error of the deconvolution was found at decay time τ = 1 . 5 s ( n = 39 ROIs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 00410 . 7554/eLife . 07224 . 005Figure 1—figure supplement 2 . Performance of the OOPSI deconvolution algorithm at different temporal resolutions and noise levels . ( A ) Surrogate intracellular calcium traces ( ΔR/R; black ) and corresponding λ estimates ( red ) at three temporal resolutions Δt . ΔR/R was derived by convolution of an in vivo spike train ( bottom ) with the instantaneous calcium response ( τ = 1 . 5 s; 5% peak amplitude for one AP; for details see ‘Materials and methods’ ) . ( B ) Peak λ linearly increases with the number of APs/Δt . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 00510 . 7554/eLife . 07224 . 006Figure 2 . Spatial and temporal clustering in ongoing spiking activity in vivo . ( A ) Probability distributions of λnorm = λ/λavg ( left ) and distribution of normalized quiescent time intervals , IBInorm = IBI/IBIavg , ( right ) in the AW state for a single neuronal group of PNs ( Δt = 250 ms ) . Gray: distributions for individual ROIs , black: average . Dotted lines , λnorm = 1 and IBInorm = 1 . The arrow is pointing at the relatively high-probability function value for λnorm << 1 , that is , no spike within Δt = 250 ms . Note that the transition from ‘no spike’ to spiking is rather abrupt in the distribution , which indicates the high signal-to-noise ratio in our recorded data ( cf . Figure 1—figure supplement 1A , B ) . ( B ) Average λnorm distributions over all PN groups for all three conditions ( left; AW; WK , waking; AN , anesthetized ) and temporal resolutions ( right ) . Inset: probability of quiescence Pq ( number of recordings is indicated in parentheses ) . ( C ) Distribution of IBInorm for all three conditions ( left ) and temporal resolutions ( right ) . Inset: Coefficient of variation ( CV ) for IBI ( p < 0 . 05 ) . ( D ) Average autocorrelation function for λ across PNs for all three conditions ( left; Δt = 250 ms ) and temporal resolutions ( right ) . Inset: Power-law exponent β ( p < 0 . 05 ) . Note steeper power-law decay for AW indicating increased temporal clustering . ( E ) Distribution of pairwise crosscorrelation , R , in λ for all PN groups and different states . Broken lines: Corresponding independent model by shuffling λ for each ROI . ( F ) AW shows steeper distance decay in R when compared to AN indicating higher spatial clustering . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 00610 . 7554/eLife . 07224 . 007Figure 2—figure supplement 1 . CV in λ remains larger than one for increasing λthr for all conditions ( A ) and temporal resolutions ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 007 To further differentiate the high irregularity encountered in the AW resting state from WK and AN conditions , we studied the temporal and spatial clustering of PN firing . Previous work on the AW resting state revealed temporal AP clustering during large intracellular membrane potential fluctuations ( Poulet and Petersen , 2008 ) and spatial clustering of AP firing for L2/3 PNs ( Greenberg et al . , 2008 ) similar to the correlation profile found during tasks ( Komiyama et al . , 2010 ) . Indeed , we found that λ was more correlated in time during AW compared to WK and AN , as demonstrated by a significantly steeper decay in the autocorrelation for periods <10 s ( Figure 2D , left; ANOVA , F ( 2 , 35 ) = 14 . 77 , p < 0 . 001 ) . On the other hand , correlated firing between pairs of neurons was weak , in line with the notion of an ‘asynchronous state of cortex’ ( Poulet and Petersen , 2008; Ecker et al . , 2010; Renart et al . , 2010 ) , and the average did not differ between states ( AW: 0 . 06 ± 0 . 06; WK: 0 . 05 ± 0 . 04; AN: 0 . 06 ± 0 . 04 , mean ± SD; ANOVA , F ( 2 , 35 ) = 0 . 29 , p = 0 . 75; Figure 2E ) . While neighboring neurons tended to be correlated more than distant neurons , in line with previous reports ( Sato et al . , 2007; Greenberg et al . , 2008 ) , this spatial profile was largely similar across all three states ( Figure 2F ) . In the preceding section , we quantified how irregular spontaneous firing in individual PNs and their pairwise correlations change as the animal transitions from the AN to the AW state . None of these measures , though , allows us to identify neuronal avalanches , which reflect a scale-invariant relationship of neuronal group activities . In fact , we recently demonstrated that event rate and pairwise correlation R are insufficient to predict neuronal avalanches in cortical activity ( Yu et al . , 2011 ) . In a first approach , we therefore identified spatiotemporal activity clusters in the neuronal population . This was done by concatenating firing events of neurons that co-occurred either within Δt or within consecutive periods of Δt ( Figure 3A; gray areas ) and separating clusters by quiescent periods of at least Δt , the original approach to identify avalanche dynamics ( Beggs and Plenz , 2003 ) . For a given neuronal population and 2-PI , this process has two free parameters: ( 1 ) the temporal resolution Δt , which is fixed by the scanning frame rate of 2-PI and ( 2 ) the activity threshold , λthr , of a firing event . In general , if λthr is low , most firing events will be concatenated into few large clusters . Similarly , if λthr is high , the few remaining firing events will group into few clusters . Thus , a maximal number of clusters is expected at an intermediate threshold λthrmax . We first studied this relationship in the AW state . Indeed , for a given recording at Δt , the cluster rate increased with λthr and was maximal at an intermediate threshold λthrmax ( Figure 3B , arrows ) . As expected , λthrmax shifted towards smaller λthr values at higher temporal resolutions due to the improved resolution of fast λ fluctuations . Next , we studied the cluster size s , that is , the sum of all firing events within a cluster normalized by the predicted cluster size limit Λ , which is determined by the number of ROIs and their respective average firing rate ( see ‘Materials and methods’ ) . If the activity of neurons was rather independent from each other , as one might assume from the low average pairwise correlation in λ ( Figure 1 ) , the distribution in cluster size should be close to an exponential function . On the other hand , if interactions between neurons contribute significantly to spontaneous firing , then the cluster size distribution deviates from an exponential function , and , in the case of avalanche dynamics , should follow a power law ( Plenz and Thiagarajan , 2007 ) . Importantly , we found that cluster sizes distributed according to a power law over approximately two orders of magnitude ( Figure 3C: α = 1 . 63 ± 0 . 13 , LLR = 25 . 7–201 . 4 favors power law over exponential , p < 0 . 003 for all n = 10 experiments , Δt = 88 ms at individual λthrmax ) . To determine whether the AP activities in PN groups that resulted in power-law distributed cluster sizes were indeed a result of spatiotemporal correlations , we performed , as a control , time-shuffling of the corresponding λ events . As shown in Figure 3D , time-shuffled λ events did not yield power-law size distributions , and instead , cluster size distributions were better fit by an exponential ( LLR = −66 . 6 to −6 . 9 , favors exponential over power law , p < 0 . 05 for 7/10 experiments , Δt = 88 ms; thresholded at λthrmax obtained for each distribution individually ) . 10 . 7554/eLife . 07224 . 008Figure 3 . Ongoing spiking in local PNs organizes as neuronal avalanches in vivo . ( A ) Sketch of cluster formation at given Δt and chosen λthr = 1 . Gray boxes delineate clusters of activity ( i . e . , consecutive time bins with λ > λthr ) . ( B ) Maximal cluster rate at intermediate λthr for different Δt in the AW condition . Vertical arrows indicate the respective λthr=λthrmax at which cluster rate is maximal . ( C ) Individual distributions of normalized cluster sizes , s , in AW ( top; Δt = 88 ms , n = 10 recordings , threshold at λthrmax ) . Dotted line , predicted cut-off at s = 1; dashed line , power law with α = −1 . 5 . ( D ) Corresponding distributions after shuffling λ . Shuffling destroys spatiotemporal correlations in activity and abolishes the power law in cluster sizes . ( E ) Relationship between α and branching ratio σ for all three temporal resolutions , Δt . Note the systematic change for increasing Δt as shown previously for avalanche dynamics based on the LFP . ( F ) Distribution of cluster lifetimes , T , for different Δt . Dashed line , slope = −2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 008 For LFP-based avalanche dynamics , it has consistently been shown that power-law exponent and branching ratio increase systematically with temporal resolution Δt ( Beggs and Plenz , 2003; Petermann et al . , 2009 ) . Importantly , for critical avalanche dynamics and negligible finite-size effects , the temporal resolution for which the power-law exponent , α , is −1 . 5 yields a branching ratio , σ , close to 1 . As shown in Figure 3E , a similar relationship between α and σ was also found empirically for AP avalanches in vivo . Furthermore , the temporal organization of neuronal avalanches , that is , the avalanche life time , was shown to distribute according to a power law with exponent close to −2 ( experimentally: [Beggs and Plenz , 2003]; simulations and theory: [Harris , 1989; Eurich et al . , 2002] ) . Similarly , we found that the cluster lifetime , T , distributed according to a power law with slope close to −2 ( Figure 3F; Δt = 250 ms , slope γ = 1 . 7 ± 0 . 1; Δt = 167 ms , γ = 1 . 9 ± 0 . 2; Δt = 88 ms , γ = 2 . 2 ± 0 . 2 , LLR = 100 . 4–344 . 1; p < 0 . 001 ) . To study the robustness of the power-law size distributions with respect to threshold , we systematically varied λthr around λthrmax . In Figure 4 , we show that the body of the distributions followed a power law up to the predicted cluster size limit ( s = λ/Λ = 1 for the normalized distributions ) beyond which a cut-off was observed . This cut-off was more pronounced at lower temporal resolutions and higher thresholds as shown previously for LFP-based avalanches ( Yu et al . , 2014 ) . The threshold robustness and cut-off are in line with previous reports on avalanche dynamics in vitro ( Beggs and Plenz , 2003 ) and in non-human primates ( Petermann et al . , 2009 ) and humans ( Shriki et al . , 2013 ) . 10 . 7554/eLife . 07224 . 009Figure 4 . Avalanche dynamics is robust to changes in λthr . ( A–C ) Cluster size distributions for Δt = 250 , 167 , and 88 ms ( from left to right ) . The green distributions correspond to the respective thresholds , λthr = λthrmax , at which the cluster rate was maximal . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 009 Avalanche dynamics were unique to the AW state ( Figure 5 ) . During AN , cluster size distributions at corresponding λthrmax ( Figure 5—figure supplement 1A ) were slightly bimodal ( Figure 5A , arrow ) , in line with a progressively worse fit to a power law for WK and AN compared to AW ( Figure 5B; ANOVA , p < 0 . 05; cf . Figure 4 , Figure 5—figure supplement 1B ) . 10 . 7554/eLife . 07224 . 010Figure 5 . Avalanche dynamics is abolished under anesthesia . ( A ) Overplot of size distributions for the anesthetized state ( n = 22 recordings ) showing a slight increase in the probability of large clusters ( arrow ) . ( B ) Average KS distance , DKS , between individual PDFs and best-fit power-law distributions for all three states ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 01010 . 7554/eLife . 07224 . 011Figure 5—figure supplement 1 . ( A ) Maximum cluster rate is observed at intermediate threshold levels for all three conditions . ( B ) Average cluster size distributions as a function of λthr for WK . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 011 An alternative approach to obtain avalanches , in which periods of integrated population activity above a population threshold were extracted ( Poil et al . , 2012 ) , also yielded power-law size distributions with exponent close to −1 . 5 and cut-off that were robust to changes in λthr ( Figure 6A–D ) . Similar to what was found when using the original definition of avalanches , cluster size distributions obtained by population thresholding deviated from avalanche dynamics under isoflurane anesthesia ( Figure 5B , Figure 6E , p < 0 . 01; Kruskal–Wallis test on Kolmogorov–Smirnov distances , DKS ) . Furthermore , cluster size and lifetime were correlated , and the corresponding exponent scaled as suggested by the theory of critical systems ( Sethna et al . , 2001 ) ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 07224 . 012Figure 6 . Identifying avalanche dynamics , that is , power law in clustering , using thresholding of the population rate vector ( Poil et al . , 2012 ) . ( A , B ) Cluster size distributions for individual recordings ( n = 12 ) following thresholding of the population rate vector at λthr = 1 ( A ) and 2 . 5 ( B ) . Dashed line: slope = −1 . 5 . ( C ) Rate-preserved shuffling of λ in individual ROIs prior to calculation of population rate vector destroys the power law . ( D ) Power-law exponent , α , is relatively threshold-invariant . ( E ) Deviation from power-law dynamics at the population rate vector level increases with the transition from AW to AN ( p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 01210 . 7554/eLife . 07224 . 013Figure 6—figure supplement 1 . Scaling relationship between lifetime and size of spontaneous AP clusters supports neuronal avalanche dynamics ( Sethna et al . , 2001; Friedman et al . , 2012 ) . ( A ) Double logarithmic plot of avalanche duration and corresponding avalanche size for one local L2/3 PN group in vivo in the AW state . Duration and size scale according to a power law with exponent 1/c . In this example , c was found to be 0 . 75 based on regression analysis ( red line ) . ( B ) For increasing temporal resolution , the scaling relationship between life time exponent and size exponent approaches c . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 013 To summarize , ongoing AP firing of local groups of L2/3 PNs in the AW state displayed the five hallmarks of neuronal avalanche dynamics: first , a power law in size distributions with exponent close to −1 . 5; second , a critical branching parameter close to 1; third , threshold robustness; fourth , a lifetime distribution with exponent close to −2; and fifth , scaling of lifetimes and sizes . These results , for the first time , demonstrate the emergence of neuronal avalanches in the spiking of PN groups from superficial cortical layers in the AW animal . LFP recordings in cortex slice cultures ( Beggs and Plenz , 2003; Stewart and Plenz , 2007; Gireesh and Plenz , 2008 ) have shown avalanche dynamics to emerge spontaneously in superficial layers . Similarly , spike avalanches have been identified in extracellular unit recordings from dissociated cultures of hippocampus ( Mazzoni et al . , 2007 ) and cortex ( Pasquale et al . , 2008; Tetzlaff et al . , 2010; Vincent et al . , 2012 ) , although the mesoscopic organization of the tissue was not preserved . Yet , these studies are limited by the unknown composition of the LFP population signal ( see ‘Introduction’ ) and cell types recorded from . For extracellular unit activity , strongly bursting interneurons can dominate large spike clusters in the neuronal population , in which case heavy-tailed cluster size distributions reflect neuronal differences rather than neuronal interactions . In order to demonstrate that avalanche dynamics also capture spatiotemporal activity in L2/3 PNs in vitro , we conducted studies in GECI-expressing cortical slices , co-cultured with VTA to ensure proper maturation of superficial cortical layers ( Gireesh and Plenz , 2008 ) ( Figure 7A–D ) . We recorded AP activity from local groups of L2/3 PNs in vitro ( n = 15–80 ROIs ) monitored with YC2 . 60 ( Δt = 250 ms; n = 129 movies , n = 35 cultures; Figure 7B–D ) and compared the activity to conditions when GABAA ( 5 µM PTX , n = 8 ) or AMPA and NMDA-receptor mediated ( 0 . 5 µM DNQX , 5 µM AP5 , n = 6 ) synaptic transmission were slightly reduced . Neuronal activity was stable throughout the recording for each condition ( Figure 7—figure supplement 1A , B ) . At the single neuron level , AP firing was irregular , in line with our in vivo results ( Figure 7E , F; Figure 7—figure supplement 1C , D ) . Temporal clustering was present under normal conditions ( ACSF ) but was reduced during disinhibition or disfacilitation ( Figure 7G , PTX and DNQX/AP5 , respectively ) . An intermediate level in correlated AP firing was found under normal conditions ( Figure 7H ) . Correlations between neighboring and distant neurons were highly similar and as expected increased during disinhibition but decreased during disfacilitation ( Figure 7H ) . As described in our in vivo results , the mean rate smoothly declined with increase in λthr ( Figure 7—figure supplement 1E ) , and the number of AP cascades of PN groups peaked in rate at an intermediate threshold λthr ( Figure 7—figure supplement 1F ) for all three conditions . When processed at the corresponding λthrmax , cascade sizes under normal conditions distributed according to a power law that was robust to changes in λthr ( Figure 7I , left ) . As expected , the power law was destroyed when spiking activity was shuffled ( Figure 7I , right ) . As previously shown for LFP-based analysis ( Plenz , 2012 ) , AP-based cluster size distributions became strongly bimodal during pharmacological disinhibition and slightly bimodal during disfacilitation ( Figure 7J , PTX and DNQX/AP5 , respectively; Figure 7—figure supplement 2 ) . YC2 . 60 , while being sensitive to single APs , tends to saturate for very strong spike bursts ( Yamada et al . , 2011 ) . In contrast , the GECI GCaMP3 ( Tian et al . , 2009 ) naturally has a higher threshold for AP detection ( >3 APs ) but reports even strong bursts linearly ( Yamada et al . , 2011 ) ( Figure 7—figure supplement 3A , B ) . In line with our expectation of threshold invariance for LFP-based avalanches in the AW monkey ( Petermann et al . , 2009 ) and our YC2 . 60 measurements , we found that AP bursts measured with GCaMP3 were irregular at the single neuron level ( Figure 7—figure supplement 3C–H ) , while AP cascades formed a clear power law ( Figure 7—figure supplement 3I , J; n = 9 cultures ) . These in vitro results demonstrate neuronal avalanches to describe the spatiotemporal spike activity in L2/3 PN groups , that is , sensitive to the balance of excitation and inhibition and can be detected using high-threshold GECIs . 10 . 7554/eLife . 07224 . 014Figure 7 . Spatiotemporal clustering in ongoing spiking activity recorded from groups of L2/3 PNs in vitro . ( A ) Organotypic cortex ( ctx ) -ventral tegmental area ( vta ) co-culture . YC2 . 60-expression in PNs from superficial ( super ) but not deep ( deep ) cortical layers . wm: white matter border location . ( B ) Single imaging plane containing a group of PNs with significant changes in R over time ( colored ROIs ) . ( C ) Time course of ΔR/R for individual ROIs . ( D ) Top: Binary raster display of instantaneous spike rate estimate λ ( λthr = 0 . 1 ) . Middle: Expanded period showing λ amplitudes . Bottom: Overplot of λ time course for individual ( color coded ) ROIs . ( E ) Distributions of λnorm for different pharmacological conditions ( ACSF , DNQX/AP5 , PTX ) . Dotted line , λnorm = 1 . ( F ) Distributions of IBInorm . Dotted line , IBInorm = 1 . Inset: Pq ( number of recordings is indicated; p < 0 . 05 ) . ( G ) Mean λ autocorrelation function for individual PNs and different conditions . ( H ) Distance dependence of pairwise crosscorrelation in λ and different conditions . ( I ) Left: Power-law distributions in s for different λthr ( color scale ) for normal condition ( ACSF ) . Right: λ shuffling destroys power-law organization for normal condition . ( J ) Probability distributions in s for different λthr ( color scale ) under PTX ( left ) and DNQX/AP5 ( right ) . Dashed lines in I and J: slope = −1 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 01410 . 7554/eLife . 07224 . 015Figure 7—figure supplement 1 . ( A , B ) Mean λ rate and pairwise crosscorrelation are stationary over entire imaging session . ( C , D ) Distribution of normalized λ and IBIs for a single PN group ( n = 42 ROIs ) . ( E ) Mean λ rate smoothly decreases with increasing threshold . ( F ) Cluster rate as function of λthr shows peak at intermediate λthr . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 01510 . 7554/eLife . 07224 . 016Figure 7—figure supplement 2 . Cluster size distributions under normal condition , disfacilitation ( DNQX/AP5 ) , and disinhibition ( PTX ) for YC2 . 60 . ( A ) Example of simultaneously recorded fluorescent intensity ∆R/R over time for a subset of ROIs for all conditions using YC2 . 60 expressed in superficial cortical PNs . ( B ) Distributions of normalized cluster sizes for all conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 01610 . 7554/eLife . 07224 . 017Figure 7—figure supplement 3 . Neuronal avalanche dynamics recorded in vitro using GCaMP3 in organotypic cortex cultures . ( A ) GCaMP3 covers a wide dynamic range of spike bursts but fails to reliably capture burst sizes <3 APs . Loose-patch recordings combined with 2-PI of cortex slice cultures grown for ∼2 weeks . Fluorescence traces triggered on spontaneously occurring AP bursts and sorted by the number of spontaneous APs/250 ms . Note the relative insensitivity of GCaMP3 to small bursts establishing a natural threshold of λthr in the data acquisition . Single PN . ( B ) Summary of change in fluorescent intensity ΔF/F , which increases linearly with spontaneous APs/250 ms , but is undetectably low at sizes <3 APs ( n = 8 cells; color codes ) . Broken lines: regression for individual neurons . ( C ) Average λnorm distribution for all ROIs . Vertical line , λnorm = 1 . ( D ) Corresponding average distribution in normalized quiescent time intervals , IBInorm . Vertical line , IBInorm = 1 . ( E ) Mean λ autocorrelation function . Note the strong decay in autocorrelation demonstrating temporal correlations for up to 10 s . ( F ) CV in AP firing is much larger than 1 and independent of λthr . ( G ) Representative traces of changes in fluorescent intensity ( ΔF/F ) simultaneously recorded over time from GCaMP3-expressing L2/3 PNs in vitro . Note the relatively sparse activity compared to YC2 . 60 recordings and large fluctuations in burst amplitudes . ( H ) Cluster rate as a function of λthr . ( I ) Cluster size distributions of individual cultures based on GCaMP3 recordings . ( J ) Average power-law distributions in normalized cluster size s for different λthr ( color scale ) . Observed cascade sizes with the burst indicator GCaMP3 also follow a power law up to the cut-off of s = 1 . Dashed line , slope = −1 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07224 . 017 GECIs derived from YC are fluorescence resonance energy transfer sensors , that is , the binding of calcium ions decreases fluorescence for short wavelengths while it increases fluorescence in the longer wavelength range ( Nagai et al . , 2004 ) . The anti-correlated change at two wavelengths , captured in the ratiometric signal , increases the signal-to-noise ratio and naturally reduces global signal artifacts , such as those caused by movement of the animal . Our cell-attached recordings also demonstrate linear reporting of AP bursts , measured for up to 28 Hz ( i . e . , 7 APs/250 ms; Figure 1—figure supplement 1B ) . This observation suggests that the estimation of AP burst strength in our recordings is only weakly effected by saturation ( Lütcke et al . , 2010 ) . The lack of impact of any potential saturation effect for YC2 . 60 on our main results was also demonstrated by our GCaMP3 in vitro recordings . GCaMP3 is relatively insensitive to small AP bursts and does not show saturation for relatively strong bursts , which introduces a natural λthr of ∼3 APs in our data acquisition ( Figure 7—figure supplement 3A ) . To summarize , our choice of YC2 . 60 provided us with a GECI that exhibited an excellent dynamic range , that is , high signal-to-noise ratio , reasonable temporal resolution , high probability of single AP detection , and linear mapping of instantaneous firing rates up to ∼20 Hz . The identification of neuronal avalanche dynamics has been found to be robust to a wide range of thresholds for local event detection . As demonstrated in awake non-human primates , the amplitude of the negative LFP ( nLFP ) , which identifies local synchronization , monotonically increases with the number of extracellular spikes recorded at the same microelectrode ( Petermann et al . , 2009 ) . Thus , when identifying nLFPs on the microelectrode array using a threshold , a high threshold includes activity from local neurons that fire strongly , whereas a low threshold also includes weak local neuronal activity . It was found that for LFP-based avalanches , the threshold to identify nLFPs can vary over many standard deviations of the ongoing LFP fluctuations , yet the essential power law in avalanche sizes is maintained , despite a large decline in the nLFP and cascade rates with increase in threshold ( Petermann et al . , 2009 ) . A similar threshold robustness was found for human avalanche dynamics based on the MEG ( Shriki et al . , 2013 ) or BOLD signal in the fMRI signal ( Tagliazucchi et al . , 2012 ) . This is in line with simulations demonstrating that thresholding does not affect avalanche size distributions ( Lasse et al . , 2009 ) . Threshold independence allows large local events ( which are less frequent than small events ) to be properly identified for avalanche analysis even at low temporal resolutions . In the current study , by increasing the local firing threshold , we isolated strong , local events in time and space , thus allowing for concatenation of significant local events into avalanches . At higher temporal resolution , the firing threshold can be lower as even small local firing can be properly separated in time ( cf . Figure 3B ) . As was shown in the original publication on avalanches , at a fixed spatial resolution , prolonging/shortening Δt increases/decreases concatenation leading to a systematic change in the power-law slope of avalanches' sizes while maintaining the power-law signature ( Beggs and Plenz , 2003 ) . Similarly , the distribution of ROIs within an imaging frame establishes a spatial sampling matrix and correspondingly , the power-law slope was found to change accordingly with change in the frame rate ( cf . Figure 3E ) . Both our in vitro and in vivo recordings demonstrate avalanche dynamics to capture the organization of spontaneous spiking in a coarsely sampled sub-network of L2/3 PNs . The plane of imaging captured about 25–35 labeled PNs within a cortical area of 0 . 15 ± 0 . 05 mm2 . At a neuronal density of ∼100 , 000/mm3 for L2/3 in the adult rat ( Meyer et al . , 2010 ) , we therefore expect ∼300 PNs within a focal plane . Thus , we are recording from ∼10% of PNs within our field of view . Simulations of spike avalanches using a branching process ( Ribeiro et al . , 2010 , 2014 ) or models of self-organized criticality ( Priesemann et al . , 2009 ) indicate that coarse spatial subsampling affects proper sampling of avalanche size distributions . Previous attempts to identify spike avalanches in the awake monkey found log-normal distributions , despite robust power laws in nLFP-based avalanches ( Petermann et al . , 2009 ) . Such log-normal distributions have been suggested to indicate slightly subcritical dynamics in spike avalanches ( Priesemann et al . , 2014 ) . However , our results demonstrate a power law in spike avalanches in the awake , but not anesthetized state , within our imaging frame . Because both states should be affected similarly from subsampling , we conclude that subsampling within our cortical field of view is not a major factor in our recordings to identify the power law in avalanche sizes . This could be because , in the above studies , the sampling density of spikes achieved by microelectrode arrays was orders of magnitudes lower than in the current study . On the other hand , because our cranial window only captured a small region of the cortex , we expect to see a cut-off in the avalanche size distribution , as reported in numerous experimental findings on avalanches based on the nLFP ( Beggs and Plenz , 2003; Petermann et al . , 2009; Yu et al . , 2014 ) or MEG ( Shriki et al . , 2013 ) and simulations ( Priesemann et al . , 2009; Ribeiro et al . , 2010; Priesemann et al . , 2013; Ribeiro et al . , 2014; Yu et al . , 2014 ) . Here , we show for the first time that this cut-off also holds for AP avalanches . Traditionally , irregular AP firing in individual PNs has been viewed as evidence that cortical dynamics are noisy and thus require averaging in time or across neurons to separate signal from noise ( Softky and Koch , 1993; Shadlen and Newsome , 1998; London et al . , 2010 ) . Our results confirm that indeed single neuron firing is irregular in both the anesthetized and AW state . We found a lower firing rate , less irregularity , and less temporal clustering under anesthesia compared to the AW state . Yet , these differences did not specify any particular organization of the AW state , in particular , given that the average pairwise correlation between PNs did not differ between the two states . This latter finding suggested at first glance no difference in the average neuronal interactions . Only when temporally contingent spatial spike clusters in the neuronal population were taken into account , did the specific scale-invariant avalanche organization of the AW state become evident . Thus , irregularity of single neuron firing by itself does neither preclude nor determine avalanche dynamics . Several neuronal network simulations have demonstrated the coexistence of avalanche dynamics and highly irregular firing at the single neuron level . Using a stochastic model of spiking neurons , Benayoun et al . show the coexistence of irregularity and avalanche dynamics close to a critical transition ( Benayoun et al . , 2010 ) . Recent deterministic network models that incorporate synaptic plasticity also demonstrate how such irregularity robustly co-exists with avalanche dynamics , even in the absence of stochasticity ( Stepp et al . , 2014 ) , yet demonstratively at a phase transition , where the addition of a single spike led to exponentially deviating network trajectories . Several computational studies in non-leaky as well as leaky integrate and fire networks have demonstrated the emergence of spike avalanches at or near a phase transition , where irregular spikes form clusters whose size distributions follow a power law ( Chen et al . , 1995; Eurich et al . , 2002; Levina et al . , 2007 , 2009; Millman et al . , 2010 ) . Thus , while irregular firing might be ubiquitous in many systems and during different brain states , it is in the AW state where irregularity combines with avalanche dynamics suggestive of a system residing near a phase transition . Our work , by providing detailed parameters on firing rate distributions , event count , interburst statistics , and correlations , should provide new experimental guidance to inform more realistic neuronal network models on critical spiking dynamics . Computational models of cortical networks typically derive irregular spiking by establishing a balance in fast E/I ( Shadlen and Newsome , 1998; Renart et al . , 2010 ) , which allows inhibitory and excitatory currents to track each other closely in time , resulting in an active de-correlation in spiking ( Renart et al . , 2010 ) . However , since independent , stationary Poisson processes are insufficient to explain the high variability of spiking ( CV > 1 ) observed typically in vivo ( Softky and Koch , 1993; Shadlen and Newsome , 1998 ) , alternative mechanisms beyond external Poisson inputs ( Brunel , 2000 ) have been proposed to further increase variability , such as intrinsic chaotic dynamics ( Van Vreeswijk and Sompolinsky , 1996; Sussillo and Abbott , 2009; Ostojic , 2014 ) , conductance-based synapses ( Kumar et al . , 2008 ) , clustered network architecture ( Litwin-Kumar and Doiron , 2012 ) , external synchronous inputs ( Stevens and Zador , 1998 ) , and ‘doubly stochastic’ approaches using non-stationary Poisson processes ( Churchland et al . , 2010 ) , among others . Our results confirm the high variability of single neuron firing in the AW state with CV > 1 . We suggest that the significant increase in CV beyond 1 for the AW state ( cf . Figure 2C ) is in line with the general notion that critical systems operate in a fluctuation-dominated regime , that is , high irregularity encountered at the single neuron level in the AW state might arise from large fluctuations that naturally occur when system dynamics approach a critical point ( Chialvo , 2010; Fraiman and Chialvo , 2012; Meisel et al . , 2015 ) . Most in vivo studies reporting avalanche dynamics have been conducted in the AW animal for example , non-human primates ( Petermann et al . , 2009; Yu et al . , 2011 ) or AW human subjects ( Fraiman and Chialvo , 2012; Palva et al . , 2013; Shriki et al . , 2013 ) . However , it was not clear whether avalanches also arise under anesthesia . In fact , avalanche analysis in deeply anesthetized cats or rodents reveals typical deviations , such as unusually shallow power laws based on the LFP and log-normal distributions of extracellular unit clusters ( Hahn et al . , 2007; Ribeiro et al . , 2010 ) . In these latter studies , however , the effect of anesthesia was difficult to separate from technical aspects , such as subsampling , which can affect spike clusters . In the present study , we ( 1 ) increased the number of neurons typically recorded with microelectrodes within an area of 200 × 200 µm by 1–2 orders of magnitude using 2-PI , ( 2 ) extracted spike clusters from a well-defined population of PNs , ( 3 ) used an exceptionally sensitive GECI to identify even single spikes , and ( 4 ) transitioned the animal between the anesthetized and the AW state . This approach provided us with the necessary precision and sensitivity to demonstrate the increasing deviation from a power law in size distribution that occurs even under light anesthesia ( cf . Figures 5B , 6E ) . Our results suggest that avalanche dynamics might provide a precise ‘fingerprint’ to delineate the transition from the anesthetized state to the fully AW state . This delineation might be helpful in further quantifying different aspects of the AW state . For example , when studying avalanche dynamics in normal subjects , the degree of sleep deprivation was found to correlate positively with deviations from avalanche dynamics ( Meisel et al . , 2013 ) . Intracranial recordings in human subjects have also shown small changes in ‘vigilance’ with changes in avalanche dynamics ( Priesemann et al . , 2013 ) . Avalanche dynamics were originally defined by ( 1 ) cluster sizes in LFP activity in vitro that follow a power law with slope close to −1 . 5 , ( 2 ) a power law in life time distribution with slope of −2 , and ( 3 ) a critical branching parameter of 1 ( Beggs and Plenz , 2003 ) . All three aspects have been demonstrated in spike clusters from layer 2/3 PNs in the present study . Detailed correspondences with avalanche work include the cut-off in size distributions beyond system size , originally identified in the LFP ( Klaus et al . , 2011 ) , and clearly visible in the steep drop in spike cluster distributions beyond system size in the present study ( cf . Figures 3C , 4 ) . We obtained a branching parameter slightly smaller than 1 at the power law slope of −1 . 5 , which might be due to the small neuronal group size recorded from . We also confirmed recent demonstrations of scale-invariance based on the collapse of avalanche waveforms ( Sethna et al . , 2001; Papanikolaou et al . , 2011; Friedman et al . , 2012 ) . These measures combined strongly suggest that AP output of PN groups reflects critical dynamics in the AW state . Avalanche measures have to be carefully evaluated for potential pitfalls . For example , power laws that arise from non-critical dynamics have been reported ( Touboul and Destexhe , 2010 ) , yet those power laws exhibit slopes of −10 to −50 . In contrast , the slope of power laws for avalanche dynamics is typically more shallow than −2 , that is , these distributions do not have a mean and display unbounded variance , that is , non-existing first and second moments . Similarly , the upper cut-off in avalanche size distributions has sometimes been included into statistical fits . However , this upper cut-off arises from finite-size effects and needs to be disregarded for fitting ( Yu et al . , 2014 ) , otherwise , statistical tests ( Langlois et al . , 2014 ) can be misdirected to fit the cut-off only ( Clauset et al . , 2009; Dehghani et al . , 2012 ) . While our results clearly demonstrate that cortical dynamics approach a scale-invariant , that is , power-law organization in the AW state , the precise distance to the critical point is not known . In simulations , complex–hierarchical modular architectures of cortical networks have been shown to support critical dynamics ( Wang et al . , 2011 ) . Such architectures , however , when combined with near-critical dynamics , can ‘trap’ activity ( Rubinov et al . , 2011; Friedman and Landsberg , 2013 ) inducing heavy-tail size distributions approximating power laws , or in general , extend the region in which critical-like behavior is observed by establishing so-called Griffiths phases ( Moretti and Muñoz , 2013 ) . Whether the layer 2/3 network can exhibit ‘true’ critical dynamics has also been called into question on the grounds that this would require a system to be placed exactly at the critical point , which is only possible for fine-tuned , ‘conservative’ systems ( Juan et al . , 2010 ) . Cortical networks , while being critical in the long-term , could show subcritical transients . Subcritical dynamics have been invoked to explain findings from intracranial recordings in humans ( Priesemann et al . , 2013 ) . In network simulations , a transition from subcritical to critical dynamics has been shown to benefit information processing ( Tomen et al . , 2014 ) . Our results demonstrate that in the relatively fast transition from the anesthetized to the AW state , cortical dynamics more closely approaches or enters a critical regime . Finally , a dimension not employed in the current work is the specific temporal correlation structure of avalanches , which reveals scale-invariance and differs significantly from disinhibited or disfacilitated networks ( Lombardi et al . , 2012 , 2014; Plenz , 2012 ) . Our in vitro results for the first time demonstrate that avalanche dynamics also describe the organization of AP patterns in a well identified neuronal population , local groups of L2/3 PNs in isolated cortex preparations . This overcomes previous limitations of in vitro studies in which cell identities and signal composition were largely unknown . The dependency of L2/3 avalanches on a GABAA antagonists and glutamate antagonists supports theories that indeed the E/I balance is important to establish avalanche dynamics in cortical networks . Our findings further demonstrate that in isolated cortex , avalanche dynamics is the natural organization that describes PN spiking in the absence of any inputs . Our in vitro recordings resulted in the collection of spike activity over longer periods in time compared to in vivo . This allowed us to demonstrate that GECIs , such as GCaMP3 , with a natural threshold for spike burst imaging and which are less prone to saturation can also be used for avalanche detection ( Figure 7—figure supplement 3 ) , a direct experimental confirmation of simulations that showed thresholding does not affect avalanche size distributions ( Lasse et al . , 2009 ) . Our in vitro results also provide additional benchmarks to which to compare in vivo avalanche dynamics . For example , pairwise correlations in vitro for avalanche dynamics are significantly higher compared to in vivo . While these values take an intermediate position compared to those found for the disinhibited or disfacilitated state ( cf . Figure 7H ) , it is clear that absolute pairwise correlation values do not predict avalanche dynamics . Our findings transcend previous reports on the spontaneous formation of stable activity patterns in isolated cortical networks , such as the acute slice ( Sanchez-Vives and McCormick , 2000; Beggs and Plenz , 2003; Cossart et al . , 2003 ) and slice culture ( Beggs and Plenz , 2003 ) or in vivo ( Miller et al . , 2014 ) , which suggest the presence of ‘attractor’ states ( Beggs and Plenz , 2004; Ikegaya et al . , 2004; Miller et al . , 2014 ) . In fact , our work shows that AP patterns of L2/3 PNs form a specific subset of patterns in which sizes relate to each other in a scale-invariant manner . It is expected that a strongly fluctuating yet specific spatiotemporal organization in spikes will translate into correspondingly precise inputs in nearby PNs . Fast , subthreshold fluctuations in the intracellular membrane potential have been shown to translate into precisely timed action potentials ( Bryant and Segundo , 1976; Mainen and Sejnowski , 1995 ) and that such spike precision carries information about the input ( Cecchi et al . , 2000 ) . It is , therefore , reasonable to expect that spike avalanches translate into fast voltage-fluctuations , which in turn generate precise spike outputs , thereby maintaining avalanche dynamics in the cortical microcircuit . Recent findings using voltage-sensitive dyes in layer 2/3 PNs indeed demonstrate neuronal avalanche dynamics to emerge in the AW mouse ( Scott et al . , 2014 ) . Signals from voltage-sensitive dyes are proportional to the surface-to-volume ratio of the cellular compartment in which the dye is localized , and thus they preferentially report subthreshold intracellular membrane potential fluctuations ( Plenz and Aertsen , 1993 ) . This is in contrast to intracellular calcium reporters , which mainly report suprathreshold activity . Our study thus complements reports of neuronal avalanche dynamics in the input to PNs by demonstrating neuronal avalanches in the spike output of PNs in superficial layers in the AW state . The emergence of neuronal avalanches at the neuronal group level does not exclude a role for critical dynamics at the cellular and subcellular level . The intermittent bursting behavior of single isolated neurons in response to stimulation ( Gal et al . , 2010; Marom and Wallach , 2011; Gal and Marom , 2013 ) suggests critical dynamics in the form of a low dimensional phase transition to control spike generation . Recent experimental demonstrations of critical slowing down as the membrane potential approaches spike threshold demonstrate critical dynamics to profoundly affect the AP generation ( Meisel et al . , 2015 ) . Similarly , long-term fluctuations and power-law relationships have been reported for sodium channel gating ( Toib et al . , 1998 ) . Thus , while spike avalanche dynamics emerge at the neuronal group level , the underlying mechanisms are likely to involve specific dynamical properties at the single cell and subcellular level . The identification of avalanches in the main excitatory cell type that constitutes the mammalian cortex establishes the strongest proof to date that avalanche dynamics provide the guiding principles for propagation of cortical activity . This finding should have a number of consequences for a cellular understanding of cortical network activity . First , optimization principles in information processing identified for avalanche dynamics at the population level of cortex should be directly applicable to the interaction of PNs ( Beggs and Plenz , 2003; Bertschinger and Natschlager , 2004; Kinouchi and Copelli , 2006; Rämö et al . , 2007; Nykter et al . , 2008; Shew et al . , 2009; de Arcangelis and Herrmann , 2010; Shew et al . , 2011 ) . Specifically , local layer 2/3 networks should exhibit maximal dynamic range to process layer 4 inputs and maximize mutual information between patterns elicited in layer 4 and superficial layers . Second , the nature of spontaneous , irregular firing in PNs profoundly influences theories on cortical function . For example , when these fluctuations in firing are considered to reflect noise , spatiotemporal averaging over neuronal populations can be used to enhance response encoding at the expense of temporal precision and neuronal identity ( Shadlen and Newsome , 1998; London et al . , 2010; Renart et al . , 2010 ) . On the other hand , we demonstrated that fluctuations in single neuron firing , amount to a scale-invariant order in active neuronal groups , suggestive of critical dynamics guiding single neuron firing . Accordingly fluctuations that arise from long-range spatiotemporal correlations between neurons should not be averaged ( Fraiman and Chialvo , 2012 ) but instead need to be taken into account , for example , for theories on cortical population coding ( Averbeck et al . , 2006 ) . Finally , it is well known that resting or ongoing activity profoundly influences stimulus responses ( Arieli et al . , 1996; Sato et al . , 2007; Luczak et al . , 2009; Womelsdorf et al . , 2012 ) . For example , evoked visual responses correlate strongly with ongoing activity shortly preceding the stimulus ( Arieli et al . , 1996 ) . Long-term temporal correlations have been soundly established at the population level , such as the EEG ( Linkenkaer-Hansen et al . , 2001 ) and ECoG ( He et al . , 2008 ) . Our findings suggest that the occurrence of a single ‘spontaneous’ spike or spike burst in the AW state correlates with activity in other PNs over time and distance in cortex , as quantified by the scale-invariant correlation structure established by neuronal avalanches . We suggest that this will be of particular importance in the context of ‘noise correlations’ , which capture the non-stimulus induced correlation structure and tend to affect the decoding capability of a neuronal population ( Averbeck et al . , 2006; Insabato et al . , 2014 ) . The emergence of scale-invariant order from the interaction of local elements is a hallmark of systems at criticality ( Plenz and Thiagarajan , 2007; Chialvo , 2010; Plenz and Niebur , 2014 ) . By demonstrating such scale-invariance to exist at the neuron level , we suggest that neuronal avalanches provide a unifying framework of cell assembly formation in cortex that ranges from local groups of neurons to the global scale of the brain . Timed-pregnant rats ( Sprague Dawley , embryonic day 15 . 5 ± 0 . 5 , Taconic Farms ) underwent a laparotomy ( 1 . 5–4 % isoflurane anesthesia ) during which 5–8 µg of purified plasmid DNA ( Endofree Maxiprep kit , Qiagen , Germantown , MD ) , consisting of Yellow Cameleon 2 . 60 ( YC2 . 60 ) ( Mikoshiba Lab , RIKEN , Japan; [Yamada et al . , 2011] ) or GCaMP3 ( Tian et al . , 2009 ) subcloned into a pCAG backbone , was pressure-injected through the uterine wall into the frontal ventricle of one hemisphere using a fine point glass capillary . DNA was electroporated into cells of the subventricular zone ( Saito and Nakatsuji , 2001; Saito , 2006 ) using platinum tweezertrodes ( 5 mm diameter; 5 square pulses , 45 V amplitude , 50 ms duration; ECM-830 , Harvard Apparatus , Holliston , MA ) , predominantly labeling PNs in superficial cortical layers 2 and 3 ( L2/3; see [Saito , 2006]; Figure 1A ) . In utero electroporated pups ( postnatal day [P] 1–3 ) were checked for expression of YC2 . 60 in dorsolateral cortex and used for the preparation of organotypic co-cultures , consisting of cortex and ventral tegmental area ( VTA ) , as described previously ( Gireesh and Plenz , 2008 ) . In brief , coronal sections of cortex and midbrain were cut using a vibratome ( VT100 S , Leica ) under sterile conditions at 350 µm and 500 µm , respectively . Regions of cortex ( up to 2 mm wide ) containing all layers , and midbrain tissue containing the VTA , were excised and attached adjacent to each other on a glass coverslip . Co-cultures were grown under sterile conditions in standard culture medium in a roller tube arrangement and were used for electrophysiology and 2-photon imaging ( 2-PI ) after 14–20 days in vitro ( DIV ) . For 2-PI in the AW animal , in utero electroporated rats were first identified by transcranial YC2 . 60 fluorescence observation at P1–3 . Animals expressing YC2 . 60 were fitted with a custom-made , T-shaped stainless steel head bar at ∼P21 . To this end , animals were anesthetized ( isoflurane: 4% induction , 1 . 5–2% maintenance ) and mounted in a stereotaxic frame . After a midline incision was made in the scalp , the skull surface was cleared of membranes; adhesive luting cement ( C&B Metabond , Parkell , Inc . , Edgewood , NY ) was applied contralaterally to the YC2 . 60-expressing hemisphere , followed by attachment of the head bar using Grip cement ( Dentsply International Inc . , Milford , DE ) . Rats were given an analgesic ( Ketoprofen , 5 mg/kg s . c ) for up to 2 days post surgery . Rats were habituated to the recording condition for up to 5 sessions post surgery . In each session , the rat was briefly anesthetized ( <5 min of 2–3% isoflurane ) and installed in the head fixation apparatus , which consisted of a plastic tube which loosely confined the rat's limbs without restricting breathing , a platform and a custom-made steel beam , which screwed into the head bar at one end and a fixed post at the other end , allowing horizontal , vertical , and axial freedom of movement to position the rat's head under the 2-PI objective . After awakening , rats were left in the apparatus for 10–20 min per session . Rats became comfortable with the recording condition after 3–5 habituation sessions . On the day of imaging , rats were subjected to craniotomy and cranial window implantation . Rats were anesthetized ( isoflurane: 4% induction , 1 . 5–2% maintenance ) and mounted in a stereotaxic frame . The location of the craniotomy was determined by observation of transcranial YC2 . 60 fluorescence and was usually found within sensorimotor and frontal cortex ( from bregma: AP 0 . 5 ± 1 . 0 mm , ML 3 . 0 ± 0 . 5 mm ) . A section of the skull ( ∼3–4 mm diameter ) was removed using a dental drill and the underlying dura was resected . Care was taken not to damage any subdural blood vessels . The exposed brain was continuously irrigated with sterile saline . Finally , a glass coverslip was cut to the size of the craniotomy using a stylus , mounted on the opening using low melting point agarose ( 1–2% in sterile saline ) , and secured with Grip cement . Rats were given an analgesic ( Ketoprofen , 5 mg/kg s . c ) and allowed to recover for at least 3–6 hr before undergoing 2-PI . For in vivo 2-PI , rats ( P27–35 ) that had undergone head bar implantation , habituation , and craniotomy were anesthetized ( isoflurane: 4% induction , 1–2% maintenance ) , head-fixed , and placed under a 2-photon microscope ( 25× objective , 1 . 05 NA , 1000 MPE , Olympus , Center Valley , PA ) . YC2 . 60 was excited at 840 nm ( Chameleon Vision II , Coherent , Santa Clara , CA ) , and cpVenus and ECFP fluorescent emission were collected using 460–500 nm and 520–560 nm bandpass filters , respectively , separated by a 505 nm dichroic mirror . Once an imaging field containing up to ∼40 neurons was located , 5–15 min long movies were recorded at a temporal resolution , Δt , of ∼250 ms . Higher temporal resolutions ( 167 and 88 ms ) were achieved by 2× and 4× line skipping , respectively ( Olympus Fluoview software ) . To obtain movies in the waking ( WK ) and AW animal , isoflurane was turned off . Movies recorded >5 min after turning off isoflurane were classified as WK , and subsequent movies ( >20 min after turning off isoflurane ) were classified as AW . During WK and AW conditions , the behavioral state of the animal was monitored with an infrared ( IR ) camera ( c525 , IR filter removed , Logitech , Newark , CA ) . Periods of animal movement ( which were minimized by habituation ) generated an easily identifiable artifact in the ∆R/R calcium signal ( see below ) and were manually removed before analysis . For in vitro 2-PI , cultures were submerged in oxygenated artificial cerebrospinal fluid ( ACSF , bubbled with 95% O2 and 5% C02 ) containing ( in mM ) 124 NaCl , 3 . 5 KCl , 10 D-glucose , 26 . 2 NaHCO3 , 0 . 3 NaH2PO4 , 1 MgSO4 , and 2 CaCl2 warmed to 32°C at a flow rate of 1 ml/min . Intracellular calcium dynamics of 15–80 spontaneously active PNs were imaged continuously within a 250 µm by 50–100 µm wide region for 5–20 min with a temporal resolution Δt = ∼250 ms . For our experiments , we chose YC2 . 60 over related GECIs , such as D3cpv and YC3 . 60 , for the following reasons . In general , single AP detection in vivo is still below 100% for yellow chameleons and related GECIs ( Lütcke et al . , 2010; Margolis et al . , 2012 ) . We excluded D3cpv , which shows single AP sensitivity , due to its saturation for small AP bursts ( Wallace et al . , 2008 ) . YC3 . 60 is an interesting alternative to YC2 . 60 due to its higher KD and shorter decay time constant ( 0 . 5 s in vivo at physiological temperature; 0 . 8 s in vitro at room temperature [Yamada et al . , 2011] ) . While a short decay constant allows for higher temporal resolution in imaging , YC3 . 60 is about 50% less sensitive to single APs compared to YC2 . 60 . Given the relatively low spontaneous AP rate for neurons in superficial cortical layers in vivo ( for review see [Barth and Poulet , 2012] ) , we , therefore , opted for YC2 . 60 with its somewhat longer decay constant of ∼1–2 s . Our observation of 4–5% ΔR/R for single APs using YC2 . 60 in layer 2/3 ( L2/3 ) PNs in vitro at 32°C is within the range reported for D3cpv in vitro ( 8 . 3% at room temperature ) and in vivo ( 3 . 5% ) ( Wallace et al . , 2008 ) . It is also in line for YC2 . 60 single AP detection at 33°C in the acute slice , which shows a ΔR/R of ∼4–5% and a decay time constant of ∼2 s for a 10 AP burst ( Yamada et al . , 2011 ) . For YC3 . 60 , sensitivity and decay time constant were shown to be similar in vivo ( Lütcke et al . , 2010 ) and in vitro when measured at physiological temperature ( Yamada et al . , 2011 ) . The range in similarities for YC GECIs suggests that our YC2 . 60 in vitro characterization at physiological temperature similarly predicts its performance in vivo . This is further supported by the insensitivity of coefficient of variation ( CV ) in our in vivo data to the changes in λthr < 1 ( cf . Figure 2—figure supplement 1 ) , which is in line with the binary detection operation of single APs for that range ( cf . Figure 1—figure supplement 1A ) . YC2 . 60-expressing PNs were visually identified either by high-average somatic fluorescence or high-somatic fluorescence CV , which captures relatively quiescent neurons with intermittent , sparse spiking activity and whose average somatic fluorescence remained low . Boxcar regions of interest ( ROIs ) were manually drawn around the somatic region of labeled neurons for which the nucleus was clearly visible within the cross-sectional somatic area . The boxcar was aligned with the outer perimeter of the neuron , and all pixels within the boxcar were taken for analysis . For YC2 . 60 , the ratio , R , of cpVenus fluorescence to ECFP fluorescence was calculated for each ROI and each frame . The ratio measurement requires neuronal signals to be anti-correlated in the two wavelength bands , which allows for easy identification of non-signal artifact , for example , from small animal movements . A continuous function of fluorescence was calculated as ∆R/R = ( RROI − R0 ) /R0 , where RROI denotes the average fluorescence ratio within the ROI . The baseline fluorescence ratio , R0 , was defined as the median of R . ∆R/R was low-pass filtered ( Kampa et al . , 2011 ) ( 3Δt; symmetrical Gaussian kernel ) , and the instantaneous firing rate estimate , λ , in arbitrary units was obtained for each ROI using fast , non-negative deconvolution ( Vogelstein et al . , 2010 ) ( OOPSI package , Matlab , MathWorks Inc . , Natick , MA ) . The decay time constant of the non-negative deconvolution for YC2 . 60 was ∼1–1 . 5 s , as estimated from simultaneous cell-attached recording and 2-PI ( n = 8 neurons ) . This value was also obtained by minimizing the summed square error between R and a reconstructed R′ , made by convolving an exponential decaying kernel with λ over the range of τ = Δt − 15 × Δt for neuronal populations recorded in vivo and in vitro ( Figure 1—figure supplement 1C ) and is in line with previous reports ( Yamada et al . , 2011 ) . The a priori rate estimate for spiking was set to 1 Hz based on a meta-study ( Barth and Poulet , 2012 ) , and lower rates were examined systematically in the form of thresholding on λ ( λ > λthr ) . The standard deviation parameter was estimated for each ∆R/R using calcium amplitudes less than 5% ∆R/R , corresponding to the average amplitude for single spike detection of YC2 . 60 . Deconvolution parameters were determined , first , from loose-patch recordings ( Figure 1—figure supplement 1A , B ) and second , from minimizing the residual error in reconstructed ΔR/R traces from λ ( Figure 1—figure supplement 1C ) . Both methods yielded an optimal deconvolution time constant of ∼1 . 5 s , in line with previous reports ( Yamada et al . , 2011 ) . ROIs with λavg < 0 . 016 ( approx . less than 1 AP/min ) were removed from the analysis ( O'Connor et al . , 2010 ) . The average instantaneous spike rate estimate of λ for each ROI was calculated for a range of λthr = 0 . 1 to 8 and included zero estimates for time bins Δt with λ ≤ λthr . The average burst strength of λ was the mean of all bins with λ > λthr . Interburst intervals ( IBIs ) were defined as consecutive bins of length Δt for which λ ≤ λthr . Pairwise crosscorrelations were calculated in Matlab ( Mathworks Inc . ) using the function corrcoef . All other analyses , unless stated otherwise , were performed in Matlab using custom routines . In order to identify the relationship between the number of spontaneous APs and ∆R/R ( YC2 . 60 ) or ∆F/F ( GCaMP3 ) , loose-patch , voltage-clamp recordings were performed in GECI-expressing PNs in organotypic cortex-VTA cultures ( Gireesh and Plenz , 2008 ) at 14 DIV or later . Cultures were submerged in regular ACSF ( flow rate , 1 ml/min ) at 32°C . After forming a loose-patch on a visually identified YC2 . 60 ( n = 8 ) or GCaMP3 ( n = 8 ) labeled PN in L2/3 , spontaneous , extracellular AP currents were recorded simultaneously with intracellular calcium transients using 2-PI . For each imaging frame with duration Δt = 250 ms , the number of spontaneous APs was correlated with ∆R/R as well as the corresponding peak amplitude and integrated area of the firing rate estimate λ ( Figure 1—figure supplement 1A , B ) . This relationship was also true for ∆F/F ( data not shown ) . To obtain comparable λ and IBI distributions for neurons with different average rates , λavg , the normalized rate λnorm = λ/λavg and normalized IBInorm = IBI/IBIavg were used . The relationship of λ and the real spike count for different temporal resolutions was evaluated using spike trains from 26 extracellularly recorded single units in superficial layers of somatosensory cortex from adult rat during the AW resting state in a separate experiment . For these recordings , we used a Neuronexus array with 8 short shanks and 4 electrodes per shank separated by 200 µm . The array was lowered into the cortex under visual control until the last electrode entered layer 1 , anchored using dental cement , and the cranial opening was closed for chronic in vivo recordings . The array configuration and array insertion targeted unit activity from within the first 600–800 µm of cortical depth , which largely covers superficial layers in the adult rat . Unit activity was sampled at 30 kHz and sorted offline ( Offline sorter , Plexon , Dallas , TX ) . Simultaneous LFP recordings from the same electrodes demonstrated nLFP-based avalanche dynamics in the AW state , further supporting superficial layers as the main recording sites ( see e . g . , [Stewart and Plenz , 2006; Gireesh and Plenz , 2008]; data not shown ) . To obtain surrogate calcium traces at different temporal resolutions , the following three steps were performed: ( 1 ) spike trains were convolved using an impulse function with instantaneous 5% peak amplitude and an exponential decay of 1 . 5 s , parameters obtained from 2-PI ( Figure 1—figure supplement 1 ) . ( 2 ) The resulting calcium traces were sampled at 100 Hz , and uniform noise was added . The noise level for Δt = 250 ms was set to ±8% . Similar results were found for Gaussian noise . ( 3 ) Calcium traces were down-sampled to a final temporal resolution of Δt = 250 , 167 , or 88 ms . In order to simulate the lower signal-to-noise ratios for smaller Δt , which resulted from line skipping during in vivo imaging , we adjusted noise levels by a factor of √2 and √4 for Δt = 167 and 88 ms , respectively . From the resulting calcium traces ( Figure 1—figure supplement 2A ) , λ was estimated using fast , non-negative spike deconvolution ( Vogelstein et al . , 2010 ) . This analysis showed that the mapping of λ to number of APs/Δt is linear and has the same slope ( Figure 1—figure supplement 2B; all R2 = 0 . 99 ) . Spatiotemporal clusters of L2/3 PN activity were defined by spiking activity in at least one ROI above a given threshold ( λ > λthr ) within the same or consecutive time bins of duration ∆t ( Beggs and Plenz , 2003 ) . By definition , a cluster is flanked by empty bins , that is , all ROI have λ ≥ λthr ( Figure 3A ) . The value of ∆t was determined by the temporal resolution of 2-PI . The value of λthr was chosen such that the number of cascades was maximized for a given recording ( Poil et al . , 2012 ) ( Figure 3B ) . The size , sλ , of a cluster was defined as the sum of λ across all active neurons , i = 1 , … , k within the cluster , that is , sλ=∑i=1kλi . We found that sλ was proportional to the number of active neurons in a cluster , sk=∑i=1k1 , by a factor given by the average rate across all ROIs , λavgpop ( R2 = 0 . 98 ) . Therefore , sk and sλ provide similar information about avalanche sizes , as reported previously for nLFP-based avalanches ( Beggs and Plenz , 2003 ) . To compare across experiments with different number of neurons , N , ( i . e . , ROIs ) cluster sizes sλ were normalized by the predicted cluster size limit , Λ=Nλavgpop ( Klaus et al . , 2011; Yu et al . , 2014 ) . For visualization , cluster size distributions were logarithmically binned ( 30 bins ) . Neuronal avalanches were defined by their distribution in cluster size that follows a power law with an exponent close to −1 . 5 , up to the cluster size limit . The branching parameter σ , the average ratio of the number of descendants to ancestors within a cascade ( Beggs and Plenz , 2003 ) , was estimated using participation counts from the first ( ancestor ) and subsequent ( descendent ) Δt bins within each cascade . Lifetime , T , was defined as the number of active time bins during the avalanche multiplied by Δt . A population thresholding approach on the instantaneous integrated population vector on λ , λpop , has recently been introduced as an alternative method to obtain the distribution of cascade sizes ( Poil et al . , 2012 ) . For a given λpop based on a chosen λthr , population thresholds were placed to search for the maximum number of cascade sizes . By including only those Δt bins whose member events sum above a minimum threshold , this method avoids issues in cascade concatenation when periods of empty Δt bins may be rare due to insufficient temporal resolution or when monitoring a high number of units . If not stated otherwise , power-law exponents were estimated by minimizing the Kolmogorov–Smirnov distance , DKS , between the cumulative distribution functions ( CDFs ) of the data , Cdata ( s ) , and the power-law model , Cα ( s ) ( Klaus et al . , 2011 ) :α^=arg minαDKS , andDKS=maxs\|Cdata ( s ) −Cα ( s ) \| . The power-law model for smin <s <smax is given by Pα ( s ) =csα , where c= ( α+1 ) / ( smaxα+1−sminα+1 ) is the normalization constant . The corresponding CDF is then defined by Cα ( s ) =∫sminsPα ( s ) ds . smin was set to the smallest observed avalanche size . The upper bound , smax , was set to the predicted cluster size limit Λ=Nλavgpop at which cluster size distributions started to deviate from a power law ( Figure 3C; see [Yu et al . , 2014] ) . DKS was also used to compare the goodness of the power law fit across different conditions ( Figure 5B; Figure 6E ) . For the comparison of the power law vs the exponential distribution , the log-likelihood ratio ( LLR ) was calculated and parameter estimates were obtained by likelihood maximization ( Clauset et al . , 2009; Klaus et al . , 2011 ) :LLR ( x ) = l ( α\|x ) −l ( γ\|x ) , where l ( α\|x ) =∑i=1nlnPα ( xi ) is the log-likelihood of observing the sample vector , x = x1 , . . . , xn , assuming the power-law model Pα ( s ) , and l ( γ\|x ) =∑i=1nlnPγ ( xi ) is the log-likelihood of observing x assuming an exponential model Pγ ( s ) =ce−γs with c being the corresponding normalization constant . The LLR obtains positive values if the data are better fit by a power law compared to an exponential distribution , and negative values if the exponential distribution yields the better fit . The p-value for determining statistical significance is given by ( Clauset et al . , 2009; Klaus et al . , 2011 ) :p=erfc ( \|LLR\|2nσ2 ) , whereσ2=1n∑i=1n[ ( l ( α\|xi ) −l ( α\|x ) /n ) − ( l ( γ\|xi ) −l ( γ\|x ) /n ) ]2 . One-way analysis of variance ( ANOVA ) was used for multiple comparisons with Bonferroni post hoc test if not stated otherwise . Error bars and shaded areas around averages denote standard error of the mean .	Even when we are not engaged in any specific task , the brain shows coordinated patterns of spontaneous activity that can be monitored using electrodes placed on the scalp . This resting activity shapes the way that the brain responds to subsequent stimuli . Changes in resting activity patterns are seen in various neurological and psychiatric disorders , as well as in healthy individuals following sleep deprivation . The brain's outer layer is known as the cortex . On a large scale , when monitoring many thousands of neurons , resting activity in the cortex demonstrates propagation in the brain in an organized manner . Specifically , resting activity was found to organize as so-called neuronal avalanches , in which large bursts of neuronal activity are grouped with medium-sized and smaller bursts in a very characteristic order . In fact , the sizes of these bursts—that is , the number of neurons that fire—are found to be scale-invariant , that is , the ratio of large bursts to medium-sized bursts is the same as that of medium-sized to small bursts . Such scale-invariance suggests that neuronal bursts are not independent of one another . However , it is largely unclear how neuronal avalanches arise from individual neurons , which fire simply in a noisy , irregular manner . Bellay , Klaus et al . have now provided insights into this process by examining patterns of firing of a particular type of neuron—known as a pyramidal cell—in the cortex of rats as they recover from anesthesia . As the animals awaken , the firing of individual pyramidal cells in the cortex becomes even more irregular than under anesthesia . However , by considering the activity of a group of these neurons , Bellay , Klaus et al . realized that it is this more irregular firing that gives rise to neuronal avalanches , and that this occurs only when the animals are awake . Further experiments on individual pyramidal cells grown in the laboratory confirmed that neuronal avalanches emerge spontaneously from the irregular firing of individual neurons . These avalanches depend on there being a balance between two types of activity among the cells: ‘excitatory’ activity that causes other neurons to fire , and ‘inhibitory’ activity that prevents neuronal firing . Given that resting activity influences the brain's responses to the outside world , the origins of neuronal avalanches are likely to provide clues about the way the brain processes information . Future experiments should also examine the possibility that the emergence of neuronal avalanches marks the transition from unconsciousness to wakefulness within the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Irregular spiking of pyramidal neurons organizes as scale-invariant neuronal avalanches in the awake state
Primary cilia are vital signaling organelles that extend from most types of cells , including neurons and glia . These structures are essential for development of many tissues and organs; however , their function in adult tissues , particularly neurons in the brain , remains largely unknown . Tau tubulin kinase 2 ( TTBK2 ) is a critical regulator of ciliogenesis , and is also mutated in a hereditary neurodegenerative disorder , spinocerebellar ataxia type 11 ( SCA11 ) . Here , we show that conditional knockout of Ttbk2 in adult mice results in degenerative cerebellar phenotypes that recapitulate aspects of SCA11 including motor coordination deficits and defects to Purkinje cell ( PC ) integrity . We also find that the Ttbk2 conditional mutant mice quickly lose cilia throughout the brain . We show that conditional knockout of the key ciliary trafficking gene Ift88 in adult mice results in nearly identical cerebellar phenotypes to those of the Ttbk2 knockout , indicating that disruption of ciliary signaling is a key driver of these phenotypes . Our data suggest that primary cilia play an integral role in maintaining the function of PCs in the adult cerebellum and reveal novel insights into mechanisms involved in neurodegeneration . Primary cilia are organelles that serve as compartments that mediate and integrate essential signaling pathways , including Hedgehog ( HH ) signaling . Because of their critical roles in developmental signaling pathways ( Goetz and Anderson , 2010 ) , disruptions to cilium assembly , structure , or function are associated with a number of hereditary developmental syndromes , collectively termed ciliopathies . Among the more common pathologies associated with ciliopathy are a variety of neurological deficits ( Reiter and Leroux , 2017 ) . During development , ciliary signals drive proliferation and patterning of neural progenitor populations ( Guemez-Gamboa et al . , 2014 ) . Cilia then persist on post-mitotic neurons through adulthood ( Sterpka and Chen , 2018 ) . However , we have only a limited understanding of the roles for primary cilia on differentiated , post-mitotic neurons , particularly within the adult brain . Newly emerging evidence suggests that cilia and ciliary signaling are important in adult neurons: Cilia are required for the establishment of synaptic connectivity in hippocampal dentate granule neurons ( Kumamoto et al . , 2012 ) and in striatal interneurons ( Guo et al . , 2017 ) . Neuronal cilia also concentrate a wide array of G-protein coupled receptors ( GPCRs ) and other neuropeptide and neurotrophin receptors that are important for complex neurological functions ( Berbari et al . , 2008; Domire et al . , 2011; Green et al . , 2012; Guadiana et al . , 2016 ) . Defects in cilia structure have also been observed in patient samples and animal models of several neurodegenerative and neuropsychiatric conditions ( Chakravarthy et al . , 2012; Dhekne et al . , 2018; Keryer et al . , 2011; Muñoz-Estrada et al . , 2018 ) . We previously showed that Tau tubulin kinase 2 ( TTBK2 ) , a kinase causally mutated in the hereditary neurodegenerative disorder spinocerebellar ataxia type 11 ( SCA11 ) ( Houlden et al . , 2007 ) , is an essential regulator of ciliogenesis ( Goetz et al . , 2012 ) . These mutations are frameshift-causing indels that result in premature truncation of TTBK2 at ~AA 450 . SCA11 is characterized by a loss of Purkinje cells ( PC ) in the cerebellum , causing ataxia and other motor coordination deficits ( Houlden et al . , 2007; Seidel et al . , 2012 ) . Recently , we demonstrated that SCA11-associated alleles of Ttbk2 act as dominant negatives , causing defects in cilium assembly , stability , and function ( Bowie et al . , 2018 ) . Given the association between the SCA11-associated truncations of TTBK2 and ciliary dysfunction , we set out to test whether loss of TTBK2 function within the adult brain is associated with degeneration of cerebellar neurons . The cerebellum is the region of the brain responsible for controlling motor coordination , learning , and other cognitive functions . The development and morphogenesis of the cerebellum depends on primary cilia , which are critical for expansion of granule neuron progenitors ( Chizhikov et al . , 2007; Spassky et al . , 2008 ) . PCs , granule neurons , and interneurons , as well as Bergmann glia ( BG ) , are ciliated in the adult cerebellum as well as during development . However , the roles of cilia and ciliary signaling in the adult cerebellum are unknown . In this study , we show that global conditional knockout of Ttbk2 during adulthood as well as genetic targeting of cilia using Ift88 conditional knockout mice , cause similar degenerative changes to cerebellar connectivity . These cellular changes are accompanied by motor coordination phenotypes in the mice . We demonstrate that loss of Ttbk2 and cilia leads to altered intracellular Ca++ in PCs , loss of VGLUT2+ synapses on PC dendrites , and general dysfunction of these cells . We provide strong evidence that primary cilia and ciliary signals are important for maintaining connectivity of specific neurons within the brain , and we demonstrate that dysfunction of primary cilia can cause or contribute to neurodegeneration within the mammalian brain . Mutations within TTBK2 cause the adult-onset , neurodegenerative disease SCA11 . However , the etiology of SCA11 is poorly defined . SCA11 is somewhat unusual among SCAs , in part because the reported causal mutations are base pair insertions or deletions within the coding region of TTBK2 ( Houlden et al . , 2007; Johnson et al . , 2008; Lindquist et al . , 2017 ) , rather than the expansion of CAG repeats , which is the genetic cause of most SCA subtypes ( Hersheson et al . , 2012 ) . To test the requirements for TTBK2 in maintaining neural function within the adult brain , we obtained a conditional allele of Ttbk2 ( Ttbk2tm1c ( EUCOMM ) Hmgu ) from the European Mutant Mouse Cell Repository , ( referred to from here as Ttbk2fl ) . We then crossed Ttbk2fl mice to a mouse line expressing tamoxifen-inducible Cre recombinase driven by a ubiquitously expressed promoter , Ubc-Cre-ERT2 ( Ruzankina et al . , 2007 ) . Using this model , we induce recombination of Ttbk2 in all tissues of the mouse , including the brain , upon injection with tamoxifen ( TMX ) . Because morphogenesis of the mouse cerebellum is complete by P21 ( Marzban et al . , 2014 ) , we chose this time to begin our TMX injections . For all of our experiments , Control animals are either siblings with the same genotype ( Ttbk2fl/fl;Ubc-Cre-ERT2+ ) injected with oil vehicle only , or Ttbk2fl/fl;Ubc-Cre-ERT2- sibling mice injected with the same dose of TMX . We found no phenotypic differences between Control condition animals or pre-induction Ttbk2fl/fl;Ubc-Cre-ERT2+ animals at P21 ( Figure 1—figure supplement 1A-C ) . Consistent with other conditional mutants where cilia are globally removed in adulthood ( Davenport et al . , 2007 ) , 4-month-old Ttbk2c . mut mice exhibit obesity ( Figure 1—figure supplement 1D , D’ , E: 32 . 29 g ± 1 . 86 for Control vs . 46 . 33 g ± 2 . 04 for Ttbk2c . mut ) as well as cystic kidneys ( Figure 1—figure supplement 1F ) . Loss of TTBK2 protein was confirmed with western blot analysis on cerebellum lysates from Ttbk2c . mut animals and littermate Controls ( Figure 1—figure supplement 2G ) . Because the cerebellum is critical for motor coordination and SCA11 is associated with motor deficits , we evaluated locomotor behavior in the Ttbk2fl/fl;Ubc-Cre-ERT2+ , TMX treated animals ( referred to from here as Ttbk2c . mut ) relative to littermate Controls . Within 3 weeks following induction of recombination with TMX , Ttbk2c . mut mice exhibited apparent locomotor deficiencies when observed in their cage ( Figure 1—video 1 ) . To further examine motor coordination in Ttbk2c . mut mice , we employed a rotarod performance test . Ttbk2c . mut mice exhibited a shorter latency to fall compared to the littermate Controls in each trial , for both the accelerating rotarod analysis as well as the steady speed rotarod analysis ( Figure 1A , B ) . These results indicate that Ttbk2c . mut mice are impaired in their motor coordination , consistent with motor deficits observed in multiple mouse models of SCA ( Lalonde and Strazielle , 2019; Klockgether et al . , 2019 ) . To assess whether the motor behavioral changes we observed in the Ttbk2c . mut animals are a consequence of changes to neuronal architecture in the adult brain , we examined Ttbk2c . mut mice at 4 months of age ( 3 months post TMX ) . The brains of Ttbk2c . mut mice have slightly smaller olfactory bulbs , but the overall gross morphology of the cortex and cerebellum was unchanged ( Figure 1—figure supplement 1H ) . SCA11 pathology is associated with degeneration of the cerebellar neurons . We therefore examined the architecture and connectivity of neurons within the cerebellum to assess whether the Ttbk2c . mut animals exhibited phenotypes similar to those described for mouse models of other subtypes of SCA . Within the cerebellum , PCs are the major source of functional neuronal output , and receive excitatory inputs primarily from parallel fibers and climbing fibers . Parallel fibers extend from the granule neurons , a population of densely packed neurons found directly beneath PCs ( Ichikawa et al . , 2016 ) . Climbing fibers extend from neurons of the Inferior Olivary Nuclei ( ION ) in the medulla ( Kano et al . , 2018 ) . These connections are essential for PC function , and dysfunction or loss of these connections , particularly the VGLUT2+ excitatory synapses from the climbing fibers , has been shown in various mouse models of SCA to be linked to pathology and disease progression ( Duvick et al . , 2010; Ebner et al . , 2013; Furrer et al . , 2013; Smeets and Verbeek , 2016; Smeets et al . , 2015 ) . To assess the changes throughout Ttbk2c . mut cerebella , we looked at the different layers of the cerebellar folia . We observed thinning of the molecular layer of the cerebellum , which comprises elaborate dendrites extended from the PCs ( Figure 1C , D . 175 μm ± 3 . 422 for Control vs . 160 . 9 μm ± 2 . 527 for Ttbk2c . mut ) . More dramatically , on examination of the synaptic marker VGLUT2 , we found a marked reduction in these puncta throughout the Ttbk2c . mut cerebellum compared to Controls ( Figure 1C , E 527 . 3 puncta ± 12 . 68 for Control vs . 297 . 8 puncta ± 15 . 65 for Ttbk2c . mut ) despite a lack of obvious gross morphological changes in the dendritic arbors of Ttbk2c . mut PCs ( Figure 1—figure supplement 2A , B ) . Additionally , Ttbk2c . mut animals injected with tamoxifen beginning at P45 exhibit the same loss of VGLUT2+ synapses ( Figure 1—figure supplement 3A-B ) , showing that loss of TTBK2 after all connections and development of the cerebellum are complete , leads to degeneration of these synapses . We confirmed that this loss of VGLUT2 synapses was not caused by loss of dendritic spines , with use of a Golgi stain to view individual dendrites and spines on Ttbk2c . mut PCs ( Figure 1F , G; 1 . 236 ± 0 . 22 spines per μm for Control vs . 1 . 368 ± 0 . 32 spines per μm for Ttbk2c . mut ) . To further explore the role of TTBK2 in PC function , we examined calcium receptor abundance in these neurons . PCs require an intracellular calcium modulation network as a means of signaling function . Within this network inositol 1 , 4 , 5-trisphosphate receptors ( IP3Rs ) are key calcium channel regulators needed for calcium release from the surrounding endoplasmic reticulum ( ER ) throughout the PC ( Sarkisov and Wang , 2008 ) . Precise regulation of IP3R activity is critical , and a balance of calcium channel release is imperative to the overall function of the cerebellum . Mutations in the IP3R1 gene have been linked to SCA15 and SCA29 , while overexpression of IP3R1 underlies phenotypes within SCA2 and SCA3 ( Tada et al . , 2016 ) . Because of these links to other SCA-related phenotypes , we therefore examined IP3R expression throughout the Ttbk2c . mut animals . We found that levels of IP3R in PCs are reduced in Ttbk2c . mut compared to Controls starting at P45 , with expression strongly reduced throughout the PCs 3 months after TMX ( Figure 1H ) . Thus , like mouse models of other SCA subtypes , the PCs of Ttbk2c . mut animals exhibit defects in calcium modulation consistent with dysfunction of these cells . Next , we tested whether loss of Ttbk2 affects other cell types linked to the pathology of SCA , in addition to the PCs . Climbing fibers extend from neurons of the inferior olivary nucleus ( ION ) in the medulla . These fibers traverse the brain stem , enter the cerebellar cortex , and innervate the PC dendrites ( Watanabe and Kano , 2011 ) . As we saw a reduction of the VGLUT2+ synaptic terminals between these climbing fibers and PC dendrites , we examined the soma of the ION neurons from which these climbing fibers extend . In several subtypes of SCA , including SCA1 , 2 , 3 , 6 , and 7 ( Seidel et al . , 2012 ) , the pathology of the disorder is characterized in part by the loss of ION soma; a characteristic also observed in mouse models of these diseases . Neurons within the ION can be identified by dual expression of Calbindin and NeuN in the medial ventral region of the medulla ( Figure 2—figure supplement 1A , B ) . When we looked at this population of neurons , we did not notice a loss of these cells . However , we did find that the perikarya of the neurons within the ION were smaller in Ttbk2c . mut animals compared to Controls . We therefore used NeuN to label the perikarya of neurons within the ION and measured the area of the somata of ION neurons , and found a significant reduction in the area of ION neuron soma in Ttbk2c . mut mice at 4 months of age compared to Controls ( Figure 2A , B; 180 . 4 μm2 ± 1 . 93 for Control vs . 109 μm2 ± 1 . 01 for Ttbk2c . mut ) . Neuronal shrinkage is a phenotype that has been noted in patients with SCA1 as well as Friedrich’s ataxia , and is thought to precede neuronal apoptosis ( Nagaoka , 2003; Dell'Orco et al . , 2015; Kemp et al . , 2016 ) . This implies that , in addition to the PCs themselves , the neurons sending critical inputs to the PCs are perturbed in Ttbk2c . mut mice . Throughout the brain , astrocytes and glia also play important roles in maintaining synaptic connectivity and strength . In the cerebellum , the processes of the BG are interspersed with PC dendrites in the molecular layer , with BGs enwrapping the excitatory synapses of the PCs ( Leung and Li , 2018 ) . As defects in BG morphology have been linked to the etiology of SCA7 ( Furrer et al . , 2011 ) , we examined the BGs in Ttbk2c . mut . To assess the morphology of BGs in the Ttbk2c . mut animals and evaluate whether defects in these cells may contribute to the phenotype , we used GFAP to visualize BG fibers that extend throughout the cerebellar folia . We found that the numbers of glial fibers were moderately reduced in Ttbk2cmut cerebellar folia compared to littermate Controls ( Figure 2C , D; 11 . 44 BG fibers ± 0 . 29 for Control vs . 7 . 64 BG fibers ± 0 . 22 for Ttbk2cmut ) , suggesting that loss of Ttbk2 has modest effects on the morphology of BGs . Taken together , these data suggest that loss of Ttbk2 affects several cell types in the cerebellum and medulla , underscoring the widespread importance of Ttbk2 within these tissues . Dysfunction and eventual atrophy of the PCs in the cerebellum is the primary pathology underlying SCA11 in human patients ( Houlden et al . , 2007 ) . In our conditional Ttbk2 mutant mice , the most prominent phenotype is altered connectivity of the PCs with additional cellular changes seen in the BGs as well as ION neurons . To determine the degree to which these defects are the result of cell autonomous vs . non-cell-autonomous requirements for Ttbk2 in the PCs , we used the PC-specific Cre line Pcp2-Cre , which drives recombination specifically in PCs within the cerebellum beginning at P6 ( Zhang et al . , 2004 ) . At P30 , Ttbk2fl/fl;Pcp2-Cre+ ( referred to from here as Ttbk2Pcp2 ) animals have normal cerebellar structure throughout , with molecular layer thickness comparable to that of littermate Control animals ( Figure 3A , C . 202 . 7 μm ± 3 . 51 in P30 Control vs . 191 . 5 μm ± 3 . 21 in P30 Ttbk2Pcp2 ) . The VGLUT2+ synapses between climbing fibers and PCs are not significantly changed between P30 Control and Ttbk2Pcp2 animals ( Figure 3A , D; 548 . 2 puncta ± 13 . 36 in P30 Control vs . 538 . 9 puncta ± 18 . 14 in P30 Ttbk2Pcp2 ) . This indicates that despite postnatal loss of Ttbk2 , initial connections between PCs and climbing fibers are established normally . By P90 , however , Ttbk2Pcp2 animals exhibited phenotypes largely recapitulating those observed in the Ttbk2c . mut animals . At P90 , numbers of primary cilia are significantly reduced on PCs of Ttbk2Pcp2 animals ( Figure 3—figure supplement 1A , B; 48 . 93 ± 7 . 86 percent PCs ciliated in Control vs . 18 . 67 ± 11 . 07 percent PCs ciliated in P90 Ttbk2Pcp2 ) . While molecular layer thickness between the P90 Control and Ttbk2Pcp2 was not changed ( Figure 3B , C; 202 . 9 μm ± 2 . 11 in P90 Control vs . 197 . 4 μm ± 4 . 18 in P90 Ttbk2Pcp2 ) , we see a significant decrease in VGLUT2 puncta throughout the cerebellum ( Figure 3B , D; 549 . 9 puncta ± 15 . 47 in P90 Control vs . 476 . 9 puncta ± 15 . 82 in P90 Ttbk2Pcp2 ) , indicating that PCs have started to lose these important connections from the climbing fiber synapses . We then assessed motor coordination of P30 and P90 Ttbk2Pcp2 animals using the rotarod performance test and did not observe significant changes in P30 animals . However , by P90 , the Ttbk2Pcp2 animals consistently exhibited reduced latency to fall on both the accelerating rotarod as well as the steady speed rotarod performance tests ( Figure 3E and F ) . These data show that loss of Ttbk2 , specifically from PCs , causes neurodegenerative phenotypes . Our data from the Ttbk2 global conditional knockouts revealed that the morphology of BGs was modestly perturbed ( Figure 2C , D ) . As defects in BGs have been shown to non-cell autonomously contribute to the degenerative phenotypes observed in SCA7 ( Furrer et al . , 2011 ) , we tested whether deletion of TTBK2 specifically from these cells could also result in loss of synapses and other degenerative changes to the PCs . We crossed Ttbk2fl/fl animals to a Slc1a3-CreER mouse ( Wang et al . , 2012 ) to produce Ttbk2fl/fl; Slc1a3-CreER+ mice to induce recombination of the Ttbk2 allele specifically within glial cells . Following the same TMX injection protocol used for the Ttbk2c . mut experiments , we did not see changes to the VGLUT2+ synapses on PC dendrites ( Figure 3G–I ) . These data indicate that the PC phenotypes observed in the Ttbk2c . mut mice are primarily cell autonomous . In the first 3 months following TMX injections , the phenotypes exhibited by the Ttbk2c . mut mice consisted mainly of altered synaptic connectivity between PC and ION climbing fibers , and accompanying deficits in motor coordination ( Figure 1 ) . However , when we assessed the cerebellar phenotypes of animals at 6 months of age ( 5 months following TMX injection ) , we found gaps in the molecular layer where PCs appear to be absent ( Figure 4A ) . We quantified this observation by counting PC soma within a defined region of the primary fissure , and confirmed that the number of PCs is reduced in 6-month-old Ttbk2c . mut mice compared to littermate Controls of the same age , as well as compared to 4-month-old Ttbk2c . mut mice ( Figure 4B; 18 . 5 ± 0 . 29 PCs per 500 μm for 4-month Control vs . 18 . 42 PCs ± 0 . 34 for 4-month Ttbk2c . mut; 18 . 67 PCs ± 0 . 43 for 6-month Control vs . 11 . 92 PCs ± 0 . 74 for 6-month Ttbk2c . mut ) . We found more PC gaps in folia of 6-month-old Ttbk2c . mut animals compared to Controls , and that most gaps are enriched at the inner folia . PC gaps were not found on folia X , which is consistent with data showing folia X being resistant to neurodegeneration ( Figure 4C and Figure 4—figure supplement 1A ) ( Tolbert et al . , 1995 ) . A postmortem examination of a SCA11 affected individual revealed Tau aggregates in regions of the brain outside of the cerebellum ( Houlden et al . , 2007 ) . We therefore looked for pathological Tau aggregates in the cortex in both the 4-month-old and 6-month-old Ttbk2c . mut animals . We could not detect the accumulation of phosphorylated Tau in the Control or 4-month-old Ttbk2c . mut cortex . However , in the cortex of the 6-month-old Ttbk2c . mut animals , we noticed a small number of neurons with some accumulation of phosphorylated Tau , recognized by an antibody specific to Ser202 and Thr205 phosphorylated tau ( Figure 4—figure supplement 2A ) . This mild accumulation is compared to neurons in a mouse model of Alzheimer’s , JNPL3 ( P301L ) , which expresses a mutated form of Tau ( Lewis et al . , 2000 ) and is therefore positive for accumulation of phosphorylated Tau ( Figure 4—figure supplement 2B ) . Thus , dysfunction of PCs after loss of Ttbk2 , accompanied by Tau accumulation outside of the cerebellum , recapitulates currently described SCA11 phenotypes . Our prior work demonstrated that mutations associated with SCA11 , which result in the production of a truncated protein , interfere with the function of full-length TTBK2 . In particular , these mutations dominantly interfere with cilia formation in embryos and cultured cells ( Bowie et al . , 2018 ) . Throughout the adult cerebellum and other regions of the hindbrain , neurons possess primary cilia ( Figure 5A–C ) . Within 20 days following administration of TMX to induce recombination ( P45 ) , the number of ciliated cells in the cerebellum declined dramatically in Ttbk2c . mut mice: from a mean of 22 . 46 cilia per 32 mm2 field ± 0 . 7626 in Control animals to 2 . 36 cilia per 32 mm2 field ± 0 . 3103 in Ttbk2c . mut animals ( Figure 5D , E ) . This loss of cilia was observed throughout the cerebellum , brain stem , and other areas of the brain such as the hippocampus and the cortex ( Figure 5—figure supplement 1A , B ) . Thus , loss of cilia coincides with the behavioral changes we identified in Ttbk2c . mut mice , yet precedes the cellular changes ocurring throughout the cerebellum of Ttbk2c . mut mice as they age . Ttbk2 is essential both for the initiation of cilium assembly as well as the structure and stability of cilia ( Bowie et al . , 2018; Goetz et al . , 2012 ) . Given this critical link between TTBK2 and primary cilia in all cell types examined in both developing and adult tissues , we tested whether loss or dysfunction of cilia via a different genetic mechanism causes convergent phenotypes to those of the Ttbk2c . mut mice . For these studies we turned to conditional mutants of another key ciliary protein , Intraflagellar Transport Protein 88 ( IFT88 ) . IFT88 is a component of the IFTB particle required for assembly of the ciliary axoneme as well as anterograde trafficking within the cilium ( Pazour et al . , 2000 ) . Our previous work shows that IFT88 functions downstream of TTBK2 in cilium initiation ( Goetz et al . , 2012 ) , with TTBK2 being required for IFT recruitment . In the developing and postnatal brain , IFT88 is important for cilia structure in the hippocampus and cortex ( Willaredt et al . , 2008 ) and when knocked out in these specific neuron populations results in memory deficits ( Berbari et al . , 2014 ) . Additionally , Ift88 null mutants exhibit nearly identical embryonic phenotypes to those of Ttbk2 null mutants ( Murcia et al . , 2000 ) . When we knocked out Ift88 using the same approach described for Ttbk2c . mut animals , we observed that the numbers of cilia were significantly reduced in Ift88c . mut cerebella at 3 months post TMX treatment , although more cilia remain in Ift88c . mut cerebella compared to the Ttbk2c . mut animals with the same treatment ( Figure 6A , C; 17 . 31 cilia per 32 mm2 field ± 0 . 65 for Control vs . 12 . 13 cilia per 32 mm2 field ± 0 . 50 for Ift88c . mut ) . Western blot analysis of cerebellar tissue from Ift88c . mut mice reveals that a small amount of IFT88 protein perdures in brain tissue ( Figure 6B ) . This could , in part , help to explain why we do not see a full loss of cilia throughout the cerebellum similar to that observed in the Ttbk2c . mut mice . Regardless , the cilia that do remain in Ift88c . mut animals are shorter in length than Controls ( Figure 6D; 2 . 31 μm ± 0 . 10 for Control vs . 1 . 70 μm ± 0 . 08 for Ift88c . mut ) . To further characterize the remaining cilia in the brains of Ift88c . mut mice , we examined additional markers of the ciliary membrane , including adenylate cyclase 3 ( AC3 ) ( Guadiana et al . , 2016 ) . We observed that within the WT cerebellum , there exist cilia that are AC3+ as well as AC3+/ARL13B+ ( Figure 6E , arrowhead , Figure 6F ) . In our Ift88c . mut animals , we observed that the numbers of AC3+ cilia were strongly reduced ( Figure 6G: 7 . 47 AC3+ cilia per 32 mm2 field ± 0 . 41 in Control vs . 2 . 17 AC3+ cilia per 32 mm2 field ± 0 . 22 in Ift88c . mut ) . This analysis suggests that IFT88 is required forlocalization of specific signaling molecules such as AC3 to neuronal primary cilia throughout the cerebellum . We then examined cerebellar structure and circuitry in Ift88c . mut animals . Similar to our findings in Ttbk2c . mut animals , no changes to PC number were evident at 3 months post TMX treatment . Molecular layer thickness was reduced in Ift88c . mut animals ( Figure 7A , B; 182 . 3 μm ± 2 . 5 in Control vs . 169 . 3 μm ± 3 . 04 in Ift88c . mut ) . Similarly , VGLUT2 puncta were reduced in Ift88c . mut compared to Controls ( Figure 7A , C; 515 . 7 puncta ± 20 . 58 in Control vs . 395 . 6 puncta ± 13 . 7 in Ift88c . mut ) . We also tested Ift88c . mut animals on the rotarod performance test to uncover any motor coordination deficits , given that these mice exhibit similar cellular changes to those of the Ttbk2c . mut animals . These tests revealed that Ift88c . mut animals also have a shorter latency to fall time on the steady speed rotarod performance test , but not on the accelerating rotarod performance test ( Figure 7D , E ) , while Ttbk2c . mut animals have a shorter latency to fall time on both accelerating and steady speed rotarod performance tests ( Figure 1A , B ) . Steady speed rotarod analysis is thought to more accurately detect motor coordination deficits , while the accelerating rotarod test can also be affected by mouse fatigue ( Monville et al . , 2006 ) . Taken together , these data show that loss of IFT88 from the adult brain results in impaired ciliary structure and similar defects in cerebellar architecture and locomotor behavior to those observed in the animals lacking TTBK2 . We further assessed whether the Ift88c . mut animals also lose PCs as they age , as was the case for the Ttbk2c . mut mice ( Figure 4 ) . 6-month-old Ift88c . mut animals show gaps throughout the PC layer , and have reduced numbers of PC soma ( Figure 8A , B; 17 . 5 ± 0 . 44 per 500 μm in 4-month-old Control vs . 16 . 92 PC soma ± 0 . 31 in 4-month-old Ift88c . mut . 17 . 17 PC soma ± 0 . 55 in 6-month-old Control vs . 12 . 67 PC soma ± 0 . 43 in 6-month-old Ift88c . mut ) . Coupled with these findings , the molecular layer thickness is further reduced in 6-month-old Ift88c . mut animals ( Figure 8C–E; 173 . 9 μm ± 2 . 28 in 6-month-old Control vs . 158 . 0 μm ± 1 . 63 in 6-month-old Ift88c . mut ) , as well as VGLUT2 puncta counts being diminished in 6-month-old Ift88c . mut ( Figure 8C , D , F; 645 . 9 puncta ± 26 . 83 in 6-month-old Control vs . 461 . 4 puncta ± 25 . 42 in 6-month-old Ift88c . mut ) . In this work , we tested the hypothesis that SCA11 pathology results from the requirements for TTBK2 in cilium assembly and stability . We showed that TTBK2 is essential for maintaining the connectivity and viability of PCs in the adult cerebellum . These phenotypes are similar to those reported for mouse models of other subtypes of SCA as well as consistent with many aspects of SCA11 in human patients , suggesting that the Ttbk2c . mut mice model the human condition . We further demonstrated that mice conditionally lacking the ciliary protein IFT88 in adult tissues exhibit highly similar neurodegenerative phenotypes to those observed in the Ttbk2c . mut mice , including loss of excitatory synapses from the climbing fibers and the eventual loss of PCs . The high degree of convergence of these phenotypes suggests that the neural degenerative phenotypes of the Ttbk2c . mut mice are driven primarily by the requirement for TTBK2 in mediating cilium assembly , and points to a critical role for these organelles in maintaining neuronal function during adulthood . Cilia and ciliary signals play a variety of important roles during embryonic and postnatal development of the brain and central nervous system . Cilia are linked to processes including the expansion and patterning of neural progenitors ( Guemez-Gamboa et al . , 2014 ) , the migration and laminar placement of interneurons ( Higginbotham et al . , 2012 ) , and in the establishment of neuronal morphology ( Guadiana et al . , 2016; Sarkisian and Guadiana , 2015 ) . Consistent with the varied roles of cilia in neural development , an array of neurological deficits are among the most common hallmarks of ciliopathies ( Lee and Gleeson , 2010; Youn and Han , 2018 ) , highlighting the importance of these organelles in human health . In addition to their critical developmental functions , mounting evidence supports an important role for ciliary signaling in tissue regeneration and homeostasis in adult organs , including the kidneys ( Davenport et al . , 2007 ) , skin ( Croyle et al . , 2011 ) , skeletal muscle ( Kopinke et al . , 2017 ) , and bone ( Moore et al . , 2018 ) . Within the adult CNS , dysfunction of ciliary trafficking is linked to retinal degeneration ( Wheway et al . , 2014 ) . Degeneration of photoreceptors is a feature of many human ciliopathies as well as mouse models of these disorders ( Braun and Hildebrandt , 2017; Bujakowska et al . , 2017 ) , and occurs as trafficking within the photoreceptor outer segments ( the modified cilium ) fails . This results in accumulation of rhodopsin within the cell body and leads to death of the photoreceptor neurons through mechanisms that are not completely understood ( Seo and Datta , 2017 ) . Within the brain , conditional loss of the ciliary protein ARL13B from mouse striatal interneurons , both during their development and following maturation , results in changes in their morphology and connectivity ( Guo et al . , 2017 ) . Our work extends these findings and provides strong genetic evidence that primary cilia and signals mediated by these organelles are important to maintain the morphology and function of a specific type of neuron within the brain , cerebellar PCs . Of substantial interest for future studies in our laboratory is the question of whether additional types of neurons in other regions of the brain require primary cilia to maintain their connectivity or viability . These include neurons that are affected by other more common neurodegenerative conditions , such as hippocampal neurons affected by Alzheimer’s disease , midbrain dopaminergic neurons lost in Parkinson’s disease , or medium spiny neurons affected in Huntington’s disease . In each of these cases , links have been made between cilia dysfunction and these disorders , although functional and mechanistic studies have yet to be performed ( Chakravarthy et al . , 2012; Dhekne et al . , 2018; Keryer et al . , 2011 ) . Our data show that conditional mutants of Ttbk2 and Ift88 have similar phenotypes with respect to loss of excitatory synapses to PCs from the climbing fibers , motor coordination deficits , and eventually loss of PCs . This evidence suggests that these defects are the result of ciliary loss or dysfunction . We note , however , that the ciliary phenotypes that result from loss of Ttbk2 differ from those observed in the Ift88 conditional mutants in the context of the adult brain . Ttbk2 conditional mutant mice rapidly lose cilia following administration with TMX , with nearly all cilia within the cerebellum being absent within 20 days . In contrast , the numbers of cilia in the Ift88 conditional mutants are only slightly reduced 3 months following TMX . However , these cilia exhibit significant abnormalities , including reduced length , and a near-complete loss of AC3 from the remaining cilia . This suggests that the degenerative phenotypes observed in both conditional mutants are driven by the loss of a specific ciliary signal . While the precise nature of the ciliary signals that maintain the connectivity and viability of PC neurons remains unknown , there are a number of candidates . Many different GPCRs and associated signaling cascades and second messengers have been shown to concentrate in primary cilia or to be enriched predominantly at the primary cilium ( Mykytyn and Askwith , 2017 ) . In particular , cAMP and Ca++ are highly concentrated within the cilium ( Moore et al . , 2016 ) . Misregulation of these concentrations through either ciliary loss or dysfunction could therefore result in perturbed signaling outputs to the cell bodies of these neurons . Provocatively , we found that AC3 , a molecule important for the production of cAMP , is largely absent from the cilia that remain in the brains of Ift88c . mut animals , suggesting that its loss from the cilium may be a contributing signaling mechanism to the cellular changes in the cerebellum . Additionally , primary cilia play a well-known role as essential mediators of Hedgehog ( HH ) signaling . Sonic Hedgehog ( SHH ) is secreted by PCs during development well into adulthood ( Lewis et al . , 2004; Traiffort et al . , 1998 ) , although the precise role and functional significance of SHH within the adult cerebellum is largely unclear . Ultimately , it will be important to investigate the regulation and molecular composition of neuronal cilia in a comprehensive and unbiased manner . Our observations also have interesting implications for regulation of cilium assembly and stability both generally , and in post-mitotic adult cells and neurons in particular . We found , for example , that most IFT88 protein is lost in Ift88 conditional mutants as expected . However , the cilia persist on adult neurons within the brain in the absence of IFT88 , suggesting that fully functional IFT is not required for these cilia . In contrast , in the absence of TTBK2 , cilia are rapidly lost . This suggests that , in this context , TTBK2 could be playing a more central role than the IFT machinery in maintaining the stability of cilia . The exact mechanisms by which TTBK2 regulates the stability of cilia and the degree to which this role may be specifically important in neurons will be the subject of future investigations within our lab . In particular , the dynamics of cilium assembly and disassembly in post-mitotic cells in vivo have not been characterized . For example , a recent study found that the proteins that make up the basal body of adult neurons in the mouse are very long-lived whereas those of the ciliary axoneme turn over more quickly ( Arrojo E Drigo et al . , 2019 ) . This might imply that the cilia of these adult neurons turn over at some interval , or simply that their protein components are replaced - a topic that merits further investigation . In addition to being required for the biogenesis of cilia , TTBK2 also localizes to the + tips of microtubules , via interaction with the + end binding protein EB1 ( Jiang et al . , 2012 ) . TTBK2 has also been shown to phosphorylate β-Tubulin as well as microtubule associated proteins TAU and MAP2 through in vitro assays ( Takahashi et al . , 1995; Tomizawa et al . , 2001 ) , pointing to roles for TTBK2 in regulation of the microtubule cytoskeleton beyond the cilium . In addition , TTBK2 and the closely related kinase TTBK1 both phosphorylate SV2A in vitro , a component of synaptic vesicles important for retrieval of the membrane trafficking protein Synaptotagmin one during the endocytosis of synaptic vesicles ( Zhang et al . , 2015 ) . Although our data show that the Ttbk2 mutant phenotypes strongly overlap with those of other ciliary genes , such as Ift88 , we cannot exclude the possibility that other roles of TTBK2 specifically within the brain also contribute to the degenerative phenotypes . Importantly , these two possibilities are not mutually exclusive , and indeed , one exciting possibility is that TTBK2 is important for relaying signals from the cilium to the neuronal cell body . In this work , we present evidence that loss or impaired function of TTBK2 within the brain results in degeneration of PCs largely because of the requirement for TTBK2 in mediating the assembly and stability of primary cilia . This points to ciliary dysfunction as being a major mechanism underlying the pathology of SCA11 , which is caused by truncating mutations to TTBK2 ( Houlden et al . , 2007 ) that act as dominant negatives ( Bowie et al . , 2018 ) . In addition , our work raises the possibility that cilia play an important , largely unappreciated role in maintaining neuronal connectivity within the brain , and may also be required for the viability of some types of neurons . From a clinical perspective , our findings suggest that neurodegeneration , in addition to other neurological impairments with developmental origins , may emerge in some patients with ciliopathies such as Joubert and Bardet Biedl syndromes , particularly as patients age . The use and care of mice as described in this study was approved by the Institutional Animal Care and Use Committees of Duke University ( Approval Number A218-17-09 ) . All animal studies were performed in compliance with internationally accepted standards . Ttbk2c . mut mice were produced by crossing Ttbk2tm1a ( EUCOMM ) Hmgu mice to ACTB:FLPe ( Jax stock #003800 ) . The following mice were purchased from Jackson Laboratories: Ift88flox ( stock #022409 ) , Ubc-CreER ( stock #007001 ) , Slc1a3-CreER ( stock #012586 ) and Pcp2-Cre ( Jax stock #010536 ) . Slides used from JNPL3 ( P301L ) mice were a gift from Dr . Carol Colton at Duke University . PCR genotyping was performed on all mice before experiments to confirm the presence of floxed alleles and Cre . Ttbk2-floxed allele , primers used: 5’ ATACGGTTGAGATTCTTCTCCA , 3’ AGGCTGTACTGTAACTCACAAT ( WT band 978 bp , floxed band 1241 bp ) . Ift88-floxed allele , primers used: 5’ GCCTCCTGTTTCTTGACAACAGTG , 3’ GGTCCTAACAAGTAAGCCCAGTGTT ( WT band 350 bp , floxed band 370 bp ) . Universal Cre ( Ubc-CreER , Pcp2-Cre ) , primers used: 5’ GATCTCCGGTATTGAAACTCCAGC , 3’ GCTAAACATGCTTCATCGTCGG ( transgene band 650 bp ) . Tamoxifen powder ( Sigma T5648 ) was dissolved in corn oil ( Sigma C8267 ) to a desired concentration of 20 mg/mL . Mice were given five consecutive 100 μL intraperitoneal injections of 20 mg/mL tamoxifen starting at P21 . Control mice were given corn oil vehicle only . To harvest tissues from adult mice , animals were deeply anesthetized with 12 . 5 mg/mL avertin and transcardially perfused with 10 mL of phosphate buffered saline ( PBS ) followed by 20 mL of 4% paraformaldehyde ( PFA ) . Whole brains were dissected out and left to incubate for 24 h in 4% PFA at 4°C . For cryosectioning , tissue was cryoprotected in 30% sucrose overnight and embedded in Tissue Freezing Medium ( General Data TFM-5 ) . Cerebella were then cut sagittally down the middle , and embedded in Tissue Freezing Medium ( General Data TFM-5 ) . Tissue was sectioned at 20–30 μm thickness on a Leica Cyrostat ( model CM3050S ) . Western blot methods were done as previously described ( Bouskila et al . , 2011 ) ( Bowie et al . , 2018 ) . Briefly , for tissue which was being used to quantify levels of TTBK2 , a buffer containing 50 mM Tris/HCl , pH 7 . 5 , 1 mM EGTA , 1 mM EDTA , 1 mM sodium orthovanadate , 10 mM sodium-2-glycerophosphate , 50 mM sodium fluoride , 5 mM sodium pyrophosphate , 0 . 27 M sucrose , 1 mM benzamidine and 2 mM PMSF , supplemented with 0 . 5% NP-40 and 150 mM NaCl was used . For other tissue samples a buffer containing 10 mM Tris/Cl pH 7 . 5 , 150 mM NaCl , 0 . 5 mM EDTA , 1% Triton , 1 mM protease inhibitors ( Sigma #11836170001 ) and 25 mM β-glycerol phosphate ( Sigma 50020 ) was used . Total protein concentration was determined using a BSA Protein Assay Kit ( Thermo Fisher #23227 ) . For western blots , 15 μg of protein lysate was used for detection . All quantification of cerebellar tissue was done using ImageJ software . Images taken for quantification of cilia abundance were 10 μm z-stacks taken at 63x in four distinct folia regions of the cerebellum , two rostral and two caudal ( specifically , the outer edge of folia I/II , the internal zone between folia III and IV , the tip of folia VI , and the outer edge of folia IX on a sagittal section were imaged for cilia quantification ) . The Purkinje cell layer was placed into the middle of the image with equal distance above and below for quantification . Per animal , four sections were scored each and three animals were included in all quantifications . These cilia were the same population taken for cilia length measurements as well . For the molecular layer thickness , images were taken at 20x along the entirety of a primary fissure . A line was drawn from the base of the molecular layer to the pial surface , and a measurement was recorded . For this same line , the top of the line measurement was then brought down to the distal extent of the VGLUT2 synapse area , and a measurement recorded . For consistency , only the caudal side of the folia was measured . For the VGLUT2 puncta analysis , the ‘Analyze Particles’ function in ImageJ was used . Each image for the VGLUT2 puncta analysis was taken at 63x on the caudal side of the primary fissure , a 10 μm z-stack was made , and the image quantified . For the quantification , each stack was made into a black and white image , where the VGLUT2 puncta were black against a white background . Thresholding was performed , and the Analyze Particle function used . These measurements were routinely tested against user ROI counting to confirm accuracy . Four cerebellar slices were imaged per animal , and three animals were included in the analysis . For the area measurements of the ION nuclei , the ION was identified by cells that were positive for both NeuN and Calbindin as well as location within ventral medulla in which these cells reside . Images used for the NeuN area analysis were taken at 20x . A 10 μm z-stack image was made , and using the line tool , outlines were carefully drawn around the NeuN positive neuron and the area recorded . Per animal , over 150 cells were measured and three animals were included in the quantification . Glial fibers were assessed as previously described ( Furrer et al . , 2011 ) . Briefly , a 100 μm horizontal line was drawn 50 μm below the pial surface of the primary fissure folia . Glial fibers which crossed this 100 μm were scored . Per animal , 36 measurements were made and three animals were included in the quantification . The following antibodies and dilutions were used in this study: mouse anti-ARL13B ( NeuroMabs N295B/66 , 1:500 ) , rabbit anti-ARL13B ( gift from Tamara Caspary , 1:500 , and Proteintech 17711–1-AP , 1:500 ) , mouse anti-gamma-Tubulin ( Sigma T6557 , 1:1000 ) , rabbit anti-Calbindin D28K ( Cell Signaling Technologies 13176S , 1:250 ) , guinea pig anti-Calbindin D28K ( Synaptic Systems 214–004 , 1:200 ) , guinea pig anti-VGLUT2 ( EMD Millipore AB2251 , 1:2500 ) , rabbit anti-NeuN ( Abcam ab177487 , 1:1000 ) , DAPI ( Sigma D9542 , 1x ) , rabbit anti-AC3 ( Santa Cruz SC-588 , 1:10 - discontinued ) , rabbit anti-AC3 ( Abeomics 34–1003 , 1:100 ) , chicken anti-GFAP ( EMD Millipore AB5541 , 1:500 ) , rabbit anti-FoxP2 ( Abcam , ab106046 , 1:400 ) , rabbit anti-IP3 ( Abcam , ab108517 , 1:200 ) , and mouse anti-AT8 ( Thermo Scientific MN1020 , 1:100 ) . For immunostaining cerebellar tissue , sections were rinsed in 1xPBS to remove OCT and permeabilized in 0 . 2% PBS-T ( PBS + 0 . 2% Triton X-100 ) for 10 min , and then rinsed 3 × 5 min in PBS before the blocking step . Blocking solution contained 5% serum , 1% BSA made up in 0 . 1% PBS-T , and sections were incubated at room temperature in blocking solution for 1 h . Primary antibodies were used at indicated dilutions and incubated at 4°C overnight . Following primary antibody incubation , slides were rinsed 3 × 5 min in 1xPBS and secondary antibodies were used to detect epitopes . All secondary antibodies were supplied from Life Technologies . Secondary antibodies incubated for 1–3 h at room temperature . Following secondary antibody incubation , slides were rinsed 3 × 5 min in 1xPBS and mounted with either ProLong Gold antifade reagent ( Invitrogen P23930 ) . For Golgi staining of cerebellar tissue , FD Rapid Golgistain kit ( FD Neurotechnologies ) was used , according to manufacturer’s instructions . 100 μm sections were made after staining . Imaging of Golgi stained PCs was completed using a Zeiss Axio Imager with a 40x objective . Z-stacks containing entire PCs were used for quantification , and proximal dendrites were chosen that did not intersect with other dendrites for clear quantification . Using ImageJ software , spines were counted per length of measurement and reported as number of spines per micron . A rotarod performance test was completed with help from the Duke University Mouse Behavioral and Neuroendocrine Core Facility . Testers were blind to mouse genotype before beginning any experiments . The accelerating rotarod testing was performed the day before steady state rotarod testing . All accelerating tests were conducted at a speed that increased from 4 RPM to 40 RPM over 5 min . All steady speed tests were conducted at 32 RPM . Four trials were conducted per test . A trial was stopped after 300 s maximum time had elapsed for mice that did not fall off the rotarod during testing . Mice were aborted from the trial run if they held onto the rotarod for three full rotations . Mice were given 30 min between trials to rest , and four trials were completed per test . Statistical analyses , p-values , and experimental numbers for all experiments are outlined in respective figure legends . Analyses were performed using Graph Pad Prism 8 .	Many mammalian cells have a single hair-like structure , known as the primary cilium that projects away from the surface of the cell . This small projection from the membrane regulates many signaling pathways , particularly during embryonic development . However , most of the neurons in the adult brain also have primary cilia , and it is not yet understood what the role of the primary cilium has in maintaining most adult tissues . The primary cilium needs the protein TTBK2 to assemble , and mutations in the gene that codes for this protein cause a neurodegenerative disorder that first appears in adulthood known as spinocerebral ataxia type 11 ( SCA11 ) . People with this disease have a movement disorder caused by the loss of neurons called Purkinje cells in the cerebellum . In 2018 , researchers showed that mutated versions of TTBK2 associated with SCA11 interfere with the role of normal TTBK2 in assembling the cilium . But it was unclear whether primary cilia are required for the survival of Purkinje cells in the cerebellum . Now , Bowie and Goetz ( who are two of the researchers that conducted the 2018 study ) have found that deleting the gene that codes for TTBK2 in the brain of adult mice leads to the loss of cilia , followed by impaired movement . Additionally , the connections between Purkinje cells and other neurons are lost , and Purkinje cells eventually degenerate and die . If the cilia are removed using a different mechanism , the results are the same , showing for the first time that primary cilia are important to keep Purkinje cells alive and connected to other neurons . These results shed light on the roles of primary cilia within adult tissues , and provide insight into the mechanisms underlying SCA11 , a neurodegenerative disease for which no treatment currently exists . In the future , it will be important to extend the results of this study to other types of neurons affected in different neurodegenerative conditions . Ultimately , this line of research could lead to uncovering the causes of certain neurodegenerative disorders and provide new paths to treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2020	TTBK2 and primary cilia are essential for the connectivity and survival of cerebellar Purkinje neurons
Evaluation of sepsis-induced immunoparalysis has highlighted how decreased lymphocyte number/function contribute to worsened infection/cancer . Yet , an interesting contrast exists with autoimmune disease development , wherein diminishing pathogenic effectors may benefit the post-septic host . Within this framework , the impact of cecal ligation and puncture ( CLP ) -induced sepsis on the development of experimental autoimmune encephalomyelitis ( EAE ) was explored . Notably , CLP mice have delayed onset and reduced disease severity , relative to sham mice . Reduction in disease severity was associated with reduced number , but not function , of autoantigen ( MOG ) -specific pathogenic CD4 T cells in the CNS during disease and draining lymph node during priming . Numerical deficits of CD4 T cell effectors are associated with the loss of MOG-specific naive precursors . Critically , transfer of MOG-TCR transgenic ( 2D2 ) CD4 T cells after , but not before , CLP led to EAE disease equivalent to sham mice . Thus , broad impairment of antigenic responses , including autoantigens , is a hallmark of sepsis-induced immunoparalysis . Sepsis poses both a significant health concern , affecting 1 . 7 million and killing 270 , 000 Americans yearly , and economic burden ( > $20 billion annually ) ( CDC , 2020 ) . Sepsis is characterized by cytokine storm , a maladaptive response to a mismanaged infection , that is comprised of both pro- and anti-inflammatory cytokines . Although the acute cytokine storm is a grave situation , patients who survive a septic event are susceptible to further complications with increased susceptibility to unrelated secondary infection , increased viral reactivation , and decreased 5-year survival relative to non-septic patients ( Dombrovskiy et al . , 2007; Donnelly et al . , 2015; Gaieski et al . , 2013; Kutza et al . , 1998; Walton et al . , 2014 ) . These factors have been attributed to the sepsis-induced lymphopenic state and functional deficits of the surviving cells ( Hotchkiss et al . , 2006; Hotchkiss et al . , 2005; Hotchkiss et al . , 2001 ) . This state of immunoparalysis is so profound that the majority of sepsis-associated mortality has shifted to be late deaths following secondary infections or other morbid conditions as a result of immunologic impairment ( Delano and Ward , 2016 ) . Therefore , emphasis has been strongly shifted toward determining the mechanisms by which sepsis impairs the immune response to subsequent infection or cancer ( Chen et al . , 2019; Condotta et al . , 2015; Condotta et al . , 2013; Danahy et al . , 2019a; Jensen et al . , 2018b ) . Yet , this focus on the sepsis-induced loss of beneficial host responses has come at the expense of understanding how sepsis may influence other subsequent maladaptive immune responses . As opposed to infection/cancer , wherein effectors promote disease control , autoantigen-specific effectors promote disease and are detrimental to the host . Multiple sclerosis ( MS ) is a chronic neuroinflammatory disease of the CNS and is potentially the most common non-traumatic cause of CNS disability in young adults ( Compston and Coles , 2002 ) . MS poses a major personal and economic burden; it effects mostly women and on average presents at age 30 , an age crucial for family planning ( Fox , 2004 ) . The etiology of MS involves a complex interaction between genetic and environmental components , resulting in a clinical presentation including , but not limited to , sensory , motor , and/or cognitive deficits ( Dendrou et al . , 2015; Freedman et al . , 2018 ) . MS is thought to result from autoreactive CD4 T cell responses to the myelin antigens in the CNS followed by inflammatory cellular infiltration , demyelination , and neurodegeneration . In experimental autoimmune encephalomyelitis ( EAE ) , a murine model of MS , this role for CD4 T cells is well-established and induced by immunization against myelin antigens ( Stinissen et al . , 1997 ) . While there is a dearth of knowledge regarding the impact of sepsis on autoimmunity , descriptions of how sepsis influences specific cell subsets can provide insight into how autoimmune disease may be influenced . With regard to the critical role for CD4 T cells in EAE disease , the sepsis-induced lymphopenic state is known to impact both naive and memory CD4 T cells ( Cabrera-Perez et al . , 2016; Cabrera-Perez et al . , 2015; Chen et al . , 2017; Jensen et al . , 2018a; Sjaastad et al . , 2020b ) . In particular , sepsis can diminish the number of naive CD4 T cell precursors subsequently limiting the number of cells capable of responding to a given antigen ( Cabrera-Perez et al . , 2015; Martin et al . , 2020 ) . Additionally , sepsis can impair the effector function of surviving T cells further limiting the capacity of cells to mount an antigen-specific response ( Mohr et al . , 2012; Pötschke et al . , 2013; Sjaastad et al . , 2018 ) . Sepsis can also impair antigen-specific T cell responses through various cell extrinsic factors , including diminished number/function of dendritic cells ( DCs ) and diminished capacity of endothelial cells to promote chemotaxis ( Danahy et al . , 2017; Strother et al . , 2016 ) . Cumulatively , these findings suggest that , while detrimental to control of pathogens/cancer , sepsis may alternately diminish a host’s capacity to develop autoimmunity . Herein , we demonstrate that following cecal ligation and puncture ( CLP ) , mice have diminished EAE disease severity . Autoantigen-specific effector CD4 T cells were reduced in CLP hosts with EAE due to a loss of their naive precursors , which was the determining factor in the impediment of EAE disease . These data further define how the sepsis-induced immunoparalysis reduces host capacity to generate antigen-specific responses , regardless of whether it is a foreign antigen or an endogenous autoantigen . The influence of the immunoparalysis phase of sepsis , which follows the resolution of the cytokine storm , on the ability of the host to develop autoimmunity is undefined . To address this relationship , we employed well-established models of polymicrobial sepsis and inducible autoimmunity , CLP and EAE , respectively . CLP effectively mimics the pathophysiology of acute peritonitis and corresponding septic inflammation , with the capacity to modulate disease severity through adjustment of surgical parameters ( Dejager et al . , 2011; Sjaastad et al . , 2020a ) . Likewise , EAE presents an ideal model system to study the impact of sepsis on the development of autoimmune disease since it is a robust , inducible system allowing for manipulations prior to disease initiation ( Stinissen et al . , 1997 ) . CLP , sham , and non-surgery mice were immunized with MOG35-55 peptide 5 days post-surgery , a time when the cytokine storm has resolved ( Danahy et al . , 2019b ) , to induce EAE and disease progression was monitored ( Figure 1a ) . Of note: unless otherwise stated , experimental timepoints are defined with respect to EAE immunization . Importantly , while sham and no surgery control groups of mice developed robust EAE clinical disease , CLP survivors ( day 5 CLP group ) had substantially diminished disease severity ( Figure 1b , c ) and delayed disease onset ( Figure 1d ) . Further , CLP hosts do not achieve similar disease severity even when disease course is prolonged ( 40 days ) ( Figure 1—figure supplement 1 ) indicating that diminished disease is not due to delayed disease onset . Sepsis can lead to long-term functional deficits by immune cells ( Jensen et al . , 2018a ) . Therefore , to address whether the CLP-induced reduction in EAE disease was a durable phenomenon , we included a group of mice that had undergone CLP surgery 25 days prior to EAE immunization and evaluated the development of EAE disease ( Figure 1a ) . Interestingly , even at this later time point CLP survivors continued to exhibit reduced disease severity ( Figure 1b , c ) and delayed disease onset ( Figure 1d ) relative to sham or no surgery controls , though disease severity was increased relative to day 5 CLP mice . The more severe disease relative to day 5 mice was due to a noted variability among the day 25 CLP mice , wherein some individuals developed severe disease and others had minimal disease . This difference in EAE severity suggests that the influence of the immunoparalysis state may wain with time , at least in some of the mice . Therefore , to capture the maximal state of suppression , we focused on comparisons of day five sham and CLP hosts . It is important to stress that sham surgery is the preferred control for assessing how the septic event influences the development of subsequent T-cell-mediated immune responses , since those mice experience the same surgical procedures as CLP mice ( excluding those that initiate the septic event ) ( Cuenca et al . , 2010; Sjaastad et al . , 2020a ) . In particular , anesthesia , such as the ketamine used for the surgical procedures , can influence CD4 T cell function and capacity to induce EAE ( Hou et al . , 2016; Lee et al . , 2017; Ohta et al . , 2009 ) . Thus , to ensure that differences observed were due to the septic event , rather than being imposed by surgery , we utilized sham surgery as a relevant control to CLP procedure for the remainder of our experiments . To further address how sepsis altered the systemic parameters of disease , we assessed serum cytokine/chemokine concentrations in sham and CLP hosts both prior ( day 0 ) to and during EAE disease ( day 15 post-EAE induction ) . Notably , the differences in clinical scores between sham and CLP hosts were associated with an altered pattern of expression by various systemic cytokines following EAE immunization ( Figure 1—figure supplement 2a ) ; this included numerous effector cytokines associated with EAE disease development ( Figure 1—figure supplement 2b–i ) . Given the difference observed in disease between sham and CLP mice , in conjunction with an altered cytokine response , histopathologic scoring of the parenchymal and meningeal regions of the spinal cord during active disease was performed . Corresponding with their milder disease , CLP mice with EAE had less CNS pathology as highlighted by reduced inflammatory cell infiltrate ( Figure 1e , H and E ) in both the parenchyma ( Figure 1f ) and meninges ( Figure 1g ) of the spinal cord , the primary target of EAE inflammation . Further , CLP mice had milder axonal demyelination ( Figure 1e , LFB ) and vacuole formation , factors associated with EAE-induced paralysis . The difference in infiltrate observed extended to a reduced frequency and number of both microglia and infiltrating monocytes and macrophages present within the CNS in CLP hosts , relative to their sham counterparts ( Figure 1—figure supplement 3a–e ) . A significant reduction in MHC II expression was also observed on both microglia and infiltrating monocytes/macrophages in CLP hosts compared to sham-treated mice ( Figure 1—figure supplement 3f , h ) . This reduction compounded with the already reduced number of myeloid cells leading to ~100 fold reduction in MHC II-expressing cells in CLP hosts ( Figure 1—figure supplement 3g , i ) . Prior studies have demonstrated alterations in MHC expression by myeloid cells which are dysregulated following sepsis ( Jensen et al . , 2020; Monneret et al . , 2006; Siegler et al . , 2018 ) . MHC II expression is not only a marker of activation of these cell populations but also serves to present antigen to CD4 T cells , including autoreactive T cells that are critical mediators of disease in EAE . Therefore , this global infiltrate reduction in CLP hosts may reflect the lack of an autoimmune response in the CNS for which CD4 T cells are critical mediators . To address whether the reduction in histological inflammatory cell infiltrate reflected a difference in the accumulation of pathogenic antigen-specific CD4 T cells , total CNS ( brain and spinal cord ) was harvested from sham and CLP mice , following perfusion , at day 15 post-EAE induction ( Figure 2a ) . Flow cytometric analysis of the CNS revealed a decreased frequency ( Figure 2b , c ) and number ( Figure 2d ) of antigen-experienced CD4+ CD11a+ cells in the CNS ( Christiaansen et al . , 2017 ) . Further , tetramer staining of CD4 T cells reactive to the MOG antigen revealed a further reduction in the frequency ( Figure 2e , f ) and number ( Figure 2g ) of this pathogenic effector population . These data suggest that the reduction in disease severity was potentially due to lack of autoantigen-specific CD4 T cells infiltrating the CNS; however , sepsis is also known to influence T cell function . Thus , direct ex vivo analysis of the MOG-specific CD4 T cells in the CNS was evaluated by intracellular cytokine staining at the same timepoint . No differences were observed in the frequencies of MOG-specific CD4 T cells producing pathogenic effector cytokines IFNγ , IL-17A , and TNFα without further stimulation directly ex vivo ( Figure 3a , b; Figure 3—figure supplement 1; Lee et al . , 2019 ) . However , the numerical deficit in MOG-specific CD4 T cells led to a significant reduction in the number of IFNγ- and TNFα-producing cells with a trending decrease in IL-17A-producing cells ( Figure 3c ) . This finding indicates the autoantigen-specific CD4 T cells present in the CNS after CLP surgery are functionally competent and potentially were the least impacted by the septic event . In summary , the data presented so far suggest that the numerical deficit in CNS-infiltrating pathogenic effectors might stem from insufficient priming and/or expansion of encephalitogenic CD4 T cells in the draining lymph node in the periphery . To address whether the CLP hosts had a priming deficit for autoantigen-specific CD4 T cells , MOG-specific CD4 T cells were evaluated in the draining inguinal lymph node ( iLN ) , 7 days after EAE induction ( Figure 4a ) . This is a time during which CD4 T cells are being primed and expanding that precedes the development of clinical disease ( Bischof et al . , 2004 ) . Similar to the CNS during peak disease , both the frequency ( Figure 4b ) and number ( Figure 4c ) of MOG-specific CD4 T cells were decreased in the iLN of CLP hosts . Since lymphopenia is a hallmark of sepsis and differentially affects CD4 T cell precursor populations ( Cabrera-Perez et al . , 2015 ) , a reduction in precursor numbers is one possibility for the decreased number of autoantigen-specific CD4 T cells in the iLN at the day 7 priming timepoint . Thus , the precursor frequency of MOG-specific naive CD4 T cells was interrogated through tetramer enrichment of splenic CD4 T cells 5 days after surgery ( i . e . the day of EAE induction , Figure 5a ) . This analysis revealed a significant numerical loss of MOG-specific CD4 T cell precursors in CLP hosts . These data suggest CLP-induced lymphopenia may protect against the development of EAE by reducing the number of naive autoantigen-specific CD4 T cells . The subsequent loss of these naive cells would explain the corresponding reduction in the number of effector cells present in the both the lymph node and CNS following immunization . To address whether the prior observation , that CLP hosts have a lasting decrease in the severity of EAE disease ( Figure 1b–d ) , is also associated with a lasting numerical deficit , the number of MOG-specific CD4 T cells was enumerated in sham and CLP hosts 25 days post-surgery . Consistent with our observation of disparate disease development among CLP hosts , we observed a corresponding bifurcation in the number of MOG-specific CD4 T cells in individual mice ( Figure 5d ) further suggesting that the ability to initiate EAE disease might correspond to the number of MOG-specific CD4 T cells available . Additionally , similar to prior reports ( Cabrera-Perez et al . , 2016; Jensen et al . , 2018a; Skirecki et al . , 2020 ) , an increased frequency of activated ( CD44hi ) CD4 T cells was observed in CLP hosts ( Figure 5e ) . Importantly , this extended to an increase in the frequency of memory-like CD44hi cells among MOG-specific CD4 T cells in CLP hosts ( Figure 5f ) . This increase in CD44 expression could occur either be through homeostatic proliferation following the severe lymphodepletion that occurs following a septic event or potentially through antigen release due to sepsis-induced tissue damage ( Cabrera-Perez et al . , 2016; Cabrera-Perez et al . , 2015 ) . Thus , even for those cells that do recover in numbers they are phenotypically and functionally distinct , which may further contribute to the differences in EAE disease development and progression following immunization . To further delineate the contributions of the sepsis influence directly on CD4 T cells or their environment , we utilized an adoptive transfer system wherein naive MOG-specific , TCR-transgenic 2D2 CD4 T cells were transferred into congenically distinct recipient mice . By having the fixed TCR of the 2D2 cells , the function of the 2D2 cells could be equally assessed between sham and CLP hosts . Mice receiving 2D2 cells 1 day prior to surgery ( 6 days prior to EAE induction ) were part of the ‘pre-transfer’ group . Alternatively , a second cohort of mice received 2D2 cells 4 days post-surgery ( 1 day prior to EAE induction ) , a time at which sepsis-associated inflammation has resolved and the immunoparalysis state is established ( Danahy et al . , 2019b ) , were in the ‘post-transfer’ group . 2D2 cells in the pre-transfer group are exposed to both the intrinsic and extrinsic changes that sepsis induces . In contrast , the 2D2 cells in the post-transfer experiments were only influenced by the sepsis-induced changes in the environment , such as reduced DC function ( Strother et al . , 2016 ) , reduced capacity to traffic across endothelium ( Danahy et al . , 2017 ) , and alterations in the proportion of regulatory CD4 and CD8 T cells ( Sharma et al . , 2015; Sinha et al . , 2015 ) . These impairments are all likely to be present in the context of post-septic environment and may influence the development of EAE . Thus , the post-septic environmental factors which influence CD4 T cells can be assessed by transferring cells into the host after the immunoparalysis state has been established . Further , by comparing the pre- and post-transfer cells from sham and CLP mice the intrinsic and extrinsic influences of sepsis can be parsed . Thus , to address the intrinsic and extrinsic impact of sepsis on priming of CD4 T cells , 2D2 cell number , proliferation , as assessed by recent proliferation marker Ki67 ( Miller et al . , 2018 ) , and apoptosis , as assessed by presence of active caspase 3/7 ( FLICA+ ) and membrane depolarization ( PI+ ) , were evaluated in the iLN 7 days post-EAE induction ( Figure 6a , b ) . Notably , cells that were transferred prior to surgery were numerically diminished in CLP hosts relative to sham counterparts ( Figure 6c ) recapitulating the observations with endogenous MOG-specific CD4 T cells ( see Figure 5 ) . Strikingly , 2D2 cells that were transferred after surgery were numerically equivalent between sham and CLP hosts , suggesting the septic environment ( e . g . diminished DC number/function , increased proportion of regulatory CD8 and/or CD4 T cells ) did not limit their capacity to expand ( Figure 6c ) . Further , the transferred 2D2 cells from all groups had a similar proportion of recently proliferated as well as apoptotic ( FLICA+ PI+ ) cells ( Figure 6d , e ) . Given that the post-transfer groups had equivalent cell numbers , proliferation , and apoptosis; these data demonstrate that autoantigen CD4 T cells are not limited by the post-septic environment in their capacity to be primed and expand . Thus , while sepsis notably induces a variety of T-cell-extrinsic immunologic impairments ( Danahy et al . , 2017; Sharma et al . , 2015; Sinha et al . , 2015; Strother et al . , 2016 ) , these impairments can be overcome in the context of EAE immunization . In addition , there was no observed difference in the expression of several canonical extrinsic death-inducing proteins known to limit the expansion of CD4 T cells ( i . e . Fas , FasL , TRAIL ) between sham and CLP hosts for either the pre- or post-transfer groups ( Figure 6—figure supplement 1 ) further supporting the notion that survival differences do not account for the numerical deficit in CD4 T cells exposed to sepsis . Therefore , pre-transfer 2D2 cells proliferate and die equivalently in both sham and CLP hosts , indicating that sepsis does not intrinsically impair the capacity of these cells to proliferate in response to cognate antigen recognition . To directly compare the survival and expansion potential of autoantigen-specific CD4 T cells , congenically distinct Thy1 . 1 2D2 mice underwent either sham or CLP surgery . The same number of 2D2 cells from sham and CLP mice ( 1:1 mix ) were then adoptively transferred into naive Thy1 . 2 C57Bl/6-recipient mice . A day later , recipients were either left unimmunized or immunized to induce EAE to address the survival and expansion potential of both subsets of 2D2 CD4 T cells , respectively . Survival potential was assessed in the lymph node 5 days post-transfer , while expansion was assessed at both 5- and 7 days post-transfer ( Figure 7a ) . Importantly , no difference was observed in either the survival or expansion potential of 2D2 cells acquired from CLP mice relative to those acquired from sham controls ( Figure 7b , c ) . Thus , these data demonstrate that CLP does not change the capacity of autoantigen-specific CD4 T cells to exert their effector function ( s ) upon transfer into a non-septic environment . Given the aforementioned differences in the systemic cytokine milieu between sham and CLP hosts following immunization ( Figure 1—figure supplement 1 ) , it is possible that autoantigen-specific CD4 T cells may be unable to form the relevant effector cell populations of Th1 , Th17 , and Treg cells known to influence EAE disease development ( Dendrou et al . , 2015 ) . To address this possibility , pre- and post-transfer groups were generated as in Figure 6 and the number of RORγT ( Th17 ) , Tbet ( Th1 ) , and FoxP3 ( Treg ) 2D2s were enumerated in the lymph node 7 days post-immunization ( Figure 6—figure supplement 2a ) . As shown in Figure 6 and Figure 6—figure supplement 1 , the numerical loss of 2D2 cells during sepsis influenced the number of each of these effector populations within the pre-transfer group but did not impact the number of post-transfer 2D2 cells for each of the effector populations ( Figure 6—figure supplement 2b , c ) . Therefore , relative to the pre-transfer group , the post-transfer 2D2s recovered their capacity to develop each of these populations ( Figure 6—figure supplement 2d ) . In contrast , endogenous CD4 T cells exposed to septic event remained impaired in their capacity to generate these effector populations ( Figure 6—figure supplement 2e ) . Finally , to determine whether the sepsis-induced loss of MOG-specific CD4 T cell precursors is causal in the reduced disease severity of EAE mice , the same 2D2 transfer and surgery groups were used as before and disease progression was monitored ( Figure 8a ) . Coinciding with the cellular expansion seen in the iLN , mice that received 2D2 CD4 T cells prior to surgery developed less severe disease than their sham counterparts ( Figure 8b , c ) , similar to results in Figure 1 . However , CLP mice that received 2D2 CD4 T cells post-surgery developed equivalent disease as sham counterparts ( Figure 8b , c ) . This result demonstrates that a determining factor by which sepsis limits EAE stems from a reduction in number of naive MOG-specific CD4 T cell precursors . Importantly , there was a numerical reduction in the CNS infiltration of both transferred 2D2 ( Figure 8d ) and endogenous MOG-specific CD4 T cells ( Figure 8f ) of CLP mice that received transfer prior to surgery demonstrating that both populations of cells were similarly influenced by the septic event . Conversely , there was not a numerical deficit in 2D2 CD4 T cells when transferred into CLP hosts post-surgery ( Figure 8e ) , while the number of endogenous MOG-specific CD4 T cells in the CNS in the same hosts were significantly reduced ( Figure 8g ) . These data indicate the impact sepsis has on the disease-causing capacity on MOG-specific CD4 T cells , depending on whether the given population was exposed to the septic event . Cumulatively , these data indicate that numerical loss , due to sepsis-induced lymphopenia , of naive autoantigen-specific CD4 T cell precursors is sufficient to explain the protective effect of CLP on EAE disease . Although the sepsis-induced cytokine storm remains a life-threatening condition , it has also become apparent that the aftermath of the septic event leads to significant changes in the immune systems of individuals who survive . Prior work has established how the sepsis-induced immunoparalysis state negatively impacts a host’s subsequent capacity to control infection/cancer . Yet , the consequence of sepsis on other disease states , in particular autoimmunity , remains poorly defined . Herein , we have characterized the septic impact on the development of an autoimmune disease . Although the impact of sepsis on autoimmunity is understudied , the known influences of sepsis on CD4 T cells and their environment provided a pivotal framework to begin interrogating this relationship ( Cabrera-Perez et al . , 2016; Cabrera-Perez et al . , 2017; Cabrera-Perez et al . , 2014; Cabrera-Perez et al . , 2015; Chen et al . , 2017; Hotchkiss et al . , 2001 ) . Our observation of reduced EAE disease severity in CLP mice , associated with the loss of MOG-specific naive CD4 T cell precursors , further extends the characterization of the immunoparalysis state as being non-permissive to the formation of antigen-specific T cell responses . However , the use of cell transfers may have aided in overcoming the T cell extrinsic factors , such as DC number/function and capacity of endothelia to promote cell trafficking . Therefore , these other factors may still be relevant to autoimmune disease development during the sepsis-induced immunoparalysis state . Pivotal to our findings were the enumeration of MOG-specific CD4 T cell precursors and use of TCR-transgenic 2D2 CD4 T cells . With these tools , we established that sepsis diminishes the number of MOG-specific CD4 T cell precursors but does not alter capacity of the surviving cells to proliferate ( either by intrinsic or extrinsic mechanisms ) or promote disease if numerically bolstered . This does not , however , suggest that intrinsic and extrinsic factors are non-consequential to CD4 T cells in individuals that survive a septic event . Rather , this is likely reflective of the system of disease induction used . While a large bolus of antigen and adjuvant in the immunization strategy is likely able to overcome the intrinsic and extrinsic defects observed in other systems , this proven experimental model has given us the ability to further interrogate how the sepsis-induced cytokine storm influences the naive CD4 T cell repertoire and development of autoimmunity . In the context of EAE , our findings suggest that the number of MOG-specific naive CD4 T cell precursors is a critical determining factor in the development of disease . The significance of naive myelin antigen-specific CD4 T cell precursors in the development of MS is highlighted by studies showing that genetic and environmental factors influence the MS incidence and severity through modulation of autoreactive naive T cell precursor frequency . Specifically , allelic differences in HLA-expression may influence the development of MS by differential generation of autoreactive naive CD4 T cells ( Patsopoulos , 2018 ) . Environmental factors can also influence the precursor frequency either by influencing central tolerance ( i . e . vitamin-D-regulated HLA DRB11501 expression ) or peripheral tolerance ( i . e . gut-microbe-regulated Treg frequency and function ) ( Freedman et al . , 2018; Ramagopalan et al . , 2009 ) . As antigen-specific CD4+ T cell responses play a significant role in the pathogenesis of MS , it is plausible that sepsis can modulate the development of MS and/or disease relapses . However , there are no clinical data on the effect of sepsis on the disease development or relapses in patients with MS . Additionally , we observed a numerical recovery of MOG-specific CD4 T cells with time after sepsis in some animals . Yet , the recovered MOG-specific CD4s from CLP hosts may still exhibit inherent differences from MOG-specific CD4s within sham counterparts ( e . g . potential reduced avidity of ‘recovered’ cells ) . Indeed , the acquisition of an antigen-experienced phenotype by MOG-specific CD4 T cells from CLP mice demonstrates one such difference that may influence the function of these CD4 T cells . Auto-reactive T cells from MS patients have several functional differences , including increased survival capacity , distinct transcriptomic profiles , and increased avidity for autoantigen ( Bieganowska et al . , 1997; Bielekova et al . , 2004; Cao et al . , 2015; Hong et al . , 2004 ) . Therefore , evaluating how sepsis alters the survival capacity , transcriptomic profile , and auto-antigen avidity of cells may further define/elucidate novel underlying mechanisms associated with the development of autoimmune disease . However , many other prominent questions from the sepsis perspective also remain unanswered . Foremost is the durability of sepsis-induced suppression on the development of autoimmunity and whether the causal impairment remains consistent over that time period . The reduced number of MOG-specific CD4 T cell precursors seems to be a factor at a time proximal to sepsis induction . Additionally , this appeared to extend to the later day 25 timepoint in some mice . A potential explanation for the duality in disease development at day 25 may be the severity of the septic event , though this remains unclear . However , previous work has described how some CD4 T cell precursors eventually recover to pre-sepsis numbers with time , whereas others remain numerically decreased or can even be numerically expanded ( Cabrera-Perez et al . , 2015 ) . In particular , those cells that had numerically expanded had encountered cognate antigen , so the numerical ‘recovery’ may alternately represent an antigen-specific response due to release of antigen during sepsis-induced cell death . This would be consistent with the increase in CD44 expression by MOG-specific cells . Further , T cells that expand via homeostatic proliferation in lymphopenic states , such as the post-septic environment , can gain an antigen-experienced phenotype and effector function ( Hamilton et al . , 2006 ) . Along with decreased barrier integrity after sepsis ( Honig et al . , 2016 ) , the potential of sepsis promoting autoimmune disease development remains an intriguing alternative to a continuation in the impediment of autoimmunity described here . Yet , given that not all mice develop severe disease this may indicate that at this late timepoint additional and potential novel factors may be at play in the suppression of autoimmune disease development . Interrogating the impact of sepsis on other autoimmune diseases , in particular those with different effector cells , rather than the CD4 T cells described here , could provide further robust characterization of the sepsis-induced immunoparalysis state . Characterization of how sepsis also influences cells that function to suppress autoimmune disease , such as CD4 Tregs which are more resistant to sepsis-induced numerical loss than other CD4 T cell subsets ( Cavassani et al . , 2010; Monneret et al . , 2003; Scumpia et al . , 2006; Sharma et al . , 2015 ) , would also provide an interesting angle for interrogating the immunoparalysis state . In addition , neuro-regulatory CD8 T cells have recently been described to contribute to protection against EAE ( Sinha et al . , 2015 ) . While undescribed in the context of sepsis , the influence of sepsis on these cells may have divergent outcomes for the host . If , similar to CD4 Tregs , these neuro-regulatory CD8 T cells are more resistant to sepsis-induced lymphopenia this may be an additional mechanism by which autoimmune disease could be suppressed . Alternately , if their function is impaired by sepsis , similar to known impacts on effector CD8 T cells ( Duong et al . , 2014 ) , this could potentiate or exacerbate disease when precursor loss is not the dominant factor . Another factor to consider is that sepsis alters the composition of the host intestinal microbiome ( Alverdy and Krezalek , 2017; Krezalek et al . , 2016; Zaborin et al . , 2014 ) . Environmental factors that the host experiences , such as the microbiome , account for ~70% of MS disease risk ( Kuusisto et al . , 2008 ) , and there is growing evidence that the host’s intestinal microbiome actively influences MS disease ( Chen et al . , 2016; Freedman et al . , 2018 ) . Thus , interrogation into how the sepsis-induced changes in intestinal microbiota correspond with microbiomes associated with MS may elicit further insight into the interaction between these disease states . Finally , the mechanism/timing of the initiation of other autoimmune model systems should also be a strong consideration as it could inform on different aspects of the immunoparalysis state and generate distinctions between newly developed and established autoimmunity . The novel characterization of how infection , in the form of a septic insult , can dramatically influence the development of autoimmunity , presented here , reframes the complexity of the immunoparalysis state . By interrogating the impact of sepsis on inflammatory states beyond infection/cancer , additional mechanisms of sepsis-induced impairment may become apparent . Further , by understanding how sepsis influences these diseases , insights into the mechanisms that underlie their pathologies might also be illuminated . Thus , a final pertinent direction for future study would be to understand the interaction between sepsis and autoimmunity in patient cohorts . While it appears that patients with autoimmunity , and MS in particular , have a higher incidence of sepsis ( Capkun et al . , 2015; Nelson et al . , 2015 ) , the reverse scenario is still unexamined at this time . The added burden of sepsis in patients with autoimmunity is likely associated with the inflammatory status of the host , similar to the increased susceptibility of ‘dirty’ mice to a septic event ( Huggins et al . , 2019 ) . However , similar to our results , the immunoparalysis state may , in fact , benefit those patients who survive the cytokine storm by reducing the function of the pathogenic autoreactive cells . Such an outcome would be further instructive in understanding the interplay between autoimmunity and the infectious events ( e . g . sepsis ) . Inbred C57Bl/6 ( B6; Thy1 . 2/1 . 2 ) were purchased from the National Cancer Institute ( Frederick , MD ) and maintained in the animal facilities at the University of Iowa at the appropriate biosafety level . 2D2 ( C57BL/6-Tg ( Tcra2D2 , Tcrb2D2 ) 1Kuch/J ) mice on the C57BL6/J background were purchased from Jackson Laboratories and bred with C57BL6/J ( Thy1 . 1/1 . 1 ) , purchased from the National Cancer Institute ( Frederick , MD ) , to generate heterozygotes for 2D2 and Thy1 . 1 . F1 mice were bred together to generate 2D2 Thy 1 . 1/1 . 1 and 2D2 Thy1 . 1/1 . 2 mice . Expression of the 2D2 TCR and Thy1 . 1/1 . 1 were confirmed by flow cytometric staining . Mice were anesthetized with ketamine/xylazine ( University of Iowa , Office of Animal Resources ) , the abdomen was shaved and disinfected with Betadine ( Purdue Products ) , and a midline incision was made . The distal third of the cecum was ligated with Perma-Hand Silk ( Ethicon ) , punctured once using a 25-gauge needle , and a small amount of fecal matter extruded . The cecum was returned to abdomen , the peritoneum was closed with 641G Perma-Hand Silk ( Ethicon ) , and skin sealed using surgical Vetbond ( 3M ) . Following surgery , 1 mL PBS was administered s . c . to provide post-surgery fluid resuscitation . Lidocaine was administered at the incision site , and flunixin meglumine ( Phoenix ) was administered for postoperative analgesia . This procedure created a septic state characterized by loss of appetite and body weight , ruffled hair , shivering , diarrhea , and/or periorbital exudates with 0–10% mortality rate . Sham mice underwent identical surgery excluding cecal ligation and puncture . EAE was induced and evaluated as shown previously ( Mangalam et al . , 2009 ) . Briefly , mice were immunized s . c . on day 0 on the left and right flank with 100 µg of MOG35-55 emulsified in Complete Freund's Adjuvant followed by 80 ng of pertussis toxin ( PTX ) i . p . on days 0 and 2 . Disease severity was scored as follows: 0 , no clinical symptoms; 1 , loss of tail tonicity; 2 , hind limb weakness; 3 , hind limb paralysis; 4 , fore limb weakness; 5 , moribund or death . Single-cell suspensions from lymph nodes and spleens were generated after mashing tissue through 70-μm cell strainer without enzymatic digestion . To isolate CNS leukocytes , mice were anesthetized with CO2 and quickly perfused through the left ventricle with cold PBS . Brains were removed from the skull and spinal cords were flushed through the vertebral canal with cold RPMI media . To isolate immune cells from the CNS , brain and spinal cords were combined , homogenized , and isolated by Percoll gradient centrifugation . Following centrifugation , CNS leukocytes were collected from the interface , washed , and prepared appropriately for further use . Mice were euthanized using CO2 and intravascularly perfused using a gravity fed system with 10% neutral buffered ( 10% NBF ) formalin via intracardiac puncture . Spinal cords were then emersion fixed in 10% NBF for another 24–48 hr . Spinal cords were left in situ , demineralized with 14% EDTA for ~4 days and then embedded in paraffin and routine processed . Sections ( 4 µm thick ) were stained with hemotoxylin and eosin ( H and E ) and Luxol fast blue ( LFB ) and analyzed by a board-certified veterinary pathologist . Spinal cord sections were scored for cord pathology and meningeal inflammation . The meningeal score was a 0 to 4 scale where 0 = no pathology; 1 = rare , scattered , mild meningeal inflammatory cell infiltrates; 2 = mild , multifocal and obvious meningeal inflammatory cell infiltrates; 3 = multifocal to coalescing meningeal inflammatory cell infiltrates; 4 = marked , diffuse , thick bands of meningeal inflammatory cell infiltrates . The spinal cord score identifies how much of the cord at that level was affected and was also a 0 to 4 scale where 0 = no pathology; 1 = 1–25% of the spinal cord is affected with pathology consistent with EAE; 2 = 30–50% of the spinal cord is affected with pathology consistent with EAE; 3 = 60–90% of the spinal cord is affected with pathology consistent with EAE; 4 = >90% of the spinal cord is affected with pathology consistent with EAE . Flow cytometry data were acquired on a FACSCanto ( BD Biosciences , San Diego , CA ) and analyzed with FlowJo software ( Tree Star , Ashland , OR ) . To determine expression of cell surface proteins , mAb were incubated at 4°C for 20–30 min and cells were fixed using Cytofix/Cytoperm Solution ( BD Biosciences ) and , in some instances followed by mAb incubation to detect intracellular proteins . The following mAb clones were used: CD4 ( GK1 . 5 , Biolegend ) , CD11a ( M17/4 , Biolegend ) , IFNγ ( XMG1 . 2; eBioscience ) , IL-17A ( eBio17B7 , eBioscience ) , TNFα ( MP6-XT22 , eBioscience ) , CD8a ( 5H10-1 , Biolegend ) , Ki67 ( B56 , BD Pharmingen ) , Thy1 . 1 ( HIS51 , eBioscience ) , Thy1 . 2 ( 53–2 . 1 , eBioscience ) , CD44 ( IM7 , eBioscience ) , CD45 ( 104 , eBioscience ) , F4/80 ( BM8 , Biolegend ) , CD11b ( M1/70 , eBioscience ) , IA-b ( M5/114 . 15 . 2 , eBioscience ) , CD3e ( 145–2 C11 , eBioscience ) , CD19 ( MB19-1 , eBioscience ) , FAS ( SA367H8 , Biolegend ) , FASL ( MFL3 , Biolegend ) , TRAIL ( N2B2 , Biolegend ) , RORγT ( AFKJS-9 , eBioscience ) , Tbet ( eBio4B10 , eBioscience ) , FoxP3 ( FJK-16S , Invitrogen ) , and Dump [CD11b ( M1/70 ) , CD11c ( N418 ) , B220 ( RA3-6B2 ) , F4/80 ( BM8 ) , BioLegend] . Enrichment and detection of endogenous MOG-specific CD4 T cells: Biotinylated I-Ab molecules containing the MOG40-48 epitope covalently linked to the I-Ab β-chain were produced in Drosophila melanogaster S2 cell along with the I-Ab β-chain ( Moon et al . , 2007 ) . The monomers were purified , and then made into tetramers with streptavidin-phycoerythrin ( SA-PE; Prozyme ) . To enrich for Ag-specific CD4 T cells , tetramers ( 10 nM final concentration ) were then added to single-cell suspensions in 300 µl tetramer staining buffer ( PBS containing 5% FBS , 2 mM EDTA , 1:50 normal mouse serum , and 1:100 anti-CD16/32 mAb ) . The cells were incubated in the dark at room temperature for 1 hr , followed by a wash in 10 ml ice cold FACS Buffer . The tetramer-stained cells were then resuspended in 300 µl FACS Buffer , mixed with 25 µl of anti-PE mAb-conjugated magnetic microbeads ( StemCell Technologies ) , and incubated in the dark on ice for 30 min . The cells were washed , resuspended in 3 ml cold FACS Buffer , and passed through an EasySep Magnet ( StemCell Technologies ) to yield the enriched tetramer positive population . The resulting enriched fractions were stained with a cocktail of fluorochrome-labeled mAb: Thy1 . 2 , CD4 , CD8 , CD44 , ‘dump’ ( CD11b , CD11c , B220 , F4/80 ) , and tetramer . Cell numbers for each sample were determined using AccuCheck Counting Beads ( Invitrogen ) . Samples were then analyzed using a Fortessa flow cytometer ( BD ) and FlowJo software . Intracellular cytokine staining: For direct ex vivo staining , cells were incubated for one additional hour in the presence of Brefeldin A ( BFA ) before surface and intracellular cytokine staining . Ki67 staining: Following surface staining cells were fixed overnight with FoxP3 fixation/permeabilization buffer then stained with Ki67 . Propidium Iodide and active Caspase 3/7 staining: Vybrant FAM Caspase-3 and −7 assay kit ( Thermo-Fischer ) was used to identify apoptotic cells via expression of active caspase3/7 and propidium iodide according to the manufacturer’s instructions . Briefly cells were incubated with FLICA reagent for 30 min at 37°C followed by surface staining with antibodies as well as propidium iodide at 4°C for 20 min . Cells were immediately analyzed after staining without fixation by flow cytometry . For 2D2 transfers 200 μl of blood was collected from Thy1 . 1/1 . 1 or Thy1 . 1/1 . 2 2D2 mice in heparin-coated capillary tubes or spleens were harvested and homogenized . Red blood cells were lysed and the frequency of naive CD4 T cells was determined by flow cytometric analysis of a portion of the samples . Remaining cells were then enumerated and then adjusted to transfer 5 × 103 naive CD4 T cells per mouse prior to immunization or 5 × 106 naive CD4 T cells per mouse to assess cell survival in non-immunized mice . Cells were transferred via retroorbital injection . Multiplex cytokine analysis was performed via BioRad Bio-plex Pro Mouse Cytokine 23-plex according to the manufacturer’s instructions for plasma cytokine analysis . Multiplex was analyzed on BioRad Bio-Plex ( Luminex 200 ) analyzer in the university of Iowa Flow Cytometry core facility . Unless stated otherwise data were analyzed using Prism eight software ( GraphPad ) using two-tailed Student t-test ( for two individual groups , if variance was unequal variance then Mann-Whitney U test ) , one-way ANOVA with Bonferroni post-hoc test ( for >2 individual groups , if variance was unequal variance then Kruskal-Wallis with Dunn’s post-hoc test was used ) , two-way ANOVA ( for multiparametric analysis of two or more individual groups , pairing was used for samples that came from the same animal ) with a confidence interval of >95% to determine significance ( p<0 . 05 ) . Log-rank ( Mantel-Cox ) curve comparisons was used to determine significant difference in time to disease EAE disease onset ( *p<0 . 05 ) . Data are presented as standard error of the mean .	Sepsis is a life-threatening condition that can happen when the immune system overreacts to an infection and begins to damage tissues and organs in the body . It causes an extreme immune reaction called a cytokine storm , where the body releases uncontrolled levels of cytokines , proteins that are involved in coordinating the body’s response to infections . This in turn activates more immune cells , resulting in hyperinflammation . People who survive sepsis may have long-lasing impairments in their immune system that may leave them more vulnerable to infections or cancer . But scientists do not know exactly what causes these lasting immune problems or how to treat them . The fact that people are susceptible to cancer and infection after sepsis may offer a clue . It may suggest that the immune system is not able to attack bacteria or cancer cells . One way to explore this clue would be to test the effects of sepsis on autoimmune diseases , which cause the immune system to attack the body’s own cells . For example , in the autoimmune disease multiple sclerosis , the immune system attacks and destroys cells in the nervous system . If autoimmune disease is reduced after sepsis , it would suggest the cell-destroying abilities of the immune system are lessened . Using this approach , Jensen , Jensen et al . show that sepsis reduces the number of certain immune cells , called CD4 T cells , which are are responsible for an autoimmune attack of the central nervous system . In the experiments , mice that survived sepsis were evaluated for their ability to develop a multiple sclerosis-like disease . Mice that survived sepsis developed less severe or no autoimmune disease . After sepsis , these animals also had fewer CD4 T cells . However , when these immune cells were reinstated , the autoimmune disease emerged . The experiments help explain some of the immune system changes that occur after sepsis . Jensen , Jensen et al . suggest that rather than being completely detrimental , these changes may help to block harmful autoimmune responses . The experiments may also hint at new ways to combat autoimmune diseases by trying to replicate some of the immune-suppressing effects of sepsis . Studying the effect of sepsis on other autoimmune diseases in mice might provide more clues .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2020	Sepsis impedes EAE disease development and diminishes autoantigen-specific naive CD4 T cells
Colorectal cancer ( CRC ) organoids can be derived from almost all CRC patients and therefore capture the genetic diversity of this disease . We assembled a panel of CRC organoids carrying either wild-type or mutant RAS , as well as normal organoids and tumor organoids with a CRISPR-introduced oncogenic KRAS mutation . Using this panel , we evaluated RAS pathway inhibitors and drug combinations that are currently in clinical trial for RAS mutant cancers . Presence of mutant RAS correlated strongly with resistance to these targeted therapies . This was observed in tumorigenic as well as in normal organoids . Moreover , dual inhibition of the EGFR-MEK-ERK pathway in RAS mutant organoids induced a transient cell-cycle arrest rather than cell death . In vivo drug response of xenotransplanted RAS mutant organoids confirmed this growth arrest upon pan-HER/MEK combination therapy . Altogether , our studies demonstrate the potential of patient-derived CRC organoid libraries in evaluating inhibitors and drug combinations in a preclinical setting . One of the great challenges in targeted cancer treatment has been the development of effective RAS-targeting drugs . RAS mutations occur in about 15% of all human tumors ( Bos , 1989 ) and so far all attempts to selectively interfere in mutant RAS signaling have failed in the clinic ( Stephen et al . , 2014; Cox et al . , 2014 ) . Progress has long been impeded by the fact that the currently used model systems to pre-test drugs are insufficient: cell lines , on the one hand , have very limited genetic diversity , while mouse models on the other hand , may not represent human tumors ( Sachs and Clevers , 2014; Gould et al . , 2015 ) . Moreover , until recently , personalized medicine required large-scale in-vitro screening on short-term cultures of tumor sections ( Centenera et al . , 2013 ) , or alternatively , resource-intensive in-vivo screens using xenotransplantation of tumors into immunodeficient mice ( Jin et al . , 2010; Tentler et al . , 2012 ) . Recently , stem-cell based organoid technology was introduced to establish long-term cultures of both normal and tumor tissues from various organs ( Sato et al . , 2009 , 2011; Bartfeld et al . , 2015; Boj et al . , 2015; Huch et al . , 2015; Karthaus et al . , 2014; Gao et al . , 2014 ) . The advantage of this technology is that it can capture the genetic diversity of both normal and tumor tissues . Indeed , for colorectal cancer ( CRC ) a genetically diverse Biobank of patient-derived CRC organoids was established and used to integrate genomic data and monotherapy drug responses at the level of individual patient-derived organoid lines ( van de Wetering et al . , 2015 ) . We employed this biobank to further explore potential strategies to target mutant RAS , including the combination therapy of pan-HER and MEK inhibition , which is currently tested in clinical trials . We confirm the strong correlation between the presence of mutant RAS and resistance towards EGFR inhibition . Our data reinforce the notion that an oncogenic mutation in RAS is sufficient to confer this resistance independent of cellular status , whether it concerns normal or tumorigenic cells . Moreover , real-time imaging of the resistant drug response at the cellular level reveals predominant cell-cycle arrest in RAS mutant organoids , in contrast with the complete induction of cell death in CRC organoids with WT RAS . In vivo drug response of xenotransplanted RAS mutant CRC organoids confirmed the arrest in tumor growth upon dual inhibition of the EGFR-MEK-ERK pathway . Finally , efficient inhibition by dual targeting of the mutant RAS pathway strongly sensitizes for the induction of cell death , as illustrated by minimal addition of BCL inhibition . Our studies demonstrate the strong potential of patient-derived CRC organoid libraries in evaluating inhibitors and drug combinations in a preclinical setting . To explore drug responses of patient-derived CRC organoids towards combination therapies of targeted inhibitors of the EGFR-RAS-ERK pathway , we applied a drug sensitivity screen using EGFR-family and MEK inhibitors ( EGFRi and MEKi resp . ) either as mono or combination therapy on two cancer organoids from a previously established biobank of CRC organoids ( van de Wetering et al . , 2015 ) . To start , we chose cancer organoids from the individuals P8 and P26 , which share a similar composition of frequent cancer mutations such as functionally inactive APC and TP53 . However , they differ in their KRAS status . P8T contains wild-type ( WT ) KRAS , while P26T contains an oncogenic mutant version of KRAS ( G12V ) . 3D-organoids were challenged with drugs for 72 hr and drug responses were determined by quantifying cell viability through measurements of ATP levels using CellTiter-Glo ( van de Wetering et al . , 2015 ) . We observed the expected sensitivity of P8T towards afatinib ( irreversible EGFR/HER2 inhibitor ) and insensitivity of KRAS mutant P26T ( Figure 1A ) . Selumetinib ( MEKi ) as a monotherapy showed little efficacy in both P8T and P26T , but combination therapy confirmed previous findings that MEKi sensitizes RAS mutant tumor cells to EGFR/HER2 inhibition ( Figure 1A ) ( Sun et al . , 2014 ) . However , the KRAS mutant P26T organoids were less sensitive to the combination therapy than the KRAS WT P8T organoids . 10 . 7554/eLife . 18489 . 003Figure 1 . Drug responses of patient-derived CRC organoids with and without mutant KRAS . ( A ) Dose-response curves of patient-derived CRC organoids P8T ( KRASWT; APC and TP53 mutant ) and P26T ( KRASG12V; APC and TP53 mutant ) treated with the dual EGFR/HER2 inhibitor afatinib , MEK inhibitor selumetinib or a combination thereof . Cell viability was measured by an ATP-based assay after 72 hr of drug treatment . ( B ) Stills from representative time-lapse imaging ( three days ) of CRC organoids P8T and P26T treated with vehicle ( DMSO ) or a combination of targeted inhibitors afatinib and selumetinib ( both 1 µM ) ( see also Video 1 ) . In every panel , upper images show color-coded depth of maximum-projected z-stacks of H2B-mNeonGreen fluorescent organoids . Lower panels: corresponding transmitted light images . Time interval: 15 min . Scale bars: 20 µm . Representative time-lapse of two experiments is shown ( total six organoids/condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 00310 . 7554/eLife . 18489 . 004Figure 1—source data 1 . ImageJ/Fiji macro script: ‘Organoid movie macro’ . Converts XYZT confocal data sets into analyzable 2D-movies , consisting four quadrants: depth coding , maximum projection in ‘glow’ , transmitted light image and a merge between transmitted light and fluorescence . All supplementary movies were generated using this method ( Figures 1 and 3 show 2 of 4 quadrants only ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 00410 . 7554/eLife . 18489 . 005Figure 1—figure supplement 1 . Stills from representative time-lapse imaging ( three days ) of CRC organoids P8T and P26T treated with vehicle ( DMSO ) or a combination of targeted inhibitors afatinib ( 33 nM ) and selumetinib ( 200 nM ) ( see also Video 2 ) . In every panel , upper images show color-coded depth of maximum-projected z-stacks of H2B-mNeonGreen fluorescent organoids . Lower images: corresponding transmitted light images . Time interval: 15 min . Scale bars , 20 µm . Representative time-lapse of five organoids per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 005 To monitor drug response on a cellular level , we stably introduced DNA constructs encoding fluorescently-labeled H2B and performed real-time confocal imaging on the 3D-organoids for 72 hr in the presence and absence of drugs . We performed EGFR-RAS-ERK pathway inhibition with relatively high concentrations of afatinib ( 1 µM ) in combination with selumetinib ( 1 µM ) . In P26T ( mutant KRAS ) we only observed cell cycle arrest with very limited cell death induction . This was in stark contrast with the very rapid induction of cell death in P8T ( WT KRAS ) ( Figure 1B , Video 1 ) . When we repeated these imaging experiments using much lower drug concentrations , we noticed a general shift to resistance for both organoid lines . Under these conditions , also P8T predominantly showed cell cycle arrest rather than cell death , and the cancer cells in P26T organoids even continued to proliferate ( Figure 1—figure supplement 1 , Video 2 ) . Taken together , our data indicate that 72 hr of combination treatment with afatinib and selumetinib ( EGFRi/HER2i and MEKi ) effectively kills KRAS WT P8T organoids , while the mutant KRAS P26T organoids are significantly less sensitive . 10 . 7554/eLife . 18489 . 006Video 1 . Real-time imaging of cellular drug responses in tumor organoids using high concentrations targeted inhibitors . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 00610 . 7554/eLife . 18489 . 007Video 2 . Real-time imaging of cellular drug responses in tumor organoids using low concentrations targeted inhibitors . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 007 In order to validate the observed drug response of in vitro cultured organoids in an in vivo model , we xenotransplanted P18T and P26T tumor organoids in immunodeficient mice . In line with a previous report where only engineered tumor progression organoids with increasing number of cancer mutations ( APC , KRAS , P53 and/or SMAD4 ) showed efficient engraftment ( Drost et al . , 2015 ) , we only obtained reliable engraftment using P26T CRC organoids . We initially started using concentration schedules of afatinib and selumetinib that had previously been reported ( Sun et al . , 2014 ) , but we observed no significant effect of the drug combination on tumor growth over time ( Figure 2A ) . To exclude that the tumors had acquired resistance during the in vivo drug treatment , we isolated the tumors to re-establish secondary organoids and subjected these to identical drug tests . Dose-response curves on these secondary organoids were identical to the parental organoid line P26T , independent of the type of drug treatment that the tumors underwent in the mice ( Figure 2—figure supplement 1 ) . Indeed , in agreement with lower drug concentrations that proved to be ineffective in blocking proliferation in vitro ( Figure 1—figure supplement 1 , Video 1 ) , we speculate that the in vivo drug concentrations were insufficient to effectively block the EGFR-MEK-ERK pathway . To confirm this hypothesis , we further increased the drug levels to high but tolerable doses . This indeed induced significant growth stabilization ( but no regression ) of P26T xenotransplanted tumor in mice ( Figure 2B ) , in agreement with loss of proliferative activity as was also detected in vitro ( Figure 1B ) . The fact that in vivo xenografted CRC organoids yields similar drug responses as in vitro organoid cultures and identical to previous reported drug response of KRAS mutant PDX models of CRC ( Sun et al . , 2014 ) , validates the testing and evaluation of targeted inhibitors in CRC organoids . 10 . 7554/eLife . 18489 . 008Figure 2 . In vivo drug response of xenotransplanted CRC organoids . ( A ) P26T CRC organoids were subcutaneously transplanted in immunodeficient mice . Once tumors have grown to a volume of 300 mm3 , animals were treated for 28 days with vehicle , afatinib ( 12 , 5 mg/kg; five days on , two days off ) , selumetinib ( 20 mg/kg; five days on , two days off ) or both drugs in combination . The mean percentage change in tumor volume relative to initial tumor volume is shown . Error bars represent standard deviation . n . s . , not significant . ( B ) Same experimental setup as in A , but with increased drug concentrations for afatinib ( 20 mg/kg; five days on , two days off ) and selumetinib ( 25 mg/kg; five days on , two days off ) ; as well as in combined treatment . Error bars represent standard deviation . p<0 , 05; p<0 , 01; *p<0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 00810 . 7554/eLife . 18489 . 009Figure 2—figure supplement 1 . Secondary organoid cultures were derived from xenografted P26T tumors ( organoid-derived xenograft , ODX ) of mice that have been treated with vehicle , afatinib ( 12 , 5 mg/kg; five days on , two days off ) , selumetinib ( 20 mg/kg; five days on , two days off ) or both . Dose response curves were determined from these secondary post-xenograft cultures , as well as from the parental P26T organoid culture . Regardless of the in vivo applied drug treatment , the drug sensitivity phenotype of the organoids remained unaltered . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 009 P8T and P26T CRCs are microsatellite-stable ( MSS ) and belong to the same molecular subtype classification based on RNA expression data ( TA , also referred to as canonical CMS2 according to consensus classification ) ( van de Wetering et al . , 2015; Guinney et al . , 2015 ) . Genomic characterization of these patient-derived CRC organoids in comparison to their matched normal tissue revealed many additional mutations within the protein coding sequence of the genome ( van de Wetering et al . , 2015 ) . For P8T and P26T , 230 and 163 of such cancer specific mutations were detected respectively ( van de Wetering et al . , 2015 ) . To exclude potential contributions of all these additional mutations to the effect that oncogenic KRAS imposes on drug responses , we introduced an oncogenic KRAS mutation in patient-derived CRC organoid P18T via CRISPR/Cas9-mediated homologous recombination ( Drost et al . , 2015 ) . Like P8T , original P18T is WT for the entire downstream EGFR signaling pathway . P18T-KRASG12D mutant cells were generated as reported previously for normal colon organoids ( Drost et al . , 2015 ) and genotyping of clonally expanded organoids confirmed that the clones contained the KRASG12D mutation ( Figure 3A ) , as well as a Cas9-mediated inactivation of the second allele by introducing an 86 bp deletion . Upon addition of oncogenic KRAS , no overall differences in morphology or growth rates were observed during normal culture conditions . 10 . 7554/eLife . 18489 . 010Figure 3 . CRISPR genome editing in CRC organoids reveals effect of KRASG12D on drug response . ( A ) Schematic representation of the CRISPR/Cas9-induced homologous recombination strategy to introduce the KRASG12D mutation in the KRASWT patient-derived CRC organoid P18T . Green bar: start codon . Red bar: G12D mutation . Parental and mutant sequences are shown on the right . ( B ) Extensive dual-inhibitor dose-response assay of patient-derived CRC organoids P18T and P18T-KRASG12D treated for 72 hr . 14×14 drug concentrations of afatinib and selumetinib were chosen with logarithmic intervals covering a 5 nM–5 μM range . The results of the full matrix screen are represented as a heat map ( left ) , where red represents 0% ATP levels ( no viability ) and green represents 100% ATP levels ( max viability ) . The dose-response curves to the right represent the horizontal ( afatinib monotherapy ) , vertical ( selumetinib monotherapy ) and diagonal ( afatinib/selumetinib combination therapy ) lines in the heat maps . Dashed lines are P18T; solid lines are P18T-KRASG12D . ( C ) Stills from representative time-lapse imaging ( three days ) of CRC organoids P18T and P18T-KRASG12D treated with vehicle ( DMSO ) or afatinib + selumetinib ( both 1 µM ) ( see also Video 1 ) . In every panel , upper images show color-coded depth of maximum-projected z-stacks of H2B-mNeonGreen fluorescent organoids . Lower panels: corresponding transmitted light images . Time interval: 15 min . Scale bars: 20 µm . Representative time-lapse of 2 ( total eight organoids/condition ) and four experiments ( total 20 organoids/condition ) for P18T and P18T-KRASG12D resp . ( D ) Mitotic and apoptotic events in the organoid drug response movies ( C and Video 1 ) were manually marked and quantified ( see Materials and methods and Figure 3—figure supplement 3 ) . In comparison with vehicle ( - ) , drug treatment of p18T with afatinib and selumetinib ( a+s ) results in both proliferation block and apoptosis induction , while p18T-KRASG12D only shows reduced proliferation but unchanged apoptosis rates . Error bars represent standard deviation . p<0 , 05; **p<0 , 001; n . s . = not significant ( p=0 , 4 ) DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01010 . 7554/eLife . 18489 . 011Figure 3—figure supplement 1 . Original heat map of viability and heat map of calculated scores for p18T and p18T-KRASG12D . Positive Bliss scores ( red hues ) indicate combinations where the effect is greater than expected based on additive effects . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01110 . 7554/eLife . 18489 . 012Figure 3—figure supplement 2 . Quantifying life and death during real-time imaging of drug response . ( A ) Time-lapse XYZT acquisitions were visualized using color-coded depth projections and mitotic and apoptotic analysis were manually marked . ( B ) Increases ( mitoses ) and decreases ( apoptosis ) in cell numbers were chronologically ranked , thus reconstructing organoid size evolution in time . ( C ) Separate contributions of mitoses and apoptosis to development of P18T and P18T-KRASG12D while incubated with vehicle ( − ) or afatinib + selumetinib ( a+s ) . Whereas in P18T drug treatment ( a+s ) results in both proliferation block and apoptosis induction , P18T-KRASG12D only show reduced proliferation but unchanged apoptosis rates . p<0 , 05; **p<0 , 001; n . s . = not significant ( p=0 , 4 ) DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01210 . 7554/eLife . 18489 . 013Figure 3—figure supplement 3 . Drug response of CRC organoids as examined by Western blot . Combined Pan-HER/ MEK inhibition results in reduction of ERK phosphorylation in KRASWT and KRASG12D CRC organoids . Organoids were treated for 24 hr with MEK inhibitors selumetinib ( 1 μM ) , trametinib ( 10 nM ) , and the pan-HER inhibitor afatinib ( 1 μM ) as indicated . WB is representative of four independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01310 . 7554/eLife . 18489 . 014Figure 3—figure supplement 4 . Stills from representative time-lapse imaging ( three days ) of CRC organoids P18T and P18T-KRASG12D treated with vehicle ( DMSO ) or a combination of targeted inhibitors afatinib ( 33 nM ) and selumetinib ( 200 nM ) ( see also Video 2 ) . In every panel , upper images show color-coded depth of maximum-projected z-stacks of H2B-mNeonGreen fluorescent organoids . Lower images: corresponding transmitted light images . Time interval: 15 min . Scale bars , 20 µm . Representative time-lapse of five organoids per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 014 To investigate the exclusive effect of oncogenic KRAS on a combination therapy that targets the EGFR-RAS-ERK pathway , we performed a full matrix screen of 14 drug concentrations over a 5 nM to 5 µM range of both the targeted inhibitors afatinib ( EGFR/HER2i ) and selumetinib ( MEKi ) ( Figure 3B ) . Notably , their combined administration is currently used in a clinical trial for patients with RAS mutant CRCs ( NCT02450656 ) . While the original P18T demonstrated high sensitivity to EGFR/HER2 inhibition by monotherapy , a single introduced oncogenic point mutation in KRAS provided resistance to EGFR/HER2 inhibition . Moreover , we analyzed combination effects using the Bliss independence model . Positive Bliss scores indicate combinatorial effects that exceed additive effects . The heat map of Bliss scores for P18T and P18T-KRAS shows that a large range of concentrations for both compounds show positive scores , but that presence of oncogenic KRAS renders the loss of viability and positive Bliss range towards higher drug concentrations indicating resistance ( Figure 3—figure supplement 1 ) . Next , we again studied the cellular drug response by real-time imaging . Reminiscent of the patient-derived CRC organoid with an endogenous RAS mutation ( P26T ) , we noticed that the introduction of oncogenic KRAS renders a CRC organoid less sensitive to the afatinib/selumetinib combination therapy ( Figure 3C , Video 1 ) . More specifically , quantifications of all mitotic and apoptotic events during the filmed drug response revealed both loss of proliferation and apoptosis induction in P18T , while P18T-KRASG12D only showed reduced proliferation but unchanged apoptosis rates ( Figure 3D and Figure 3—figure supplement 2 ) . Despite the phenotypic difference in drug response , pERK levels in both tumor organoids were severely reduced ( Figure 3—figure supplement 3 ) . Since suboptimal suppression of ERK activity might permit tumor growth in BRAF mutant cancers ( Bollag et al . , 2010; Corcoran et al . , 2015 ) , we determined the cellular effects of drug response when lowering drug concentrations . Since significant differential effects were observed between P18T and P18T-KRASG12D during the matrix screen around 33 nM afatinib + 200 nM selumetinib ( Figure 3B ) , we repeated real-time imaging of drug response using these lower drug concentrations . As with P8T and P26T , we noticed a general shift from sensitivity towards resistance when drug concentrations were reduced . More specifically , the RAS WT cancer organoids showed cell cycle arrest rather than cell death , while the RAS mutant organoids appeared unaffected and sustained proliferation ( Figure 3—figure supplement 4 , Video 2 ) . Considering the isogenic CRC organoids P18T and P18T-KRASG12D as our gold standard to reveal the specific effects of KRASG12D on drug responses , we expanded our focus at targeting the linear EGFR-RAS-ERK pathway with the ultimate aim to find a targeted therapy that is specifically effective against RAS mutant CRCs . Multiple targeted inhibitors against identical targets were used to exclude artifacts and to increase the mechanistic significance behind the rationale of potential therapies ( Figure 4A; and Figure 4—source data 1 and Supplementary file 1 for all dose-response curves ) of which few combination therapies are in clinical trial ( Figure 4B ) . 10 . 7554/eLife . 18489 . 015Figure 4 . Differential drug sensitivities upon combination therapies including EGFR inhibition . ( A ) Heat map of dose-response measurements ( cell viability ) in CRC organoids P18T ( top panel ) and P18T-KRASG12D ( bottom panel ) . Organoids were treated ( 72 hr ) with vehicle ( DMSO ) or inhibitors targeting the EGFR-RAS-ERK pathway ( 5 nM – 20 μM range , in 22 logarithmic intervals ) . Red represents 0% ATP levels ( max cell death ) and green represents 100% ATP levels ( max viability ) . Drug names and their nominal targets are indicated in the left panel . Combination therapies that are currently in clinical trial for patients with RAS mutant CRCs are indicated in red font . See Figure 4—source data 1 and Supplementary file 1 for all dose-response curves . ( B ) Dose-response curves of CRC organoids P18T ( dashed lines ) and P18T-KRASG12D ( solid lines ) treated with combination therapies that are currently in clinical trial for patients with RAS mutant CRCs . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01510 . 7554/eLife . 18489 . 016Figure 4—source data 1 . Dose-response curves for patient-derived tumor organoids P18T and P18T KRASG12D as indicated . A number of biological replicates for each dose-response curve are indicated between parenthesis ( first monotherapy/ second monotherapy/ combination therapy ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01610 . 7554/eLife . 18489 . 017Figure 4—source data 2 . Dose-response curves for patient-derived tumor organoids P18T and P18T KRASG12D as indicated . A number of biological replicates for each dose-response curve are indicated between parenthesis ( first monotherapy/ second monotherapy/ combination therapy ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01710 . 7554/eLife . 18489 . 018Figure 4—figure supplement 1 . Upper panel: Heat map of all IC50 values ( Log10-scale ) for P18T and P18T-KRASG12D , determined from dataset shown in Figure 4A ( and Figure 4—source data 1 and Supplementary file 1 ) . IC50 values are color-coded: blue for highest and red for lowest efficacies ( see scale bar below ) . IC50 values account for both drugs when added in combination , except combinations that include trametinib ( IC50 values marked with ; here trametinib concentrations are two log decades lower as indicated ) . Lower panel: Differential IC50’s comparing CRC organoids P18T-KRASG12D and P18T ( ΔIC50 = IC50 ( P18T-KRAS ) – IC50 ( P18T ) . ΔIC50 values are color-coded , where higher efficacies in P18T are coded red and higher efficacies in P18T-KRASG12D are coded blue ( see scale bar ) . This data representation clearly illustrates that P18T organoids are much more sensitive to targeted therapies that include EGFR inhibition than KRAS mutant P18T . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 01810 . 7554/eLife . 18489 . 019Figure 4—figure supplement 2 . Drug combinations on P18T and P18T-KRASG12D organoids targeting EGFR-RAS-ERK and PI3K-AKT pathways . ( A ) Heat map of dose-response measurements ( cell viability ) in CRC organoids P18T ( top panel ) and P18T-KRAS ( bottom panel ) . Organoids were treated ( 72 hr ) with vehicle ( DMSO ) or targeted inhibitors over a 5 nM – 20 μM range ( in 22 logarithmic intervals ) . Red represents 0% ATP levels ( max cell death ) and green represents 100% ATP levels ( max viability ) . Drug names and their nominal targets are indicated in the left panel . ( B ) Heat map of all IC50 values ( Log10-scale ) for P18T and P18T-KRAS , determined from dataset shown in A ( and Figure 4—source data 2 and Supplementary file 1 ) . IC50 values are color-coded: blue for highest and red for lowest efficacies ( see scale bar below ) . IC50 values account for both drugs when added in combination . Differential IC50’s comparing CRC organoids P18T-KRAS and P18T ( ΔIC50 = IC50 ( P18T-KRAS ) – IC50 ( P18T ) . ΔIC50 values are color-coded , where higher efficacies in P18T are coded red and higher efficacies in P18T-KRAS are coded blue ( see scale bar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 019 First , we noticed a much lower sensitivity of P18T-KRASG12D for pan-HER inhibitors afatinib , lapatinib and dacomitinib in contrast to the parental P18T ( much lower IC50 , Figure 3—figure supplement 1 ) . Second , within P18T hardly any additive sensitivity could be observed when EGFRi was complemented with MEK or ERK inhibition ( Figure 4—figure supplement 1 ) . In contrast , dual-targeting strategies strongly enhanced efficacies in P18T-KRASG12D regardless which specific inhibitor combination was used ( Figure 4—figure supplement 1 ) . Nevertheless , all tested combinations that included EGFR inhibition revealed stronger negative effect on cellular viability in P18T than in P18T with mutant KRAS ( positive ΔIC50’s , Figure 4—figure supplement 1 ) . In contrast , most mono- and combination therapies against MEK and/or ERK that excluded EGFRi showed on average similar efficacies in P18T-KRASG12D as in P18T ( Figure 4—figure supplement 1 ) . In parallel , we tested dual-targeting strategies involving PI3K-AKT and EGFR-RAS-ERK pathways considering the interconnectivity between these pathways ( Figure 4—figure supplement 2A ) . Like MEK or ERK inhibition , we observed that pharmacological inhibition of PI3K or AKT in combination with anti-EGFR therapy did not enhance efficacy in a KRAS mutant background ( Figure 4—figure supplement 2B ) . In line with this , clinical studies focusing on combining MEK inhibitors with PI3K , AKT or mTOR inhibitors in KRAS mutant CRCs did not yield satisfactory results ( Shimizu et al . , 2012 ) . Next , we aimed to further establish whether the effects of oncogenic KRAS on drug response is dependent on a tumorigenic background or could manifest independent of cellular state . We therefore used normal colon organoids and a derivative of that line in which the oncogenic KRASG12D mutation was introduced via similar CRISPR/Cas9-mediated genome-editing strategy as in P18T ( Drost et al . , 2015 ) . In analogy with mouse studies ( Snippert et al . , 2014 ) , we observed no morphological alteration nor induction of senescence upon introduction of oncogenic KRAS ( Figure 5—figure supplement 1 ) . Strikingly , drug response profiles of normal organoids to targeted inhibitors against the EGFR-RAS-ERK pathway ( Figure 5 and Figure 5—figure supplement 2 ) revealed a similar trend as in CRC organoid P18T ( Figure 4 and Figure 4—figure supplement 1 ) . Thus , the effect that oncogenic KRAS imposes on drug response appears independent of cellular status and the presence of additional cancer mutations . 10 . 7554/eLife . 18489 . 020Figure 5 . Comparable drug response profiles in normal and tumorigenic background . ( A ) Heat map of dose-response measurements of cell viability in normal colon organoids ( top panel ) and in normal colon organoids with an oncogenic KRAS mutation ( bottom panel ) after 72 hr drug treatment with inhibitors targeting the EGFR-RAS-ERK pathway . Same concentration range and color-coding as in Figure 4 . Combination therapies that are currently in clinical trial for patients with RAS mutant CRCs are indicated in red . See Figure 5—source data 1 and Supplementary file 1 for all dose-response curves . ( B ) Dose-response curves of normal organoids ( dashed lines ) and normal organoids + KRAS ( solid lines ) treated with combination therapies that are currently in clinical trial for patients with RAS mutant CRCs . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02010 . 7554/eLife . 18489 . 021Figure 5—source data 1 . Dose-response curves for normal and normal KRASG12D organoids as indicated . Number of biological replicates for each dose-response curve are indicated between parenthesis ( first monotherapy/ second monotherapy/ combination therapy ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02110 . 7554/eLife . 18489 . 022Figure 5—figure supplement 1 . Comparison of normal organoids and normal organoids with an introduced oncogenic G12D mutation within the endogenous KRAS locus . No overall phenotypic differences are observed based on morphology and proliferative activity . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02210 . 7554/eLife . 18489 . 023Figure 5—figure supplement 2 . Upper panel: Heat map of all IC50 values ( Log10-scale ) for normal colon organoids with and without mutant KRAS , determined from dataset shown in Figure 5A ( and Figure 5—source data 1 and Supplementary file 1 ) . IC50 values are color-coded: blue for highest and red for lowest efficacies ( see scale bar below ) . IC50 values account for both drugs when added in combination , except combinations that include trametinib ( IC50 values marked with ; here trametinib concentrations are two log decades lower as indicated ) . Lower panel: Differential IC50’s comparing normal colon organoids with and without mutant KRAS . ΔIC50 values are color-coded , where relative higher efficacies in normal colon organoids are coded red , while relative higher efficacies in normal colon organoids +KRASG12D are coded blue ( see scale bar ) . The similarities between the ΔIC50 matrices in CRC and in normal organoids illustrate that the effect that KRAS imposes on drug response is independent of cellular background . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 023 Next , we aimed to extend our analyses towards a wider collection of CRC organoids that is more representative for the clinic . We screened 10 additional patient-derived CRC organoids for combinatorial therapies against the EGFR-RAS-ERK signaling pathway . Since all the organoid lines are fully characterized in terms of genome information , we could select CRC organoids with and without a mutant RAS pathway . Based on EGFR/HER2 dual inhibition by afatinib , we could clearly discriminate the organoid panel in two types of responders , namely the sensitive versus the resistant ones ( Figure 6A , green versus red lines respectively ) . Indeed , drug sensitivity towards all tested EGFR inhibitors clearly correlated with the mutational status of KRAS . However , there were two notable exceptions ( Figure 6A and Figure 6—figure supplement 1 ) . The first was P25T , which , although WT for KRAS , turned out to contain an oncogenic mutation in NRAS ( Q61H ) , thereby fully explaining the resistant behavior . The second exception was organoid line P19bT that , unlike the other CRC organoids in our panel , is characterized as microsatellite instable ( MSI ) including the hyper-mutator phenotype ( van de Wetering et al . , 2015 ) . Most importantly , P19bT tumor contains a BRAF ( V600E ) mutation , providing resistance towards the targeted drugs ( Di Nicolantonio et al . , 2008 ) . Thus , these two cases underscore that drug screening on human organoids is able to evaluate the functionality of entire oncogenic pathways beyond the scope of the most frequent candidate mutations . 10 . 7554/eLife . 18489 . 024Figure 6 . Screening multiple human CRC organoids confirm RAS mutational status for outcome EGFR inhibition . ( A ) Dose-response curves of 11 different patient-derived CRC organoids and one engineered CRC organoid ( P18T-KRASG12D ) treated for 72 hr with single targeted inhibitors or combinations thereof , namely afatinib ( dual EGFR/HER2 inhibitor ) , selumetinib ( MEK inhibitor ) and SCH772984 ( ERK inhibitor ) . 5 CRC organoids contain WT KRAS ( green lines ) , 5 CRC organoids contained annotated oncogenic KRAS mutations ( red lines ) , P19bT contains an oncogenic version of BRAF and P25T contains an oncogenic mutation in NRAS ( purple and blue lines , resp . ) . See Figure 6—source data 1 and Supplementary file 1 for all dose-response curves . ( B ) CRC and normal organoids classified based on the mutational status of the RAS-RAF-MEK-ERK signaling pathway . Responses to afatinib , selumetinib , SCH772984 and combinations thereof , are shown in scatter plots of IC50 values ( μM; 10log scale ) . Each colored dot represents an individual organoid line . Note that the experiment included normal organoids from the colon as well as the small intestine ( three individual persons ) . Color corresponds as indicated in the legend . Black bar is the geometric mean . n . s . , not significant . p<0 , 05 . *p<0 , 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02410 . 7554/eLife . 18489 . 025Figure 6—source data 1 . Dose-response curves for panel of patient-derived tumor organoids as indicated . A number of biological replicates for each dose-response curve are indicated between parenthesis ( first monotherapy/ second monotherapy/ combination therapy ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02510 . 7554/eLife . 18489 . 026Figure 6—figure supplement 1 . Top panel: Heat map of all IC50 values ( Log10-scale ) for multiple drug responses in CRC organoids with and without mutant RAS signaling , determined from datasets shown in Figure 6—source data 1 and Supplementary file 1 . IC50 values are color-coded: blue for highest and red for lowest efficacies ( see scale bar below ) . IC50 values account for both drugs when added in combination . Bottom panel: Heat map of all IC50 values ( Log10-scale ) for multiple drug responses in CRC organoids with and without mutant RAS signaling , determined from datasets shown in Figure 6—source data 1 and Supplementary file 1 . IC50 values are color-coded: blue for highest and red for lowest efficacies ( see scale bar below ) . IC50 values account for both drugs when added in combination , except combinations that include trametinib ( IC50 values marked with ; here trametinib concentrations are two fold ( log ) lower as indicated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 026 Next , we aimed to evaluate the effects of drug combinations on CRCs organoids with WT and mutant RAS pathways , as well as on non-tumorigenic normal organoids . In all seven independent CRC patients with a mutant RAS pathway , we observed synergistic activity when combining EGFRi and MEKi ( Sun et al . , 2014 ) or EGFRi and ERKi ( Figure 6B ) . Nevertheless , RAS WT organoids , either of CRC or normal origin , reveal higher sensitivities to combination therapies that include EGFR inhibition than RAS mutant CRCs . In contrast , therapies that do not include EGFR inhibition , but target MEK ( selumetinib ) and ERK ( SCH772984 ) , were similarly effective in all organoid lines , regardless of the mutational status of the RAS pathway or cellular state ( Figure 6B and Figure 6—figure supplement 1 ) . Another important observation after comparative analysis on this wide CRC panel concerns patient-to-patient variability in the response to anti-EGFR monotherapy , even within the RAS WT and mutant subgroups ( Figure 6B , upper left ) . Combining EGFRi with either MEK or ERKi results in a more consistent inhibitory effect over this set of CRCs ( Figure 6B , lower panels ) , thereby not only improving individual responses but also augmenting success rates on a population scale . These findings are supportive to the previously published concept that proposes to combine EGFR with MEK inhibition directly at the start of therapy in patients with WT RAS tumors with the rationale of preventing sub-clones with acquired resistance to anti-EGFR monotherapy from reigniting tumor growth ( Misale et al . , 2015 ) . Besides the direct effects of therapeutic treatments on tumor mass , the ability of cancer cells to recover from the treatments and restart tumor growth is of utmost relevance . We therefore studied the recovery of CRC organoids after release from ( i . e . washout of ) targeted inhibitors . More precisely , we monitored cellular viability , proliferation and cell death induction by quantifying viable nuclei ( marked green ) and dead nuclei ( marked red ) in 3D confocal tiled-scans at multiple time-points before and after treatment of targeted inhibitors ( Figure 7A and Materials and methods ) . 10 . 7554/eLife . 18489 . 027Figure 7 . Therapy surviving cancer cells reignite proliferation after release of targeted inhibition . ( A ) Scheme of image-processing workflow . Multiple z-stacks were acquired in a tile-scan mode . H2B-mNeonGreen and bright field images were recorded of >10 organoids ( left panel ) over multiple days . Lower half of the imaged z-planes were selected of 3D-organoids that were fully recorded on each day ( second panel ) . Live nuclei and dead nuclear remnants were marked for each z-plane , as identified by nuclear size ( third panel , see Materials and methods section ) , measured and integrated per lower half of the 3D scanned organoid as an absolute measure for the amount of viable cells , while summed dead nuclei represent the amount of dead cells ( fourth panel ) . ( B ) Bar diagrams showing proliferation and/or death of organoid cells following drug treatment and during recovery after drug removal . 3D tile-scans were acquired at the beginning and end of the therapy ( day −3 and 0 ) , as well as 3 and 7 days after the end of the therapy ( i . e . drug removal ) ( day 3 and 7 ) . All bars report pixel count from H2B-NeonGreen in living ( color ) as well as dead ( black ) organoid cells . Color corresponds to targeted inhibitor ( see legend ) . All values are means ± s . e . m . of 12–15 organoids , normalized to ‘alive H2B’ prior to treatment . One representative z-plane is provided of a P18T and P18T-KRASG12D CRC organoid during and after afatinib ( dual EGFR/HER2 inhibitor ) therapy . Green , alive nuclei . Red , marked nuclear remnants of dead cells . Color code legend is provided at the bottom of panel C . ( C ) Patient-derived CRC organoids P8T and P26T were treated and analyzed as described in B . In general , cancer cells that survived drug therapy rapidly reignited cell proliferation after drug release . veh , vehicle ( DMSO ) ; sel , selumetinib; afa , afatinib; afa+sel , afatinib+ selumetinib; SCH , SCH772984; SCH+sel , SCH772984+selumetinib . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02710 . 7554/eLife . 18489 . 028Figure 7—source data 1 . ImageJ/Fiji macro script: ‘Macro Drug&release experiment’ . Guides the user through XYZ stacks of organoids , acquired at various time points ( days apart ) . Helps to find back individual organoids and , per z-slice , lets the user indicate dead H2B particles by manual drawing . All output data are summarized in excel output file . For more detail , see Materials and methods section . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 02810 . 7554/eLife . 18489 . 029Figure 7—figure supplement 1 . The custom-made image analysis software for quantifications in Figure 7 is extensively described in the Materials and methods Section . In short: alive nuclei were discriminated from dead fragments by particle recognition and manual drawing based on their size and morphology . Due to long culture periods ( >10 days ) , we decided to avoid potential toxic biases of non-permeable DNA staining dyes such as propidium iodide ( PI ) . To validate this manual discrimination strategy , we acquired an additional ( single time point ) dataset of P18T organoids with variable degrees of apoptosis , this time including PI staining of dead nuclear fragments . ( A ) Four example organoids showing different degrees of apoptosis . From left to right: images were acquired in green ( H2B-Neon ) and red ( PI ) channel ( XYZ , but XY slices are shown ) and analyzed using two custom-made macros: ( I ) using manual selection , separating alive H2B from dead fragments as described above ( used for Figure 7 ) and ( II ) using automated thresholding on H2B-Neon and PI channels ( used for Figure 8 ) . Both methods yield surface ratios of dead/total H2B ( plotted in B ) . ( B ) 16 z-stacks were analyzed by the two analysis methods ( see A ) to yield surface ratios ( dead/total H2B ) , that show a strong correlation ( Pearsson coefficient of 0 . 76 , p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 029 In P18T , MEK or ERK inhibition ( selumetinib or SCH772984 resp . ) did not induce significant cell death after three days of monotherapy . Afatinib ( EGFR/HER2i ) , either alone or in combination with selumetinib ( MEKi ) , induced significant degrees of cell death , in line with our 72 hr monitoring of drug response . Similar effects were obtained by combined inhibition of MEK and ERK ( Figure 7B , left panel ) . Additionally , this experiment shows that the described effects persisted for seven days after drug washout . As expected for P18T-KRASG12D , suppressing EGFR/HER2 activity upstream of mutant KRAS using afatinib proved ineffective , while monotherapy of MEKi or ERKi inhibited proliferation only to a minor extent ( Figure 7B , right panel ) . Only the inhibitor combinations EGFRi/MEKi and MEKi/ERKi induced complete proliferative stagnation , albeit with minor induction of cell death . Importantly , independent of a therapeutic strategy , the CRC organoids quickly restored proliferative activity after drug release . Comparable results were obtained in CRC organoids P8T ( KRAS WT ) and KRAS mutant P26T , with the exception that cell death induction in P8T was less pronounced than in P18T ( Figure 7C ) . In summary , these data indicate that inhibition of the EGFR-RAS-ERK pathway , independent of inhibitor combination , predominantly inhibits cell-cycle progression in KRAS mutant CRC organoids . To characterize the induced cell cycle arrest in RAS mutant tumor cells in more detail , we performed cell cycle analysis by flow cytometry using a 3 hr EdU pulse in combination with DNA staining to discriminate between the different cell cycle phases ( G1 , S and G2 respectively ) in P18T-KRASG12D and P26T . Indicative of a G1 arrest , we observed a sharp decline in the amount of RAS mutant tumor cells in S-phase using inhibitor combinations EGFRi/MEKi and MEKi/ERKi , while a similar fraction of cells accumulated in G1 ( Figure 8A ) . 10 . 7554/eLife . 18489 . 030Figure 8 . Cell cycle arrest upon dual inhibition of EGFR-MEK-ERK pathway . ( A ) Representative cell cycle analysis of P18T-KRASG12D and P26T by flow cytometry ( n = 2 ) . DNA was stained with DAPI and DNA-synthesis was detected using a 3 hr EdU pulse to clearly discriminate between G1 , S and G2 stages of the cell cycle . Dual inhibition of the EGFR-MEK-ERK pathway significantly changes the distribution of cells between stages of the cell cycle ( Chi2: all p values<0 , 0001 ) with a predominant increase in G1 at the expense of cells in S-phase . EGFRi + MEKi = afatinib + selumetinib . MEKi + ERKi = selumetinib + SCH772984 . ( B ) Almost no incorporation of EdU ( red ) is detected during the last 24 hr of drug treatment using dual inhibition of the EGFR-MEK-ERK signaling pathway , indicative of halted proliferative activity . Nuclei are counterstained with Hoechst ( white ) . EGFRi + MEKi = afatinib + selumetinib . EGFRi + ERKi = afatinib + SCH772984 . Scale bar is 100 µm . ( C ) Virtually all cancer cells incorporate EdU ( red ) when provided after release from targeted inhibition of the EGFR-MEK-ERK pathway . Nuclei are counterstained with Hoechst ( white ) . EGFRi + MEKi = afatinib + selumetinib . EGFRi + ERKi = afatinib + SCH772984 . Scale bar is 100 µm . ( D ) Chronological ranking of mitotic and apoptotic events extracted from live-cell imaging data of tumor recovery reconstructs the organoid size evolution over time . In contrast to vehicle treated organoids ( blue lines ) , afatinib + selumetinib treated organoids ( red lines ) show first mitotic activity again from 20–24 hr onwards after drug withdrawal . Typical snapshots of live-cell imaging data are provided . White circles indicate mitotic events . Arrows indicate the organoid and moment of snapshot . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03010 . 7554/eLife . 18489 . 031Figure 8—source data 1 . ImageJ/Fiji macro script: ‘Score Events macro’ . Guides the user through the analysis of the event-rich organoid movies ( e . g . as generated with the Organoid movie macro ) , by numbering and drawing indicated events ( mitosis , apoptosis ) in the movie and generating an overview excel file . Graphs in Figure 8D and Figure 3—figure supplement 2 were generated using this method . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03110 . 7554/eLife . 18489 . 032Figure 8—figure supplement 1 . Average growth speeds of the organoids were determined by linear fitting of the traces shown in Figure 8D . The two time frames roughly correspond to the first half and the second half of the experiment . Directly after drug removal , afatinib- and selumetinib-treated organoids show a significant reduction in growth speed as compared to vehicle-treated organoids . After 22–24 hr of recovery , growth rates return to the level of vehicle-treated organoids ( possibly even slightly faster ) . p<0 , 05; p<0 , 01; **p<0 , 001; n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 032 To further characterize drug-induced arrest , we investigated whether the regrowth after drugs washout involves all therapy-surviving tumor cells or only a minor subpopulation that fuels tumor relapse . For this , we performed two functional assays . First , we performed EdU incorporations at various time points during the drug response and during organoid recovery after drug withdrawal . In agreement with the previous cell cycle analysis , almost no EdU incorporation was detected in the presence of inhibitor combinations EGFRi/MEKi and MEKi/ERKi ( no cells in S-phase ) ( Figure 8B ) . However , during the first three days after drug withdrawal , the vast majority of growth-arrested tumor cells incorporated EdU again , suggesting renewed proliferative activity in virtually all tumor cells , thereby excluding the presence of senescence or minor subpopulations being responsible for restored growth ( Figure 8C ) . In addition , we performed live-cell imaging on tumor organoids after drug withdrawal and quantified the number of mitotic and apoptotic events over time ( see Materials and methods ) . Indeed , in line with a G1 arrest , we observed a delay of about 20–24 hr after withdrawal of the drugs ( EGFRi/MEKi ) before observing numerous mitotic events again in all regions of the arrested organoids ( Figure 8D and Figure 8—figure supplement 1 and Video 3 ) . ( Similar results were obtained for MEKi/ERKi , data not shown ) . 10 . 7554/eLife . 18489 . 033Video 3 . Real-time imaging of cellular behavior in tumor organoids surviving treatment with afatinib and selumetinib . P18T-KRASG12D and P26T organoids were treated for 72 hr with afatinib ( 1 μM ) and selumetinib ( 1 μM ) , similarly to Figures 1 and 3 and Video 1 . After the subsequent washout of the drugs , organoids were imaged for ~40 hr to visualize cell behaviour in surviving organoids . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 033 As described above , none of the therapies targeting the EGFR-RAS-ERK pathway could induce the desired degree of cell death in RAS mutant CRC organoids . As opposed to this , it has been reported that combined inhibition of anti-apoptotic BCL2 family members and effectors of the RAS pathway can effectively induce cell death in KRAS mutant cancers ( Corcoran et al . , 2013; Hata et al . , 2014 , 2015; Tan et al . , 2013 ) . These results prompted clinical trials to evaluate combined targeting of MEK and BCL2/BCLXL in KRAS mutant solid tumors ( NCT02079740 ) . Unfortunately , clinical application of BCL2/BCLXL inhibitors is hindered by on-target toxicity of BCLXL inhibition in blood platelets ( Hata et al . , 2015 ) and might therefore strongly benefit from strategies that allow minimized doses . Here , we explored the use of navitoclax , a clinically tested BCL2/BCLXL inhibitor , in targeted therapies against RAS mutant CRC organoids . Indeed , straightforward ATP-based screening confirmed that navitoclax , when combined with afatinib ( Figure 9A , left panel ) , selumetinib ( Figure 9A , middle panel ) , or both ( Figure 9A , right panel ) is far more efficient in antagonizing tumor organoid growth than any of the related monotherapies . Importantly , regarding the dose-limiting effects of navitoclax , we show that robust dual inhibition of the EGFR-RAS-ERK pathway ( 1 µM afatinib/1 µM selumetinib ) is exceptionally potent in sensitizing navitoclax-induced effects ( Figure 9A , right panel ) . Such strong sensitization could not be achieved by afatinib ( 1 µM ) alone or selumetinib ( 1 µM ) alone . Similar results were obtained in other patient-derived CRC organoids that harbor a KRAS mutation ( Figure 9—figure supplement 1 ) or , alternatively , with different combinations of inhibitors targeting the EGFR-RAS-ERK pathway ( Figure 9—figure supplement 2 ) . 10 . 7554/eLife . 18489 . 034Figure 9 . Robust inhibition of the EGFR-RAS-ERK pathway sensitizes for navitoclax-induced cell death . ( A ) Dose-response curves of patient-derived CRC organoids P18T-KRASG12D treated with the dual EGFR/HER2 inhibitor afatinib , MEK inhibitor selumetinib , BCL2/BCLXL inhibitor navitoclax or a combination thereof . Cell viability was measured by an ATP-based assay after 72 hr of drug treatment . Inhibition of the EGFR-RAS-ERK pathway using high concentrations of afatinib and selumetinib ( 1 µM ) strongly sensitizes for navitoclax-induced reduction of cellular viability ( right panel , black line ) . Such strong sensitization could not be achieved by afatinib ( 1 µM ) alone ( left panel , orange line ) or selumetinib ( 1 µM ) alone ( middle panel , purple line ) . Dose-response curves are averages of n = 2 . ( B ) Representative images taken from CRC organoids P18T-KRASG12D treated for 72 hr with above described drug combinations . Low amounts of navitoclax ( 65 nM ) only induce cell death in combination with effective inhibition of the RAS pathway using a high concentrations of afatinib and selumetinib . Bar diagram at the right shows quantifications of cell death by scoring propidium-iodide stained nuclei ( dead ) over viable H2B-labeled nuclei ( see Materials and methods section ) of minimal 15 organoids per condition ( signals pooled prior ratioing , hence no standard deviation calculated: see Materials and methods section ) . Representative experiment of n = 2 . ( C ) Bar diagrams representing cellular viability ( alamarBlue assay ) of organoids that have been recovered for six days after 72 hr drug treatment at different concentrations ranging from 5 µM ( left ) to 5 nM ( right ) . Color corresponds to targeted inhibitor as indicated in the legend . All values are normalized to control samples ( DMSO ) . Targeting of the RAS pathway does not provide long-lasting effects after drug removal , even combination treatments at high concentrations ( green bars ) . However , it does sensitize for cell death induction using low amounts of navitoclax ( black bars ) . An average of two independent experiments is shown . Bar diagrams are averages of n = 2 . ( D ) Extensive dual or triple-inhibitor dose-response assay of patient-derived CRC organoids P18T-KRASG12D treated for 72 hr . 9×9 drug concentrations of selumetinib ( MEKi ) or afatinib/selumetinib ( 1/1 ) versus navitoclax ( BCL2/BCLXL ) or venetoclax ( BCL2 ) were chosen with logarithmic interval covering a 5 nM–20 µM range . The results of the full matrix screen are represented as heat maps , where red represents 0% ATP levels ( no viability ) and green represents 100% ATP levels ( max viability ) . Exploring optimal drug concentrations reveal that the more effective inhibition of the RAS pathway is achieved ( dual targeting and high concentrations ) , the less navitoclax is required . Venotoclax , a BCL2-specific inhibitor , is not able to copy the effects of navitoclax ( BCL2/BCLXL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03410 . 7554/eLife . 18489 . 035Figure 9—source data 1 . ImageJ/Fiji macro script: 'Macro PI versus H2B' . Analyzes H2B-mNeon expressing organoids , that have been labelled with propidium iodide to mark dead cells/fragments . Measure for death induction is the ratio between PI-positive pixels ( dead ) and H2B-positive pixels ( total ) . Thresholding values for both channels can be tweaked . Thresholded images ( z-stacks ) are stored and all readouts are summarized in excel data file . For more detail , see Materials and methods section . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03510 . 7554/eLife . 18489 . 036Figure 9—figure supplement 1 . Dose-response curves of patient-derived KRAS mutant CRC organoids P9T and P26T treated with the dual EGFR/HER2 inhibitor afatinib , MEK inhibitor selumetinib , BCL2/BCLXL inhibitor navitoclax or a combination thereof . Cell viability was measured by an ATP-based assay after 72 hr of drug treatment . Inhibition of the EGFR-RAS-ERK pathway using high concentrations for both afatinib and selumetinib ( 1 µM ) strongly sensitizes for navitoclax-induced reduction of cellular viability ( black line ) . Horizontal colored-dashed lines represent the level of cellular viability of the respective anchor treatments without navitoclax . Representative images of the combined anchor treatments with and without a low dose of navitoclax are depicted on the right . Note that organoids remain viable with only robust EGFR-RAS-ERK inhibition . Dose-response curves are averages of n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03610 . 7554/eLife . 18489 . 037Figure 9—figure supplement 2 . Dose-response curves of patient-derived CRC organoids P18T-KRAS treated with different combination therapies against the EGFR-RAS-ERK pathway with addition of BCL2/BCLXL inhibitor navitoclax and the corresponding mono and dual therapy controls . Top panel: EGFR-RAS-ERK pathway inhibition with dual EGFR/HER2 inhibitor afatinib and MEK inhibitor trametinib . Lower panel: EGFR-RAS-ERK pathway inhibition with MEK inhibitor selumetinib and ERK inhibitor SCH772984 . Cell viability was measured by an ATP-based assay after 72 hr of drug treatment . Inhibition of the EGFR-RAS-ERK pathway using fixed high concentrations of inhibitors against EGFR-RAS-MEK pathway effectors strongly sensitizes for navitoclax-induced reduction of cellular viability ( black line ) . Horizontal colored-dashed lines represent the level of cellular viability of the respective anchor treatments without navitoclax . The representative images of the combined anchor treatments with and without a low dose of navitoclax are depicted on the right . Note that organoids remain viable with only robust EGFR-RAS-ERK inhibition . Dose-response curves are averages of n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 03710 . 7554/eLife . 18489 . 038Figure 9—figure supplement 3 . Drug response of P18T-KRASG12D and P26T CRC organoids examined by Western bot after 24 hr . Most effective reduction of p-ERK is detected when the organoids were treated with high concentrations of inhibitors afatinib ( 1 μM ) and selumetinib ( 1 μM ) and not at lower concentrations ( 65 nM of both drugs or 33 nM afatinib + 200 nM of selumetinib ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18489 . 038 To ensure that the observed effects represent cell death rather than cell-cycle arrest , we performed qualitative and quantitative microscopic analyses ( Figure 9B ) . Indeed , cell death could be induced by low concentrations of navitoclax when combined with high concentrations of afatinib and selumetinib . Next , we designed a medium-throughput assay to monitor the persistence of drug response after wash-out of the above-mentioned inhibitor combinations ( Figure 9C ) . As expected , even dual inhibition with afatinib ( EGFR/HER2i ) and selumetinib ( MEKi ) at high concentrations does not induce sufficient cell death , since these cultures can recover to the levels of untreated controls within six days ( Figure 9C ) . In contrast , the addition of low concentrations navitoclax to high concentrations of inhibitor combination afatinib/selumetinib potently induces cell death , as shown by the sustained inhibitory effects on culture growth ( Figure 9C , black bars ) . In light of the dose-limiting toxicity of navitoclax in blood platelets , we performed full matrix-screens to explore optimal combinations of drug concentrations ( Figure 9D ) . In agreement with previous results , the more efficient inhibition of the RAS pathway , i . e . high concentrations and/or dual targeting ( Figure 9—figure supplement 3 ) the lower the concentration of navitoclax that is necessary to affect cellular viability . Furthermore , venetoclax , a BCL2-specific inhibitor , is unable to reproduce the effects of navitoclax ( BCL2/BCLXLi ) ( Figure 9D ) , suggesting that in agreement with the reported findings in lung cancer ( Corcoran et al . , 2013 ) , it is BCLXL that protects against apoptosis upon targeted inhibition of a mutant RAS pathway in CRC organoids ( Figure 9D ) . Patient-derived CRC organoids were recently introduced as a model system in cancer research that is complementary to cell lines and PDX models ( van de Wetering et al . , 2015 ) . We assembled a panel of normal and CRC organoids with either WT or mutant RAS that had been derived from different patients ( van de Wetering et al . , 2015 ) . Moreover , we included normal ( Drost et al . , 2015 ) and tumor organoids in which the oncogenic G12D mutation was introduced in the endogenous KRAS locus by CRISPR-Cas9-induced homologous recombination . These engineered organoid lines , in combination with patient-derived CRC organoids of different genetic backgrounds , allowed us to study the effect of mutant KRAS on drug response to targeted inhibition of the EGFR-RAS-ERK pathway . Moreover , real-time imaging allows the monitoring of exact cellular fates in drug-challenged CRC organoids with spatial ( 3D ) and temporal resolution . Using this panel , we show that the presence of mutant RAS is sufficient to confer resistance to EGFR inhibitors . Moreover , we confirmed the synergistic effect of the clinically tested combination of pan-HER and MEK inhibition on mutant RAS organoids . However , we found that RAS mutant organoids remained largely resistant to apoptosis and became largely arrested in proliferation . Importantly , for KRAS mutant tumor organoid P26T we observed similar drug sensitivities in vitro as in vivo upon xenotransplantation . Moreover , similar growth arrested responses were reported in PDX mouse models of KRAS mutant CRC cancers ( Sun et al . , 2014 ) , underscoring the notion that tumor organoids are a reliable model system to test drug responses . We report , for the first time , how normal tissue organoids respond to inhibitors targeting the EGFR-RAS-ERK pathway . Intriguingly , drug effects were almost identical in normal organoids and patient-derived CRC organoids when WT for RAS . This may imply that the sensitivity of RAS WT colon cancer for EGFRi is not an acquired oncogene-addiction , but merely represents the dependency of normal colon ( stem ) cells on EGFR signaling activity ( Wong et al . , 2012 ) . Indeed , normal organoids from the human colon consist predominantly of proliferative stem and progenitor cells due to the WNT ligands in the culture medium . Therefore , the observed toxicity in the normal organoids may very well be most representative for direct effects on the stem cell compartments of the normal colon . Indeed , one of the direct side effects of anti-EGFR monotherapy is diarrhea ( Miroddi et al . , 2015 ) . In analogy to drug responses with WT RAS , normal organoids harboring a CRISPR-introduced oncogenic KRAS mutation showed resistance profiles towards targeted therapies that closely resemble those of RAS mutant CRC organoids , again underscoring the dominance of the RAS mutational status on drug response . Anti-EGFR therapy in patients with KRAS WT colon tumors is standard of care , whereas patients with RAS mutant tumors are excluded . Our results confirm the drug sensitivity profile of colorectal cancers with and without mutant RAS , as has been established both in other model systems as well as in the clinic ( Sun et al . , 2014; Karapetis et al . , 2008; Amado et al . , 2008 ) . For RAS mutant tumors , a number of different drugs and drug combinations have been tested , but thus far this has been without significant effect ( Ryan et al . , 2015 ) . Our analyses confirm that drug treatments targeting the EGFR-RAS-ERK and the PI-3K/AKT cascades , including combinations thereof , are largely ineffective in RAS mutant CRC organoids . In contrast , for CRC patients with RAS WT tumors combination therapies that target the EGFR-MEK-ERK pathway may be an improved alternative over anti-EGFR monotherapy . First , we observed that this combination treatment induced cell death more systematically over a panel of individual patient-derived CRC organoids with WT RAS ( i . e . with decreased variability ) than monotherapy with a pan-HER inhibitor . Furthermore , combination treatments may decrease the potency of low-abundant RAS mutant subclones to initiate tumor-relapse during therapy against a predominantly KRAS WT tumor ( Misale et al . , 2015 ) . The combined inhibition of pan-HER and MEK is currently tested in patients with RAS mutant cancers in several clinical trials ( e . g . NCT02450656 , NCT02230553 and NCT02039336 ) . Also in our hands , this combination showed a clear synergistic effect in suppressing growth of RAS mutant CRC organoids , as determined by a straightforward ATP-based assay . Importantly however , our data revealed that this inhibitor combination induced a cell cycle arrest in mutant RAS organoids but no cell death . As a result , the cells rapidly restored proliferative activity after withdrawal of these drugs . The inability to induce cell death likely affects the long-term effectiveness of this combination in CRC patients with mutant RAS . An alternative combination that is currently in clinical trials is the combined inhibition of MEK and ERK ( NCT02457793 ) . The rationale behind this combination is the notion that resistance to targeted inhibition of RAF and MEK often involves reactivation of ERK ( Ryan et al . , 2015 ) , while suboptimal suppression of ERK activity in RAF mutant cancers may underlie the limited efficacy ( Bollag et al . , 2010; Corcoran et al . , 2015 ) . Although our RAS mutant CRC organoids showed sensitivity to dual inhibition of MEK and ERK , also this drug combination induced cell-cycle arrest rather than cell death , questioning whether it will be sufficient for the treatment of RAS mutant CRC . With respect to clinical applications , we here report that effective inhibition of the EGFR-MEK-ERK pathway through combinatorial targeting does significantly prime the cytostatic RAS mutant cancer cells for apoptosis . This can be utilized by low concentrations of navitoclax , one of the most clinically advanced inhibitors of anti-apoptotic BCL family members . Minimizing navitoclax concentrations would be beneficial due to its on-target effects on BCLXL in circulating platelets , thereby causing thrombocytopenia ( Hata et al . , 2015 ) . However , triple combination therapy with low concentrations of navitoclax ( 50 mg/kg; five days on , two days off ) proved to be too toxic for the mice ( data not shown ) . Nevertheless , we consider the navitoclax-induced apoptosis as a proof-of-principle that EGFR-MEK-ERK pathway inhibition in combination with alternative signaling nodes holds great promise in identifying therapeutic drug combinations that kill RAS mutant tumor cells while being tolerated by the patient . In summary , we show drug responses of a wide panel of patient-derived CRC organoids to multiple clinically advanced targeted inhibitors , either alone or in combinations , against the EGFR-RAS-ERK pathway . Importantly , the drug phenotypes that we observe in the organoids appear representative for previous reported responses in vivo . We believe that organoid collections will facilitate the identification and optimization of effective targeted therapies , since drug screens can be performed at a scale that is currently unprecedented when using resource-intensive PDX models . Due to the reliability and scalability of tumor organoids as a model system , we advocate that novel drugs should be tested on a panel of tumor organoids before their use in clinical trials . The patient-derived organoids used in this study were previously established and characterized ( van de Wetering et al . , 2015 ) . Human CRC and healthy colon organoids were cultured as described previously ( van de Wetering et al . , 2015 ) . In short , organoids were cultured in drops of Basement Membrane Extract ( BME; Amsbio ) and medium was refreshed every two days . The CRC culture medium contained advanced DMEM/F12 ( Invitrogen ) with 1% Penicillin/Streptomycin ( P/S , Lonza ) , 1% Hepes buffer ( Invitrogen ) and 1% Glutamax ( Invitrogen ) , 20% R-spondin conditioned medium , 10% Noggin conditioned medium , 1X B27 ( Invitrogen ) , 1 . 25 mM n-Acetyl Cysteine ( Sigma-Aldrich ) , 10 mM Nicotinamide ( Sigma-Aldrich ) , 50 ng/ml EGF ( Invitrogen ) , 500 nM A83-01 ( Tocris ) , 10 μM SB202190 ( ApexBio ) and 100 µg/ml Primocin ( Invitrogen ) . The medium of healthy colon organoids had additional Wnt conditioned media . Organoids were splitted through shear stress ( pipetting ) and/or Trypsin-EDTA ( Sigma-Aldrich ) treatment . Culture medium after splitting was supplemented with 10 µM Y-27632 dihydrochloride . Organoid cultures have repeatedly been tested negative for Mycoplasma . Western blots are performed as described previously ( Drost et al . , 2015 ) . Antibodies used: ERK ( RRID:AB_390779 ) , pERK ( RRID:AB_331646 ) and GAPDH ( RRID:AB_2107445 ) . CRISPR guide RNAs ( sgRNAs ) were generated as described by Drost et al . ( 2015 ) . The KRAS target sequences used were: 5′-GAATATAAACTTGTGGTAGTTGG-3′ and 5′-GTAGTTGGAGCTGGTGGCGTAGG-3′ . Transfections of p18T and p26N organoids with sgRNAs and subsequent selections by withdrawing EGF and adding the EGFR inhibitor gefitinib to the culture medium were performed as previously described ( Drost et al . , 2015 ) . The presence of KRAS G12D mutation was verified by sequencing the PCR product obtained using the primers KRAS_for , 5′-TGGACCCTGACATACTCCCA-3′ and KRAS_rev , 5′-AAGCGTCGATGGAGGAGTTT-3′ ( Drost et al . , 2015 ) . Organoids were infected with lentivirus encoding histone2B fused to mNeonGreen ( bright monomeric GFP variant ) linked to a puromycin-resistance gene ( pLV-H2B-mNeonGreen-ires-Puro ) ( Shaner et al . , 2013 ) to visualize and track nuclei . Infected organoids were selected using 2 µg/ml puromycin . Five days after organoid typsinization , 1 mg/ml dispase II ( Invitrogen ) was added to the medium of the organoids and these were incubated for 15 min at 37°C to digest the BME . Subsequently , organoids were mechanically dissociated by pipetting , filtrated using a 40 μm nylon cell strainer ( Falcon ) , resuspended in 2% BME/growth medium ( 15–20 , 000 organoids/ml ) prior plating of 30 µl ( MultidropTM Combi Reagent Dispenser ) on BME pre-coated 384-well plates . The drugs and their combinations were added 3 hr after plating the organoids using the Tecan D300e Digital Dispenser . Drugs were dispensed in a randomized manner and DMSO end concentration was 0 . 4% in all wells . 72 hr after adding the drugs , ATP levels were measured using the Cell-Titer Glo2 . 0 ( Promega BV ) according to the manufacturer's instructions . Results were normalized to vehicle ( DMSO = 100% ) and baseline levels ( multi drug ATP plateau at high concentrations = 0% ) that were manually determined per organoid type and screening day . Multiple identical drug combinations were averaged . Heatmaps were smoothened using a moving average . Bliss scores were calculated as described previously ( Tan et al . , 2013 ) . Afatinib , Dacomitinib , Lapatinib , Selumetinib , Trametinib , BYL719 , MK2206 and GDC-0994 , navitoclax and venetoclax were purchased from Selleck Chemicals . SCH772984 was obtained from MedChem Express and Cobimetinib from Active Biochem . These compounds were dissolved in dimethylsulfoxide ( DMSO , Sigma-Aldrich ) and stored as 10 mM aliquots . Data analyses were performed using GraphPad software by applying the nonlinear regression ( curve fit ) and the equation log ( inhibitor ) vs . normalized response ( variable slope ) . Five days after trypsinization , H2B-mNeonGreen-expressing organoids were plated in matrigel in glass-bottom 96-well plates and mounted on an inverted confocal laser scanning microscope ( Leica SP8X ) under controlled conditions ( 37°C , 6% CO2 ) . Drugs were added to the organoids just prior to imaging . For 72 hr , organoids were imaged every 15 min in XYZT-mode using a 40x objective ( 1 . 1NA ) and a 506 nm laser excitation light from a tunable white light laser for 72 hr . The images were converted using ImageJ/Fiji software into manageable and maximally informative videos , combining z-projection , depth color-coding and merging with transmitted light images ( see source code files , ‘Organoid Movie Macro’ ) . Five days after trypsinization , H2B-mNeon-expressing organoids were plated in glass-bottom 96-well plates . Prior to drug addition ( day −3 ) , the organoids were imaged on a Leica SP8X . One 3D tile scan ( merging 3×3 images , ~175 µm in Z in total , 5 μm Z-step ) was acquired per well , allowing the visualization of 10–20 organoids per well . On day 0 , 3 and 7 , exactly the same fields of organoids were imaged again and medium was refreshed . A custom-made ImageJ/Fiji macro ( see source code files , ‘Macro Drug&Release’ ) was developed to analyze 12–15 organoids per well/condition in a paired manner , i . e . individual organoids were tracked over the entire experiment ( >10 days ) . Per organoid , a pseudo-quantitative measure for absolute numbers of living and dead cells was established as follows: ( 1 ) Thresholding on H2B-Neon fluorescence to select H2B-positive pixels ( total ) . ( 2 ) Marking of ‘dead’ pixels in each slice ( initially automated , based on particle size and eventually manually by selection ) . ( 3 ) Dividing pixels in dead and alive ( total minus dead ) . ( 4 ) Integrating the pixel areas representing alive and dead H2B , respectively , from the slices that make up for the lowest 50% of the organoid volume . This was done to avoid analysis on the upper 50% of the volume , which is inevitably of lower image quality . The analysis was performed in 12 to 15 organoids per well/condition . No biological replicates . Of note , this method is independent of non-permeable DNA dyes such as PI to avoid their potential long-term effect on organoid growth . In order to validate the current method , a single time point data set was acquired with the use of PI ( see Figure 7—figure supplement 1 ) , validating the robustness of the strategy . For the analysis of organoid recovery after drug withdrawal , organoids were treated with the indicated drugs for 72 hr as described . Subsequently , drugs were washed out through aspiration of the drug-containing medium , followed by a washing step ( 10’ incubation at 37°C with basal DMEM/F12 ( +++ ) ) . After washing , organoids were incubated with CRC-medium containing 2%BME for recovery . After 48 hr , the medium was replaced by CRC-2%BME combined with 10% AlamarBlue ( AB ) cell viability reagent ( ThermoFisher Scientific ) according to the manufacturer's instructions . The increase of AB fluorescence ( excitation 544 nm , emission 590 nm ) was monitored over a time course of 2 hr ( with measurements at 15’ intervals ) at 37°C , using a SpectraMax M5 microplate reader ( Molecular Devices ) . Fluorescence kinetics were plotted over time ( as relative fluorescent units ( RFU ) per hour ) to define the linear range of the assay . Cell viability was then defined as the maximum AB fluorescence ( RFUmax , within the linear range of the assay ) , corrected for background fluorescence ( RFU at time point 0 ) . Viability data from drug treated organoids were normalized to vehicle ( DMSO ) treated controls . Upon the AB time course , organoids were washed two times with basal DMEM/F12 and incubated with CRC-2%BME for another 96 hr , after which the AB assay was repeated . H2B-mNeon-expressing organoids were plated in 384-well plates and provided with drugs as described for the viability assay . To selectively stain dead cells , propidium iodide ( PI ) was added 2 hr before starting imaging ( 2 µM ) . 3D stacks of 150 μm ( 7 , 5 μm per plane ) were acquired on a Leica SP8 scanning confocal microscope , using a 10x dry lens for large field-of-view ( 3000×3000 pixels ) . Being an endpoint assay , high laser intensities could be applied for optimal imaging quality ( and hence analysis ) . Green ( mNeon ) and red ( PI ) signals were sequentially acquired to avoid spectral mixing . Signals were pooled for 15–20 organoids before ratio calculation , hence no error bars . Custom-made software ( ImageJ/Fiji , see source code files , 'Macro PI versus H2B' ) was designed to determine surfaces ( i . e . numbers of pixels ) representing H2B-mNeon and PI , respectively; the ratio of these surfaces ( PI/H2B ) is the quantitative measure for cell death in the drug-challenged organoids . Central to the unambiguous determination of these surfaces is setting of the threshold . Our algorithm initially measures surfaces and corresponding mean intensities with ramping threshold , and from these data mathematically derives the threshold-optimum by combining the highest mean signal and most confined surface area . Depth-coded projection movies were analyzed for life and death in time: mitotic and apoptotic events were marked with help of custom-made ImageJ/Fiji macro ( see source code files , ‘ScoreEvents’ ) . Indicated events were automatically drawn in the movie ( essential when aiming to mark all events ) and data were automatically sorted into Excel-files containing a ( chronologically sorted ) list of events . 3 hr prior trypsinization ( TriplE , 5 min at 37C ) to a single cell , organoids were incubated with 500 nM EdU . Single cells were fixed with ethanol ( 5% ) . EdU click-it reaction was performed according to manufacturer's protocol . DNA was stained using 1 µg/ml DAPI . Cells were analysed using a FACSCanto II ( BD ) . Approval for this study was obtained by the local animal experimental committee at The Netherlands Cancer Institute ( DEC-NKI; OZP = 12012 and WP5727 and WP5689 ) . P26T patient-derived organoids were trypsinized and 200 , 000 cells were resuspended in 50 µl medium/Matrigel ( BD Biosciences ) mixture at a 1:1 ratio and injected subcutaneously into NSG mice ( JAX stock no: 005557 ) . Mice with established tumors ( average volume of 300 mm3 ) were treated with afatinib ( 12 . 5 mg/kg; five days on , two days off ) , selumetinib ( 20 mg/kg; same schedule ) or with a combination of both drugs for four weeks . After three weeks recovery from the drug treatment , mice were sacrificed . For the second in vivo experiment , P26T organoids ( ~300 . 000 cells ) were resuspended in 50% matrigel/medium with 10% collagen type I ( BD Biosciences ) and injected subcutaneously into NSG mice ( JAX stock no: 005557 ) . Mice with established tumors ( average volume of 200 mm3 ) were treated with afatinib ( 20 mg/kg; five days on , two days off ) , selumetinib ( 25 mg/kg; same schedule ) or with a combination of both drugs for four weeks . Tumor volumes were evaluated three times per week by caliper and the approximate volume of the mass was calculated using the formula Dxd2/2 , where D is the major tumor axis and d is the minor tumor axis . For in vivo dosing , afatinib was dissolved in 1 . 8% hydroxypropyl-b-cyclodextrin ( Sigma ) , 5% of a 10% acetic acid stock and aqueous natrosol ( 0 , 5% ) . Selumetinib was resuspended in 0 , 5% hydroxypropylmethylcellulose ( Sigma ) and 0 . 4% Tween-80 in distilled water . All agents were administered via oral gavage . The results presented are representative of three independent experiments run in triplicate , unless otherwise indicated . Student’s t test and two-way ANOVA were carried out using GraphPad Prism to calculate significance . Differences were considered significant at p<0 . 05 . Results are expressed as mean ± standard error ( S . D . ) .	Recent technical advances mean that miniature replicas of many tissues can be grown in the laboratory . These so-called organoids provide scientists with model systems that are not as limited as simple , two-dimensional sheets of cells growing in a petri dish , and less labor and resource intensive than studies using laboratory animals . In particular , organoids grown from tumor cells from cancer patients have been suggested as having numerous advantages over both laboratory-grown cancer cells and mice when it comes to testing potential new anticancer drugs . Mutations in a gene called KRAS are common in many types of cancer including colon cancer . Tumors with these mutations are difficult to treat and so far virtually all attempts to generate compounds that selectively interfere with the KRAS protein encoded by the mutant gene have failed . Instead , drugs that indirectly inhibit this protein’s effects by targeting other proteins in the same signaling pathway are currently being tested on patients . However , there is still a need for better ways to pre-test whether these drugs will be effective in humans without having to expose the patient to side effects or an ineffective drug . Now , Verissimo , Overmeer , Ponsioen et al . have tested clinically-used KRAS pathway inhibitors and drug combinations against normal colon organoids and colon cancer organoids derived from patients with colon cancer . Gene editing techniques were used to introduce KRAS mutations into some of the normal organoids grown from healthy tissue , and into cancer organoids grown from tumors that had a normal copy of the KRAS gene . In all cases , only those organoids with mutant forms of the KRAS gene were resistant to the treatments . Furthermore , when organoids with the KRAS mutation were treated with some combination therapies that are currently being tested in clinical trials , the tumors stopped growing but the tumor cells failed to die . Similar drug treatments on mice carrying human colon cancer organoids confirmed these results , which is in line with previous studies where tumor tissue from human patients was transplanted into mice . These findings show that collections of tumor organoids from multiple patients could help researchers to quickly identify and optimize targeted anticancer therapies before they are incorporated into clinical trials . In the future , clinical studies are needed to verify how accurately the testing of cancer drugs on organoids predicts whether the drug will or will not work in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cancer", "biology" ]	2016	Targeting mutant RAS in patient-derived colorectal cancer organoids by combinatorial drug screening
Adaptive immunity vitally depends on major histocompatibility complex class I ( MHC I ) molecules loaded with peptides . Selective loading of peptides onto MHC I , referred to as peptide editing , is catalyzed by tapasin and the tapasin-related TAPBPR . An important catalytic role has been ascribed to a structural feature in TAPBPR called the scoop loop , but the exact function of the scoop loop remains elusive . Here , using a reconstituted system of defined peptide-exchange components including human TAPBPR variants , we uncover a substantial contribution of the scoop loop to the stability of the MHC I-chaperone complex and to peptide editing . We reveal that the scoop loop of TAPBPR functions as an internal peptide surrogate in peptide-depleted environments stabilizing empty MHC I and impeding peptide rebinding . The scoop loop thereby acts as an additional selectivity filter in shaping the repertoire of presented peptide epitopes and the formation of a hierarchical immune response . Nucleated cells of higher vertebrates provide information about their health status by presenting a selection of endogenous peptides on MHC I molecules at the cell surface . By sampling these peptide-MHC I ( pMHC I ) complexes , CD8+ T lymphocytes are able to detect and eliminate infected or cancerous cells ( Blum et al . , 2013; Rock et al . , 2016 ) . In a process called peptide editing or proofreading , peptides derived from the cellular proteome are selected for their ability to form stable pMHC I complexes . This peptide editing is known to be catalyzed by the two homologous MHC I-specific chaperones tapasin ( Tsn ) and TAP-binding protein-related ( TAPBPR ) ( Fleischmann et al . , 2015; Hermann et al . , 2015; Morozov et al . , 2016; Neerincx and Boyle , 2017; Tan et al . , 2002; Thomas and Tampé , 2019; Wearsch and Cresswell , 2007; Wearsch et al . , 2011 ) . The selection of high-affinity MHC I-associated peptide epitopes is of pivotal importance not only for immunosurveillance by effector T lymphocytes , but also for priming of naïve T cells and T cell differentiation . As an integral constituent of the peptide-loading complex ( PLC ) in the endoplasmic reticulum ( ER ) membrane , the ER-restricted Tsn functions in a ‘nanocompartment’ characterized by a high concentration of diverse , optimal peptides . The peptides are shuttled into the ER by the heterodimeric ABC ( ATP-binding cassette ) transporter associated with antigen processing TAP1/2 , the central component of the PLC ( Abele and Tampé , 2018 ) . In the ER , most peptides are further trimmed by the aminopeptidases ERAP1 and ERAP2 to an optimal length for binding in the MHC I groove ( Evnouchidou and van Endert , 2019; Hammer et al . , 2007 ) . In contrast to Tsn , TAPBPR operates independently of the PLC and is also found in the peptide-depleted cis-Golgi network ( Boyle et al . , 2013 ) . Fundamental insights into the architecture and dynamic nature of the Tsn-containing PLC have come from a recent cryo-EM study of the fully-assembled human PLC ( Blees et al . , 2017 ) , while the basic principles underlying catalyzed peptide editing have been elucidated by crystal structures of the TAPBPR-MHC I complex ( Jiang et al . , 2017; Thomas and Tampé , 2017a ) : TAPBPR stabilizes the peptide-binding groove in a widened conformation primarily through the MHC I α2–1 helix , distorts the floor of the binding groove , and shifts the position of β2-microglobulin ( β2m ) . Furthermore , one of the two TAPBPR-MHC I complex structures revealed a remarkable structural feature in TAPBPR named the scoop loop ( Thomas and Tampé , 2017a ) . In TAPBPR , this loop is significantly longer than the corresponding region in Tsn , which was not resolved in the X-ray structure of Tsn ( Dong et al . , 2009 ) . Notably , the scoop loop of TAPBPR is located in the F-pocket region of the empty MHC I binding groove ( Figure 1A , B ) . By anchoring the C-terminal part of the peptide , the F pocket region is crucially involved in defining pMHC I stability ( Abualrous et al . , 2015; Hein et al . , 2014 ) . The scoop loop occupies a position that is incompatible with peptide binding and displaces or coordinates several key MHC I residues responsible for binding the C terminus of the peptide . We therefore proposed that the scoop loop can be regarded as a surrogate for the C terminus of the displaced peptide , stabilizing the inherently labile empty MHC I molecule ( Thomas and Tampé , 2017a ) . At the same time , by occupying a region critical to peptide binding , the scoop loop might allow only high-affinity peptides to re-enter the MHC I binding groove after displacement of sub-optimal peptide . The proposed importance of the scoop loop for TAPBPR function has recently been scrutinized in a study by Ilca et al . investigating TAPBPR scoop-loop variants using immunopeptidomics and cell-based assays ( Ilca et al . , 2018 ) . Ilca et al . found that a specific leucine residue in the scoop loop facilitates peptide displacement on MHC I allomorphs favoring hydrophobic peptide side chains in their F pocket . Here , we aimed to clarify the role of the scoop loop during TAPBPR-catalyzed peptide editing using in vitro interaction and peptide-exchange studies with defined , purified components . We demonstrate that the scoop loop is of critical importance for TAPBPR-mediated stabilization of empty MHC I clients in peptide-depleted environments and contributes to peptide quality control during editing by impeding released peptide to rebind in the MHC I groove . Collectively , our data support a crucial role for the TAPBPR scoop loop in establishing a hierarchical immune response . To investigate the function of the scoop loop , we prepared two human TAPBPR variants: TAPBPRTsn-SL , in which the TAPBPR scoop loop was replaced with the corresponding shorter loop of Tsn , and TAPBPRΔSL , in which the original scoop loop was essentially deleted by replacing it with three glycine residues to preserve proper folding of the MHC I chaperone ( Figure 1C ) . The ER-lumenal domains of wildtype ( wt ) TAPBPR and the variants , each harboring a C-terminal histidine tag , were expressed in insect cells and purified from the cell culture supernatant via immobilized-metal affinity chromatography ( IMAC ) and size-exclusion chromatography ( SEC ) . As MHC I chaperone clients , we chose mouse H2-Db and human HLA-A02:01 , which are known to interact with TAPBPR ( Hermann et al . , 2013; Ilca et al . , 2019; Morozov et al . , 2016 ) . HLA-A02:01 , the major MHC I allomorph in the Caucasian population and found in more than 50% of the global population , presents a diverse spectrum of immunodominant autoimmune , viral , and tumor epitopes and is therefore medically highly relevant ( Boucherma et al . , 2013 ) . The MHC I allomorphs were expressed in E . coli as inclusion bodies and refolded in the presence of β2m and fluorescently-labeled or photo-cleavable peptide ( Rodenko et al . , 2006 ) . The highly pure TAPBPR variants and pMHC I complexes eluted as monodisperse samples at expected size during SEC ( Figure 1D–F ) . During peptide exchange , MHC I molecules go through a peptide-free high-energy intermediate state after peptide release and before re-entry of a new peptide . A hallmark of peptide editors like TAPBPR is their ability to recognize and chaperone this intermediate until it is located in a peptide-rich environment where a high-affinity peptide ligand can enter the MHC I binding groove ( Thomas and Tampé , 2019; Thomas and Tampé , 2017b ) . To scrutinize the role of the scoop loop in chaperoning empty MHC I , we tested the ability of our TAPBPR variants to stabilize peptide-free H2-Db . Hence , H2-Db ( 10 µM ) loaded with a photo-cleavable peptide was incubated with TAPBPR ( 3 µM ) under UV exposure . Subsequent SEC analysis revealed that both TAPBPRTsn-SL and TAPBPRΔSL are , in principle , competent to form complexes with MHC I ( Figure 2A ) . However , in comparison to TAPBPRwt ( Figure 2A , B ) , the amount of H2-Db complex detected for TAPBPRTsn-SL and TAPBPRΔSL during SEC was reduced by around 40% and 90% , respectively ( Figure 2C ) . After reanalysis of the MHC I chaperone complexes by SEC , the mutant complexes were mostly dissociated , indicating kinetic instability ( Figure 2—figure supplement 1A ) . In contrast , isolation and reinjection of the wt complex showed that it remained stable for the duration of the experiment ( Figure 2—figure supplement 1A , B ) . Yet , in the presence of a high-affinity peptide , even the TAPBPRwt-MHC I complex dissociated , in accordance with the role of TAPBPR as a chaperone that stabilizes the MHC I as long as no optimal peptide is present ( Figure 2—figure supplement 1B ) . Taken together , these findings demonstrate that the scoop loop is crucial to an extended lifetime of the chaperone-client complex , enabling the escorting of empty MHC I by TAPBPR in a peptide-deficient environment . After investigating the chaperone activity of the TAPBPR scoop-loop mutants , we tested their ability to displace MHC I-bound peptide . To this end , we employed an in-vitro peptide exchange assay similar to the one previously described for measuring the activity of Tsn ( Fleischmann et al . , 2015; Chen and Bouvier , 2007 ) . Dissociation of medium-affinity fluorescent peptide from refolded and purified pMHC I ( p denotes fluorescently-labeled peptide ) was monitored by fluorescence polarization after addition of a 1000-fold molar excess of unlabeled high-affinity competitor peptide in the absence or presence of TAPBPR ( Figure 3A ) . The large molar excess of unlabeled competitor peptide ensures that once a fluorescent peptide dissociates , it does not rebind , but is replaced by an unlabeled competitor-peptide molecule . The observed rate constant is thus solely determined by the dissociation rate constant of the fluorescent peptide . The condition of this assay mimics the environment of the PLC , where optimal , high-affinity peptides abound . For the mouse MHC I allomorph H2-Db , TAPBPRwt and the scoop-loop variants accelerated the uncatalyzed peptide release ( 2 . 53 ± 0 . 37 × 10−3 s−1 ) to a similar extent . The TAPBPRΔSL mutant lacking the entire scoop loop exhibited slightly reduced activity ( 7 . 68 ± 1 . 17 × 10−3 s−1 ) compared to the wt protein ( 10 . 41 ± 0 . 54 × 10−3 s−1 ) , whereas TAPBPRTsn-SL was slightly more active ( 12 . 64 ± 1 . 03 × 10−3 s−1 ) ( Figure 3B , C ) . When we performed the experiment at a much lower TAPBPR concentration ( 75 nM ) , the TAPBPRs retained their activity , and the gradual activity differences between the variants remained ( Figure 3—figure supplement 1 ) . This suggests that TAPBPRwt and the scoop-loop mutants have similar affinities for H2-Db . TAPBPRwt was even able to catalyze displacement of a high-affinity peptide from H2-Db , although the catalytic effect was considerably smaller ( 1 . 8-fold acceleration ) than for H2-Db loaded with the medium-affinity peptide ( 4 . 1-fold acceleration ) ( Figure 3—figure supplement 2A , B ) . In a second set of experiments , we analyzed peptide dissociation from the human MHC I allomorph HLA-A02:01 . Similar to the experiments with H2-Db , in a peptide-rich environment ( 1000-fold molar excess of peptide ) , the highest catalytic activity towards HLA-A02:01 was observed for TAPBPRTsn-SL , followed by TAPBPRwt and TAPBPRΔSL; yet , the differences in activity between the three TAPBPRs were more pronounced , and the acceleration of the uncatalyzed peptide dissociation from HLA-A02:01 ( 1 . 90 ± 0 . 04 × 10−3 s−1 ) by TAPBPRTsn-SL ( 26 . 31 ± 2 . 59 × 10−3 s−1 ) and TAPBPRwt ( 15 . 79 ± 0 . 71 × 10−3 s−1 ) was significantly higher than for H2-Db , while the activity of TAPBPRΔSL ( 8 . 52 ± 1 . 18 × 10−3 s−1 ) remained almost the same ( Figure 3D , E ) . The validity of our peptide exchange assay was confirmed by two interface mutants of TAPBPRwt , TN3-Ala and TN6 . The TN3 ( E72K ) and TN6 ( E185K , R187E , Q189S , Q261S ) mutants were initially described for Tsn to significantly reduce or abolish MHC I binding ( Dong et al . , 2009 ) . The impact of the TN6 mutations on MHC I interaction was later confirmed for TAPBPR ( Morozov et al . , 2016 ) . According to the TAPBPR-MHC I crystal structures ( Jiang et al . , 2017; Thomas and Tampé , 2017a ) , the residue in TAPBPR ( E105 ) corresponding to the mutated residue in Tsn-TN3 forms a hydrogen bond with the swung-out Y84 of the MHC heavy chain , which is involved in coordinating the C-terminus of the peptide in liganded MHC . We reasoned that a mutation to Ala instead of Lys might increase the mutational effect and therefore generated the TN3-Ala mutant . Two of the mutated residues in TN6 ( R210 and Q212 ) are part of the jack hairpin of TAPBPR and form several interactions with MHC I heavy-chain residues , while Q275 lies in the interface with the α2–1 helix and the β8 sheet in the floor of the MHC I binding groove . Consequently , TN3-Ala and TN6 displayed drastically reduced activity towards H2-Db in our peptide-exchange experiment , with peptide dissociation rate constants close to the value of the uncatalyzed reaction ( Figure 3C , F ) . In summary , the results of our exchange assays demonstrate that under peptide-rich condition , the tested TAPBPR variants differ gradually in their displacement activity in an allomorph-dependent manner . But even the TAPBPRΔSL mutant lacking the scoop loop is still able to substantially accelerate peptide dissociation from MHC I . In the TAPBPR-MHC I crystal structure , the scoop loop binds in the F pocket region of the MHC binding groove and appears to act as a surrogate for the peptide C terminus ( Thomas and Tampé , 2017a ) . This notion is corroborated by our SEC analyses , which show that the scoop loop stabilizes peptide-free MHC I . We therefore wondered if the scoop loop impedes rebinding of displaced peptide and functions ‘in cis’ as a tethered , internal peptide competitor in the F pocket with extremely high effective concentration . To test this hypothesis , we modified the peptide exchange assay for H2-Db and HLA-A02:01 by adding in a first step only TAPBPR without competitor peptide , which allowed us to monitor the change in free and bound fluorescent peptide under the influence of peptide rebinding in the presence of TAPBPR ( Figure 4A ) . This condition mimics the physiological environment TAPBPR is operating in , where optimal replacement peptides are scarce . Strikingly , after addition of the different TAPBPRs to H2-Db loaded with fluorescent peptide , the polarization changes , which correspond to the changes in the ratio of free to bound peptide , diverged dramatically ( Figure 4B ) . Peptide dissociation was most pronounced for TAPBPRwt with the native scoop loop , reaching ~ 60% peptide release , whereas only ~ 12% of the peptide population was released from H2-Db by TAPBPRTsn-SL , and almost no decrease in polarization was caused by TAPBPRΔSL . Similar to our original peptide exchange assay ( Figure 3 ) , differences between the two MHC I allomorphs were observed: In comparison to H2-Db , TAPBPRTsn-SL-induced peptide dissociation from HLA-A02:01 was significantly stronger , approaching the level of peptide release induced by TAPBPRwt ( Figure 4—figure supplement 1A ) . Peptide release was also peptide-dependent , as H2-Db loaded with a high-affinity peptide led to a significantly smaller decline in bound peptide ( Figure 3—figure supplement 2C ) . After addition of competitor peptide ( 2nd step ) , the observed dissociation rate constants were in the same range as the values determined for the one-step experiment . Moreover , the level of released peptide after TAPBPR addition was titratable and reached saturation at 3 µM TAPBPR ( Figure 4C–E , Figure 4—figure supplement 1B ) . Under the given conditions , TAPBPRwt was able to dissociate 70% ( H2-Db ) and 80% ( HLA-A02:01 ) of total MHC I-associated peptide , respectively ( Figure 4C , Figure 4—figure supplement 1B ) . These results suggest that the scoop loop interferes with re-binding of displaced peptide . It can only be completely dislodged from the MHC I binding pocket by a high-affinity peptide . The scoop loop thus acts as a crucial selectivity filter during peptide editing on MHC I . Tsn and TAPBPR are MHC I-dedicated chaperones , which facilitate loading and selective exchange of antigenic peptides and thereby generate stable pMHC I complexes that shape a hierarchical immune response . The molecular underpinnings of their chaperone and peptide proofreading activities have only recently been uncovered by crystal structures of the TAPBPR-MHC I complex ( Jiang et al . , 2017; Thomas and Tampé , 2017a ) . Notably , one of the X-ray structures resolved a loop structure , termed the scoop loop , that is wedged into the F-pocket region of the empty MHC I binding groove and has been postulated to play an important role during peptide exchange ( Thomas and Tampé , 2017a ) . Here , we show that the TAPBPR scoop loop is indeed critically important in chaperoning intrinsically unstable empty MHC I clients in a peptide-depleted environment . This is illustrated by the reduced chaperone activity of TAPBPRTsn-SL , which harbors the shorter Tsn scoop loop , and by the dramatically reduced lifetime of the TAPBPRΔSL complex . In a peptide-rich , PLC-like environment , emulated by our one-step displacement experiments , the TAPBPRTsn-SL mutant displays the highest activity , while TAPBPRΔSL retains the ability to displace peptide . The latter observation appears to be in contrast to the study by Ilca et al . which found that TAPBPR with a mutated , but full-length scoop loop loses its ability to effectively mediate peptide dissociation ( Ilca et al . , 2018 ) . In addition to stabilizing the chaperone-MHC I complex , we demonstrate that the TAPBPR scoop loop acts as an internal peptide competitor , and thus , as a selectivity filter in the discrimination between low- and high-affinity peptides . Although a direct competition appears to be the most obvious explanation for the effect on peptide rebinding , we cannot exclude that the scoop loop exerts its influence on peptide rebinding by an allosteric mechanism . The peptide-filtering activity seems to be allomorph-dependent for TAPBPRTsn-SL . Our current interpretation of this allomorph specificity is that the Tsn scoop loop interacts more strongly with the F-pocket region of HLA-A02:01 and is therefore able to impede peptide rebinding more efficiently than in the case of H2-Db . In contrast , TAPBPRwt shows a strong peptide release activity towards both MHC I allomorphs . Based on the new insights , we propose the following model of TAPBPR-catalyzed peptide optimization on MHC I ( Figure 5 ) : The large concave surface formed by the N-terminal domain of TAPBPR mediates its initial encounter with a suboptimally-loaded MHC I , assisted by the C-terminal domain of TAPBPR , which contacts the α3 domain of the MHC I heavy chain and β2m . TAPBPR facilitates the release of low- to medium-affinity peptides primarily by widening the peptide-binding groove through the MHC I α2–1-helix , fastening the peptide-coordinating Tyr84 , distorting the floor of the binding groove , and shifting the position of β2m ( Jiang et al . , 2017; Thomas and Tampé , 2017a ) . This remodeling is made possible by the intrinsic plasticity of MHC I molecules ( Bailey et al . , 2015; Garstka et al . , 2011; McShan et al . , 2019; Natarajan et al . , 2018; Thomas and Tampé , 2017b; van Hateren et al . , 2017; van Hateren et al . , 2015; Wieczorek et al . , 2017 ) , and it appears to be induced primarily by structural elements of TAPBPR that lie outside the scoop loop . As a result , the TAPBPRΔSL mutant lacking the scoop loop is still able to catalyze peptide displacement . Once the suboptimal peptide has been released , the scoop loop occupies the position of the peptide C terminus in the F-pocket region . The scoop loop thereby contributes to the stabilization of the peptide-deficient binding groove . Our two-step peptide exchange — mimicking a peptide-depleted environment — demonstrates that the scoop loop functions at the same time as a peptide selectivity filter by impeding re-binding of the replaced peptide , either through direct competition with the C terminus of the incoming replacement peptide or through an allosteric mechanism . Hence , the scoop loop contributes to the significant affinity decrease of incoming peptides for the MHC I groove in the presence of TAPBPR ( McShan et al . , 2018 ) . Assuming a mode of direct competition , the replacement peptide would dock in the MHC I groove first with its N terminus , before it competes with the TAPBPR scoop loop over the F pocket region ( Hafstrand et al . , 2019; Thomas and Tampé , 2017a ) . Negative allosteric coupling between different parts of the MHC I molecule might play a role in the final release of TAPBPR ( McShan et al . , 2018 ) . The shorter scoop loop in Tsn suggests that its selective pressure on the replacement peptide is weaker than in TAPBPR . Indeed , our fluorescence polarization and SEC analyses show that the tapasin scoop loop in TAPBPRTsn-SL is less efficient in preventing re-binding of dissociated peptide . Physiologically , these observations might be explained by the fact that Tsn functions within the PLC , a ‘nanocompartment’ characterized by an abundant and diverse supply of optimal peptides , reaching a bulk concentration of up to 16 µM before the TAP transporter is arrested by trans-inhibition ( Grossmann et al . , 2014 ) . Moreover , Tsn is supported by other PLC chaperones in stabilizing empty MHC I clients . In contrast , TAPBPR operates as a single MHC I-dedicated chaperone outside the PLC in environments where the concentration of high-affinity peptides is drastically lower and MHC I clients have to be stabilized in a peptide-receptive state for extended periods of time . Long-term stabilization of suboptimally-loaded or empty MHC I by TAPBPR also allows the major ER/cis-Golgi glycoprotein folding sensor UGGT1 ( UDP-glucose:glycoprotein glucosyltransferase 1 ) to re-glucosylate the MHC I molecule in order to feed it back into the calnexin/calreticulin cycle and/or allow recruitment of the MHC I to the PLC ( Neerincx et al . , 2017; Thomas and Tampé , 2019 ) . In conclusion , the evidence provided by our study indicates that the scoop loop is evolutionarily fine-tuned to enable Tsn and TAPBPR to accomplish their dual function as chaperone and proofreader in the specific subcellular location they operate in . By serving both as a stabilizing element and as selectivity filter in TAPBPR , the scoop loop influences peptide editing and impacts the repertoire of MHC I-associated epitopes presented on the cell surface . The DNA constructs of human β2m , the ectodomain of mouse H2-Db , and TAPBPRwt were identical to the ones previously described ( Thomas and Tampé , 2017a ) , except for position 97 in TAPBPRwt , which contained the native cysteine . The TAPBPR scoop loop mutants TAPBPRTsn-SL and TAPBPRΔSL were generated by overlap extension PCR , the TN3-Ala and TN6 mutants were generated by site-directed mutagenesis . The TN3-Ala and TN6 mutants harbored the same mutations that were described for the corresponding mutants of Tsn ( Dong et al . , 2009 ) , except that in TN3-Ala E105 was mutated to alanine . TAPBPRTsn-SL , TAPBPRΔSL , TN3-Ala , and TN6 all contained the C97A mutation . Human HLA-A02:01 ( amino acids 1–278 ) was cloned into pET-28 ( Novagen , Merck Millipore , Darmstadt , Germany ) and ended in a C-terminal His6-tag preceded by a linker ( sequence: HE ) . The amino acid numbering of TAPBPR is based on the mature protein as defined by N-terminal sequencing ( Zhang and Henzel , 2004 ) . Human β2m and the ectodomains of mouse H2-Db and human HLA-A02:01 were expressed as inclusion bodies in Escherichia coli BL21 ( DE3 ) as described before ( Rodenko et al . , 2006; Thomas and Tampé , 2017a ) . TAPBPR proteins were expressed in Spodoptera frugiperda ( Sf21 or Sf9 ) insect cells according to standard protocols for the Bac-to-Bac system ( Thermo Fisher Scientific , Waltham , MA ) . A high-titer recombinant baculovirus stock was used to infect the insect cells at a density of 1 . 5–2 . 0 × 106 cells/mL , which were cultivated in Sf-900 III SFM medium ( Thermo Fisher Scientific ) at 28°C . The cell culture medium containing secreted TAPBPR was harvested 72 hr after infection . β2m was refolded by dialysis essentially as described previously ( Rodenko et al . , 2006 ) and purified by SEC on a Superdex 75 column ( GE Healthcare , Piscataway , NJ ) in HEPES-buffered saline ( 1xHBS: 10 mM HEPES pH 7 . 2 , 150 mM NaCl ) . Purified protein was concentrated by ultrafiltration ( Amicon Ultra 3 kDa MWCO , Merck Millipore ) . The following peptides were used: the photo-cleavable peptide photo-P18-I10 ( RGPGRAFJTI ) ( H2-Db ) [J=3-amino-3- ( 2-nitro ) phenyl-propionic acid] , the non-fluorescent competitor peptides ASNENMETM ( H2-Db ) and FLPSDEEPYV ( HLA-A02:01 ) , as well as the fluorescently labeled peptides TQSCNTQSI ( H2-Db ) , FLPSDCFPSF ( HLA-A02:01 ) , ASNCNMETM ( H2-Db ) ( C* denotes TAMRA-labeled cysteine ) . Non-natural peptide epitopes were designed based on their theoretical affinities according to the NetMHCpan server ( Jurtz et al . , 2017 ) . While TQSCNTQSI and FLPSDCFPSF were constructed to have medium affinity ( 500–600 nM ) , ASNCNMETM and the competitor peptides were designed to be high-affinity ( 8–10 nM ) ligands . Peptides were synthesized using standard Fmoc solid-phase chemistry and purified by C18 reversed-phase HPLC . The identity of peptides was verified either by matrix-assisted laser desorption/ionization mass spectrometry ( MALDI-MS ) or by electrospray ionization-mass spectrometry ( ESI-MS ) . In order to site-specifically label peptides with fluorophores , 10 . 5 µM peptide were incubated with 26 µM TAMRA-5-maleimide ( single isomer , Thermo Fisher Scientific ) or TAMRA-6 C2 maleimide ( Lumiprobe , Hannover , Germany ) ( used for labeling of FLPSDCFPSF ) overnight at 4°C . Labeled peptides were purified by C18 reversed-phase HPLC , and their identity was confirmed by ESI-MS . H2-Db and HLA-A*02:01 were refolded from inclusion bodies by rapid dilution in the presence of purified β2m and peptide according to established protocols ( Rodenko et al . , 2006 ) . Refolded MHC I complexes were purified by SEC ( Superdex 200 Increase 10/300 , GE Healthcare ) in 1xHBS and concentrated by ultrafiltration ( Amicon Ultra , Merck Millipore ) . TAPBPR proteins were purified from the insect cell culture medium by IMAC according to a protocol published earlier ( Thomas and Tampé , 2017a ) , polished by SEC ( Superdex 200 Increase 10/300 , GE Healthcare ) in 1xHBS , and concentrated by ultrafiltration ( Amicon Ultra , Merck Millipore ) . Dissociation of fluorescently labeled peptide from MHC I was monitored at 23°C in 1xHBS by fluorescence polarization ( Fluorolog-3 spectrofluorometer , Horiba Jobin Yvon , Bensheim , Germany ) with λex/em of 530/560 nm . One-step and two-step dissociation assays were carried out with 300 nM MHC I loaded with TAMRA-labeled peptide , 1 µM TAPBPR , and 300 µM competitor peptide . Dissociation rate constants were determined in GraphPad Prism using a one-phase exponential decay regression . In the presence of purified TAPBPR ( 3 µM ) , photo-P18-I10-loaded H2-Db ( 10 µM ) was irradiated with UV light ( 36 nm , 185 mW/cm2 , 120 s ) on ice and afterwards incubated for 10 min at room temperature . Samples were subsequently centrifuged at 10 , 000xg for 10 min and analyzed by analytical SEC on a Superdex 75 ( 3 . 2/300 ) column ( GE Healthcare ) . SEC runs were conducted in 1xHBS and monitored by absorbance at 280 nm . Chromatograms were deconvoluted into three Gaussian functions using the program Fityk 1 . 3 . 1 ( Wojdyr , 2010 ) . The amount of complex was assessed by the area of the complex peak . Purified peptide-deficient TAPBPRwt-H2-Db , TAPBPRTsn-SL-H2-Db , and TAPBPRΔSL-H2-Db complexes were analyzed via analytical SEC either on a Superdex 75 ( 3 . 2/300 ) or a Superdex 200 ( 3 . 2/300 ) column ( GE Healthcare ) at a flow rate of 0 . 075 mL/min . A separate sample of purified TAPBPRwt-H2-Db complex was incubated with a 100-fold molar excess of high-affinity peptide prior to re-analysis by SEC .	Cells in the body keep the immune system informed about their health by showing it fragments of the proteins they have been making . They display these fragments , called peptides , on MHC molecules for passing immune cells to inspect . That way , if a cell becomes infected and starts to make virus proteins , or if it becomes damaged and starts to make abnormal proteins , the immune system can ‘see’ what is happening inside and trigger a response . MHC molecules each have a groove that can hold one peptide for inspection . For the surveillance system to work , the cell needs to load a peptide into each groove before the MHC molecules reach the cell surface . Once the MHC molecules are on the cell surface , the peptides need to stay put; if they fall out , the immune system will not be able to detect them . The problem for the cell is that not all peptides fit tightly into the groove , so the cell needs to check each one before it goes out . It does this using a protein called TAPBPR . TAPBPR has a finger-like structural feature called the "scoop loop" , which fits into the end of the MHC groove while the molecule waits for a peptide . It was not clear , however , what this loop actually does . To investigate , Sagert et al . mutated the scoop loop of TAPBPR to see what happened to MHC loading in test tubes . The experiments revealed that the scoop loop plays two important roles . The first is to keep the MHC molecule stable when it is empty , and the second is to hinder unsuitable peptides from binding . The scoop loop sticks into one side of the groove like a tiny hairpin , so that pushed-out , poorly fitting peptides cannot reattach . At the same time , it holds the MHC molecule steady until a better peptide comes along and only releases when the new peptide has slotted tightly into the groove . Understanding how cells choose which peptides to show to the immune system is important for many diseases . If cells are unable to find a suitable peptide for a particular illness , it can stop the immune system from mounting a strong response . Further research into this quality control process could aid the design of new therapies for infectious diseases , autoimmune disorders and cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "immunology", "and", "inflammation" ]	2020	A loop structure allows TAPBPR to exert its dual function as MHC I chaperone and peptide editor
The hologenome concept proposes that microbes and their host organism are an independent unit of selection . Motivated by this concept , we hypothesized that thermal acclimation in poikilothermic organisms , owing to their inability to maintain their body temperature , is connected to their microbiome composition . To test this hypothesis , we used a unique experimental setup with a transgenerational selective breeding scheme for cold tolerance in tropical tilapias . We tested the effects of the selection on the gut microbiome and on host transcriptomic response . Interestingly , we found that host genetic selection for thermal tolerance shapes the microbiome composition and its response to cold . The microbiomes of cold-resistant fish showed higher resilience to temperature changes , indicating that the microbiome is shaped by its host's selection . These findings are consistent with the hologenome concept and highlight the connection between the host and its microbiome's response to the environment . Cold temperature is an environmental challenge that greatly affects metabolic and physiological processes . Thus , adaptations to temperature fluctuations are expected to be found across all organisms . When exposed to cold temperatures , animals undergo remarkable physiological adjustments in order to maintain homeostasis . Such adjustments may occur through genetic and/or non-genetic mechanisms that increase their fitness . Recent work in mice revealed that the gut microbiome facilitates key adaptations during cold exposure by promoting energy demand-driven regulation ( Chevalier et al . , 2015; Rosenberg and Zilber-Rosenberg , 2018 ) . This is not surprising , as many studies have already proven the vital importance of gut microbial communities for host survival , homeostasis , development and functioning ( McFall-Ngai et al . , 2013; Shabat et al . , 2016 ) . In fact , the universality of host–microbe associations , either transient or tight , inspired the hologenome concept ( Bordenstein and Theis , 2015; Brucker and Bordenstein , 2013; Rosenberg and Zilber-Rosenberg , 2013; Rosenberg and Zilber-Rosenberg , 2016; Rosenberg and Zilber-Rosenberg , 2018; Theis et al . , 2016 ) , which proposes that within the holobiont , units at different levels , such as genes , chromosomes or biont combinations ( i . e . , host and microbes ) , are subject to selection or neutrality . Within this concept , host–microbe interactions have an important role in the host's physiology , whilst microbiome composition may be affected by host selection . Taking this into account , we hypothesized that host selection may facilitate changes in the host-associated microbial species and in their response to environmental selection pressure . Poikilothermic organisms , such as fish , must develop strategies to maintain homeostasis within diverse temperature gradients because environmental temperature changes would otherwise disrupt homeostasis and cause deleterious effects on vital physiological functions ( Guschina and Harwood , 2006; Vinagre et al . , 2012 ) . The physiological response to stress may vary among populations or individuals and can be affected by various factors ( Nitzan et al . , 2016; Somero , 2010 ) . Microbial communities in aquatic environments have also been reported to be affected by abiotic factors that act as a habitat-filtering force , such as temperature ( Fuhrman et al . , 2008 ) . In poikilothermic organisms , the gut microbiome experiences temperature fluctuations that do not exist in homeotherms . There is evidence of a temperature effect on the microbial dynamics associated with invertebrate species ( Beleneva and Zhukova , 2009; Carlos et al . , 2013; Erwin et al . , 2012; Lokmer and Mathias Wegner , 2015; Mahalaxmi et al . , 2013; Preheim et al . , 2011; Zurel et al . , 2011 ) , but such information is largely missing for aquatic poikilothermic vertebrates . Taking the hologenome concept into account , we asked whether host transgenerational selection for thermal tolerance , which is instrumental to poikilothermic organisms’ fitness , involves the host–gut microbiome axis . More specifically , how does microbiome composition respond to temperature changes and does host tolerance shape this response ? Here , we hypothesized that the thermal tolerance of poikilothermic vertebrates affects their associated gut microbial species , as well as the response of these microbes to temperature alterations . We focused specifically on the association of changes in the microbiome with the transgenerational low-temperature response in tropical fish , such as tilapia species . The growth of these fish species is highly affected by temperature; they are subject to growth inhibition and mortality when exposed to low temperatures ( Cnaani et al . , 2000 ) . However , recent work revealed transgenerational inheritance of cold tolerance in these species ( Nitzan et al . , 2016 ) . Therefore , we used a unique setup , composed of 66 fish families that are part of a transgenerational selective-breeding scheme for enhanced cold tolerance ( Figure 1 ) . In this setup , we chose fish progeny from families with the most extreme phenotypes , selected on the basis of the survival rate of their siblings at low temperatures and characterized as cold-resistant or cold-sensitive . Hence , these fish progenies had never experienced low temperature exposure before they were challenged . We used this setup to understand how selection on host thermal tolerance corresponds to microbiome shifts . Our results suggest that host selection for cold tolerance is followed by changes in the gut microbiome , which are manifested by higher resilience to temperature shifts and which may be a key factor in orchestrating cold acclimation in poikilothermic animals . We analyzed microbiome populations with respect to two potential habitat-filtering forces: temperature and host thermal tolerance through genetic selection . Temperature is known to affect microbial composition in aquatic habitats and to act as a habitat filter ( Fuhrman et al . , 2008 ) . There is some limited information on the effect of temperature on the gut microbiome , showing compositional changes in homeotherms ( Chevalier et al . , 2015 ) . However , such information is largely missing in poikilothermic vertebrates , in which physiology is strongly influenced by environmental temperature . To explore whether the thermal tolerance of poikilothermic vertebrates affects their associated gut microbial species and their response to temperature alterations , we used the blue tilapia , Oreochromis aureus , as our experimental model . The optimal temperatures for this species , similar to those for other tilapiine species , are 24–28°C ( Trewavas , 1983 ) , but this species is one of the most cold-tolerant among this group of fish ( Cnaani et al . , 2000 ) . An ongoing selective-breeding process , which has been running for the past few years in Israel , is aimed at improving growth and cold-tolerance traits in this species ( Zak et al . , 2014 ) . In a recent study , we found that the fish's tolerance to temperatures below 16°C is transgenerational and potentially connected to maternal effects ( Nitzan et al . , 2016 ) . Thus , the present study focused on understanding host–microbiome interactions in relation to temperature stress and cold acclimation . To evaluate the responses of the fish that are resistant and sensitive to cold exposure , we challenged a total of 84 fish , sampled at 24°C or 12°C , originating from 3 tilapia families with high tolerance and 3 with low tolerance to cold temperatures ( 7 fish per family ) . For our experiments , we chose fish originating from mothers that had the most extreme phenotypes , selected on the basis of their sibling’s survival rate at low temperature , as determined by previous cold-challenge trials ( Figure 1; Figure 1—figure supplement 1 ) . All of the fish within each generation were kept in the same tank to eliminate confounding effects . The test groups , consisting of 42 fish each from resistant and sensitive families , were challenged by temperature reduction to 12°C over a period of 2 weeks and compared to a control group , consisting of 42 fish from the same families kept at 24°C . At the end of the temperature-challenge trial , we analyzed the fish gut microbiomes from the test and control groups using 16S rRNA gene sequencing , and we also looked at the host's response using liver transcriptomic analysis ( Figure 1 ) . Linear mixed-effects model analysis indicated that temperature was the major factor shaping microbial community diversity and richness within the fish gut ( Table 1; FShannon = 46 . 43 , FRichness = 45 . 79 , p = 0 . 001 ) . Indeed , we found a dramatic decrease ( Figure 2Ai; Wilcoxon t-test , two-sided , 95% confidence interval ( CI ) , p = 7 . 2E-02 ) in microbial diversity ( Shannon index H’ ) and richness ( Figure 2—figure supplement 1; Wilcoxon t-test , two-sided , 95% CI , p = 2 . 7E-08; Figure 2—figure supplement 2 , rarefaction curves ) when both groups were exposed to cold temperature . This decrease was in agreement with the low-temperature decrease in the 16S gene copy numbers ( real-time PCR; Figure 2—figure supplement 3 ) . Furthermore , the microbial communities in the fish that were subjected to cold temperatures exhibited lower variability in richness between samples , whereas greater individual variation was observed in the high temperature group ( Figure 2—figure supplement 1i ) , suggesting that low-temperature conditions are very selective and potentially constrain microbial communities , decreasing their inter-individual variation . Microbial community structure ( β-diversity ) was also dramatically affected by cold exposure , being the major factor associated with increasing similarity between individuals in the communities ( Table 2; FTemperature = 6 . 5072 , p = 0 . 002 ) , after gut physiology effects ( FPart = 28 . 8936 , p = 0 . 001 ) . Orders from the Proteobacteria phylum , and more specifically Vibrionales and Alteromonadales , were enriched at 12°C ( Figure 2B; Supplementary file 1 , Table S1; Figure 2—figure supplement 4 ) , with prior studies reporting these taxa's potential for survival in cold temperatures ( Lauro et al . , 2011; Math et al . , 2012; Raymond-Bouchard and Whyte , 2017 ) . Overall , however , as microbial diversity and richness decreased ( Figure 2Ai; Figure 2—figure supplement 1i ) , several microbial phyla , such as Planctomycetes , Bacteroidetes and Verrucomicrobia , were significantly depleted after cold-exposure ( Figure 2B; Figure 2—figure supplement 5; Supplementary file 1 , Table S1 ) . These findings indicate that in poikilothermic vertebrates such as the blue tilapia , gut microbial communities are strongly affected by environmental temperature . After observing a drastic decrease in gut microbial diversity during cold exposure ( Figure 2Ai ) , we sought to evaluate how the host's thermal tolerance affects microbiome composition . Moreover , if host thermal tolerance is connected to cold temperature acclimation , then we would expect the microbiome response to relate to different tolerance phenotypes . We tested this hypothesis by evaluating the response of the cold-sensitive and cold-resistant selected families to the changing environmental conditions . We compared the gut microbial compositions of the sensitive and resistant families and found that host thermal tolerance has a significant effect on gut microbial diversity ( Table 2; FTolerance = 2 . 2269 , p = 0 . 04 ) . Furthermore , we also observed a significant interaction between host thermal tolerance and temperature , indicating a different microbiome response to the lower temperature in the resistant vs . sensitive hosts ( FTxT = 3 . 6113 , p = 0 . 012 ) . Interestingly , when fish from both groups were kept under their optimal temperature conditions of 24°C , the sensitive fish exhibited higher microbial diversity ( Shannon H′ ) than the resistant ones , as well as higher individual variability ( Figure 2Aii; Wilcoxon t-test , two-sided , 95% CI , p = 0 . 048 ) . Indeed , when we compared the β-diversity of the two groups under these optimal conditions , we found that cold-resistant fish had a higher within-group microbiome similarity than the sensitive fish ( Figure 2Ci; non-parametric Bonferroni-corrected p = 0 . 01 , using 1000 Monte Carlo permutations ) , potentially because of the selection process to which resistant fish have been subjected with regard to the cold-resistance trait , thus showing host control of microbiome composition . We further aimed to gain better insight into the microbiome composition of these host groups . We first compared the microbiome composition of cold-resistant and cold-sensitive hosts under each temperature condition; more specifically , we asked whether microbiomes of similar ( Shannon H′ ) diversity also share the same microbiome compositions . To answer this , we stratified our data and examined individuals from each host group with similar Shannon H′ diversity ( individuals from each group with the 10 highest or 10 lowest Shannon H′ diversity values; Supplementary file 1 , Table S2 ) . We found that the individuals from both high- and low-diversity microbiomes ( Principal coordinate analysis; Figure 2—figure supplement 6 ) clustered significantly according to host group ( Permanova analysis; Supplementary file 1 , Tables S3 and S4 ) , suggesting that host genetic background ( genetic selection for cold tolerance ) had a strong effect on shaping the microbiome composition . Our next step was to explore the microbial taxa that are associated with genetic background effects . Indicator species analysis ( which identifies habitat-associated species on the basis of their fidelity and relative abundance in different environments; see 'Materials and methods' ) revealed several microbial taxa that were significantly associated with host genetic background and had been described previously as naturally occurring strains in the tilapia gut ( Giatsis et al . , 2016; Haygood and Jha , 2016 ) ( Supplementary file 1 , Tables S5 and S6; Figure 2—figure supplements 7–8 ) . Taxa including microbial species such as Cetobacterium somerae ( Tsuchiya et al . , 2008 ) and species of the order Vibrionales ( i . e . , Vibrio cholerae; Figure 2—figure supplement 5F ) , which were also described as microbes surviving in environments with a broad temperature range ( Raymond-Bouchard and Whyte , 2017; Townsley et al . , 2016 ) , were enriched in the resistant fish gut ( Figure 2D; Figure 2—figure supplement 8; Supplementary file 1 , Tables S5 and S6 ) . On the other hand , a higher diversity of taxa , including Prevotella sp . , Streptococcus luteciae and Bacteroidetes species , as well as species from the families Christensenellaceae , Succinivibrionaceae and Clostridiaceae , was enriched in the sensitive fish gut ( Figure 2D; Supplementary file 1 , Tables S5 and S6 ) . Altogether , our results indicate that host genetic background shapes microbiome composition by selecting for specific microbes , with some of them potentially carrying fitness traits for low-temperature tolerance . As host fish were specifically selected according to their thermal tolerance , and as our findings indicated that temperature and host genetic background affect the fish gut microbiome , we further asked whether the microbiome's response to temperature stress is also affected by its host's tolerance . When exposed to the cold temperature of 12°C , both resistant and sensitive hosts exhibited a decrease in their microbiome diversity and richness ( Figure 2A; Figure 2—figure supplement 1 ) , as well as changes in microbiome composition ( Figure 2B; Figure 2—figure supplements 4–8 ) . When we compared the β-diversity within each of these host groups in relation to changing temperature , however , we found that after exposure to the stressful low-temperature conditions , the microbiomes of cold-resistant fish families were less affected than those of sensitive ones ( Figure 2Cii; non-parametric Bonferroni-corrected p = 0 . 02 , using 1000 Monte Carlo permutations ) . More specifically , when we compared the similarity of the microbial compositions between fish at 24°C and 12°C , we found it to be significantly greater in the resistant fish families than in the sensitive ones , showing a stronger effect of temperature change in the sensitive fish ( Figure 2Cii ) . We next asked whether the microbiome's differential response to temperature in the two host groups also applies to microbes that are shared between the two host groups . We assessed microbiome resilience by looking at 11 selected taxa that were present in >50% of the individuals from each temperature environment ( the 'core microbiome' ) . After renormalizing the core microbes' abundance ( expressed as percentage within the core microbiome community; Figure 2E ) , we evaluated the fold changes in the abundance of these taxa upon change from 24°C to 12°C in both the cold-resistant and cold-sensitive fish . Similar to the overall microbiome response ( Figure 2Cii ) , the fold change in the relative abundance of the core microbiome in resistant fish was significantly lower than that in sensitive ones ( Figure 2Eii ) . We further designed specific primers to measure the absolute copy numbers of the two most abundant and prevalent ( >95% of the individuals ) core taxa ( Pseudomonas veronii and Janthinobacterium lividum ) in both host groups and quantified their abundance by quantitative PCR ( see 'Materials and methods' ) . Our results were in agreement with the sequencing data ( Figure 2Ei , ii ) , showing a significantly less pronounced temperature effect on the abundance of the core microbes in the gut of the resistant fish compared to that in the gut of the sensitive ones ( Figure 2Eiii ) . These results thus supported our hypothesis that host thermal tolerance modulates the microbiome's response to temperature changes . Since we found that microbiome composition and response to temperature changes are connected to the host's thermal tolerance , we next evaluated the microbial community composition and attributes in relation to temperature and host cold acclimation . We utilized the linear discriminant analysis ( LDA ) effect size ( LEfSe ) tool ( Segata et al . , 2011 ) and the PICRUSt ( Phylogenetic Investigation of Communities by Reconstruction of Unobserved States ) tool ( Langille et al . , 2013 ) . During cold exposure , we found a significant change ( enrichment or depletion ) in several pathways in all host microbiomes , resistant and sensitive . More specifically , a decrease in the abundance of genes belonging to metabolic pathways was observed , whereas genes in pathways related to cellular processes and signaling increased significantly in abundance ( Figure 3A ) . In line with the changes in microbial composition , the overall fold change in microbial coding capacity between warm and cold temperature was significantly greater in the cold-sensitive fish ( Figure 3B; p < 0 . 0001 , Wilcoxon rank-sum test , two-sided ) , showing a more pronounced response in this group . Specifically , microbiome composition changes resulting from cold exposure seem to relate to functional shifts , which overall correspond to a stress-related response and to cell defense , in line with prior studies suggesting reduction of metabolic activity and transcription , and increases in motility , chemotaxis and membrane transport in response to cold ( Nachin et al . , 2005 ) . Therefore , host thermal tolerance seems to contribute to and to enhance the differences in microbiome functions with respect to thermal stress , potentially suggesting that not only microbiome composition , but also certain attributes of the microbiome may be influenced by host genetic background and response to low temperatures . To this point , our findings indicated that selection for host thermal tolerance shapes microbiome composition and response to decreased temperatures . We then asked whether the host response to the change in temperature in the resistant and sensitive families agrees with the microbiome response , as it is known that cold exposure greatly affects metabolic and physiological processes in poikilothermic vertebrates , such as fish , toward maintaining fitness and adaptation ( Schulte et al . , 2011 ) . Specifically , we asked whether host response in the resistant fish families is more moderate than that in the sensitive families . We sequenced the transcriptome in the liver , which is an important and sensitive organ for stress and immune regulation in fish ( Kokou et al . , 2016; Möller et al . , 2014 ) . In agreement with our findings of a differential microbiome response to cold exposure between the resistant and sensitive fish , the transcriptomic response in the liver showed that the overall fold change in gene expression between warm and cold temperature conditions was significantly higher in the cold-sensitive fish ( Figure 4A; p < 0 . 0001 , Wilcoxon rank-sum test , one-sided ) , showing a more pronounced response in this group . Moreover , when we compared the similarity in the transcript presence/absence patterns of the host response between fish in 24°C and 12°C conditions , we found it to be significantly higher in the resistant families than in the sensitive ones , showing a stronger effect of temperature change in the sensitive fish ( Figure 4B; Figure 4—figure supplement 1 ) , as was also the case for the microbiome response ( Figure 3B ) . Similarly , when we analysed the transcriptomic response for host-associated genes ( sensitive vs . resistant; see 'Materials and methods' ) , the response of the resistant hosts to a change in temperature from 24°C to 12°C changed significantly less than that of the sensitive fish ( Figure 4C; Figure 4—figure supplement 2 ) . To further test such agreement between host and microbiome responses , we compared the posterior gut microbiome structure and the liver transcriptome of resistant and sensitive fish . Specifically , we performed a Principal Component Analysis ( PCA ) for both the microbiome and transcriptome and we correlated the top five microbiome principal components ( PCs ) to the top five transcriptome PCs by means of Canonical Correlation Analysis ( CCA ) , which aims to maximize and measure the correlation between two multivariate datasets . Interestingly , the measured correlation coefficient was significant when we compared it to correlation coefficients achieved after randomly shuffling the sample labels ( Figure 4B; p = 0 . 0044 ) . This analysis showed that more than 80% of the variance observed in the microbiome response to temperature could be explained by the variance observed in the transcriptomic response ( Figure 4B; actual variance was 83% vs . permuted 44% ) , thus supporting a host–microbiome interaction in response to environmental changes . Notably , when we looked at the host transcriptome response to cold exposure , we found several metabolic pathways ( Figure 3C ) that were also affected in the microbiome ( Figure 3A ) , with most of the changes occurring in the same direction ( depletion or enrichment ) . Looking at the functions that are commonly shared between the host and the microbiome response to cold exposure ( Figure 3C ) , we found that some of these functions , namely transcription- and cytochrome P450-related pathways ( Figure 3C; Figure 3—figure supplement 1 ) , were also reported as key cell metabolism cold-induced adaptations in homeotherms ( Shore et al . , 2013 ) . Such findings led us hypothesize that these functions may be conserved with regard to host and gut microbiome responses to temperature changes . Our insight into the microbial dynamics of the blue tilapia gut microbiome in response to environmental conditions shows that temperature , host genetic background and tolerance to thermal stress are major determinants of community structure and dynamics . Microbial community diversity and richness severely decreased during cold exposure; but microbial response was related to host thermal tolerance , which affected microbiome stability ( Figure 5 ) . These results indicate that host cold tolerance shapes not only gut microbiome composition but also the sensitivity of core microbial communities to temperature , and that host cold tolerance modulates the thermal acclimation of the gut microbiome . These finding are consistent with the hologenome concept , which proposes that associations exist between the microbes and their hosts , suggesting that together they consist a unit of natural selection . Thus , taking this concept into account , we highlight potential connections between host acclimation and its microbiome response to environmental stress . Three sensitive and three resistant families of blue tilapia ( Nitzan et al . , 2016 ) , selected for different survival rates during cold exposure ( Figure 1—figure supplement 1 ) , were raised at the Dor Aquaculture Research Station , Israel , and transferred to a climate-controlled room in the Institute of Animal Science , Agricultural Research Organization ( Rishon LeZion , Israel ) . Progenies of each mother are termed family in our dataset ( we used 6 families in total - 3 resistant and 3 sensitive ) . Seven fish from each family were cold-challenged by gradual temperature decrease from 24°C to 12°C at a rate of 1 °C/day and then held at 12°C for 2 days , while their full-siblings were maintained at 24°C . No mortality was observed during this period . Liver tissue from three fish per family in each treatment was dissected and kept in Ambion RNAlater buffer at −20°C for transcriptome analysis . In addition , two gut compartments were sampled , the anterior and posterior intestine , from the 21 tolerant and 21 sensitive fish ( 7 from each family ) in each treatment ( in total 84 fish in cold and warm conditions; 168 samples from both anterior and posterior intestine ) . After dissection , the gut parts , including content and mucosa , were removed and the connective tissue cleaned; each sample was then ground , frozen and stored at −80°C for further analysis of microbiome composition . This study was approved by the Agricultural Research Organization Committee for Ethics in Using Experimental Animals and was carried out in compliance with the current laws governing biological research in Israel ( Approval number: 146/09IL ) . Bacterial DNA was isolated from gut samples using the protocol described by Roeselers et al . ( 2011 ) with some modifications ( Sun et al . , 2013 ) . Excised intestines were combined in 2 . 0 ml screw-cap tubes with 0 . 5 mm and 1 mm silica beads , 400 ml 50 mM Na-phosphate buffer ( pH 8 . 0 ) and 200 ml of lysis solution containing 5% w/v sodium dodecyl sulfate , 0 . 5 M Tris-HCl ( pH 8 . 0 ) and 0 . 1 M NaCl . Samples were homogenized in a bead-beater for 5 min on high speed and centrifuged at 16 , 000 g for 5 min . The supernatant was transferred to new tubes and lysozyme ( Sigma , St . Louis , MO ) was added to a final concentration of 2 mg/ml followed by incubation at 42°C for 1 hr and then at 37°C for 1 hr . Following this step , the solution was sequentially extracted with TE ( 10 mM Tris-HCl ( pH 8 . 0 ) and 1 mM EDTA ) , saturated phenol , phenol-chloroform ( 1:1 v/v ) , and chloroform-isoamyl alcohol ( 24:1 v/v ) . Finally , DNA in the aqueous phase was precipitated with 0 . 1 vol 3 M sodium acetate ( pH 5 . 2 ) and 0 . 7 vol isopropanol . The concentration of DNA in the solution was measured using a UV-Vis Spectrophotometer ( Thermo Scientific , Waltham , MA ) ( Cnaani et al . , 2000 ) and stored at −20°C for further analysis . Only samples that resulted in a high yield of quality DNA were used for subsequent analyses . Quantitative real-time PCR analysis was performed to measure the total number of bacteria in the anterior and posterior parts of both resistant and sensitive fish at both temperatures , through amplification of their 16S rRNA gene . Real-time PCR was performed in a 10 µl reaction mixture containing 5 µl Absolute Blue SYBR Green Master Mix ( Thermo Scientific ) , 0 . 5 µl of each primer ( 10 mM working concentration; Forward 5’-ACTCCTACGGGAGGCAGC-3’ and Reverse 5’-GTATTACCGCGGCTGCTGGCA-3’ ) , 3 µl nuclease-free water and 2 µl of 100 ng DNA template . Amplification involved one cycle held at 95°C for 15 min for initial denaturation and activation of the hot-start polymerase system , and then 40 cycles at 95°C for 10 s followed by annealing for 15 s at 60°C and extension at 72°C for 20 s . Bacteria were quantified using a standard curve for the 16S rRNA gene at different concentrations ( 102–108 copies/µl ) and the results were expressed as 16S gene copy numbers per microliter . Sequencing of the PCR-amplified V4 region of 16S rRNA was performed using the Next Generation system ( Illumina , California , United States ) ( Kanehisa and Goto , 2000 ) . First , amplification of the V4 region , using primers 515F ( 5’-GTGCCAGCMGCCGCGGTAA-3’ ) and 806R ( in which each R contained a different 12 bp index ) , was performed under the following conditions: 94°C for 15 min , followed by 35 cycles of 94°C for 45 s , 50°C for 60 s and 72°C for 90 s , and a final elongation step at 72°C for 10 min . The PCR product ( 380 bp ) was cleaned using DNA Clean and Concentrator ( Zymo Research , California , United States ) and the fragments containing the Illumina adaptors were quantified . Amplification involved one cycle held at 95°C for 15 min for initial denaturation and then 40 cycles at 95°C for 10 s , followed by annealing at 60°C for 20 s and extension at 72°C for 30 s . The product was quantified using a standard curve with serial DNA concentrations ( 0 . 1–10 nM ) . Finally , the samples were equimolarly diluted to a concentration of 0 . 4 nM and prepared for sequencing according to the manufacturer’s instructions . An open-source software package , DADA2 ( Callahan et al . , 2016 ) , was applied to model and correct Illumina-sequenced amplicon errors , following the tutorial suggestions ( https://github . com/benjjneb/dada2 ) . DADA2 resolves differences at the single-nucleotide level and the end product is an amplicon sequence variant table , which is a higher-resolution analog of the traditional OTU table , recording the number of times each exact sequence variant ( ESV ) was observed in each sample ( 100% sequence identity ) . Taxonomy was assigned using the Ribosomal Database Project Classifier ( Wang et al . , 2007 ) against the 16S gene reference Greengenes database ( 13 . 8 version ) ( McDonald et al . , 2012 ) . Owing to the variation in sequence depths between samples , all samples were normalized to the lowest depth by subsampling ( 6000 read/sample ) and sequences that were present in fewer than two samples ( doubletons ) were discarded from the final table ( Beleneva and Zhukova , 2009 ) . Sequences were submitted under the accession number SRP131209 in the Sequence Read Archive ( SRA ) . Richness ( number of observed species ) and Shannon α-diversity were calculated using QIIME . To assess β-diversity , cluster analyses exploring the similarities between the compositions of gut communities from different samples were performed using Bray–Curtis similarity , unless otherwise stated . Adonis implementation of Permanova ( Anderson , 2001 ) ( non-parametric permutational multivariate analysis of variance ) was used for comparison between groups . Distances between the different groups were calculated using the QIIME script ‘make_distance_plots . py’ and their significance was assessed using a non-parametric t-test with Monte Carlo simulation . Moreover , to examine taxonomical composition statistically and to identify taxa that were associated with different tolerance groups and temperature conditions , we applied the Dufrene–Legendre Indicator Species Analysis ( Dufrêne and Legendre , 1997 ) ( function indval with 1000 permutations , package ‘labdsv’ in R ) . This analysis calculates the indicator value ( fidelity and relative abundance ) of species in clusters or types , given by the following formula: IndValij = Specificityij × Fidelityij×100 where IndValij is the indicator value of species 'i' in relation to site type 'j'; Specificityij is the proportion of sites of type 'j' with species 'i' , and Fidelityij is the proportion of the number of individuals ( abundance ) of species 'i' that are in site type 'j' . Core communities were identified as the most resilient microbes present in >50% of the individuals at either 24°C or 12°C . To measure the absolute abundance of the two most prevalent core microbes ( Pseudomonas veronii and Janthinobacterium lividum ) and to validate our sequencing data , we designed specific primers to amplify their copy numbers using the Primer-BLAST tool ( Ye et al . , 2012 ) . After validation of the primers , real-time PCR was performed using the specific primers ( P . veronii , Forward 5′-CCGCGGTAATACAGAGGGTG-3′ and Reverse 5′-ACCCTCTACCATACTCTAGTCAGT-3′; J . lividum , Forward 5′- CGCAGGCGGTTTTGTAAGTC-3′ and Reverse 5′- GTCAATCTTGACCCAGGGGG-3′ ) . The fold change in abundance of each microbe was calculated relative to that in the warm conditions as the control . To predict the functional content of the gut microbiome originating from the different groups , we used the PICRUSt tool ( Langille et al . , 2013 ) . PICRUSt ( http://picrust . github . com/picrust/ ) is a software package designed to infer metagenome functional content from 16S metagenomic data . The paired-end merged 16S sequences were used for closed-reference OTU picking using QIIME , and the resulting OTU table was then fed into PICRUSt and functional predictions assigned to KEGG pathways ( Kanehisa and Goto , 2000 ) were made according to the metagenome inference workflow described by the developers . PICRUSt results were normalized , and then analyzed using the LEfSe tool ( Segata et al . , 2011 ) . LEfSe is an algorithm for high-dimensional biomarker discovery and explanation that identifies genomic features ( i . e . , taxa ) characterizing the differences between two or more biological conditions . It uses statistical significance and biological relevance to identify differentially abundant features , using the non-parametric factorial Kruskal–Wallis rank-sum test and the Wilcoxon rank-sum pairwise test to compare biologically significant categories . Then , it uses LDA to estimate the effect size of each differentially abundant feature and to perform dimension reduction to assess whether the differences are consistent with the expected biological behavior . We thus reanalyzed our sequences using the QIIME closed reference protocol against the 97% similarity to the GreenGenes database . The nonparametric t-test ( using Monte Carlo simulation ) was used to compare differently enriched pathways between groups . A heat map was created using the R package ‘heatmap3’ function . The non-relevant categories , that is human diseases and functions that are exclusive to eukaryotic organisms , were not presented in the heat map . Total RNA from the liver of selected individuals was extracted using TRIzol reagent ( Invitrogen , California , USA ) , according to the manufacturer's instructions and treated with DNAse ( TURBO-DNase , Ambion ) . Selection of the individuals was based on their microbiome composition ( principal coordinates analysis using Bray–Curtis metric to select for individuals with discriminating microbiomes; Figure 4—figure supplement 3 ) . RNA concentration and quality were determined in each sample and samples were equimolarly diluted and sent to the Technion Genome Center ( The Technion , Haifa , Israel ) for library preparation and single-end sequencing using Illumina HiSeq . The resultant raw short reads were subjected to a filtering and cleaning procedure as follows: the SortMeRNA tool ( Kopylova et al . , 2012 ) was used to filter out rRNA , and then the FASTX Toolkit ( Gordon and Hannon , 2010 ) ( version 0 . 0 . 13 . 2 ) was used for ( a ) trimming read-end nucleotides with quality scores <30 using 'fastq_quality_trimmer' and ( b ) removing reads with less than 70% base pairs with quality score ≤30 using 'fastq_quality_filter' . The Bowtie2 version 2 . 1 alignment tool was used to map cleaned reads onto the reference transcriptome of Oreochromis niloticus extracted from the NCBI database ( ftp://ftp . ncbi . nlm . nih . gov/genomes/Oreochromis_niloticus/; ASM185804v2 ) ( Supplementary file 1 , Table S7 ) . Then , transcript quantification was performed using the expectation-maximization method ( RSEM ) , by estimating maximum likelihood expression levels ( Li and Dewey , 2011 ) . Differential expression analysis was performed with the R package ‘DESeq2’ ( Love et al . , 2014 ) and transcripts that were more than 2-fold differentially expressed with false discovery rate ( FDR ) -corrected p < 0 . 05 ( Benjamini and Hochberg , 1995 ) were considered . The reference transcriptome sequences of O . niloticus were searched against the Danio rerio transcriptome reference ( from the NCBI database ) for homology identification . The results were further analyzed using the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Huang et al . , 2009 ) to determine the Gene Ontology ( GO ) and KEGG pathways . Last , to examine host-associated gene expression patterns statistically and to identify genes that are associated with different tolerance groups , we applied the Dufrene–Legendre Indicator Species Analysis ( Dufrene and Legendre ( 19977 ) ; function indval with 1000 permutations , package ‘labdsv’ in R ) , as described in the 'Comparison of gut communities' section . Transcriptome raw sequences are submitted under the accession number SRP164378 ( PRJNA419688 ) in the SRA . Non-parametric tests ( Wilcoxon test ) and linear mixed-effect models ( nlme R package [Bates and Pinheiro , 1998] ) were used to assess α-diversity , while Adonis implementation of Permanova ( vegan R package [Anderson , 2001] ) was used for comparisons between groups in the analysis of β-diversity using the Bray–Curtis distance matrix . LEfSe ( Segata et al . , 2011 ) was used to estimate the effect size of each differentially abundant PICRUSt feature and to perform dimension reduction using the Galaxy online tool ( http://huttenhower . sph . harvard . edu/galaxy/ ) . CCA ( Butts , 2012 ) , was performed between the matrices of the microbiome composition table ( OTUs ) and the transcriptomic data using principal component analysis . The total fraction of OTU variance accounting for each component by the transcriptomic analysis ( expression ) variables , through all canonical variates , was calculated . For that , the first four principal components ( PCs ) from the OTU table were extracted ( R prcomp ) . In addition , the first four expression matrix PCs were extracted using R prcomp . The actual value of the total variance was then compared to that of 1000 random permutations , in which the order of the host line PCs was shuffled .	Animals and plants host diverse microbial communities that are vital for their survival . In fact , the host organisms and their associated ‘microbiome’ are so closely linked that they are often described as a single entity: the holobiont unit . This suggests that when the host adapts to cope with stressful conditions , similar changes should also occur in its microbiome . Fish are unable to maintain a stable body temperature and can be greatly affected by temperature fluctuations . Some fish are better able to tolerate cold conditions than others , but it was not known if their gut microbes are similarly affected by changes in temperature . To investigate , Kokou et al . selectively bred tropical blue tilapia to create families of fish that could either tolerate the cold well , or that were highly sensitive to the cold . The gut microbiomes of cold-resistant fish were different from the cold-sensitive ones , even though the fish lived in the same tank . Moreover , the gut microbiomes of the cold-tolerant fish showed higher resilience to temperature changes than the microbes in the guts of the cold-sensitive fish . It remains to be determined whether the response of the microbiome directly affects how its host fish responds to temperature changes . However , the results presented by Kokou et al . show that there are links between how the host and its microbes adapt to environmental stress . As well as helping us to understand how holobionts evolved , this knowledge could also potentially be applied broadly in clinical sciences or agriculture , for example to select for efficient crops .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2018	Host genetic selection for cold tolerance shapes microbiome composition and modulates its response to temperature
The spindle- and kinetochore-associated ( Ska ) complex is essential for normal anaphase onset in mitosis . The C-terminal domain ( CTD ) of Ska1 binds microtubules and was proposed to facilitate kinetochore movement on depolymerizing spindle microtubules . Here , we show that Ska complex recruits protein phosphatase 1 ( PP1 ) to kinetochores . This recruitment requires the Ska1 CTD , which binds PP1 in vitro and in human HeLa cells . Ska1 lacking its CTD fused to a PP1-binding peptide or fused directly to PP1 rescues mitotic defects caused by Ska1 depletion . Ska1 fusion to catalytically dead PP1 mutant does not rescue and shows dominant negative effects . Thus , the Ska complex , specifically the Ska1 CTD , recruits PP1 to kinetochores to oppose spindle checkpoint signaling kinases and promote anaphase onset . Microtubule binding by Ska , rather than acting in force production for chromosome movement , may instead serve to promote PP1 recruitment to kinetochores fully attached to spindle microtubules at metaphase . Anaphase onset and mitotic exit are driven by proteasome-mediated destruction of Securin and Cyclin B , which are targeted for ubiquitylation by the E3 ligase , the Anaphase-promoting complex/cyclosome ( APC/C ) ( Sivakumar and Gorbsky , 2015 ) . Until metaphase , APC/C activity is restrained by the spindle checkpoint . Recent studies indicate that a key element in extinguishing checkpoint signaling when chromosomes align at metaphase is the displacement of the critical checkpoint signaling kinase , Mps1 , from its substrate , Knl1 , at kinetochores ( Aravamudhan et al . , 2015; Hiruma et al . , 2015; Ji et al . , 2015 ) . Then , to allow anaphase onset and mitotic exit , Mps1 phosphorylation events must be reversed by phosphatases . Previously , we showed that depletion of the Ska complex leads to a strong metaphase arrest or delay ( Daum et al . , 2009; Sivakumar et al . , 2014 ) . Even when Mps1 was directly inhibited with a strong chemical inhibitor , cells depleted of the Ska complex exited mitosis more slowly ( Sivakumar et al . , 2014 ) . These results suggested that the Ska complex played some role in opposition to , or downstream of , checkpoint signaling . Here , we provide evidence that the Ska complex recruits protein phosphatase 1 ( PP1 ) to kinetochores , consistent with a role in opposing checkpoint signaling by kinetochore-associated kinases . Centromere- and kinetochore-associated kinases play key roles in regulating chromosome attachment to the spindle and cell cycle progression in mitosis . As chromosomes align to the metaphase plate , PP1 becomes concentrated at kinetochores stabilizing kinetochore-microtubule interactions , and promoting anaphase onset by opposing spindle checkpoint kinase signaling ( Liu et al . , 2010; Nijenhuis et al . , 2014 ) . Recruitment of PP1 to the kinetochore then leads to substrate dephosphorylation , stabilization of kinetochore-microtubule attachments , and anaphase onset ( Liu et al . , 2010; Nijenhuis et al . , 2014; Meadows et al . , 2011; Rosenberg et al . , 2011; Zhang et al . , 2014 ) . The heterotrimeric spindle- and kinetochore-associated ( Ska ) protein complex consisting of Ska1–3 is required for timely anaphase onset ( Daum et al . , 2009; Welburn et al . , 2009; Gaitanos et al . , 2009 ) . Some laboratories have reported that Ska depletion strongly compromises chromosome alignment ( Welburn et al . , 2009; Gaitanos et al . , 2009; Raaijmakers et al . , 2009 ) . However , our detailed video analyses show that depletion of Ska components , individually or in combination , causes delays in chromosome alignment followed by a robust metaphase arrest ( Daum et al . , 2009; Sivakumar et al . , 2014 ) . Metaphase arrest is often then followed by cohesion fatigue and asynchronous chromatid separation without progression to anaphase ( Stevens et al . , 2011; Daum et al . , 2011 ) . Cells that have undergone cohesion fatigue remain arrested in mitosis with a terminal phenotype in which mixtures of separated chromatids and intact chromosomes are scattered across the spindle . The C-terminal domain ( CTD ) of Ska1 binds to microtubules and has thus been termed the microtubule-binding domain or MTBD ( Abad et al . , 2014; Schmidt et al . , 2012 ) . HeLa cells expressing Ska1 lacking the CTD showed phenotypes similar to Ska depletion . Cells delayed at metaphase , and cold stable kinetochore fibers were reportedly decreased ( Abad et al . , 2014; Schmidt et al . , 2012 ) . Furthermore , in vitro the Ska complex was shown to track depolymerizing microtubule ends ( Schmidt et al . , 2012 ) . These results have led to a model in which Ska directly promotes microtubule stability and kinetochore movement on microtubules in conjunction with other kinetochore components , such as the Ndc80 complex ( Abad et al . , 2014; Schmidt et al . , 2012 ) . In this study , we report that Ska recruits PP1 to kinetochores . Compromising this recruitment increases phosphorylation of Knl1 and increases recruitment of the checkpoint kinase Bub1 . We find that Ska1 CTD is required for binding to PP1 in vivo and in vitro . The metaphase arrest or delay seen after Ska depletion is strongly rescued by expressing a chimeric Ska1 protein with its CTD replaced either by a PP1-binding motif or by a direct fusion to PP1 . Thus , a major function of the Ska complex is to recruit PP1 to the kinetochore . Several mitotic kinases , including Mps1 , Aurora B , Bub1 and Plk1 accrue to high levels at centromeres and kinetochores in early mitosis , during prophase and prometaphase ( Funabiki and Wynne , 2013 ) . An opposing phosphatase , PP1 , also accumulates on kinetochores as microtubules attach , reaching maximal levels at metaphase ( Liu et al . , 2010; Nijenhuis et al . , 2014; Suijkerbuijk et al . , 2012; Foley et al . , 2011 ) . We tested if the recruitment of PP1 to kinetochores was affected in Ska3-depleted cells . We found that the kinetochore levels of PP1 were diminished in Ska3-depleted cells at prometaphase and metaphase ( Figure 1A ) . In contrast , the pool of PP1 on the mitotic spindle appeared unaffected . Thus , the Ska complex is required to recruit PP1 to the kinetochore . 10 . 7554/eLife . 12902 . 003Figure 1 . Ska complex is required for PP1 recruitment to the kinetochore . ( A ) HeLa cells were transfected with control or Ska3 siRNA at 50nM final concentration . Thirty hours after transfection immunofluorescence was done and PP1 at the kinetochore was quantified . Ska3 antibody staining shows efficiency of depletion . PP1 at kinetochores increases from prometaphase to metaphase . Ska3-depleted cells are inefficient in PP1 recruitment to kinetochores in both prometaphase and metaphase . ( B ) HeLa cells were transfected with Mis12-GFP , Ska1-GFP or Mis12Ska1-GFP to increase Ska complex accumulation at kinetochores . Thirty-six hours after transfection , MG132 was added for 1 hr to accumulate cells at metaphase . Immunofluorescence of PP1 at kinetochores was quantified . PP1 accumulates at kinetochores in Mis12Ska1GFP-expressing cells to a greater extent than in Mis12GFP- or Ska1GFP-expressing cells . ( C ) In cells treated with 3 . 3 μM nocodazole , PP1 accumulated to higher levels at kinetochores of cells expressing Mis12Ska1-GFP compared to cells expressing Mis12GFP or Ska1GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 003 Since Ska depletion reduced PP1 recruitment to the kinetochore , we asked if overexpression of Ska would increase it . We had previously shown that expressing a fusion of the KMN component , Mis12 , to Ska1 ( Mis12-Ska1 ) increased total Ska recruitment to the kinetochore ( Sivakumar et al . , 2014 ) . Cells expressing Mis12-Ska1 had higher levels of PP1 at kinetochores than control cells expressing either Mis12 or Ska1 . The cells in this experiment were arrested at metaphase with the proteasome inhibitor , MG132 ( Figure 1B ) . Expression of Mis12-Ska1 also increased PP1 levels at kinetochores in cells treated with nocodazole ( Figure 1C ) , suggesting that targeting Ska complex to the kinetochore leads to accumulation of kinetochore PP1 in a microtubule-independent manner . Although expression of Mis12-Ska1 increased PP1 levels at the kinetochore , it was not sufficient to accelerate anaphase onset in normal cells or induce mitotic exit in nocodazole-treated cells ( Sivakumar et al . , 2014 ) . To better understand how the Ska complex recruits PP1 to the kinetochore , we first tested if the Ska complex physically interacted with PP1 in mitotic HeLa cell lysates . We found that Ska3 could indeed be co-immunoprecipitated ( co-IPed ) with PP1 ( Figure 2A ) . As a positive control , Knl1 , a kinetochore protein known to directly bind PP1 ( Liu et al . , 2010; Meadows et al . , 2011; Rosenberg et al . , 2011; Zhang et al . , 2014 ) , also co-IPed with PP1 . To test whether the interaction between Ska and PP1 might be bridged by Knl1 , we compared Ska3 IPs from control and Knl1-depleted cells ( Figure 2B ) . The Ska complex interacted with similar amounts of PP1 in control and Knl1-depleted cells . Knl1 also co-IPed similar amounts of PP1 from control and Ska3-depleted cells , indicating that Ska depletion did not affect the Knl1-PP1 interaction ( Figure 2C ) . By antibody staining and by using a HeLa-PP1γGFP stable cell line , we found that kinetochore PP1 levels were low and were slightly above the background signals of soluble , non-kinetochore PP1 . Individual depletions of Knl1 or Ska3 reduced the kinetochore PP1 levels to the background levels . Thus , we were unable to determine the individual contributions of Ska and Knl1 toward kinetochore recruitment of PP1 by immunostaining . From our IP experiments , we conclude that the Ska complex and Knl1 are capable of independent binding to PP1 . 10 . 7554/eLife . 12902 . 004Figure 2 . Ska1 interacts with PP1 . ( A ) HeLa cells synchronized in S phase with thymidine were then released and arrested in mitosis using 0 . 33 μM nocodazole . The cells were subsequently released from nocodazole into MG132 ( 25 μM ) for 1 hr to allow metaphase chromosome alignment . Cell pellets were lysed and IgG , Ska3 , and Knl1 IPs were performed to detect interaction with PP1 . Both Ska3 and Knl1 IPs , but not control IgG IP , pulled down PP1 from cell lysates . ( B ) HeLa cells were transfected with mock siRNA , Knl1 siRNA , or Ska3 siRNA . HeLa cells were then arrested at metaphase as decribed above in A . Ska3 co-precipitated with similar amounts of PP1 in extracts prepared from control and Knl1-depleted cells . ( C ) HeLa cells transfected with mock , Knl1 , or Ska3 siRNA were synchronized at metaphase as described above in A . Knl1 co-precipitated with similar amounts of PP1 in extracts prepared from control and Ska3-depleted cells . ( D ) GST-Ska proteins purified from bacteria were incubated with in vitro translated S35-labeled Myc-PP1 . GST-Ska1/2 and GST Ska1/2+GST Ska3 interacted with S35-labeled Myc-PP1 . E . HeLa cells were transfected with mCherry-PP1 and control , GFP Ska1 , GFP Ska2 or GFP Ska3 plasmids . Endogenous Ska components were depleted using siRNA; cells were synchronized in S phase using thymidine then released and arrested in mitosis using 3 . 3 μM nocodazole . GFP-Ska1 precipitated from cells depleted of Ska2 and Ska3 showed the strongest interaction with mCherry-PP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 004 To determine if the Ska complex and PP1 could interact in vitro and to analyze which Ska component was responsible for the interaction , we purified GST-Ska1/2 and GST-Ska3 proteins from bacteria and conducted binding assays with in vitro translated Myc-PP1 . GST-Ska1/2 bound efficiently to Myc-PP1 , with or without GST-Ska3 ( Figure 2D ) . As positive controls , GST-Ska1/2 interacted with in vitro translated Ska3 , and vice versa . Thus , Ska1/2 physically interacted with PP1 in vitro . We next tested whether GFP-Ska proteins could interact with PP1 in HeLa cells depleted of endogenous Ska components . We found that GFP-Ska1 , but not GFP-Ska2 or GFP-Ska3 , interacted with mCherry-PP1 in lysates prepared from HeLa cells depleted of the other two endogenous Ska components ( Figure 2E ) . Thus , PP1 binding by the Ska complex is mainly mediated by Ska1 . The N-terminal coiled coil region of Ska1 is required to form a complex with Ska2 and Ska3 , whereas its CTD has been well characterized as a microtubule-binding domain ( Abad et al . , 2014; Schmidt et al . , 2012 ) . To determine which region of Ska1 was responsible for binding PP1 , we performed in vitro binding assays with bacterially purified GST-tagged fragments of Ska1 and in vitro translated Myc-PP1 . The Ska1 CTD was found to interact with PP1 ( Figure 3A ) . To further test if PP1 and Ska complex bind each other directly , gel-filtration experiments were carried out with purified , bacterially expressed Ska and PP1 proteins . The Ska complex containing full-length Ska1 bound PP1 , while the Ska complex containing Ska1 lacking the CTD showed reduced complex formation ( Figure 3B ) . Using microscale thermophoresis , we found that purified recombinant Ska1 CTD and PP1 proteins interacted with a Kd of 1 . 5 μM ( Figure 3C , Figure 3—figure supplement 1 ) . Thus , Ska1 CTD and PP1 directly interact with moderate affinity . Other parts of Ska or associated proteins may enhance the Ska1 CTD-PP1 interaction in vivo . 10 . 7554/eLife . 12902 . 005Figure 3 . Ska1 C terminal domain ( CTD ) physically interacts with PP1 . ( A ) GST fusions to Ska1 fragments , GST-Ska1 ( 1–91 ) , GST-Ska1 ( 92–132 ) and GST-Ska1 ( 133–255 ) were purified from bacteria . In vitro translated S35 labeled Myc-PP1 was added to GST-Ska1 fragments and in vitro binding assays were performed to detect Myc-PP1 interaction with GST-Ska1 fragments . GST-Ska1 ( 133–255 ) bound strongly with Myc-PP1 while other fragments did not . ( B ) ( Left ) Purified 6xHis-PP1α7–330 ( PP1 ) , untagged Ska1ΔCTD/Ska2/Ska31-343 and Ska1FL/Ska2/Ska31-343 , visualized on coommassie stained gel . ( Right ) Immunoblots showing the elution profile of PP1 , PP1 incubated with Ska1FL/Ska2/Ska31-343 ( blue ) or PP1 incubated with Ska1ΔCTD/Ska2/Ska31-343 ( orange ) run on a Superose 12 size-exclusion column . Ska complexes run in the same fraction in the absence of PP1 ( not shown ) . Densitometry quantifications of PP1 signal in fractions eluted from size-exclusion column shows a reduction in binding to Ska complex lacking the Ska1 CTD . ( C ) Microscale thermophoresis ( MST ) was done to analyze the direct binding interaction between PP1 and the Ska1 CTD proteins purified from bacteria . The top panel shows thermophoretic time traces of 16 samples in three independent experiments . The middle panel shows the T-Jump data ( circles ) and the fit to the data ( line ) . The residuals between the data and the fit line are indicated in the bottom panel . The Kd of PP1 binding to the Ska1 CTD was calculated to be 1 . 5 μM . Fn represents ratio ( expressed in per-mille units ) of the fluorescence readings in the time traces as measured just after ( pink region , top panel ) and before ( blue region , top panel ) activation of the MST laser; ΔFn is calculated by subtracting the refined Fn of the free PP1 from all Fn values and thus represents the change in T-Jump response as a function of ligand concentration . This part was rendered using the program GUSSI ( Brautigam , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 00510 . 7554/eLife . 12902 . 006Figure 3—figure supplement 1 . Purification of PP1 and Ska1CTD from bacteria His-PP1γ and GST-Ska1CTD were individually purified from bacteria . The tags were cleaved using Thrombin and 3C protease respectively . Coomasie stained gel shows the two proteins after size exclusion chromatography . The proteins were then used to perform MST assay . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 006 In HeLa cells depleted of endogenous Ska complex components and expressing RNAi-resistant Ska1 fragments , GFP-Ska1 CTD bound to Myc-PP1 ( Figure 4A ) . We then constructed doxycycline ( Dox ) -inducible HeLa cell lines stably expressing RNAi-resistant , GFP fusions of full-length Ska1 or Ska1ΔCTD . As reported previously ( Abad et al . , 2014; Schmidt et al . , 2012 ) , exogenously expressed full-length Ska1 localized to centrosomes , spindle microtubules , and kinetochores during mitosis while Ska1ΔCTD localized predominantly to kinetochores ( Figure 4—figure supplement 1 ) . We arrested these cells at metaphase with MG132 and immunoprecipitated the full-length Ska1 or Ska1ΔCTD with anti-GFP antibody . Full-length Ska1 interacted more efficiently with endogenous PP1 than did Ska1ΔCTD , with or without depletion of endogenous Ska1 ( Figure 4B ) . Additionally , when we overexpressed Myc-PP1 in the stable cell lines , we detected an interaction between Ska1 and Myc-PP1 even in nocodazole-treated cells where association of Ska with endogenous PP1 is difficult to detect , perhaps due to the reduced recruitment of Ska to kinetochores in mitotic cells lacking microtubules ( Figure 4C ) . Again , Ska1ΔCTD showed reduced Myc-PP1 binding both in nocodazole- and in MG132-treated cells ( Figure 4C-figure supplement 2 ) . 10 . 7554/eLife . 12902 . 007Figure 4 . Ska1CTD immunoprecipitates with PP1 in HeLa cells . ( A ) HeLa cells were transfected with GFP-Ska1 ( 1–91 ) , GFP-Ska1 ( 92–132 ) , GFP-Ska1 ( 1–132 ) , GFP-Ska1 ( 133–255 ) , and Myc-PP1 plasmids . Endogenous Ska1 , Ska2 , and Ska3 were depleted with siRNA . GFP IP’s followed by Western blotting was done to analyze the GFP-Ska1 fragment that associated most strongly with Myc-PP1 in vivo . GFP-Ska1 ( 133–255 ) interacted with Myc-PP1 in extracts from cells depleted of all endogenous Ska components . ( B ) HeLa cell lines expressing GFP-Ska1 or GFP-Ska1ΔCTD under Doxocycline ( Dox ) control and resistant to siRNA were generated . These were transfected with control or Ska1 siRNA and transgene expression was induced . Cells were synchronized by thymidine in S phase then released and arrested in mitosis using 0 . 33 μM nocodazole . Mitotic cells were released into MG132 ( 25 μM ) to allow metaphase chromosome alignment and collected after 1 hr . GFP IP’s were prepared from the cell extracts and the amount of PP1 associated determined by quantitative immunoblotting . IP of GFP-Ska1ΔCTD showed greatly reduced binding to PP1 both with or without depletion of endogenous Ska1 . ( C ) Dox-inducible HeLa GFP-Ska1 and GFP-Ska1ΔCTD cells were transfected with Myc-PP1 plasmid followed 6 hr later by transfection of control or Ska1 siRNA . Dox was added to induce transgene expression . Cells were synchronized in S phase with thymidine then released and collected in mitosis using 3 . 3 μM nocodazole . GFP IP’s were done followed by quantitative immunoblotting to detect Myc-PP1 association . GFP-Ska1 coprecipitated significant amonts of Myc-PP1 , while GFP-Ska1ΔCTD ( with or without depletion of endogenous Ska1 ) showed greatly reduced co-precipitatation with Myc-PP1 . Depletion of endogenous Ska1 did reduce somewhat the ability of full length GFP-Ska1 to co-precipitate Myc-PP1 . However , GFP-Ska1ΔCTD showed pronounced reduction in co-precipitation of Myc-PP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 00710 . 7554/eLife . 12902 . 008Figure 4—figure supplement 1 . Characterization of HeLa Tet-On cells expressing GFP-Ska1 or GFP-Ska1ΔCTD HeLa cells stably transfected with constructs for inducible expression of GFP-Ska1 or GFP Ska1ΔCTD cells were treated with Dox to induce transgene expression . Cells were imaged live to determine the localization of GFP-Ska1 and GFP-Ska1ΔCTD . As previously reported , GFP-Ska1 concentrates on spindle microtubules and at kinetochores while GFP-Ska1ΔCTD predominantly concentrates at kinetochores ( Abad et al . , 2014; Schmidt et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 00810 . 7554/eLife . 12902 . 009Figure 4—figure supplement 2 . GFP-Ska1ΔCTD shows reduced binding to Myc-PP1 in extracts from cells arrested at metaphase with MG132 . . HeLa GFP-Ska1 and GFP-Ska1ΔCTD stable cell lines were transfected with Myc-PP1 plasmid . Six hours later , cells were treated with mock or Ska1 siRNA . Dox was added to induce transgene expression . Cells were synchronized in S phase with thymidine and then released and collected in mitosis using 0 . 33 μM nocodazole . Cells were released into MG132 ( 25 μM ) for 1 hr to generate metaphase-arrested cells . GFP IPs were done to assess the amount of Myc-PP1 associated . In GFP IP’s from cell extracts , GFP-Ska1ΔCTD showed a decrease in PP1 co-precipitation compared to GFP-Ska1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 00910 . 7554/eLife . 12902 . 010Figure 4—figure supplement 3 . Characterization of Ska1CTD to identify residues involved in interaction with PP1 . HeLa cells were transfected with Myc-PP1 and the indicated Ska1 mutants . Endogenous Ska1 , Ska2 and Ska3 were depleted using siRNA . Cells were synchronized in S phase with thymidine then released and collected in mitosis using 3 . 3 μM nocodazole . The amount of Myc-PP1 in anti-GFP IP’s was determined by quantitative Western blotting . Whereas Ska1ΔCTD showed diminished binding to Myc-PP1 , none of the point mutants exhibited altered binding . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 010 The Ska1 CTD does not contain clear matches for the known PP1-binding motifs present in many PP1-interacting proteins . We thus made a series of point mutants targeting surface-exposed , conserved residues in Ska1 CTD , and systematically tested their interaction with Myc-PP1 in nocodazole-treated HeLa cells upon endogenous Ska1 depletion . We were unable to identify point mutants that disrupted the interaction between Ska1 and PP1 ( Figure 4—figure supplement 3 ) . Since Ska1ΔCTD was deficient in binding PP1 , we expected the cell line expressing this mutant to show reduced kinetochore localization of PP1 . To eliminate potential PP1 binding to microtubules or microtubule-associated proteins , we treated the cells with high concentration ( 3 . 3 μM ) nocodazole . PP1 localization to the kinetochore was decreased by 50% in cells expressing Ska1ΔCTD , compared to cells expressing full-length Ska1 ( Figure 5A ) , with or without depletion of Ska1 . Thus , Ska1ΔCTD could diminish PP1 kinetochore targeting in a dominant negative fashion , confirming a role for the Ska1 CTD in PP1 targeting to kinetochores . The spindle checkpoint kinase Mps1 phosphorylates Knl1 on multiple MELT motifs , and the phosphorylated MELT recruits other checkpoint components , such as Bub1 , to kinetochores ( Yamagishi et al . , 2012; Shepperd et al . , 2012; London et al . , 2012 ) . We found that phosphorylation of one such MELT motif , pMELT ( residue pT875 ) , was indeed elevated in cells expressing Ska1ΔCTD , as compared to cells expressing full-length Ska1 ( Figure 5B ) ( Ji et al . , 2015 ) . Consistent with this finding , cells expressing Ska1ΔCTD showed increased levels of Bub1 protein at kinetochores , as compared to cells expressing full-length Ska1 ( Figure 5C ) . Differences in PP1 , pMELT and Bub1 labeling were not simply due to differences in expression levels in the stable cell lines , since quantification showed that GFP signals in GFP-Ska1 and GFP-Ska1ΔCTD cells were similar ( Figure 5—figure supplement 1 ) . These results indicate that PP1 recruited by the Ska1 CTD opposes Knl1 phosphorylation by Mps1 directly or indirectly . 10 . 7554/eLife . 12902 . 011Figure 5 . Cells expressing Ska1ΔCTD recruit less PP1 and accumulate more Knl1 phosphoepitope and Bub1 protein at kinetochores . HeLa GFP-Ska1 and HeLa GFP-Ska1ΔCTD cells were grown on chambered cover slides . Dox was added to induce transgene expression . Cells were transfected with Ska1 siRNA , and thymidine was added for 18–24 hr to synchronize cells . Cells were released from thymidine and arrested in mitosis using 3 . 3 μM nocodazole . Immunofluorescence was done to detect PP1 , pMELT , Bub1 at the kinetochore . ( A ) Image panel showing PP1 localization at the kinetochore in GFP-Ska1 and GFP-Ska1ΔCTD cells . The graph depicts the decrease in PP1 localization in GFP-Ska1ΔCTD cells compared to GFP-Ska1 cells both with and without depletion of endogenous Ska1 . ( B ) Images show localization of antibody ( pMELT ) to a phosphoepitope on Knl1 ( pT875 ) at kinetochores in GFP-Ska1 and GFP-Ska1ΔCTD cells . The graph shows that cells expressing GFP-Ska1ΔCTD have increased pMELT signals at kinetochores compared to GFP-Ska1-expressing cells . ( C ) Images show the levels of Bub1 protein at kinetochores in GFP-Ska1 and GFP-Ska1ΔCTD cells . Qantification shows that GFP-Ska1ΔCTD accumulate more Bub1 at kinetochores compared to GFP-Ska1 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 01110 . 7554/eLife . 12902 . 012Figure 5—figure supplement 1 . Expression levels of cell lines stably expressing inducible GFP-Ska1 and GFP-Ska1ΔCTD are similar After induction with Dox , GFP expression levels in individual cells were quantified . At least five cells in each case was quantified . GFP-Ska1 ( with or without Ska1 depletion ) and GFP-Ska1ΔCTD ( with or without Ska1 depletion ) cells show similar amounts of GFP expressed . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 012 PP1 is recruited to the kinetochore to dephosphorylate kinetochore proteins to oppose spindle checkpoint signaling and promote anaphase onset . PP1 depletion was found to cause delays at metaphase ( Liu et al . , 2010; Nijenhuis et al . , 2014; Zhang et al . , 2014 ) . Our experiments show that cells expressing Ska1ΔCTD are deficient in PP1 recruitment to the kinetochore . We sought to determine whether the kinetochore recruitment of PP1 by Ska affected mitotic progression . Depletion of Ska proteins by RNAi causes long metaphase delays or arrest ( Sivakumar et al . , 2014 ) . As expected , expression of RNAi-resistant Ska1 rescued the arrest caused by Ska1 depletion ( Figure 6A ) . Consistent with previous reports ( Abad et al . , 2014; Schmidt et al . , 2012 ) , expression of Ska1ΔCTD did not rescue the metaphase arrest/delay caused by Ska1 depletion ( Figure 6A , B ) . 10 . 7554/eLife . 12902 . 013Figure 6 . Expression of Ska1ΔCTD fused to the PP1-binding domain ( amino acids 34–81 ) of Knl1 rescues phenotypes caused by Ska1 depletion . ( A ) HeLa cells were transfected with plasmids to express mCherry-Ska1 , mCherry-Knl1 ( 34–81 ) , mCherry-Knl1 ( 34–81 ) 4A ( where the PP1 binding motif RVSF is mutated to AAAA ) , mCherry-Ska1ΔCTD , mCherry-Ska1ΔCTD fused to Knl1 ( 34–81 ) ( mCherry-Ska1ΔCTDKnl1 ( 34–81 ) ) and mCherry-Ska1ΔCTDKnl1 ( 34–81 ) 4A . Endogenous Ska1 was depleted using Ska1 siRNA . Cells were imaged by time-lapse microscopy , and % of mitotic cells arrested in metaphase was plotted . As expected , expression of siRNA-resistant mCherry-Ska1 rescued metaphase arrest caused by Ska1 depletion while expression of mCherry-Ska1ΔCTD did not . Expression of Ska1ΔCTDKnl1 ( 34–81 ) fusion in Ska1-depleted cells rescued metaphase arrest while expression of Ska1ΔCTDKnl1 ( 34–81 ) 4A did not . ( B ) The interval from NEB to anaphase onset is plotted for cells that progressed to anaphase in cells expressing Ska1 constructs with or without depletion of endogenous Ska1 . In cells not depleted of endogenous Ska1 , dominant negative effects in delaying mitotic progression were caused by expression of Ska1ΔCTD , as previously reported , and by expression of the fusion Ska1ΔCTDKnl1 ( 34–81 ) 4A , which lacks the PP1-binding motif . Expression of other constructs including Ska1ΔCTDKnl1 ( 34–81 ) with intact PP1 binding caused no delay . In cells depleted of endogenous Ska1 , 40% of cells arrested at metaphase and did not progress to anaphase ( Figure 6A ) . The rest showed delayed mitotic progression with an average of 146 min from NEB to anaphase onset compared to control cells ( 40 min ) . As expected the delay was rescued by expression of mCherry-Ska1 but was not rescued by expression of mCherry-Knl1 ( 34–81 ) , mCherry-Knl1 ( 34–81 ) 4A or mCherry-Ska1ΔCTD . Expression of Ska1ΔCTDKnl1 ( 34–81 ) but not Ska1ΔCTDKnl1 ( 34–81 ) 4A showed significant rescue of the delay caused by Ska1 depletion . ( C ) Representative images of cells transfected with the indicated constructs treated with control siRNA or Ska1 siRNA . The green and red colors indicate DNA and mCherry expression , respectively . Only cells expressing indicated mCherry constructs were analyzed for this experiment . The time is indicated in hour:minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 013 We then tested if Ska1ΔCTD fused to a PP1 binding motif rescued timely anaphase onset . We expressed a chimeric protein consisting of Ska1ΔCTD fused to residues 34–81 of Knl1 . This region of Knl1 contains the PP1-binding RVSF motif , but importantly it lacks the microtubule-binding domain found at the extreme N terminus of Knl1 . Expression of Ska1ΔCTD-Knl1 ( 34–81 ) rescued anaphase onset in cells depleted of endogenous Ska1 ( Figure 6A–C ) . Mutation of the RVSF motif to AAAA was previously shown to block binding of PP1 to Knl1 ( Liu et al . , 2010 ) . Expression of Ska1ΔCTD fused to the Knl1 fragment with the RVSF to AAAA mutation failed to rescue metaphase arrest/delay in the majority of cells depleted of endogeneous Ska1 . This result suggests that the decreased PP1 recruitment contributes to the Ska depletion phenotype ( Figure 6A–C ) . To further test if direct PP1 recruitment by Ska1 could promote anaphase onset , we created a chimeric Ska1 protein with its CTD replaced with PP1 . Gratifyingly , expression of the Ska1-PP1 chimeric protein produced higher PP1 accumulation ( 1 . 5-fold increase ) at kinetochores ( Figure 7—figure supplement 1 ) , and resulted in complete rescue of the mitotic arrest and alignment delay phenotypes and substantially reduced the metaphase delay phenotype induced by Ska1 depletion ( Figure 7A ) . As a control , expression of PP1 alone , not fused to Ska1 , failed to rescue . Thus , expression of a Ska1-PP1 chimeric protein , lacking the CTD and its associated microtubule-binding activity , significantly alleviates the mitotic phenotypes caused by Ska1 depletion . As shown previously ( Schmidt et al . , 2012 ) and confirmed here , expression of Ska1ΔCTD caused metaphase delays even in cells without Ska1 depletion , indicating that it dominant negatively inhibits Ska function ( Figure 7B ) . This dominant-negative phenotype was not seen with the Ska1-PP1 chimeric protein in which the CTD was replaced by PP1 ( Figure 7A , B—figure supplement 2 ) . These results are consistent with a role for Ska1-recruited PP1 in mitotic progression . 10 . 7554/eLife . 12902 . 014Figure 7 . Expression of Ska1ΔCTD fused directly to PP1 but not phosphatase-dead PP1 ( pdPP1 ) rescues phenotypes caused by Ska1 depletion . ( A ) HeLa cells were transfected with plasmids to express mCherry-Ska1 , mCherry-PP1 , mCherry-Ska1ΔCTD , and mCherry-Ska1ΔCTD fused to PP1 ( mCherry-Ska1ΔCTDPP1 ) . Endogenous Ska1 was depleted using Ska1 siRNA . Hoechst 33342 was added at 25 ng/ml to visualize DNA . Cells were then imaged by time-lapse microscopy , and % of mitotic cells arrested in metaphase was plotted . As expected , expression of siRNA-resistant mCherry-Ska1 rescued metaphase arrest caused by Ska1 depletion while expression of mCherry-PP1 or mCherry-Ska1ΔCTD did not . Expression of Ska1ΔCTDPP1 fusion in Ska1 depleted cells completely rescued metaphase arrest . ( B ) The interval from NEB to anaphase onset is plotted for cells that progressed to anaphase while expressing Ska1 constructs without or with depletion of endogenous Ska1 . As expected , Ska1ΔCTD-expression showed a dominant negative effect delaying mitotic progression in control cells not depleted of endogenous Ska1 . Expression of the fusion , Ska1ΔCTDPP1 , caused no delay . When endogenous Ska1 was depleted , 38% of cells arrested at metaphase and did not progress to anaphase ( Figure 7A ) . The rest showed delayed progression from NEB to anaphase with an average of 110 min compared to control cells ( 35 min ) . As expected , the delay was rescued by expression of mCherry-Ska1 but was not rescued by expression of mCherry-PP1 or mCherry-Ska1ΔCTD . Expression of Ska1ΔCTDPP1 showed significant rescue of the delay cause by Ska1 depletion with an average time from NEB to anaphase of 62 min . ( C ) HeLa cells were transfected with the indicated plasmids and then treated with mock or Ska1 siRNA . Expression of Ska1ΔCTD fused to a phosphatase dead PP1 ( Ska1ΔCTDpdPP1 ) failed to rescue Ska1 depletion . Indeed , expression of phosphatase dead fusion , on its own , induced a potent metaphase arrest phenotype in cells not depleted of endogenous Ska1 . Moreover , it exacerbated the metaphase arrest in cells depleted of Ska1 . ( D ) Ska1ΔCTDpdPP1 causes a longer delay to anaphase onset than Ska1ΔCTD even without depletion of endogenous Ska1 . ( E ) Chromosome alignment is delayed in Ska1 depleted cells and this is recapitulated by expression of Ska1ΔCTDpdPP1 without endogenous Ska1 depletion . Delays in chromosome alignment are not observed upon expression of Ska1ΔCTDPP1 fusion . ( F ) Hypothetical model for dynamic balance of Mps1 kinase and PP1 phosphatase activities during mitotic progression . Photobleaching studies have shown that Mps1 , Ska , and PP1 all exhibit high turnover at kinetochores with a residence times of a few seconds ( Raaijmakers et al . , 2009; Howell et al . , 2004; Trinkle-Mulcahy et al . , 2003 ) . Before microtubule attachment , Mps1 concentration at kinetochores remains high due to interaction with the CH domains of the Ndc80 complex . Correspondingly , Ska-PP1 concentrations are low because of the paucity of microtubules . The high Mps1 and low PP1 concentrations maintain high phosphorylation Mps1 substrates , Knl1 and Bub1 . Microtubules compete with Mps1 for binding to the CH domains of the Ndc80 complex , resulting in depletion of Mps1 . The binding of Ska to microtubule protofilaments increases PP1 concentration . High Ska-PP1 and low Mps1 result in dephosphorylation of substrates , promoting release of Bub1–Bub3 and Mad1–Mad2 complexes . Diminished checkpoint signaling due to release of Bub1–Bub3 and Mad1–Mad2 from kinetochores promotes anaphase onset and mitotic exit . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 01410 . 7554/eLife . 12902 . 015Figure 7—figure supplement 1 . Ska1ΔCTDPP1 fusion increases PP1 concentration at kinetochores . HeLa cells were transfected with mCherry-Ska1 or mCherry-Ska1ΔCTDPP1 plasmids . Immunofluorescence was done and the amount of PP1 at the kinetochore was determined . Expression of Ska1ΔCTDPP1 plasmid increases PP1 localization to the kinetochore by about two fold . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 01510 . 7554/eLife . 12902 . 016Figure 7—figure supplement 2 . Ska1ΔCTDPP1 fusion but not Ska1ΔCTDpdPP1 ( pdPP1-phosphatase dead PP1 ) rescues mitotic phenotypes observed upon Ska1 depletion . Imagel panels show representative images of mitotic cells expressing the indicated mCherry constructs and progressing through mitosis . Expression of Ska1ΔCTDPP1 ( fusion of PP1 to Ska1ΔCTD ) rescues mitotic progression in Ska1 depleted cells while expression of Ska1ΔCTDpdPP1 does not . DOI: http://dx . doi . org/10 . 7554/eLife . 12902 . 016 To determine if the phosphatase activity of PP1 was essential for the rescue of Ska1-depleted cells expressing the Ska1-PP1 chimera , we generated an expression vector containing a fusion protein with a point mutation ( H248K ) in PP1 that abolished its catalytic activity ( Hirschi et al . , 2010 ) . Expression of PP1 H248K ( phosphatase dead PP1/pdPP1 ) without fusion to Ska1 was toxic to cells and induced cell death in interphase ( data not shown ) . Expression of Ska1ΔCTD-PP1 H248K ( Ska1ΔCTDpdPP1 ) allowed cells to enter mitosis , but failed to rescue the mitotic delay of Ska1-depleted cells ( Figure 7C , D , E—figure supplement 2 ) . The chromosome alignment delays observed upon Ska1 depletion were also rescued by expression of Ska1-PP1 fusion protein but not by expression of Ska1ΔCTDpdPP1 ( Figure 7E ) . Indeed , mere expression of Ska1ΔCTDpdPP1 without Ska1 depletion resulted in phenotypes remarkably similar to those caused by Ska1 depletion , indicating that this mutant acted in a dominant negative manner . These data indicate that a major function of the Ska complex is to recruit active PP1 to kinetochores . Before the alignment of chromosomes at the metaphase plate , the spindle checkpoint inhibits anaphase onset through signals generated by protein interactions at unattached kinetochores . This signaling requires the KMN complex , composed of the Knl1 protein , the Mis12 protein complex , and the Ndc80 protein complex . The central checkpoint kinase , Mps1 , phosphorylates several methionine-glutamate-leucine-threonine ( MELT ) motifs on Knl1 . The Bub1–Bub3 complex binds to phosphorylated MELT motifs and further recruits its binding partner , the BubR1–Bub3 complex ( Yamagishi et al . , 2012; Shepperd et al . , 2012; London et al . , 2012; Primorac et al . , 2013 ) . Mps1 also phosphorylates Bub1 to promote recruitment of the Mad1–Mad2 protein complex ( Yamagishi et al . , 2012; London et al . , 2012 ) . In the absence of microtubules , Mps1 associates with kinetochores via the Calponin homology ( CH ) domains of the Ndc80 complex ( Hiruma et al . , 2015; Ji et al . , 2015; Kemmler et al . , 2009; Nijenhuis et al . , 2013; Zhu et al . , 2013 ) . Photobleaching studies show that Mps1 binding to kinetochores is highly dynamic , with a turnover time of ~13 s ( Howell et al . , 2004 ) . The binding of microtubules to the Ndc80 complex displaces Mps1 , and in budding yeast , this results in a rearrangement of kinetochore substructure such that Mps1 loses access to Knl1 ( Aravamudhan et al . , 2015 ) . In metazoans , microtubule binding to Ndc80 displaces Mps1 from kinetochores ( Hiruma et al . , 2015; Ji et al . , 2015 ) . Thus , displacement of Mps1 from the Ndc80 complex by microtubule binding appears to be a central element of turning off kinase signaling . In metazoans , dynein 'stripping' of checkpoint proteins , including Mad1 , Mad2 and BubR1 , plays an additional role in down regulating the checkpoint signal . The most prominent phenotype of Ska complex depletion is arrest or long delay at metaphase , and these phenotypes require intact spindle checkpoint signaling . Abrogating checkpoint signaling with a chemical inhibitor of Mps1 induces mitotic exit in cells arrested in mitosis with high concentrations of microtubule poisons , but this exit is slower in cells depleted of Ska ( Sivakumar et al . , 2014 ) . This result is consistent with the idea that Ska functions in opposition to and downstream of checkpoint signaling . Here , we show that the Ska complex is required for full kinetochore recruitment of PP1 , a likely candidate for reversing checkpoint kinase phosphorylations . The fact that Ska depletion cannot completely block mitotic exit when Mps1 inhibitors are added likely reflects the fact that other pools of PP1 or other phosphatases , particularly PP2A , may also play a role in mitotic exit ( Nijenhuis et al . , 2014; Foley and Kapoor , 2013; Grallert et al . , 2015 ) . However , the strong metaphase arrest phenotype seen after Ska depletion attests to the importance of Ska-associated PP1 in regulating the metaphase-anaphase transition in normal mitosis . We found that the Ska1 protein , and specifically the Ska1 CTD , previously shown to bind microtubules , is essential for binding PP1 and recruiting it to the kinetochore . Since the C-terminal domain is involved in binding to both PP1 and microtubules , we propose to name this domain simply as the CTD , instead of the MTBD . Consistent with a role in opposing Mps1 checkpoint signaling , we find that in cells where PP1 binding to Ska1 is compromised , there is a 30% increase in phosphorylation of a MELT motif on Knl1 and a 20% increase in recruitment of Bub1 ( Figure 5B , C ) . A similar ( 30–40% ) increase in MELT phosphorylation and Bub1 levels also occurs upon mutation of Knl1 to inhibit its binding to PP1 ( Nijenhuis et al . , 2014 ) . We sugggest that the pools of PP1 at the kinetochore are distinct and may play cooperative and specific roles at different stages of mitosis . It remains possible that there are unidentified targets whose dephosphorylation is more reliant on PP1 bound to Ska . The increase in Bub1 levels that we observe is also consistent with our earlier finding that Bub1 levels on kinetochores were increased in cells depleted of Ska3 ( Daum et al . , 2009 ) . Although Ska does localize at kinetochores not attached to microtubules , it accumulates to its highest levels on kinetochores of cells at metaphase , the moment in time when quickly reversing checkpoint-dependent phosphorylation would be most useful in initiating anaphase . The CTD of Ska1 has been structurally well characterized . It has the characteristics of a winged helix domain , a fold previously implicated in DNA binding and in mediating protein-protein interactions ( Abad et al . , 2014; Schmidt et al . , 2012 ) . It was shown that the Ska1 CTD binds microtubules through multiple sites and can bind both straight and curved microtubule protofilaments ( Abad et al . , 2014; Schmidt et al . , 2012 ) . The same Ska1 CTD is also required for PP1 binding . Because the Ska1 CTD binds to both microtubules and PP1 , its functions in promoting chromosome alignment and anaphase onset might be a consequence of either or both activities . In previous work , it was shown that mutation of three conserved arginine residues in the Ska 1 CTD ( R155A/R236A/R245A ) compromised binding to microtubules in vitro and produced only a partial rescue of the metaphase delay caused by Ska1 depletion ( Abad et al . , 2014; Schmidt et al . , 2012 ) . We found that this mutant of Ska1 R3A was still able to co-precipitate PP1 in extracts from cells where both were expressed as transgenes ( Figure 4—figure supplement 3 ) . On the other hand , complete replacement of the CTD by a PP1-binding motif or by PP1 itself resulted in nearly complete rescue of all Ska1 depletion phenotypes , suggesting that microtubule binding by the Ska complex can be made dispensable . We previously reported that Ska depletion does not impair chromatid separation in cells induced to enter anaphase by application of a chemical Cdk1 inhibitor ( Sivakumar et al . , 2014 ) . Based on that work and the data presented here , we propose that the microtubule-binding properties of Ska1 do not play a strong mechanical coupling function for kinetochore movement on microtubules . We were unable to find a Ska1 point mutant that retained microtubule binding but was deficient in PP1 binding to further test this idea . Given the clear molecular evidence for Ska binding to microtubules and to microtubule protofilaments , we favor the idea that microtubule binding to Ska serves a regulatory role . We propose that Ska1 CTD binds near the ends of kinetochore microtubules , where separated protofilaments are enriched . This binding in concert with Ska complex interactions with other kinetochore components , may control the local concentration , dynamics , or substrate specificity of the Ska-PP1 complex at kinetochores ( Figure 7F ) . PP1 was first identified as an important regulator that counters spindle checkpoint signaling in budding yeast and fission yeast ( Pinsky et al . , 2009; Vanoosthuyse and Hardwick , 2009 ) . The N terminus of all Knl1 homologs from yeast to mammals contain conserved PP1-binding motifs ( Liu et al . , 2010 ) . In budding and fission yeast , expression of Knl1 mutants unable to bind PP1 impaired the ability of the cells to overcome checkpoint signaling and strongly compromised cell growth ( Meadows et al . , 2011; Rosenberg et al . , 2011 ) . In C . elegans embryos , RNAi-mediated depletion of wild type Knl1 and its replacement with a PP1-binding mutant led to slow chromosome congression , delays at metaphase , and partial embryonic lethality ( Espeut et al . , 2012 ) . The role of Knl1 binding of PP1 was studied in mammalian cells treated with nocodazole to disrupt microtubules and induce a strong spindle checkpoint arrest . Under these conditions , cells in which Knl1 was replaced with a PP1-binding mutant showed slower mitotic exit in comparison to controls when spindle checkpoint signaling was experimentally extinguished with a chemical inhibitor of Mps1 ( Nijenhuis et al . , 2014 ) . Together these studies have led to a model in which PP1 binding by Knl1 is a key factor in opposing checkpoint signaling for promoting the onset of anaphase and mitotic exit . However , one important result argues that this model does not fully explain the regulation of the metaphase-anaphase transition in normal mammalian cell mitosis . In mammalian cells with intact spindles , not treated with microtubule drugs , replacement of wild type Knl1 with a mutant Knl1 unable to bind PP1 results in only a modest , 10-min delay at metaphase ( Zhang et al . , 2014 ) . In contrast , loss of PP1 recruitment by the Ska complex during normal mitosis causes a lengthy delay or complete arrest at metaphase . Importantly , Ska homologs have not been identified in budding or fission yeast , consistent with the importance of PP1 recruitment by Knl1 in those organisms . In C . elegans , a two-protein Ska complex is present ( Schmidt et al . , 2012 ) . However , RNAi and mutant studies on Ska homologs in C . elegans embryos have not revealed an essential role in chromosome segregation ( Arshad Desai , personal communication ) . Interestingly , an elegant approach for manipulating protein interactions within kinetochores in budding yeast at nanometer resolution indicated that recruitment of the yeast PP1 homolog to outer kinetochores was important for reversing Mps1 phosphorylations of Knl1 ( Aravamudhan et al . , 2015 ) . In mammalian cells with intact mitotic spindles , our study suggests that Ska , an outer kinetochore protein complex , is a critical recruiter of PP1 in opposing spindle checkpoint kinase signaling at kinetochores . Our data indicate that binding of Ska and binding of Knl1 to PP1 are independent , suggesting that multiple pools of kinetochore-associated PP1 may cooperatively counter kinase activities at kinetochores . Their functions may be additive , recruiting PP1 to the threshold level required for anaphase onset . Interestingly , similar to the Ska1 CTD , the N-terminal region of Knl1 adjacent to its PP1-binding motif also binds microtubules in vitro ( Cheeseman et al . , 2006 ) . It is conceivable that the microtubule-binding domains of Knl1 and Ska1 may each regulate their associated PP1 pools , allowing them to be sensitive to the attachment status of the kinetochore . In addition , several other PP1-interacting proteins , including Cenp-E , SDS22 and Repo-man , have been identified as playing roles in mitosis ( Kim et al . , 2010; Posch et al . , 2010; Trinkle-Mulcahy et al . , 2006 ) . However , these proteins , when expressed at endogenous levels , do not normally accumulate at kinetochores of metaphase chromosomes ( Kim et al . , 2010; Eiteneuer et al . , 2014; Wurzenberger et al . , 2012 ) . During other stages of mitosis , prometaphase and anaphase , they may function in regulating PP1 activities on kinetochores , chromosome arms , and in the cytoplasm ( Eiteneuer et al . , 2014; Wurzenberger et al . , 2012; Qian et al . , 2013; Qian et al . , 2011 ) . In the future , it will be important to determine which specific protein phosphorylations are targeted by Ska-PP1 or by other PP1-binding proteins during mitosis . Finally , it is clear that PP2A , and possibly other phosphatases also play vital roles in regulating phosphorylation to control chromosome movement and cell cycle progression in mitosis ( Nijenhuis et al . , 2014; Foley et al . , 2011; Grallert et al . , 2015; Kruse et al . , 2013; Porter et al . , 2013; Xu et al . , 2014; Espert et al . , 2014 ) . In summary , here we make the surprising discovery that a chimeric Ska1-PP1 fusion lacking the microtubule-binding domain of Ska1 rescues nearly all the mitotic phenotypes observed upon Ska depletion , including delays in chromosome alignment and metaphase arrest . This rescue is fully dependent on the phosphatase activity of the chimera . Moreover , when expressed on its own , the phosphatase-dead Ska1-PP1 chimera has dominant phenotypes that closely mimic those of Ska depletion . Thus , rather than serving a mechanical coupling function between kinetochores and microtubules , the microtubule-binding properties of the Ska complex may primarily aid in coordinating PP1 recruitment to , or activity at , kinetochores . Our data suggest that PP1 recruitment is a critical function of the Ska complex for opposing mitotic kinases that destabilize kinetochore-microtubule attachment and that signal the spindle checkpoint . Thus , the Ska complex may integrate chromosome alignment at metaphase with full recruitment of PP1 , thus opposing spindle checkpoint kinases signaling and promoting the metaphase-anaphase transition . All cell experiments were conducted with HeLa cells . Parental HeLa cells were obtained from ATCC , Manassas , VA . HeLA Tet-On cells were obtained from Clontech , Mountatin View , CA . HeLa cells stably expressing GFP-Histone H2B were provided by Geoff Wahl ( Kanda et al . , 1998 ) . HeLa Ska1-GFP and HeLa Ska1ΔMTBD cell lines were obtained from Iain M Cheeseman ( Schmidt et al . , 2012 ) . All lines were routinely tested and found to be free of mycoplasma but were not further authenticated . All cell lines were grown in tissue culture dishes , culture flasks or chambered coverslips in 5% CO2 at 37oC using complete DMEM media supplemented with 10% FBS , penicillin and streptomycin . To synchronize HeLa cells , cultures were treated with 2 mM thymidine for 18–24 hr and released into media containing 3 . 3 µM nocodazole . Transient transfection for expression of plasmids was achieved using the Fugene 6 ( Roche , Indianapolis , IN or Promega , Madison , WI or Mirus , Madison , WI ) or Lipofectamine 3000 ( Invitrogen , Carlsbad , CA ) transfection reagent according to manufacturer’s instructions . QuikChange Lightning site directed mutagenesis kit ( Agilent technologies , Santa Clara , CA ) was used to make point mutants of Ska1 . Transfection of siRNA was done using Lipofectamine RNAi reagent ( Invitrogen ) according to manufacturer’s instructions . SiGenome siRNA against Ska1 , On target plus smartpool siRNA against Ska2 and Ska3 was obtained from DharmaconGE ( Lafayette , CO ) and these were used at 25–50 nM final concentration . To generate the stable cell lines , HeLa Tet-On cells were transfected with pTRE2 vectors expressing siRNA resistant GFP-Ska1 or GFP-Ska1ΔCTD and selected with 300 μg/ml hygromycin ( Clontech ) . Further screening of the clones to obtain stable cells was done in the presence of 150 μg/ml hygromycin . HeLa H2B-GFP or HeLa Tet-On cells were grown in Nunc chambered coverslips ( Thermo Sci . Inc . , Waltham , MA ) . In some instances , to visualize DNA in HeLa cells , a cell permeable Hoechst dye ( 33342; Invitrogen ) was used at 25–50 ng/ml . Time-lapse fluorescence images were collected every 5 min for 24–48 hr using a Leica inverted microscope equipped with an environment chamber that controls temperature and CO2 , 40X objective , an Evolve 512 Delta EMCCD camera , and Metamorph software ( MDS Analytical Technologies , Sunnyvale , CA ) . Time-lapse videos displaying the elapsed time between consecutive frames were assembled using Metamorph or ImageJ software . The first time frame denoting onset of nuclear envelope breakdown ( NEB ) , metaphase chromosome alignment and anaphase onset/mitotic exit was recorded in Microsoft Excel and the interval from NEB to metaphase ( alignment time ) , metaphase to anaphase ( metaphase duration ) or NEB to anaphase onset/mitotic exit was calculated . For every cell , mitotic duration was calculated and the data were depicted as scatter plots with mean and SEM . Only cells expressing the indicated constructs ( determined by mCherry expression ) were counted to determine mitotic duration . The indicated proteins were tagged either in N terminus or C terminus and in both cases similar results were obtained by live cell imaging . Further , to be certain that the mCherry tag was not specifically influencing mitotic progression , the proteins were also tagged with GFP and again similar results were obtained . For clarity , only images and results obtained with mCherry-tagged proteins are presented . In scatter plots , each dot represents one cell; long horizontal line depicts mean and whiskers denote SEM . Graphpad prism was used for statistical analysis . HeLa cells were grown on glass coverslips or in Nunc chambered cover slides and treated as detailed in the figure legends . Cells were pre-extracted for 5 min using 1X PBS/PHEM solution containing 1% Triton X 100 supplemented with phosphatase inhibitors ( Okadaic acid at 1 μM ) . Cells were then fixed in 2 or 4% paraformaldehyde/PHEM solution supplemented with phosphatase inhibitors for 15 min . Coverslips were washed in MBST , blocked in 20% Boiled Normal goat/donkey serum or 2% Bovine serum albumin ( BSA ) for 1 hr , and incubated overnight at 4oC or 1 hr at room temperature with primary antibodies . Samples were then incubated with secondary antibodies for 1 hr , stained with DNA dye , DAPI , and mounted using Vectashield ( Vector Laboratories , Burlingame , CA ) . The following primary antibodies were used: ACA/CREST ( Anti-Centromere antibodies from Antibody Inc , Davis , CA ) and rabbit anti-Ska3 ( Daum et al . , 2009 ) , sheep anti-PP1 ( kind gift from Dr . Brautigan; used in Figure 1A images ) , goat anti-PP1 ( Santa Cruz Biotechnology Inc . ; used in Figure 1C ) , rabbit anti-pMELT ( Ji et al . , 2015 ) and rabbit anti-Bub1 ( Tang et al . , 2001 ) . Secondary antibodies used were goat anti–rabbit or donkey anti-goat or goat anti-human antibodies conjugated to Cy3 or FITC ( Jackson ImmunoResearch , West Grove , PA ) . The images were acquired using a Zeiss Axioplan II microscope equipped with a 100X objective ( N . A . 1 . 4 ) or using 100X objective on Deltavision microscope ( GE Healthcare , Pittsburgh , PA ) . Images were assembled using image J and Coreldraw ( Corel Corp , Ottawa , Canada ) . Quantification of the immunofluorescence images was done as described previously ( Daum et al . , 2009 ) . The graphs depict mean fluorescence value with SEM obtained from at least 10 cells in each condition . Graphpad Prism ( Graphpad , La Jolla , CA ) was used to determine statistical significance among groups . The Mis12-GFP , Ska1-GFP , Mis12Ska1-GFP plasmid construction has been described previously ( Sivakumar et al . , 2014 ) . To construct fragments of Ska1 , cDNA encoding indicated regions of Ska1 were PCR amplified from full length siRNA-resistant Ska1 and inserted in pCS2-GFP or GFP-N1 plasmid . pCS2-Myc-PP1 was made by inserting PP1γ into pCS2-Myc vector . Knl1 ( 34–81 ) was made by amplying the region between 34 and 81 residues of full length Knl1 and inserting the cDNA in mCherry-N1 or pCS2-mCherry vector . Knl1 ( 34–81 ) 4A was made by mutating the RVSF motif in Knl1 ( 34–81 ) to AAAA . Ska1ΔCTDPP1 or Ska1ΔCTDKnl1 ( 34–81 ) or Ska1ΔCTDKnl1 ( 34–81 ) 4A fusion proteins were made by inserting PP1 or Knl1 ( 34–81 ) or Knl1 ( 34–81 ) 4A after Ska1ΔCTD . All plasmids were verified by DNA sequencing . Plasmid containing 6xHis-PP1α7–330 was a gift from Wolfgang Peti ( Kelker et al . , 2009 ) . Plasmids encoding GST-Ska3 and untagged Ska2/Ska1 were generous gifts from Iain M Cheeseman ( Schmidt et al . , 2012 ) . To generate Ska1ΔCTD we introduced a single stop codons using site directed mutagenesis . We used a similar strategy to obtain GST-Ska31-343 . 6xHis-PP1α 7–330 was expressed and purified essentially as previously described ( Kelker et al . , 2009 ) with following modifications . Prior to elution , Ni-NTA IMAC ( Qiagen ) was incubated with wash buffer containing additional 5 mM ATP and 5 mM MgCl2 for 2 hr and subsequently washed with wash buffer . PP1 was further purified on Superdex 200 10/300 GL size-exclusion column in 50 mM HEPES pH 7 . 5 , 150 mM KCl , 1 mM DTT . Ska1ΔCTD/Ska2/Ska31-343 and Ska1FL/Ska2/Ska31-343 for gel filtration experiments were purified according to previously established protocol for Ska complex purification ( Schmidt et al . , 2012 ) with following modification . Immediately after the cleavage of the GST tag , proteins were concentrated using Amicon Ultra centrifugal filters and subsequently purified on Superdex 200 10/300 GL size-exclusion column ( GE Healthcare/Life Sciences ) in 50 mM Tris pH 7 . 5 , 150 mM KCl , 1 mM DTT . To generate complexes 100 nM 6xHis-PP1α 7–330 was incubated on ice for 1 hr with Ska1ΔCTD/Ska2/Ska31-343 or Ska1FL/Ska2/Ska31-343 in 1:2 molar ratio before loading on Superose 12 PC 3 . 2/30 size-exclusion column ( in 50 mM HEPES , 150 mM KCl , 2 mM MnCl2 and 1 mM DTT ) and 50 µl fractions were collected . Immunoblots were performed using antibodies raised against Ska3 ( Daum et al . , 2009 ) and hPP1 ( gracious gift from David Brautigan ) . The quantification of total PP1 was determined as a sum of the densitometry measurements ( ImageJ ) of all visualized fractions . To express the Ska proteins in bacteria for in vitro binding experiments , pGEX-Duet encoding GST-Ska1 and untagged Ska2 or PGEX-Ska3 were individually transformed into BL21 ( DE3 ) pLysS cells ( Invitrogen ) and protein expression was induced using 0 . 1 mM IPTG at 16–18°C overnight . Bacterial cell pellets were lysed with lysis buffer ( 25 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl , 10% ( v/v ) glycerol , 1–5 mM DTT , 0 . 1% Triton ) and sonicated . The lysate was then cleared by centrifugation at ~25 , 000 g for 1 hr . Supernatant was incubated over glutathione sepharose 4B beads ( GE Healthcare ) for 1–2 hr at 4°C . Beads were washed with lysis buffer 3–4 times . The GST-Ska proteins were eluted using glutathione , further purified and concentrated using PD-10 desalting columns ( GE Healthcare ) and Amicon Ultra centrifugal filters ( EMD Millipore , Billerica , MA ) . Bacterially purified GST-Ska proteins were bound to glutathione sepharose 4B beads ( GE Healthcare ) . S35-labeled Myc-PP1 was obtained by in vitro translation using rabbit reticulocyte lysates ( Promega ) . The Myc-PP1 was incubated with GST-Ska protein bound beads for 1 hr at room temperature . Beads were washed using wash buffer ( 50 mM Tris HCl , pH 7 . 5 , 100 mM KCl , 0 . 05% Triton ) 3–4 times and 2X SDS sample buffer was added . The proteins were resolved using SDS-PAGE gel , and the S35 signal was analyzed using a phosphor-imager ( Fuji , Burlington , NJ ) . To perform MST analysis , the GST-Ska1ΔCTD and His-PP1 were purified from bacteria . GST-Ska1ΔCTD was purified as described above and His-PP1γ was purified as described previously ( Peti et al . , 2013 ) . Briefly pET28a-His-PP1γ and a chaperone pGro7 ( expressing GroES-GroEL ) was transformed into BL21 cells and protein expression was induced overnight at 10°C using L-Arabinose and IPTG . The bacterial pellets were lysed in lysis buffer ( 50 mM Tris pH8 . 0 , 5 mM imidazole , 700 mM NaCl , 1 mM MnCl2 , 0 . 1% Triton X-100 ) and sonicated . The lysate was cleared by centrifugation at ~25 , 000 g for 1 hr . The supernatant was incubated over Ni2+-NTA resin ( Qiagen , Valencia , CA ) to allow His-PP1γ binding . For MST analysis , the His tag was cleaved using Thrombin . The GST tag was cleaved from Ska1ΔCTD using 3C protease . The untagged proteins were eluted , further purified using Superdex size exclusion columns and concentrated using Amicon ultra centrifugal filters ( GE Healthcare ) . Finally both Ska1ΔCTD and PP1 proteins were exchanged into MST buffer ( 25 mM HEPES , 50 mM NaCl , 1 mM TCEP ) . For MST , PP1 was covalently coupled to a fluorophore by incubating 200 μl of PP1 at a concentration of 40 μM with 1 μl of 40 mM Alexa-Fluor 488-N-hydroxysuccinamide ester ( Molecular Probes/Life Technologies , Grand Island , NY ) dissolved in 100% dimethyl sulfoxide ) for 30 min in the dark at room temperature . The labeled protein was separated from free dye by applying the mixture to a G25 column that had been equilibrated with 9 ml of protein storage buffer . Serial dilutions ( 1:1 ) of Ska1 CTD were made in 15 successive 10 μl reactions , resulting in 16 samples , with the highest concentration of Ska1 CTD being 40 μM . Each of the reactions was mixed with 10 μl 800 nM labeled PP1 . Thus , the final concentration of the labeled protein was 400 nM in all samples , and the final highest concentration of Ska1 CTD was 20 μM; all reaction mixtures were supplemented with Tween-20 ( NanoTemper , LLC , Munich , Germany ) to a final concentration of 0 . 05% ( v/v ) . After incubation at room temperature for approximately 30 min . , all the reactions were loaded into standard treated capillary tubes , and the final measurements were taken in a Monolith NT . 115 instrument ( Nanotemper LLC , Munich , Germany ) . The instrument’s LED ( illumination ) power was set to 25% and the MST laser power was set to 40% . Measurements were performed at ambient temperature , ca . 23°C . The times of data acquisition were 5 s before the activation of the MST laser , 30 s with the laser on , and 5 s after extinguishing the laser . Data analysis was performed in PALMIST ( biophysics . swmed . edu/MBR/software . html ) using the T-Jump mode ( Scheuermann et al . , 2016 ) . A negative control experiment in which Soybean Trypsin Inhibitor ( Worthington Biochemical Corp . , Lakewood , NJ ) was titrated into labeled PP1 under identical conditions showed only a weak trend in T-jump behavior ( data not shown ) . Antibodies ( GFP , Ska3 or Knl1- all produced in-house ) were coupled to Affi-prep Protein A beads ( Biorad ) at a concentration of 1 mg/ml . Whole cell HeLa cell extracts were lysed in buffer ( 25 mM Tris-HCl pH 7 . 5 , 75 mM NaCl , 5 mM MgCl2 , 0 . 1% NP-40 , 0 . 4% Triton X 100 , 1–5 mM DTT , 0 . 5 μM okadaic acid , 5 mM NaF , 0 . 3 mM Na3VO4 , 10 mM β Glycerophosphate , 50 units/ml Turbo-nuclease ( Accelagen , San Diego , CA ) ) containing protease inhibitor cocktail ( Roche ) . Lysate was cleared by high-speed centrifugation at 20 , 817 g for 20 min . The supernatant after centrifugation was added to antibody coupled beads and incubated at 4°C for 2–4 hr by end-over-end rotation . After incubation , the beads were washed three to four times and the immunoprecipitation reaction was terminated by addition of 2X SDS sample buffer . The immuno-precipitated proteins were resolved by SDS-PAGE electrophoresis and blotted with specific antibodies . Primary antibodies included rabbit anti-GFP antibody ( made in house ) , mouse anti-Myc ( Roche ) , rabbit anti-Knl1 ( made in house ) and rabbit anti-Ska3 ( made in house ) . Membranes were washed in TBS/0 . 05% Tween 20 ( TBST ) , and then incubated with secondary antibodies in 5% NFDM/TBST . Secondary antibodies were Dylight fluorescent dye conjugated ( Cell Signaling , Danvers , MA ) , and the membranes were analyzed with the Odyssey CLx Infrared Imaging system ( LICOR , Lincoln , NE ) . The protein bands on the membranes were quantified using Image Studio software ( LICOR ) . Every experiment was repeated at least three times , and the quantification shows the mean with SEM in each case .	When one cell divides into two daughter cells it is critical that both new cells inherit an entire copy of the genetic material . This process is called mitosis , and it involves the duplicated chromosomes lining up in the middle of the cell before being pulled apart into the two newly forming cells . Many different proteins control mitosis because mistakes during cell division can lead to cells with too much or too little DNA , which can lead to cancers and other diseases . Mitosis is mainly regulated by enzymes called kinases and phosphatases . Kinases add phosphate groups on to other proteins , which often changes their activity or localization within the cell . Phosphatases counteract the kinases by removing the phosphate groups . During mitosis , kinases and phosphatases accumulate at a specific region of the chromosomes called kinetochores . The kinetochores play two key roles: they are the regions from which the chromosomes are pulled apart and also serve as control centers for regulating the sequence of events in mitosis . To date , mitosis is best understood in yeast cells and less is known about the more complex process in human cells . Previous research had shown that human cells need a group of proteins called the Ska complex to undergo mitosis , because without this complex the chromosomes remain in the middle of the cell and do not separate . Sivakumar et al . – who include two of the researchers involved in the previous research – have now explored the Ska complex’s role in human cells in more detail . The experiments showed that the Ska complex binds to and recruits a phosphatase called PP1 to the chromosomes; this is not how PP1 is brought to the kinetochores in yeast . When enough PP1 is concentrated around the kinetochores this gives the human cell the signal to pull the chromosomes apart and finish mitosis . Future work could ask how PP1 brings about the last stages of mitosis; for example , by finding all the proteins from which PP1 removes phosphate groups . Lastly , further studies could also explore the possibility that the Ska complex performs other tasks that are crucial for the division of human cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2016	The human SKA complex drives the metaphase-anaphase cell cycle transition by recruiting protein phosphatase 1 to kinetochores
Cyclic G/AMP synthase ( cGAS ) initiates type-1 interferon responses against cytosolic double-stranded ( ds ) DNA , which range from antiviral gene expression to apoptosis . The mechanism by which cGAS shapes this diverse signaling landscape remains poorly defined . We find that substrate-binding and dsDNA length-dependent binding are coupled to the intrinsic dimerization equilibrium of cGAS , with its N-terminal domain potentiating dimerization . Notably , increasing the dimeric fraction by raising cGAS and substrate concentrations diminishes duplex length-dependent activation , but does not negate the requirement for dsDNA . These results demonstrate that reaction context dictates the duplex length dependence , reconciling competing claims on the role of dsDNA length in cGAS activation . Overall , our study reveals how ligand-mediated allostery positions cGAS in standby , ready to tune its signaling pathway in a switch-like fashion . Whether arising endogenously or exogenously , double-stranded ( ds ) DNA in the cytoplasm of eukaryote cells indicates major problems ( Chen et al . , 2016a; Paludan and Bowie , 2013 ) . For instance , genomic instability and damaged mitochondria introduce dsDNA into the cytoplasm ( Chen et al . , 2016a; Denais et al . , 2016; Mackenzie et al . , 2017; Paludan and Bowie , 2013; Shen et al . , 2015; West and Shadel , 2017 ) , and nearby rampant necrosis or pyroptosis can lead dsDNA to the cytoplasm of neighboring cells ( Abe et al . , 2013; Ishii et al . , 2001 ) . Moreover , the invasion of pathogenic bacteria or viruses introduces foreign dsDNA into the host cytoplasm ( Chen et al . , 2016a; Paludan and Bowie , 2013 ) . In metazoans , cyclic-G/AMP synthase ( cGAS ) plays a predominant role in initiating host innate immune responses against cytoplasmic dsDNA ( Chen et al . , 2016a; Sun et al . , 2013 ) . Upon detecting cytoplasmic dsDNA , cGAS cyclizes ATP and GTP into [2’−5’ , 3’−5’]-linked cGAMP ( Gao et al . , 2013b ) , a unique host second-messenger for activating type-1 mediated stress-responses via Stimulator of Interferon Genes ( STING ) . cGAS is integral not only to the host defense against all pathogens entailing DNA for replication ( e . g . HIV , HSV , L . monocytogenes; ( Gao et al . , 2013a; Hansen et al . , 2014; Reinert et al . , 2016 ) , but also to damaged organelles ( Mackenzie et al . , 2017; West et al . , 2015 ) . Moreover , cGAS plays a major role in regulating autoimmunity ( e . g . Aicardi-Goutières syndrome and systemic lupus erythematosus [An et al . , 2017; Gao et al . , 2015; Pokatayev et al . , 2016; Vincent et al . , 2017] ) and tumor formation and growth ( Ng et al . , 2018 ) . A signature of IFN-1 signaling is its variable outcomes , which include antiviral gene expression , cellular senescence , autophagy , and apoptosis ( Li and Chen , 2018; Li et al . , 2016; Liang et al . , 2014; Sun et al . , 2013; van Boxel-Dezaire et al . , 2006; Yang et al . , 2017 ) . cGAS contributes significantly to this complex signaling landscape , with its signal strength , signaling duration , and cellular contexts influencing the type of outcomes ( Li and Chen , 2018 ) . For example , the outcome of the cGAS pathway depends on cell type ( e . g . non-apoptotic macrophages vs . apoptotic T-cells [Gulen et al . , 2017; Larkin et al . , 2017; Li et al . , 2016; Tang et al . , 2016] ) , the amount of cGAMP ( e . g . autophagy vs . apoptosis [Gulen et al . , 2017; Li et al . , 2016; Liang et al . , 2014; Tang et al . , 2016] ) , and the duration for which cells are stimulated with cGAMP ( antiviral gene expression vs . apoptosis [Gulen et al . , 2017; Larkin et al . , 2017; Li et al . , 2016; Tang et al . , 2016] ) . The goal of the present study is to understand the molecular mechanisms by which cGAS drives such a dynamic signaling landscape . Resting cGAS is thought to be an inactive monomer , and formation of a 2:2 dimer with dsDNA within the catalytic domain ( human cGAS residue 157 – 522 ) is necessary for activation ( 2 cGAS molecules on two dsDNA strands [Li et al . , 2013; Zhang et al . , 2014] ) . cGAS recognizes dsDNA independent of sequence ( Gao et al . , 2013b; Kranzusch et al . , 2013; Li et al . , 2013; Zhang et al . , 2014 ) , thus it was initially proposed that any dsDNA long enough to support the dimerization of cGAS could activate the enzyme equally well ( e . g . ~15 base-pairs , bps ( Chen et al . , 2016a; Li et al . , 2013; Zhang et al . , 2014 ) ) . However , it was long known that dsDNA of at least 45 bp was required to elicit IFN-1 responses in cells ( Chen et al . , 2016a; Stetson and Medzhitov , 2006; Unterholzner et al . , 2010 ) . Indeed , two recent studies demonstrated that cGAS discriminates against short dsDNA ( Andreeva et al . , 2017; Luecke et al . , 2017 ) . For instance , cGAS is minimally activated in cells by dsDNA shorter than 50 bps , and maximal activation requires dsDNA longer than 200 bps , with the length-dependence more pronounced at lower dsDNA concentrations ( Andreeva et al . , 2017; Luecke et al . , 2017 ) . The dependence on dsDNA length is thought to arise because cGAS dimers linearly propagate along the length of two parallel dsDNA strands without making inter-dimer contacts , consequently generating a ladder-like complex that increases the overall stability via avidity ( Andreeva et al . , 2017 ) . Together , it is believed that dsDNA length-based signal-to-noise filtration occurs at the binding/recognition stage ( i . e . different KDs for different dsDNA lengths ) , but not at the signal transduction step ( i . e . same Vmax for different dsDNA lengths ( Andreeva et al . , 2017 ) ) . Our understanding of the mechanisms by which cGAS is activated has evolved over the years , yet it remains unclear why two conflicting views on the role of dsDNA length have existed . Moreover , we noted that neither the previous ( dsDNA length-independent ) nor current ( dsDNA length-dependent ) activation model provides a robust framework for understanding how cGAS might be able to shape its diverse signaling landscape . First , the relationship between dsDNA binding and activation is poorly established . For instance , it remains to be tested whether the initial dsDNA binding step alone sufficiently explains the dsDNA length-dependent activation of cGAS in cells . Second , the ladder model implies that dimerization efficiency continuously increases with dsDNA lengths ( >1000 bps ) , while the optimal cellular response peaks with any dsDNA longer than ~200 bps ( Andreeva et al . , 2017 ) . Third , the ladder model is heavily based on structural and functional studies of the catalytic domain of cGAS ( cGAScat ) . It was recently proposed that the N-domain of cGAS binds dsDNA and plays a crucial role in its cellular function ( Tao et al . , 2017; Wang et al . , 2017 ) . Moreover , dsDNA binding by the N-domain is thought to enhance the activity of the monomeric enzyme , consequently lifting the dsDNA length restriction ( Lee et al . , 2017 ) . Thus , it is not clear whether the ladder-like arrangement applies exclusively to cGAScat , or whether it is germane to the full-length protein ( cGASFL ) . Finally , given that cGAS is the predominant sensor for cytoplasmic dsDNA ( Chen et al . , 2016a ) , it is imperative for this enzyme to amplify and attenuate its signaling cascade in a switch-like manner to ensure proper host responses . How cGAS achieves this important task remains poorly understood . We find here that human cGAS can auto-dimerize without dsDNA . dsDNA regulates this intrinsic monomer-dimer equilibrium not only in a cooperative , but also in a length-dependent manner . Also unexpectedly , substrates ( ATP/GTP ) can pull cGAS into the dimeric state without dsDNA . Because ligand binding is coupled to dimerization , the length of dsDNA not only regulates binding and dimerization ( signal recognition ) , but also the substrate binding and catalysis ( signal transduction ) . Compared to cGAScat , cGASFL auto-dimerizes more readily and also couples binding of both substrate and dsDNA to dimerization more efficiently , revealing a new function of the N-domain in potentiating the dimerization of cGAS . Dimerization is essential for dsDNA-mediated activation of both cGASFL and cGAScat , and the dimers do not arrange in an ordered configuration on long dsDNA , suggesting the role of dsDNA length is to simply regulate the probability of dimerization . Importantly , shifting the monomer-dimer equilibrium via elevated enzyme and ATP/GTP concentrations in the absence of dsDNA does not override the requirement for dsDNA to activate cGAS . Instead , these other factors prime the enzyme to be activated even by short dsDNA , indicating that the dependence on duplex length can change according to cellular reaction context . Together , our results set forth a unifying activation model for cGAS in which the intrinsic monomer-dimer equilibrium poises the enzyme to dynamically turn on or off its signaling pathway in a switch-like fashion . Human cGAScat ( denoted as cGAScat hereafter ) eluted as two peaks in size-exclusion chromatography ( SEC ) depending on protein concentration ( Figure 1A ) . With decreasing protein concentrations , the two peaks progressively merged into the one with the lower apparent molecular weight ( Figure 1A ) , suggesting that cGAScat is subject to an intrinsic monomer-dimer equilibrium without dsDNA ( Figure 1—figure supplement 1 ) . This was surprising , as previous studies showed that mouse cGAScat behaved as a monomer ( Li et al . , 2013 ) ; we speculate that mouse-cGAScat intrinsically dimerizes more weakly . To further test the intrinsic dimerization capability of cGAS , we examined the oligomeric state using small-angle-x-ray-scattering ( SAXS; Figure 1B ) . The radius of gyration ( Rg ) and the maximum diameter ( Dmax ) for apo- cGAScat at all tested concentrations aligned better with those of dsDNA-bound mouse-cGAScat dimer ( Figure 1C–D; [Li et al . , 2013] ) . We analyzed the distrbution of monomeric and dimeric species using SAXS-estimated molecular weight ( SAXS MoW2 ) and OLIGOMER in ATSAS ( Figure 1D [Mylonas and Svergun , 2007; Petoukhov et al . , 2012; Petoukhov and Svergun , 2013] ) . Here , the fraction of dimeric species was proportional to protein concentrations , and the dimerization constant was estimated to be ~20 µM ( Figure 1D ) . Together , we concluded that cGAS has an intrinsic capacity to dimerize , albeit with low affinity . In allosteric signaling enzymes , incoming signal ( activator ) and substrates either exclusively or preferentially bind to the active state and stabilize the corresponding conformation ( Koshland et al . , 1966; Monod et al . , 1965; Sohn et al . , 2007; Sohn and Sauer , 2009 ) . Such a coupling mechanism synchronizes conformational states with activity states , thereby allowing the enzymes to generate switch-like responses ( Koshland et al . , 1966; Monod et al . , 1965; Sohn et al . , 2007; Sohn and Sauer , 2009 ) . Importantly , preferential , but not exclusive ligand binding to the active state grades signaling output , as the distribution of active and inactive species is dictated by the relative binding affinity of different activators to either state ( Monod et al . , 1965; Sohn and Sauer , 2009; Tsai and Nussinov , 2014 ) . Our observation that cGAS can dimerize on its own suggests a new framework for understanding its activation mechanism ( Figure 2A ) . Here , apo-cGAS is placed in an intrinsic allosteric equilibrium where it is predominantly an inactive monomer under normal conditions . Overexpression ( Ma et al . , 2015 ) , substrate binding , and cytoplasmic dsDNA synergistically activate cGAS by promoting dimerization . Furthermore , given that monomeric cGAS binds dsDNA ( Andreeva et al . , 2017; Li et al . , 2013 ) , it is possible that dsDNA length determines the fraction of active dimers ( Figure 2B ) , thus underpinning the duplex length dependent cellular activity ( Andreeva et al . , 2017; Luecke et al . , 2017 ) . Below , we describe a series of experiments to further test and develop this allosteric framework for understanding the activation of cGAS . The cellular activity of cGAS is dsDNA length-dependent ( Andreeva et al . , 2017; Luecke et al . , 2017 ) , as if the enzyme uses duplex length as a ruler to differentiate between signal and noise . Currently , it is believed that this length-based noise filtration occurs only at the initial encounter step , with longer dsDNA invoking a ladder-like arrangement ( Andreeva et al . , 2017 ) . However , all previous binding studies entailed raising cGAS concentrations ( Andreeva et al . , 2017; Li et al . , 2013 ) , which intrinsically alters the dimer population . Thus , we re-examined the coupled relationship between dsDNA-binding and dimerization without altering the intrinsic dimerization equilibrium . First , using both direct and competition methods , we observed that cGAScat indeed binds dsDNA in a length-dependent manner ( Figure 2—figure supplement 1A–B ) . Next , to directly monitor dimerization , we conjugated a FRET donor and acceptor peptide to two populations of cGAScat via sortaseA ( FRET: Förster Resonance Energy Transfer; Figure 2—figure supplement 1C ) . The dimerization of a 1:1 mixture of donor- and acceptor-labeled cGAScat at physiologically relevant concentrations was then tracked by changes in FRET emission ratios between the donor and acceptor with increasing concentrations of dsDNA ( Figure 2—figure supplement 1C; physiological concentrations of cGAS vary between ~10 – 500 nM [Andreeva et al . , 2017; Du and Chen , 2018; Ma et al . , 2015] ) . Increasing concentrations of 24 bp dsDNA did not induce significant changes in FRET ratios ( Figure 2C ) , consistent with the previous report that such a short dsDNA binds cGAS but cannot induce dimerization ( Andreeva et al . , 2017 ) . With longer dsDNA , we observed more robust changes in FRET signals ( Figure 2C ) . Importantly , the half-maximal dsDNA concentrations necessary to induce the FRET signal ( KFRET ) decreased with longer dsDNA , with the optimum length reaching at ~300 bps ( Figure 2C–D ) . The maximal change in FRET ratio also generally increased with longer dsDNA , suggesting the dimeric fraction increased with longer dsDNA ( Figure 2C ) . The fitted Hill constants in these experiments were between 1 . 5 and 2 , indicating that dsDNA-induced dimerization is a cooperative process ( Figure 2E ) . Overall , our results confirm that dsDNA binding and dimerization are directly coupled , consistent with the idea that the intrinsic monomer-dimer equilibrium underpins the dsDNA length discrimination by cGAS ( Figure 2A–B ) . It is thought that cGAS does not bind ATP/GTP in the absence of dsDNA , as the loops surrounding the active site would block substrate entry ( Gao et al . , 2013b ) . However , cGAS can bind cGAMP in the absence of dsDNA , and multiple crystal structures indicate that the B-factors of loops surrounding the active site are 5 to 20-fold higher than the protein core , suggesting cGAS might be able to weakly interact with ATP/GTP even without dsDNA ( e . g . PDB IDs: 4k8v , 4o69 , and 4km5; ( Gao et al . , 2013b; Kranzusch et al . , 2013; Zhang et al . , 2014 ) ) . Thus , we tested whether ATP/GTP and their nonhydrolyzable analogues ( AMPcPP/GMPcPP ) induce dimerization via our FRET assay . Here , introducing substrates increased the FRET ratio , albeit to a lower extent than long dsDNA ( Figure 2F ) , suggesting that substrates alone can pull cGAScat into the dimeric state to some degree . The lower capacity of AMPcPP/GMPcPP to induce FRET changes is consistent with our observations that the analogues bind more weakly than ATP/GTP ( Ki = 280 µM ( Figure 2—figure supplement 1D ) vs . KM of ~100 for ATP/GTP with dsDNA , see Figure 3 below ) . Together , our results suggest that the fraction of active , dimeric cGAS would be partitioned according to the length of dsDNA and the availability of substrates ( Figure 2A ) . Thus , our results support that cGAS employs a strategy similar to classical allosteric enzymes to generate a graded output . All published methods that quantitatively monitor the enzymatic activity of cGAS track cGAMP , and are not ideal for mechanistic studies due to their low throughput or difficulty in saturating the enzyme with substrates ( e . g . TLC , HPLC-Mass-Spec , and fluorescently-labeled ATP/GTP; ( Andreeva et al . , 2017; Gao et al . , 2013b; Hall et al . , 2017; Vincent et al . , 2017 ) ) . cGAS generates two inorganic pyrophosphates ( PPi ) per cGAMP . Thus , we adapted a pyrophosphatase ( PPiase ) -coupled assay developed by Stivers and colleagues ( Figure 3A; ( Seamon and Stivers , 2015 ) ) . Using this assay , we found that cGAScat produces PPi most efficiently in the presence of a 1:1 mixture of ATP and GTP plus dsDNA ( Figure 3B;>90% of its NTase activity produces cGAMP when ATP and GTP are equimolar ( Gao et al . , 2013b ) ) . Moreover , no PPi production was observed from an inactive cGAS variant ( E225A-D227A-cGAScat ( Gao et al . , 2013b ) ; Figure 3B ) , and the activity of PPiase was not rate-limiting ( Figure 3—figure supplement 1 ) . Thus , we concluded that the PPiase-coupled assay provides a robust method to quantitatively monitor the enzymatic activity of cGAS . Our experiments thus far support an activation model in which dsDNA length determines the distribution between active dimers and inactive monomers ( Figure 2A–B ) . This mechanism entails different dsDNA lengths to produce graded maximal signaling output ( Vmax ) even at saturating concentrations ( Sohn and Sauer , 2009 ) . In contrast , it has been proposed that the dsDNA length-dependent activity of cGAS arises solely at the signal recognition step ( binding ) , but not at the signal transduction step ( enzymatic step; ( Andreeva et al . , 2017 ) ) . However , the authors could not conduct their studies under steady-state conditions due to the use of fluorescently-labeled substrates ( Andreeva et al . , 2017 ) . Because our coupled-assay eliminates this issue , we directly tested whether dsDNA length could regulate the enzymatic activity of cGAS . Here , we found that cGAScat has low basal activity without dsDNA ( 180 ± 30 M−1min−1 ) , which can be increased by 50-fold with >300 bp dsDNA ( Figure 3C ) . dsDNA concentrations required to induce the half-maximal activity of cGAScat increased with shorter dsDNA ( Kact; Figure 3C and Figure 3—figure supplement 2A–B ) , consistent with the previously observed length-dependent binding ( Andreeva et al . , 2017 ) . Importantly , the maximum dsDNA-induced activity ( kmax ) also decreased with shorter dsDNA ( Figure 3C and Figure 3—figure supplement 2A and C ) , which is in contrast to the previous report proposing that the role of dsDNA length is limited to binding ( Andreeva et al . , 2017 ) . Moreover , normalizing the kmax by Kact for each dsDNA length showed that the overall signaling efficiency of cGAScat ( dsDNA binding and maximum output ) changes more drastically compared to either parameter alone ( Figure 3D , see also Figure 3—figure supplement 2A–C ) . For instance , the overall signaling efficiency changes by nearly 100-fold between 24 to 339 bp dsDNA , while either binding or maximal activity alone changes only up to 10-fold ( Figure 3D , see also Figure 3—figure supplement 2A–D ) . Together , our observations indicate that cGAS discriminates against short dsDNA not only at the initial recognition step , but again at the signal transduction step , resulting in two-stage dsDNA length discrimination . We next determined substrate turnover kinetics in the presence of various dsDNA lengths . Without dsDNA , cGAScat showed measurable NTase activities ( Figure 3—figure supplement 1B ) . With saturating dsDNA longer than 300 bps , the KM of cGAScat for ATP/GTP was near 100 µM , and the kcat was 5 min−1 ( Figure 3E and Figure 3—figure supplement 2D ) . The observed KM for ATP/GTP is comparable to previously reported values measured using Surface Plasmon Resonance ( SPR ) and rapid-fire Mass-Spec for both human and mouse enzymes ( Hall et al . , 2017; Vincent et al . , 2017 ) . Moreover , the relatively slow kcat is consistent with a report indicating that human cGAS is considerably slower than mouse cGAS ( ~20 min−1 ) ( Vincent et al . , 2017 ) . Considering intracellular concentrations of ATP and GTP are >1 mM and ~500 µM , respectively ( Chen et al . , 2016b; Traut , 1994 ) , our result suggests that once cGAS encounters cytoplasmic dsDNA , one cGAMP would be generated in less than 20 s , compared to about one per 15 min in the absence of dsDNA . With shorter dsDNA , the KM increased about 2-fold , and the kcat decreased up to 4-fold ( Figure 3—figure supplement 2D ) . Combined , our results indicated that the overall catalytic efficiency of cGAS can change up to 8-fold ( kcat/KM ) by the length of bound dsDNA ( Figure 3F and Figure 3—figure supplement 3D ) . On another note , the fitted Hill constants in these experiments were near two for all dsDNA lengths ( Figure 3—figure supplement 2D ) , consistent with the observation from mouse cGAScat ( Vincent et al . , 2017 ) . Because most cGAScat populations would be dimeric with saturating long dsDNA , the observed cooperativity is likely from substrate-substrate interactions ( i . e . ATP binding enhances GTP binding or vice versa; ( Vincent et al . , 2017 ) ) . Overall , these results further support that dsDNA length can grade the enzymatic activity of cGAS . It was recently reported that the N-domain of cGAS ( residues 1 – 156 ) plays an important role in vivo by providing an additional nonspecific dsDNA binding site ( Tao et al . , 2017; Wang et al . , 2017 ) . Moreover , it was proposed that the N-domain reduces the requirement for long dsDNA , because it facilitates the activation of monomeric mouse cGAS ( Lee et al . , 2017 ) . To test whether our findings using cGAScat still apply to the full-length enzyme , we generated recombinant cGASFL . The full-length protein eluted as two peaks in SEC ( Figure 4A and Figure ) , behaved as an extended particle by SAXS ( Figure 4B , and Figure 4—figure supplement 1B–C ) , and was free from nucleic acid contamination ( Figure 4—figure supplement 1A ) . Of note , it appeared that cGASFL has a higher dimerization propensity compared to cGAScat , as indicated by broader peak distribution at 15 µM ( Figure 4A vs . Figure 1A ) . Supporting this notion , SAXS analyses also suggested that the dimerization constant of cGASFL is about 2-fold less than cGAScat at ~7 . 5 µM ( Figure 4—figure supplement 1B–C; cGAScat is 48% dimeric at 15 µM; Figure 1C–D ) . To further test that the N-domain can dimerize we generated recombinant N-domain ( cGASN ) and found that it migrated as a dimer in SEC , and also behaved as an extended dimer in SAXS ( Figure 4—figure supplement 2A–C ) . Of note , in our solution equilibrium assay , cGASN bound dsDNA much more weakly than cGAScat , which is in contrast to the non-equilibrium mobility assay used by Tao et al . ( Tao et al . , 2017; Figure 4—figure supplement 2D–E; KD >10 µM ) . These observations suggest a new role of N-domain in assisting the dimerization of cGAS . cGASFL still bound dsDNA in a length dependent manner ( Figure 4—figure supplement 3A ) , and displayed dsDNA length-dependent changes in FRET ( Figure 4C , Figure 4—figure supplement 3B ) . KFRETs for dsDNA > 72 bp were essentially the same under the minimal enzyme concentrations allowed in our assays ( Figure 4C , Figure 4—figure supplement 3B ) , indicating that the full-length protein binds and dimerizes more readily on dsDNA . Substrates and their analogues also produced more robust changes in FRET signals for cGASFL compared to cGAScat ( Figure 4D , Figure 4—figure supplement 3B–C ) , further corroborating that the full-length enzyme couples substrate binding to dimerization more efficiently due to its enhanced intrinsic dimerization activity . We also found that dsDNA length still grades Kact and kmax of cGASFL , as observed with cGAScat ( Figure 4E , Figure 4—figure supplement 3D ) ; KM and kcat for cGASFL were also graded according to dsDNA lengths ( Figure 4F , Figure 4—figure supplement 3E–F ) . Overall , our observations indicate that cGASFL and cGAScat operate within the same molecular framework , and reveal a new role for the N-domain in potentiating the dimerization of cGAS . Although 24 bp dsDNA failed to induce dimerization ( Figures 2C and 4C ) , it activated cGAS to a significant extent ( Figures 3C and 4E ) . Monomeric cGAS can also bind dsDNA , but it is thought to be poorly activated ( Andreeva et al . , 2017; Li et al . , 2013 ) . Moreover , it was proposed that the N-domain enhances the dsDNA binding of monomeric cGAS ( Tao et al . , 2017 ) , thereby activating the enzyme by lifting the dimerization requirement ( Lee et al . , 2017 ) . Nonetheless , 24 bp dsDNA bound and activated both cGAScat and cGASFL only moderately ( Figures 3C and 4E ) . Thus , our data are most consistent with the allosteric model in which the presence of ATP/GTP increased the dimeric fraction , allowing the short dsDNA to activate cGAS to some extent ( Figure 2A–B ) . To further test this idea , we characterized the activities of a cGAS mutant that binds dsDNA but fails to dimerize , K394E ( Li et al . , 2013; Zhang et al . , 2014 ) ; Figure 5A . Both K394E-cGAScat and K394E-cGASFL behaved as single monomeric species in SEC ( Figure 5B and Figure 5—figure supplement 1A ) , consistent with previous reports ( Li et al . , 2013; Zhang et al . , 2014 ) . SAXS experiments corroborated that K394E-cGAScat is predominantly monomeric at all tested concentrations ( Figure 5—figure supplement 1B–D ) . Compared to wild-type , not only did K394E-cGAScat bind dsDNA more weakly , but also without length dependence ( Figure 5—figure supplement 2A ) . The dsDNA length-dependence of K394E-cGASFL was also less pronounced compared to wild-type cGASFL ( Figure 5—figure supplement 2B–C ) . We predict that the dsDNA length dependence of K394E-cGASFL likely arise from the dimerization of the N-domain . Importantly , without dsDNA , K394E-cGAS showed similar activities as wild-type; however , dsDNA failed to stimulate the enzymatic activity of the mutants regardless of duplex length ( Figure 5C–F ) . For instance , dsDNA marginally decreased the KM of K394E-cGAS , but the kcat did not increase significantly ( Figure 5D and F ) . Our results also support the idea that monomeric cGAS can bind substrate and is basally active , yet dimerization is necessary for dsDNA- and dsDNA length-dependent activation regardless of the intact N-domain . Furthermore , our observations support the idea that short dsDNA and substrates can synergistically activate cGAS ( see also Figure 7 below ) . cGAS dimers are thought to form a ladder-like array along the length of dsDNA to maximize the stability of its signaling complex ( Andreeva et al . , 2017 ) . Given that both cGAS monomers and dimers bind dsDNA ( Andreeva et al . , 2017; Li et al . , 2013 ) , our results are better explained by a simpler mechanism in which dsDNA length regulates the fraction of cGAS dimers without invoking an ordered structure ( Figure 2A–B ) . To further test this idea , we imaged cGAScat and cGASFL with dsDNA using nsEM ( Figure 6; see also Figure 6—figure supplement 1 for zoom-in images , and additional images in Figure 6—figure supplement 2 ) . When proteins were in excess over dsDNA , we observed large clusters likely reflecting multiple cGAS dimers binding to several different dsDNA strands ( Figure 6A and E ) . It is possible that these clusters reflect the recently observed phase-shifting condensates of cGAS•dsDNA ( Du and Chen , 2018 ) . With excess dsDNA over protein , which more likely resembles in vivo events when dsDNA breaches the cytoplasm , it appeared that cGAS dimers randomly decorated dsDNA ( Figure 6B and F ) , with the particle sizes corresponding to the dimeric species of cGAScat and cGASFL , respectively ( i . e . the Dmax for these constructs are ~10 and 18 nm , respectively; Figure 1 ) . Importantly , the ladder-like arrangement of cGAS particles was rare for both cGAScat and cGASFL ( Figure 6B and F , Figure 6—figure supplement 2D–E ) , suggesting that cGAS•dsDNA does not form an ordered supra-structure . On the other hand , the size of particles resulting from excess K394E-cGAScat with dsDNA appeared smaller and corresponded to the Dmax of cGAS monomers ( Figure 6C; see also Figure 5—figure supplement 1 ) , likely reflecting monomeric cGAS randomly bound on dsDNA . For K394E-cGASFL , we observed dsDNA-bound clusters somewhat similar to wild-type ( these clusters are likely mediated by the intact N-domain that promotes dimerization ) . However , the clusters were not as expansive as those formed by wild-type ( Figure 6E vs . G ) . Moreover , we did not observe any significant decoration of dsDNA when the K394E mutants were present in sub-stoichiometric amounts ( Figure 6D and H; the particle size observed in Figure 6H also corresponds to the monomeric full-length cGAS ) . Overall our nsEM experiments support the allosteric framework of cGAS ( Figure 2A–B ) in which the role of dsDNA length is to simply bias the fraction of active dimers without necessitating supramolecular assemblies . Nevertheless , given the low-resolution imaging of nsEM , future structural studies are warranted to more fully understand the nature of these cGAS•DNA complexes . It was initially proposed that dsDNA length does not play a significant role in regulating the activation of cGAS ( Gao et al . , 2013b; Kranzusch et al . , 2013; Li et al . , 2013 ) ; however , two recent studies have contested this model ( Andreeva et al . , 2017; Luecke et al . , 2017 ) . The reason for this discrepancy is still unclear . Our results suggest that raising enzyme and substrate concentrations increases the dimeric fraction of cGAS , while binding of short dsDNA cannot ( e . g . 24 bp ) . Given the vastly different cGAS and substrate concentrations used in previous studies ( Andreeva et al . , 2017; Gao et al . , 2013b; Kranzusch et al . , 2013; Li et al . , 2013; Luecke et al . , 2017 ) , we speculated that the apparent or lack of dsDNA length-dependence is caused by the fraction of cGAS dimers formed without dsDNA ( Figure 2A ) . To test this idea , we monitored the steady-state NTase activity of cGAScat and cGASFL with saturating amounts of various dsDNA lengths and a permutation of high and low concentrations of enzyme and ATP/GTP ( Figure 7A–D ) . Increasing substrate and enzyme concentrations did not eliminate the need for dsDNA . However , the dependence on dsDNA length progressively decreased with increasing protein and substrate concentrations . For instance , with low cGAScat and sub-KM ATP/GTP concentrations ( cGAS is predominantly monomeric ) , we observed strong dsDNA length-dependent activities , with a difference of 8-fold between 24 bp and 564 bp dsDNA ( Figure 7A ) . With low cGAS and high ATP/GTP , the difference between short and long dsDNA was 4-fold ( Figure 7B ) . With high cGAS and low ATP/GTP , the difference was again reduced to 2 . 5-fold ( Figure 7C ) . Finally , with high cGAScat and high ATP/GTP ( the dimer population is significant ) , the differential activity caused by various dsDNA lengths was merely 1 . 5-fold , with short dsDNA molecules robustly activating cGAScat ( Figure 7D ) . Furthermore , we observed the same trend from cGASFL except the effect of raising substrate and enzyme concentrations was more pronounced than cGAScat ( Figure 7—figure supplement 1 ) . These observations uncover the reason for conflicting observations regarding dsDNA length-dependence ( Andreeva et al . , 2017; Kranzusch et al . , 2013; Li et al . , 2013; Luecke et al . , 2017 ) . That is , the dependence on dsDNA length can either manifest or diminish by different reaction contexts that dictate the fraction of dsDNA-free cGAS dimers . Our results in turn indicate that cGAS is primed to generate a graded signaling output depending on the overall reaction condition ( e . g . the length of cytoplasmic dsDNA , cGAS expression level , and available ATP/GTP ) , providing a molecular framework for its context-dependent and diverse stress responses ( Gulen et al . , 2017; Larkin et al . , 2017; Li and Chen , 2018; Li et al . , 2016; Tang et al . , 2016 ) As the initial receptor in a major inflammatory signaling pathway ( Chen et al . , 2016a ) , it is critical for cGAS to possess a very stringent noise filtering mechanism . Although cGAS binds dsDNA in a sequence-independent manner ( Gao et al . , 2013b; Li et al . , 2013; Zhang et al . , 2014 ) , it uses dsDNA length to distinguish signal from noise ( Andreeva et al . , 2017; Luecke et al . , 2017 ) . After all , dsDNAs arising from catastrophic conditions are significantly longer than 300 bps ( e . g . mitochondrial , genomic , and viral ) , while short dsDNAs likely indicate minor genome repair and/or resolution of infection ( i . e . the viral genome has been degraded ) . Here , we find that the allosteric coupling mechanism allows cGAS to generate a two-stage noise filter against short dsDNA . For instance , as others have reported ( Andreeva et al . , 2017 ) , we recapitulate here that cGAS binds and dimerizes on dsDNA in a length-dependent manner . Also as reported , we found that dsDNA length-dependent dimerization and binding of cGAS in vitro only gradually changes ( Figures 2–4; Andreeva et al . , 2017 ) . However , we found that dsDNA length also grades the enzymatic activity of cGAS ( Figures 3–4 ) . Thus , combined with the length-dependent complex formation of cGAS dimers ( signal recognition ) , the length-dependent enzymatic activity ( signal transduction ) would allow cGAS to further differentiate correct pathogenic dsDNA from noise ( short dsDNA ) . Of note , given that dsDNA length-dependence subsides with high concentrations of cGAS , our new model also provides an avenue for how improper clearance of pathogenic or self-dsDNA can induce spurious activity of cGAS leading to auto-inflammatory conditions ( Gao et al . , 2015; Li and Chen , 2018 ) . The interactions between cGAS and its ligands ( dsDNA and ATP/GTP ) display positive cooperativity , a hallmark of allosteric enzymes ( Figures 2–4 ) . One key feature of a cooperative system is its capacity to amplify and attenuate the output in a switch-like manner ( Monod et al . , 1965; Sohn and Sauer , 2009 ) . For instance , when the concentrations of cGAS , dsDNA , and ATP/GTP change by a factor of two , a non-cooperative system would yield a total 8-fold increase in output ( 2 × 2×2=8 ) . However , because cGAS requires dimerization for activity and displays a Hill constant near two in its interaction with both dsDNA and ATP/GTP , the same two-fold change would be further amplified by the exponent of two , leading to a 64-fold amplification in output ( 22 × 22×22=64 ) . Conversely , the same cooperative mechanism would allow cGAS to attenuate its signaling output by the same magnitude with decreasing enzyme and ligand concentrations . Together with the dsDNA-length dependent activity , the cooperativity would enable cGAS to dramatically alter its output according to the changes in input parameters , allowing the initial receptor to dynamically regulate its signaling pathway in a switch-like manner . Although cGAScat is sufficient to bind dsDNA and generate cGAMP in vitro , the intact N-domain is crucial for augmenting its function in cells ( Tao et al . , 2017; Wang et al . , 2017 ) . It has been presumed that the major role of the N-domain is to enhance dsDNA binding ( Lee et al . , 2017; Tao et al . , 2017 ) . Furthermore , it was proposed that the N-domain promotes the activation of monomeric mouse cGAS by dsDNA ( Lee et al . , 2017 ) . Here , we found that N-domain potentiates the dimerization of cGAS . Our results also indicate that dimerization is necessary for dsDNA-mediated activation by both cGAScat and cGASFL ( Figure 5 ) . It is possible that mouse cGAS operates in a different mechanism than human cGAS . Indeed , it was recently proposed that mouse-cGAS would not depend on dsDNA length as much as human-cGAS for activation , as the former binds short dsDNA more tightly ( Zhou et al . , 2018 ) . However , it was previously shown that both human and mouse-cGAS exhibit similar dsDNA length dependent activation ( Andreeva et al . , 2017 ) . Considering that dsDNA-mediated dimerization is critical for both human and mouse cGAS variants for activation ( Andreeva et al . , 2017; Li et al . , 2013; Zhang et al . , 2014; Zhou et al . , 2018 ) , we propose that our findings are likely general phenomena across different species , and different intrinsic affinity constants caused by diverse primary sequences ( Zhou et al . , 2018 ) would dictate species-specific experimental observations . Absent-in-melanoma-2 ( AIM2 ) is another major cytoplasmic dsDNA sensor in mammals ( Fernandes-Alnemri et al . , 2009; Hornung et al . , 2009; Roberts et al . , 2009 ) . The single most important goal of the AIM2-mediated dsDNA sensing pathway is to induce cell-death , a digital ( not tunable ) process that does not require a new equilibrium ( Liu et al . , 2014; Roberts et al . , 2009 ) . Indeed , once assembled on dsDNA , the AIM2 inflammasome does not disassemble and multiple positive feedback loops reinforce the assembly , consequently generating a binary signaling response ( Matyszewski et al . , 2018 ) . By contrast , the cGAS signaling pathway elicits various stress-responses ranging from viral replication restriction to apoptosis , with the signal strength and cellular contexts determining the type of outcome ( Gulen et al . , 2017; Larkin et al . , 2017; Li and Chen , 2018; Li et al . , 2016; Liang et al . , 2014; Tang et al . , 2016; Yang et al . , 2017 ) . Unlike AIM2 , we find here that cGAS can dial its own activity ( tunable ) , providing a molecular framework for eliciting various cGAMP-dependent outcomes . Furthermore , although both AIM2 and cGAS are activated in a dsDNA length-dependent manner , the former assembles into filaments ( Matyszewski et al . , 2018; Morrone et al . , 2015 ) , while the latter only requires dimerization . Likewise , although cytoplasmic dsRNA sensors preferentially target long duplexes ( >500 bps ) , MDA5 assembles into filaments while RIG-I does not require polymerization for activation ( Linehan et al . , 2018; Peisley et al . , 2011; Peisley et al . , 2013; Ramanathan et al . , 2016; Sohn and Hur , 2016 ) . Thus , we propose that the assembly of supra-structures is not universal to host nucleic acid sensors . Rather , it appears that each sensor has evolved unique mechanisms to utilize the length of nucleic acids as a molecular ruler to distinguish self ( noise ) from nonself ( signal ) . In closing , our study reconciles the conflicting views on the roles of dsDNA length and the N-domain in activating cGAS . We also provide a mechanistic framework for understanding how cGAS can shape a complex signaling landscape depending on cellular reaction contexts . Future studies will be directed in understanding how this dynamic enzyme operates in conjunction with its downstream and regulatory components to regulate host innate immune responses against cytoplasmic dsDNA . dsDNA substrates and oligonucleotides shorter than 100 bps were purchased from Integrated DNA Technologies ( IDT ) . Longer dsDNAs ( ≥150 bps ) were generated by PCR . The human cGAS cDNA were kindly provided by Dr . Dinshaw Patel . E . coli pyrophosphatase was a gift from Dr . James Stivers . The SortaseA ( SortA ) enzyme was a gift from Dr . Hidde Ploegh . Purity and length of each dsDNA was confirmed by agarose gel electrophoresis . TAMRA- and Cy5-labeled peptides were purchased from Lifetein . ATP and GTP were purchased from Sigma . GMPcPP and AMPcPP were purchased from Jena Biosciences Protein preparation . Recombinant cGAS constructs were cloned into the pET28b vector ( Novagen ) with an N-terminal MBP-tag and a TEV protease cleavage site . Proteins were expressed using 200 µM IPTG at 16°C for overnight in E . coli BL21 Rosetta 2 . Recombinant cGAS constructs were then purified using amylose affinity chromatography , cation-exchange , and size exclusion chromatography . Tag-free , purified cGAS proteins were then frozen and stored in −80°C with a buffer containing 20 mM Tris HCl at pH 7 . 5 , 300 mM NaCl , 10% glycerol , 5 mM DTT . Fluorophore labeling . The labeling procedure was adapted from ( Guimaraes et al . , 2013 ) . 20 µM MBP-TEV-cGAS-LPETGG-6xHis was incubated with 30 µM SortA , 250 µM fluorophore-peptide in Sortase reaction buffer ( 50 mM Tris HCl pH 7 . 5 , 150 mM NaCl , 10 mM CaCl2 , 5% glycerol , 2 mM DTT ) at 25 ± 2°C for 3 hr on a rotator . Reactions were directly applied to Superdex 200 10/300 GL ( 20 mM Tris HCl pH 8 . 0 , 300 mM NaCl , 2% glycerol , 10 mM BME ) . Fractions containing cGAS were applied to Ni-NTA . Flow-through was applied to heparin resin and washed with Sizing Buffer . Protein was eluted with Sizing buffer supplemented with 500 mM NaCl . Eluted fractions were adjusted to 20 mM Tris HCl pH 7 . 5 , 300 mM NaCl , 10% glycerol , 5 mM DTT and concentrated . All experiments were performed at least three times . The fits to data were generated using Kaleidagraph ( synergy ) . Reported values are averages of at least three independent experiments and report errors are standard deviations . All reactions were performed under 25 mM Tris acetate pH 7 . 4 , 125 mM potassium acetate pH 7 . 4 , 2 mM DTT , 5 mM Mg ( acetate ) 2 at pH 7 . 4 , and 5% glycerol at 25 ± 2°C . dsDNA binding assays . Increasing concentrations of cGAS were added to a fixed concentration of fluorescein-amidite-labeled ( FAM ) dsDNA ( 5 – 10 nM final ) . Changes in fluorescence anisotropy were plotted as a function of cGAS concentration and fit to the Hill equation . For competition-based experiments , unlabeled dsDNA was titrated against a fixed population of FAM-dsDNA72 and cGAS ( [protein] = KD , dsDNA72 ) . Changes in fluorescence anisotropy ( FA ) was plotted against competitor dsDNA concentration and fit to yield IC50s . FRET-based oligomerization assays . 60 nM Cy5- and TAMRA-labeled MBP-TEV-cGAS-LPET-GGGQC/K-fluorophore were incubated with TEV protease in cGAS reaction buffer at 25 ± 2°C for 2 hr . Increasing amounts of dsDNAs of different lengths or equimolar concentrations of nucleotides were added to 20 nM cleaved FRET pair , and FRET efficiency was recorded until equilibrium was reached . Pyrophosphatase-coupled cGAS activity assay . cGAS activity was assayed using the pyrophosphatase-coupled assay developed by Stivers and colleagues ( Seamon and Stivers , 2015 ) with modifications . Briefly , cGAS was incubated with 50 nM E . coli pyrophosphatase , equimolar concentrations of ATP and GTP plus dsDNAs ( where indicated ) in the reaction buffer . At indicated time points , an aliquot was taken and mixed with an equal volume of quench solution ( Reaction buffer minus Mg++ plus 25 mM EDTA ) . Quenched solutions were then mixed with 10 µl malachite green solution and incubated for 45 min at RT . Absorbance at ~620 nm was compared to an internal standard curve of inorganic phosphate to determine the concentration of phosphate in each well . Phosphate concentrations of control reactions devoid of recombinant cGAS were subtracted from reactions containing recombinant cGAS . Apparent catalytic rates were calculated from the slopes of control-subtracted phosphate concentrations over time . Reported rates were halved to reflect pyrophosphate production . Average values are listed in Tables . nsEM . Experiments were conducted using a Philips BioTwin CM120 ( FEI ) as described previously ( Morrone et al . , 2015 ) . SAXS data was collected on the BIOSAXS 2000 ( Rigaku ) at the X-ray facility of the Department of Biophysics and Biophysical Chemistry at Johns Hopkins School of Medicine . Data was collected on at least three different concentrations for each sample . SamplesBi with scatter showing significant inter-particle effects were omitted from data analysis . Buffer-subtracted scatter was processed in Scatter ( Mylonas and Svergun , 2007; Petoukhov et al . , 2012; Petoukhov and Svergun , 2013 ) and with the ATSAS package ( Mylonas and Svergun , 2007; Petoukhov et al . , 2012; Petoukhov and Svergun , 2013 ) . Particle dimensions were compared between guinier analysis and real-space fitting of the scatter to ensure internal consistency of the data and fits . Estimates of average and relative molecular weights of each sample were estimated using porod volumes ( Mylonas and Svergun , 2007; Petoukhov et al . , 2012; Petoukhov and Svergun , 2013 ) and mass-normalized I0 values . The distribution of monomeric and dimeric species was calculated using SAXS-estimated molecular weights and OLIGOMER . IN OLIGOMER , crystal structures of monomeric cGAS and dimeric cGAS were used as a reference ( PDB ID: 4LEV ) .	The human immune system protects the body from various threats such as damaged cells or invading microbes . Many of these threats can move molecules of DNA , which are usually only found within a central compartment in the cell known as the nucleus , to the surrounding area , the cytoplasm . An enzyme called cGAS searches for DNA in the cytoplasm of human cells . When DNA binds to cGAS it activates the enzyme to convert certain molecules ( referred to as ‘substrates’ ) into another molecule ( the ‘signal’ ) that triggers various immune responses to protect the body against the threat . To produce the signal , two cGAS enzymes need to work together as a single unit called a dimer . The length of DNA molecules in the cytoplasm of cells can vary widely . It was initially thought that DNA molecules of any length binding to cGAS could activate the enzyme to a similar degree , but later studies demonstrated that this is not the case . However , it remains unclear how the length of the DNA could affect the activity of the enzyme , or why some of the earlier studies reported different findings . Hooy and Sohn used biochemical approaches to study the human cGAS enzyme . The experiments show that cGAS can form dimers even when no DNA is present . However , when DNA bound to cGAS , the enzyme was more likely to form dimers . Longer DNA molecules were better at promoting cGAS dimers to form than shorter DNA molecules . The binding of substrates to cGAS also made it more likely that the enzyme would form dimers . These findings suggest that inside cells cGAS is primed to trigger a switch-like response when it detects DNA in the cytoplasm . The work of Hooy and Sohn establishes a simple set of rules to predict how cGAS might respond in a given situation . Such information may aid in designing and tailoring efforts to regulate immune responses in human patients , and may provide insight into why the body responds to biological threats in different ways .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "immunology", "and", "inflammation" ]	2018	The allosteric activation of cGAS underpins its dynamic signaling landscape
Multidimensional landscapes of regulatory genes in neuronal phenotypes at whole-brain levels in the vertebrate remain elusive . We generated single-cell transcriptomes of ~67 , 000 region- and neurotransmitter/neuromodulator-identifiable cells from larval zebrafish brains . Hierarchical clustering based on effector gene profiles ( ‘terminal features’ ) distinguished major brain cell types . Sister clusters at hierarchical termini displayed similar terminal features . It was further verified by a population-level statistical method . Intriguingly , glutamatergic/GABAergic sister clusters mostly expressed distinct transcription factor ( TF ) profiles ( ‘convergent pattern’ ) , whereas neuromodulator-type sister clusters predominantly expressed the same TF profiles ( ‘matched pattern’ ) . Interestingly , glutamatergic/GABAergic clusters with similar TF profiles could also display different terminal features ( ‘divergent pattern’ ) . It led us to identify a library of RNA-binding proteins that differentially marked divergent pair clusters , suggesting the post-transcriptional regulation of neuron diversification . Thus , our findings reveal multidimensional landscapes of transcriptional and post-transcriptional regulators in whole-brain neuronal phenotypes in the zebrafish brain . The vertebrate brain harbors highly diverse neuronal types that are specifically interconnected to form functional circuits ( Kepecs and Fishell , 2014; Armañanzas and Ascoli , 2015; Moffitt et al . , 2018 ) . The brain comprises conserved major cell types ( neurons , progenitors , glia , endothelial cells , etc . ) ( Marques et al . , 2016; Saunders et al . , 2018; Hodge et al . , 2019 ) . Previous studies have indicated that neuronal types could be well characterized by their features , including electrophysiological properties , neurotransmitter/modulator identity , synaptic connectivity , brain region identity , and cellular morphology ( Zeisel et al . , 2015; Pandey et al . , 2018; Zeisel et al . , 2018 ) . How these neuronal features are molecularly encoded at the whole-brain level is an important issue in neuroscience . With the advent of single-cell RNA-sequencing ( scRNA-seq ) technology , recent studies have begun to characterize neuronal diversity by analyzing single-cell transcriptomes of large populations in the whole brain or specific brain regions of mice and humans ( Darmanis et al . , 2015; Lake et al . , 2016; Poulin et al . , 2016; Tasic et al . , 2016; Chen et al . , 2017; Tasic et al . , 2018 ) . These studies have provided compelling evidence that individual neuronal types could be well characterized by their transcriptomes ( Poulin et al . , 2016; Tasic et al . , 2018; Sugino et al . , 2019 ) . For instance , scRNA-seq analyses of half a million cells from the mouse brain showed that neurons could be classified by genes related to neuronal connectivity , synaptic transmission , and membrane conductance ( Zeisel et al . , 2018 ) . Besides , the scRNA-seq analysis of nearly 200 genetically marked mouse neuronal populations also showed that neurons could be classified by the expression level of various transcription factors ( TFs ) , ion channels , synaptic proteins , and cell adhesion molecules ( Sugino et al . , 2019 ) . These studies provided extensive information on the molecules that could be used to define neuronal types . Generally , these molecules could be sub-divided into two primary categories: regulatory genes and effector genes . Regulatory genes encode proteins involved in gene transcription and translation ( e . g . , TFs and post-transcriptional regulators ) , while effector genes encode proteins serving specific neuronal terminal features ( e . g . , synaptic proteins , ion channels , transporters , and receptors ) ( Kratsios et al . , 2015; Zeisel et al . , 2015; Paul et al . , 2017 , Zeisel et al . , 2018; Reilly et al . , 2020 ) . Regulatory genes are critical for establishing and maintaining effector gene profiles that resulted in diverse neuronal types . Interestingly , distinct TFs can determine neuronal types with similar neurotransmitter identities in the Caenorhabditis elegans , arguing for phenotypic convergence ( Serrano-Saiz et al . , 2013; Gendrel et al . , 2016; Hobert and Kratsios , 2019 ) . Remarkably , this phenotypic convergence has also been reported in the Drosophila’s optic lobe ( Konstantinides et al . , 2018 ) . However , the multidimensional landscapes of regulatory genes in vertebrate whole-brain neuronal phenotypes remain elusive . The larval zebrafish brain comprises only about 100 , 000 cells , thus providing an outstanding vertebrate model for studying cell diversity within the entire brain , using single-cell transcriptome analysis with full cell coverage by the 10× Genomics Platform . Previous scRNA analysis of the zebrafish nervous system has elegantly demonstrated the temporal dynamics of brain cell development ( Raj et al . , 2018; Raj et al . , 2020 ) . In this study , we generated the multidimensional landscape of regulator genes in effector gene-based neuronal phenotypes ( terminal features ) at the whole-brain level by combining single-cell transcriptome data obtained from the whole brain , specific brain regions , as well as neurotransmitter- and neuromodulator-defined neuronal populations . We found that , at the transcriptional and post-transcriptional levels , glutamatergic/GABAergic neurons with the same terminal features could express different TF profiles , while those with different terminal features could express the same TF but different RNA-binding protein ( RBP ) profiles . In contrast , neuromodulator-type neurons that display particular terminal features expressed unique TF profiles . Thus , our findings reveal multidimensional landscapes of transcriptional and post-transcriptional regulators in the whole zebrafish brain . To uncover the transcriptomic profiles of diverse cell types with regional identity in the larval zebrafish whole brain at single-cell resolution , we dissected and dissociated cells from the whole brain ( n = 4 ) , four specific brain regions ( n = 2 each ) , including the forebrain ( Fore ) , optic tectum ( OT ) , hindbrain ( Hind ) , and the region underneath the optic tectum ( sub-OT ) in the 8 days post fertilization ( dpf ) zebrafish ( Figure 1—figure supplement 1A ) . We performed scRNA-seq of these cells using the 10× Genomics Chromium 3’ v2 platform . The libraries were sequenced to a mean depth of 126 , 651 reads per library , with a median of 1891 UMI and 866 genes per cell ( Figure 1—source data 1 ) . Reproducibility of transcription analysis was shown by the finding that replicates of whole-brain samples and individual brain regions were primarily overlapped in the t-distributed stochastic neighbor embedding ( t-SNE ) plots ( Figure 1—figure supplement 1B-C ) . We obtained the transcriptomes of 65 , 253 cells and selected 45 , 746 cells for further analysis after filtering out low-quality data ( Materials and methods ) . Then aggregated all cells into 68 clusters using high-variance genes ( n = 1402 ) in t-SNE plot ( Figure 1A ) . The Jaccard index-based analysis showed that most clusters had a high Jaccard index greater than 0 . 6 , indicating the robustness of these clusters ( details in Materials and methods ) ( Tang et al . , 2020 ) except for a few neuronal clusters from sub-OT ( Clusters 52 , 66 ) , and non-neuronal clusters ( neuroprogenitors: Clusters 37 , 56; radial astrocytes: 65 , Figure 1—figure supplement 1E ) . Each cluster was annotated according to cell-type-specific marker genes from the literature or ZFIN ( Zebrafish Information Network ) database ( Figure 1—source data 2 ) . To assign the brain region identity for each cluster , cells from each of the four specific regions ( Fore , OT , sub-OT , Hind ) were found to cover multiple but non-overlapping clusters and all 68 clusters could be assigned with their brain region origins ( Figure 1—figure supplement 2A-B , and Figure 1—source data 3 ) . Furthermore , to identify potential brain region-specific markers that exist in all cell types , we identified all genes that were differentially expressed in each region for all major cell types ( vglut+ , glutamatergic neurons; gad+ , inhibitory neurons; pcna+ , neuroprogenitors; cx43+ , radial astrocytes ) . The differentially expressed genes shared by all cell types were considered to represent the targeted region-specific markers ( Figure 1C ) . We indeed identified a small set of genes specific to each brain region independent of cell type . For instance , foxg1a , en2a , and hoxb3a were explicitly expressed in all cell types of the Fore , OT , and Hind , respectively ( Figure 1D ) . Interestingly , these brain region-specific genes also exhibited a conserved region-specific expression pattern in the mouse brain ( Hanks et al . , 1995 , Manzanares et al . , 2001 , Kumamoto and Hanashima , 2017 ) . We found no specific gene for the sub-OT , probably due to the diverse brain structures in this region . These region-specific genes , which may be involved in forming regional identity during brain development , could be used to study region-specific neuronal connectivity and function . To assign the neurotransmitter/modulator identity for each cluster , we used the marker genes that were specific to primary neurotransmitter/modulator phenotypes , including slc17a6b ( glutamatergic ) , gad1b ( GABAergic ) , slc6a5 ( glycinergic ) , th ( dopaminergic [DA] ) , tph2 ( serotonergic [5-HT] ) , and chata ( cholinergic [ChAT] ) ( Figure 1—figure supplement 2C , Figure 1—source data 3 ) . The ratio of glutamatergic to GABAergic neurons was the highest in the forebrain and lowest in the hindbrain , indicating that glutamatergic neurons predominantly belonged to the forebrain , whereas glycinergic neurons mainly resided in the hindbrain ( Figure 1—figure supplement 2D ) . These regional patterns of neurons expressing different neurotransmitter types were validated using the transgenic fishlines: Tg ( vglut2a:loxp-DsRed-loxp-GFP ) , Tg ( gad1b:EGFP ) , and Tg ( glyT2:GFP ) , each exhibiting distinct labeling of glutamatergic , GABAergic , and glycinergic neurons , respectively ( Figure 1—figure supplement 2E ) . Moreover , the Lawson-Hanson algorithm for non-negative least squares ( NNLS ) analysis using cluster-specific marker genes ( top 20 , Figure 1—source data 4 ) showed that these 68 clusters exhibited a high overlap with their counterparts in the juvenile zebrafish brain recently reported ( Raj et al . , 2018 ) . Meanwhile , clusters with different regional origins or cell types also exhibited a high correlation with their counterparts in the juvenile zebrafish ( Figure 1E and Figure 1—figure supplement 2F ) . Thus , our analysis indicated that the brain at 8 dpf mostly represented cellular diversity in the juvenile brain . Furthermore , Gene Ontology ( GO ) analysis of 1402 variable genes used for the classification of all 68 clusters showed that the majority of these genes ( 78 . 4% , n = 1099 ) were effector genes , which could be classified as neuropeptides , receptors , transporters , ion channels , synaptic proteins , and cell adhesion molecules ( Figure 1—figure supplement 2G ) . This result suggests the importance of effector gene profiles in brain cell classification , which has also been appreciated by previous studies in different species ( Paul et al . , 2017 , Hodge et al . , 2019 ) . We thus generated the hierarchical classification of all 68 cell clusters using the profiles of these 1099 effector genes . All 68 clusters were first segregated into two groups , neuronal cells ( 48 clusters , 37 , 880 cells ) and non-neuronal cells ( 20 clusters , 7866 cells ) ( Figure 1F ) . Among non-neuronal cells , oligodendrocytes ( Clusters 40 , 42; olig2+ , sox10+ ) , microglia ( Cluster 55; apoeb+ , mpeg1 . 1+ ) , endothelial cells ( Clusters 48 , 57; fxyd1+ , rbp4+ ) , erythrocytes ( Cluster 67; hbbe1 . 2+ , hbae5+ ) , radial astrocytes ( Clusters 21 , 46 , 65; cx43+ , glua+ ) , and neuroprogenitors ( Clusters 8 , 14 , 18 , 28 , 29 , 33 , 34 , 37 , 53 , 56; pcna+ , cdk1+ ) were identified according to putative marker genes ( Figure 1—source data 5 ) . On the other hand , among neuronal cells , the first segregation defined three classes of neurons ( branch I , cerebellum and habenula; branch IIa , glutamatergic neurons , branch IIb , inhibitory neurons ) . Branch I consisted of granule cells ( Clusters 19 ) , torus longitudinals ( Cluster 45 ) , cranial ganglions ( Clusters 19 , 50 , 63 ) , dorsal and ventral habenula neurons ( Clusters 35 and 59 ) . Branch IIa included 22 subclasses of excitatory glutamatergic neurons ( vgluta2a+-gad1b--glyt2- ) , whereas Branch IIb inhibitory neurons included 17 subclasses of GABAergic neurons ( vgluta2a-/gad1b+ ) and 1 subclass of glycinergic neurons ( vgluta2a--gad1b+-glyt2+ , Cluster17 , Figure 1—source data 5 ) . To further examine the expression of neuromodulators at the whole-brain level , we performed the scRNA-seq analysis of neuromodulator neurons sorted from the whole brain of Tg ( ETvmat2:GFP ) transgenic fish ( Figure 2A ) . Using this fishline , we could examine transcriptomes of DA neurons , 5-hydroxytryptamine ( 5-HT ) neurons , and norepinephrinergic ( NE ) neurons ( Wen et al . , 2008 ) . After the filtering procedures described above , we obtained a total of 5368 vmat2-expressing cells ( Materials and methods ) . The analysis aggregated these cells into 22 clusters in t-SNE plots ( Figure 2B ) , more than those found using whole-brain samples ( two DA , one 5-HT , and three ChAT; Figure 1—source data 2 ) . To further validate the stability of clusters , we used the Jaccard index , which showed that the majority of clusters ( 20/22 ) were stable using mean/median Jaccard index >0 . 6 as cutoff ( Figure 2—figure supplement 1A ) . According to region-specific marker genes identified above and known marker genes for neuromodulator neurons , we assigned seven clusters ( Fore: Clusters 18–19; sub-OT: Clusters 16 , 20; OT: Cluster 22; and Clusters 17 , 21 without specific regional identity ) as DA neurons , 14 clusters ( Hind: Cluster 10; sub-OT: Clusters 1 , 2 , 6 , 7; and Clusters 3–5 , 8–9 , 11–14 without regional identity ) as 5-HT neurons , one cluster ( Hind: Cluster 15 ) as NE neurons ( Figure 2—figure supplement 1C-D , Figure 2—source data 1 ) . Further examination of these neuromodulator clusters for their expression of specific neurotransmitters showed that the majority of 5-HT ( 8/14 ) and DA clusters ( 5/7 ) expressed GABAergic markers gad1b/gad2 ( Figure 2C ) . This result was further validated by Tg ( ETvmat2:GFP::gad1b:gal4::uas:mCherry ) ( Figure 2—figure supplement 1E ) . Only three 5-HT clusters ( Clusters 11–13 ) expressed glutamatergic marker vglut3 , two DA clusters ( Clusters 16 , 17 ) , and one NE cluster ( Cluster 15 ) expressed glutamatergic markers , vglut2a and vglut2b ( Figure 2C ) . These results are consistent with previous studies ( Filippi et al . , 2014 ) . Besides , we also found three clusters ( Clusters 24 , 41 , and 64 ) in whole brain showed choline and glutamate preferential co-expression ( Figure 1—source data 2 ) . Thus , our analysis provided a whole-brain characterization of the co-expression patterns of neurotransmitters and neuromodulators . In sum , DA and 5-HT neurons preferentially expressed GABAergic markers , whereas NE and ChAT neurons mostly expressed glutamatergic markers . In the hierarchical classification based on effector gene profiles ( Figure 1F ) , glutamatergic/GABAergic clusters ( n = 39 ) at the terminus pairs represented the ones with the most similar terminal features , termed ‘sister clusters’ . In addition , the certainty of this hierarchical classification was verified by the bootstrap re-sampling analysis using pvclust v . 2 . 0 ( Figure 3—figure supplement 1A; Suzuki and Shimodaira , 2006 ) . We identified 11 pairs of sister clusters , neurons in each pair exhibited the same neurotransmitter types ( Figure 3—figure supplement 2A ) . To our surprise , neurons of each sister clusters could be from either the same ( n = 6 ) or different ( n = 5 ) brain regions ( Figure 3—figure supplement 2A ) , which did not reflect the strong brain region preference . To further examine the TF profiles of these effector gene-based sister clusters , we classified glutamatergic ( IIa ) and inhibitory ( IIb ) neurotransmitter-type neurons using TF profiles . Notably , TF-based and effector gene-based trees were distinct in terms of matching node ( only one matching node: Clusters 9/61 ) and tree distances ( tree distance = 0 . 71 , Figure 3—figure supplement 1B ) , suggesting that the effector gene-based sister clusters ( Figure 3—figure supplement 2A ) might express different TF profiles . We found out of 11 effector genes-based sister clusters , only one pairs could be found in TF-based sister clusters ( Clusters 9/61 , ‘matched pattern’ , Figure 3—figure supplement 2B ) . And other 10 effector gene-based sister clusters were separated in TF-based classification , suggesting that neurons with similar terminal features mostly expressed different TF profiles ( ‘convergent pattern’; Figure 3—figure supplement 2C ) . Also , neurons in each of these 10 sister clusters could come from either the same ( n = 5 ) or different ( n = 5 ) brain regions , exhibiting no brain region preference ( Figure 3—figure supplement 2C ) . Alternatively , we performed the population-level statistical analysis to compare the landscape of TF and effector gene expression accounting for the full spectrum of cell types rather than just the most similar sister clusters . For all glutamatergic/GABAergic neuronal clusters ( n = 39 ) , we calculated the distances between every two clusters ( C392 ) based on either effector gene profiles or TF profiles , and then defined the pairs , which had the lowest 10% distances after ranking , as similar pair clusters ( Figure 3A ) . Then , we defined paired clusters that were similar in both TF and effector gene profiles as ‘matched pattern’ , those paired clusters that were similar in effector gene profiles but not in TF profiles as ‘convergent pattern’ ( Figure 3B ) . The population-level analysis identified 19 pairs of effector gene-based similar pair clusters , 5 with matched pattern and 14 with convergent pattern ( Figure 3—figure supplement 2D-G ) . Overall , similar pair clusters with either matched or convergent pattern identified by the population-level analysis showed an overlapping but distinct pattern with those identified by the hierarchical sister cluster analysis ( Figure 3B ) . This discrepancy was likely due to the following facts: ( 1 ) In the population-level statistical analysis , we arbitrarily set the lowest threshold as a criterion to identify similar pair clusters . Thus , the levels of this threshold could influence the production of similar pair clusters . ( 2 ) In population-level statistical analysis , each cluster could use for multiple times , whereas in hierarchical sister cluster analysis , once a cluster was selected as a pair with another cluster , it could not be re-used again . To overcome this discrepancy , we intersected the results from hierarchical sister cluster analysis and population-level statistical analysis , and identified eight pairs of effector gene-based glutamatergic/GABAergic similar pair clusters , one with matched pattern and seven with convergent pattern ( Figure 3B ) . We further validated these patterns of paired neuronal clusters by subsampling of genes ( 80% of total either TFs or effector genes ) and average statistics over 20 times to re-identify similar paired clusters based on either TFs or effector genes ( Materials and methods ) . Notably , re-identified paired clusters of different patterns completely recapitulated those pairs identified using the population-level statistical analysis above ( Figure 3—figure supplement 2L ) , indicating the robustness of these patterns . By combining sister cluster analysis and the population-level statistical analysis , we identified one paired clusters of ‘matched pattern’ , seven paired clusters of ‘convergent pattern’ ( Figure 3B ) . Here were some representative cases with the convergent pattern . The first case was a glutamatergic pair cluster from different brain regions: tectal glutamatergic Cluster 1 and hindbrain glutamatergic Cluster 31 shared effector gene profiles including camk2n1a/stmn4/cbln2b/olfm3a/cd63 , but differentially expressed TF profiles , atf5b/bhlhe41/lhx1a and ddit3/cebpb/lef1 , respectively ( Figure 3C ) . The second case was GABAergic pair cluster from different brain regions: GABAergic clusters in the forebrain ( Cluster 36 ) and the sub-OT ( Cluster 16 ) shared effector gene profiles mcl1a/cbln1/tspan18a/dusp5/ncaldb , but differentially expressed TF profiles , tbr1b/foxg1a/barhl2 and otpa/otpb/six6a , respectively ( Figure 3D ) . Consistently , each of the above TF profiles has previously been characterized to participate in the specification of either glutamatergic or GABAergic neurons ( Li et al . , 2007; Kala et al . , 2009; Talbot et al . , 2010; Waite et al . , 2011; Achim et al . , 2013 ) . We further examined neuromodulator-type sister clusters based on either TF or effector gene profiles ( Figure 3—figure supplement 3A ) . Using hierarchical sister cluster analysis , 5/8 of neuromodulator-type sister clusters matched with those defined by TF profiles ( ‘matched pattern’ ) , 3/8 sister cluster were found to be separated in TF-based classification ( ‘convergent pattern’ , Figure 3—figure supplement 3D-G ) . Then , we also performed the population-level statistical analysis using all genes or 80% genes ( subsampling ) to compare the landscape of TF and effector gene expression . We found eight pairs with matched pattern and one pair with convergent pattern ( Figure 3—figure supplement 3H-I ) . Thus , we intersected hierarchical sister cluster analysis and population-level statistical analysis , and identified six pairs of similar neuromodulator pair clusters , five with matched pattern , and one with convergent pattern ( Figure 3—figure supplement 3J-L ) . Together with the above results of neuromodulator-type and neurotransmitter types , our analysis showed that neuromodulator pairs with similar effector gene profiles predominantly were ‘matched’ pattern , whereas neurotransmitter pairs with similar effector gene profiles mainly were ‘convergent’ pattern ( Figure 3E ) . Moreover , we performed global tree measurement using the R package ‘TreeDist’ , and found that the distance between TF-based and effector-based hierarchical tree of neuromodulator neurons ( Tree distance = 0 . 38 , Figure 3—figure supplement 3A-B ) was lower than that of glutamatergic/GABAergic neurons ( tree distance = 0 . 71 , Figure 3—figure supplement 1B ) , suggesting distinct TF regulatory logic of effector-based phenotypes between neurotransmitter-type and neuromodulator-type at the global level . The smaller distance for neuromodulator neurons was consistent with our conclusion that neuromodulator-type clusters predominantly expressed the same TF profiles ( ‘matched’ ) . The ‘matched’ pattern suggested the generation of similar neuromodulator pairs shared specific TF programs ( ‘stereotyped programming’ ) , and the ‘convergent’ pattern suggested the generation of similar neurotransmitter pairs could be generated by different TF programs ( ‘flexible programming’ ) . These different strategies may account for the fact that across species , neuromodulator types are more conserved , whereas neurotransmitter types are much diverse and variable ( La Manno et al . , 2016; Saunders et al . , 2018; Tiklová et al . , 2019; Poulin et al . , 2020 ) . On the other hand , in glutamatergic/GABAergic neuronal classification , we surprisingly found that 13 pairs of sister clusters with similar TF profiles were separated in the effector gene-based classification . In other words , different from matched and convergent pairs mentioned above ( Figure 3B–D ) , these 13 glutamatergic/GABAergic clusters exhibited different effector gene profiles but similar TF profiles , here terms as ‘divergent’ pattern ( Figure 3—figure supplement 4A ) . Also , the population-level statistical analysis identified divergent pairs ( n = 15 , Figure 3—figure supplement 4B ) that were largely overlapped with those identified by hierarchical sister analysis ( n = 10 , Figure 3B ) . Neurons in each of these 10 divergent paired clusters could be from the same ( n = 6 ) or different ( n = 4 ) brain regions ( Figure 3—figure supplement 4A ) . More strikingly , two pairs of neuronal clusters with divergent pattern expressed different neurotransmitters . For examples , sub-tectal glutamatergic neurons ( cluster 24 ) and hindbrain GABAergic neurons ( cluster 17 ) shared similar TF profiles id2a/pax2a/zfhx4/meis1b , but expressed different effector gene profiles , cpne4a/c1ql4b/slc5a7a and gad1b/slc6a1b/slc6a5 , respectively ( Figure 3F ) ; forebrain glutamatergic neurons ( Cluster 30 ) and tectal GABAergic neurons ( Cluster 11 ) shared TF profiles , bhlhe41/bcl11ba/mef2cb , but had differentially expressed effector gene profiles , adcyap1b/syt5b/rprml/gadd45bb , and kcnip1b/cxcl4/plxdc1 , respectively ( Figure 3G ) . This result illustrated that neurons with different terminal features could also express the same TF profiles . Together , our analysis demonstrated the landscape of TFs in whole-brain-wide neuronal phenotypes in the larval zebrafish brain . Neuronal morphology and effector gene profile are two critical criteria of neuron diversity classification ( Sugino et al . , 2019; Peng et al . , 2021 ) . Also , many previous studies have provided the apparent links between effector genes and neuron morphology ( Whitford et al . , 2002 , Marcette et al . , 2014; Delandre et al . , 2016; Noblett et al . , 2019; Peng et al . , 2021 ) . Thus , after we found three patterns ( ‘matched’ , ‘convergent’ , and ‘divergent’ ) , we wondered if similar patterns present between morphological subclasses and TFs . To experimentally verify the relationship between TFs and neuronal phenotypes , we focused on morphological subtypes of tectal glutamatergic neurons . Higher-coverage transcriptomes of tectal glutamatergic neurons were further achieved by scRNA-seq of sorted tectal glutamatergic neurons in 8 dpf using Tg ( vglut2a:loxp-DsRed-loxp-GFP ) fishline ( Figure 4A ) . Canonical correlation analysis ( CCA ) of cells from two independent experiments indicated the reproducibility of the analysis ( Figure 4—figure supplement 1A ) . Analysis of 3 , 883 sorted tectal glutamatergic neurons yielded 11 clusters ( n = 11 , Figure 4A ) , more than those identified from the whole-brain samples ( n = 5 , Figure 4—figure supplement 1B ) . The majority of clusters ( 10/11 ) were stable using mean/median Jaccard index >0 . 6 as cutoff ( Figure 4—figure supplement 1D ) . Each cluster of tectal glutamatergic neurons was identified by TF marker genes ( Figure 4—figure supplement 1E-F , Figure 4—source data 1 ) . We developed a labeling strategy using three different plasmids to analyze morphological subclasses of tectal glutamatergic neurons expressing identified TFs . First , we placed Gal4FF cassette at the start-codon of each TF by bacterial artificial chromosome ( BAC ) recombination technique ( Suster et al . , 2011 ) and constructed the Gal4FF plasmids for 15 marker TF genes ( Figure 4B , Materials and methods ) . Second , we generated vglut2a:CRE BAC plasmid using a similar method . Third , plasmid uas:loxp-stop-loxp-tdTomatocaax was used with the first two plasmids to intersectionally label single tectal glutamatergic neurons , each expressing one specific TF ( Figure 4C ) . These labeled neurons were then used for morphological analysis . We labeled and defined morphological subclasses of tectal neurons by injecting all three plasmids into zebrafish embryos at 4–16 cells stage . We observed the enormous variability in the morphology of tectal glutamatergic neurons in their more refined structures . By the limited number of neurons analyzed ( n = 574 ) , we were unlikely to define morphological subclasses using a global morphological description . Instead , we used the criterion based on major morphological features , including stratification , soma position , and projection patterns , to define the morphological subclasses in the current study . In addition , similar morphological classification has also been used previously for tectal neurons ( Nevin et al . , 2010 , Robles et al . , 2011; DeMarco et al . , 2020 ) . We selected six TF marker genes ( en2b , foxb1a , zic1 , bhlhe22 , zbtb18 , and irx1a ) for further analysis based on two criteria: ( 1 ) These TFs are highly expressed and specific to individual tectal glutamatergic clusters based on scRNA-seq analysis . ( 2 ) Their BAC plasmids could reliably mark particular morphological subclasses ( at least in four animals ) . Furthermore , using confocal imaging , we reconstructed a total of 574 tectal neurons ( from 263 zebrafish , Figure 4—source data 2 ) that expressed one of these six TFs ( Figure 4—figure supplement 1G ) . These neurons were categorized into seven morphological subclasses: bi-stratified ( I , n = 36 ) , mono-stratified ( II , n = 71 ) , non-stratified ( III , n = 318 ) , necklace-like ( IV , n = 76 ) , cross-hemispheric ( V , n = 15 ) , ascending ( VI , n = 18 ) , and descending ( VII , n = 40 ) ( Figure 4D ) . Whether a specific TF could serve as a marker for a particular morphological subclass was determined by the criterion that the TF appeared in a given subclass at least four times ( from at least four fish ) . Remarkably , all six TFs marked multiple morphological subclasses in a combinatorial manner , while each of seven morphological subclasses was marked by numerous TFs ( Figure 4E ) . For example , zic1 highly marked non-stratified and bi-stratified subclasses , and all six TFs marked the non-stratified subclass . In summary , we found that single TF could mark multiple morphological subtypes in the optic tectum , and multiple TFs could mark a single morphological subtype . This observation could be inferred from ‘convergent pattern’ ( same TFs were expressed in different effector-based subtypes ) and ‘divergent pattern’ ( different TFs were expressed in similar effector-based subtypes ) , respectively . However , we could not exclude unknown indirect regulations of morphological subtypes by TFs . The above finding that neuronal clusters with different terminal features expressed similar TF profiles were likely resulted from non-TF determinants ( Figure 3F–G ) . We then performed GO analysis of differentially expressed genes between sister clusters of 10 pairs with different terminal features but similar TF profiles ( ‘divergent pattern’ ) . We surprisingly found that many differentially expressed genes encoded RBPs ( Figure 5A ) . We thus yielded a library of RBP-encoded genes that could differentially mark neurons with different terminal features ( Figure 5—source data 1 ) . Here were two examples: two forebrain glutamatergic neuronal clusters ( Clusters 7 and 38 ) shared the similar expression pattern of TF profiles eomesa/neurod1/tbr1b/foxg1a , but had different effector gene profiles and RBPs , rprma/stxbp1b/egfl6/khdrbs2/msi2b/rbm5 and nrgna/susp5/fam43b/qkia/larp6a , respectively; GABAergic neurons in hindbrain ( Cluster 12 ) and sub-tectal ( Cluster 16 ) shared TF profiles , otpa/bhlhe41/id2a , but differentially expressed effector genes and RBPs , kctd4/rgs5b/cpne2/eif5a2/celf3a and slc6a1b/cplx2l /ndrg2/msi2b/rbfox1/msi2a , respectively ( Figure 5B ) . These RBP-encoded genes exhibited wide RNA functions , including capping , splicing , polyadenylation , transport , stability , and translation ( Figure 5C , Figure 5—source data 3 ) . Besides , our analysis also yielded a comprehensive list of genes encoding RBPs marking other patterns ( Figure 5—figure supplement 1A-D , Figure 5—source data 1 , Figure 5—source data 2 ) . We found that some RBPs exhibited differential expression between paired clusters specific to one of the above three patterns ( Figure 5—figure supplement 1 ) . Also , we found that the number of RBPs that were specific to divergent pattern ( n = 52 from 10 pairs ) was much more than those specific to either matched pattern ( n = 7 from one pair ) or convergent pattern ( n = 10 from seven pairs , Figure 5—figure supplement 1E-F ) . Thus , in terms of the number of pattern-specific RBPs per pair , RBPs showed the preferential expressions in divergent pairs compared to convergent ones ( Figure 5D ) . Note that since there was only one pair of matched pattern , we did not include it in this analysis . These pattern-specific RBPs were known to be involved in a wide range of RNA functions , including capping , splicing , polyadenylation , transport , stability , and translation ( Figure 5—figure supplement 1F , Figure 5—source data 3 ) . Thus , in addition to TFs , post-transcriptional regulatory factors also play significant roles in determining transcriptome profiles of neuronal classification . Furthermore , we examined effector gene that were targeted by pattern-specific RBPs using the oRNAment database ( http://rnabiology . ircm . qc . ca/oRNAment/ ) to explore a potential causality between the divergence of RBPs and the divergence of effector genes ( Benoit Bouvrette et al . , 2020 ) . GO analysis showed that pattern-specific RBPs targeted various molecular categories including effector genes , TFs , mRNA processing , metabolism ( Figure 5—figure supplement 1H-J , Figure 5—source data 4 ) . More importantly , divergent pattern-specific RBPs targeted significantly higher proportions of effector genes than matched and convergent pattern-specific RBPs ( Figure 5E ) . These results suggested the potential causality between RBPs and effector gene profiles in divergent pairs . Interestingly , genes encoding well-known sequence-specific RBP Rbfox1-encoded gene was differentially expressed in multiple neuronal clusters , like 23/25 , 27/54 , 11/30 , 13/24 , 12/16 , 2/3 , and 7/38 ( Figure 5—source data 5 ) . Rbfox1 specifically recognizes UGCAUG motifs , which are often found at 5′- and 3′-regions of introns ( Yui Jin , 2003; Vuong et al . , 2018 ) . Previous studies have elegantly revealed the importance of activity-dependent splicing regulator Rbfox1 in the transition from neuroprogenitors to neurons as well as in the interneuron subtype-specific splicing in the mouse cortex ( Zhang et al . , 2016 ) . Moreover , we calculated the frequency of individual RBPs in marking all sister pair clusters with each of three patterns and found that single RBPs were frequently present in multiple pair clusters ( 42 . 5% , 51/120 , Figure 5F ) , indicating that clusters mostly used RBPs in a combinatorial manner . Thus , our analysis revealed the importance of RBP-encoded genes in specifying diverse neuronal phenotypes , and the identification of a comprehensive list of RBP-encoded genes in our study provides a valuable resource for future investigation . This study analyzed ~65 , 000 single cells from the whole-brain , specific brain regions , neurotransmitter/neuromodulator-defined neuronal populations of the 8 dpf larval zebrafish ( Figures 1A , 2B , , 4A ) . Notably , our transcriptome analysis offered a close-to-full coverage of all cells ( ~100 , 000 ) in the zebrafish brain , providing multidimensional landscapes of TFs and post-transcriptional regulators in vertebrate whole-brain neuron classification ( Figure 6 ) . A significant focus of our analysis is the analysis of sister clusters at the termini of effector gene-based hierarchical classification ( Figure 3A ) . These sister pair clusters represent the finest molecularly defined neuronal subclasses based on effector gene profiles . Because sister clusters of the terminus pair with highly similar terminal features , it offers us an opportunity to study the determinants of their terminal feature similarity and factors that lead to their slight diversification . Our analysis of the expression profiles of regulatory genes in sister clusters including TFs and post-transcriptional regulators ( Figure 6 ) may thus help to elucidate the molecular logic underlying neuronal diversification . Specifically , our analysis identified three patterns of TFs in specifying neuronal phenotypes , ‘matched pattern’ , ‘convergent pattern’ , and ‘divergent pattern’ ( Figures 3B and 6A ) . Glutamatergic/GABAergic sister clusters with similar terminal features mostly expressed distinct TF profiles ( ‘convergent pattern’ ) , whereas neuromodulator-type sister clusters largely expressed the same TF profiles ( ‘matched pattern’ , Figure 3E ) . The results indicated that the ‘matched’ pattern supports the notion that the same set of TFs plays a crucial role in determining a given profile of neuromodulator-type terminal features , whereas the ‘convergent’ pattern suggests that even though TFs are important for subclass determination , different TFs could still converge onto the same transcriptome phenotype . Combined with earlier findings of extensive phenotypic convergence of distinct neuron types in worms and flies ( Gendrel et al . , 2016; Konstantinides et al . , 2018; Hobert and Kratsios , 2019 ) , our findings in zebrafish strongly suggest that phenotypic convergence is highly conserved from invertebrates to vertebrates . More interestingly , we identified 10 terminus clusters exhibiting different transcriptome patterns but similar TF profiles ( ‘divergent pattern’ ) . Thus , these clusters are likely determined by factors other than TFs ( Figure 3F–G ) . Our further analyses revealed the potential importance of the expression pattern of post-transcriptional regulators , particularly RBPs , in marking these neuron clusters ( Figure 5A ) . For instance , upf3a ( the marker distinguishing subclasses of the pairs 23/25 , 12/16 , and 60/62 , Figure 5—source data 5 ) is known to regulate mRNA stability via nonsense-mediated RNA decay ( Shum et al . , 2016 ) , rbfox1/rbfox3a ( the marker distinguishing subclasses of the pair 27/54 , 13/24 , 11/30 , 23/25 , 12/16 , 2/3 , and 7/38 , Figure 5—source data 5 ) could mediate cell-type-specific splicing in cortical interneurons , assembly of axon initial segment , and synaptic transmission ( Jacko et al . , 2018; Vuong et al . , 2018; Wamsley et al . , 2018 ) . In addition to post-transcriptional regulators , other factors could also be expected to involve neuron diversification , such as epigenetic regulators , translational regulators , protein stability , which are interesting to explore in future studies . Together , our findings suggest that combinatorial TFs and post-transcriptional regulators could work in concert to determine neuronal types in the larval zebrafish brain ( Figure 6 ) . Regional identity is an essential factor for classifying brain cells . In the larval zebrafish brain , our results showed that neurons , qRG , and neuronal progenitors exhibited prominent regional characteristics ( Figure 1C ) . In the hierarchical clustering and population-level statistical analysis , similar pair clusters from the same region or different regions occur in nearly equal probabilities ( Figure 3—figure supplement 2A ) . Considering the fact that neuronal phenotypes from the same region could exhibit common regional identities , it was surprising to observe that such a higher proportion of sister subclasses with similar transcriptomic phenotypes could arise from two different brain regions . Putting this finding into the context of neuron-type evolution , it raises the possibility that different brain regions independently give rise to similar neurotransmitter phenotypes through different TF programs . Alternatively , as brain regions are functionally diversified during the evolution , these highly similar neuronal clusters of different brain regions possibly derive from ancient building blocks , which become divergent through the evolutionary acquisition of different combinatorial TF codes . scRNA-seq resolves the transcriptomes of brain cells at a given time point ( Erhard et al . , 2019 ) . On the other hand , gene expression is highly dynamic and sensitive to spontaneous or stimulus-evoked neuronal activities ( West et al . , 2002; Kim et al . , 2010; Yap and Greenberg , 2018 ) , circadian rhythm ( Panda et al . , 2002 , Takahashi , 2017 ) , as well as other extrinsic and intrinsic factors . Thus , gene profiles of individual neuron subclasses could represent dynamic cellular states at a given time rather than a static subclass ( Wu et al . , 2016 , Ofengeim et al . , 2017 ) . This issue could be resolved by comparing transcriptome profiles of cell samples from different animals or from different developmental stages . In this study , the 68 cell subclasses that we have identified for the larval zebrafish ( 8 dpf ) were confirmed in another fish brain of the same age and were largely recapitulated by cell subclasses identified from the juvenile zebrafish ( 23–25 dpf , Figure 1E and Figure 1—figure supplement 2F; Raj et al . , 2018 ) . Thus , transcriptome-defined subclasses in our study indeed largely represent stable neuronal subclasses . An alternative demonstration of stable neuronal subclasses is to perform genetic labeling of the brain using promoters of marker genes for transcriptome-defined subclasses . Reproducible labeling of the same morphological subclasses , as demonstrated in our morphological studies of tectal neurons expressing specific TFs , also supports that we have identified stable neuronal subclasses ( Figure 4D–E ) . Identification of neuronal subclasses in the present study paves the way for understanding neuronal diversity . Future studies of temporal changes of single-cell transcriptome profiles in the whole brain could provide important insights into the dynamic changes of the brain . For all experiments in this study , zebrafish were maintained , mated , and raised at 28°C according to standard protocols . Animals were staged according to dpf . Animal procedures performed in this study were approved by the Animal Use Committee of Institute of Neuroscience , Chinese Academy of Sciences ( NA-045–2019 ) . Transgenic lines used in this study include: Tg ( glyT2:GFP ) ( McLean et al . , 2007 ) , Tg ( vglut2a: loxp-DsRed-loxp-GFP ) ( Satou et al . , 2012 ) , Tg ( Etvmat2:GFP ) ( Wen et al . , 2008 ) , Tg ( gad1b:EGFP ) ( Wang et al . , 2020 ) . scRNA-seq data has been deposited on BIG ( CRA002361 ) : https://ngdc . cncb . ac . cn/gsa/browse/CRA002361 . All the code for the data analysis is available .	The brain harbors an astounding diversity of interconnected cells . Each cell contains the same basic set of genetic instructions , but only a fraction of the genome is used in each cell to assemble proteins . This selective gene expression gives rise to each cell’s characteristic properties , such as their shape and location , or whether they can activate or inhibit neighbouring cells . How these defining features are encoded on a genetic level and selectively activated in cells to produce such diversity in the brain is not fully understood . One way to study gene expression in single cells involves profiling the transcriptome , the full range of intermediary RNA molecules a cell produces from its genes to make proteins . Zhang et al . used transcriptome profiling to better understand how thousands of regulatory genes encoding regulatory proteins called transcription factors create different types of neurons in the zebrafish brain , which is similar to but much simpler than the human brain . To do so , they analysed transcriptome data extracted from cell populations located in specific brain regions and displaying different properties . Zhang et al . identified distinct clusters of neurons in the larval zebrafish brain . Mathematical models then analysed the transcriptome profiles of these neuronal clusters with characteristic features . They revealed that neurons with similar characteristics did not necessarily share the same transcription factors . In other words , distinct sets of transcription factors gave rise to the same types of cells . Zhang et al . described this observation as a ‘convergent’ pattern . On the contrary , some neurons with dissimilar features expressed the same sorts of transcription factors , suggesting a ‘divergent’ developmental pattern also exists . In summary , this work sheds light on variable gene expression patterns akin to design principles that shape neuronal diversity in the brain . It gives a new appreciation of how neuronal subtypes result from a complex set of regulatory factors controlling gene expression .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2021	The landscape of regulatory genes in brain-wide neuronal phenotypes of a vertebrate brain
Locomotion of tetrapods on land adapted to different environments and needs resulting in a variety of different gait styles . However , comparative analyses reveal common principles of limb movement control . Here , we report that a kinematic synergy involving the planar covariation of limb segment motion holds in 54 different animal species ( 10 birds and 44 mammals ) , despite large differences in body size , mass ( ranging from 30 g to 4 tonnes ) , limb configuration , and amplitude of movements . This kinematic synergy lies at the interface between the neural command signals output by locomotor pattern generators , the mechanics of the body center of mass and the external environment , and it may represent one neuromechanical principle conserved in evolution to save mechanical energy . Terrestrial locomotion of animals has evolved in vastly different designs adapted to the specific habitat of each species ( Hildebrand , 1976; Grillner , 1981 ) . Anatomically , tetrapods may differ in the number of limbs used for locomotion ( bipedal versus quadrupedal ) , limb length , shape , and mass . Functionally , locomotor styles may differ in terms of limb posture ( more flexed or more extended ) , duty factor ( percent of stride interval when each hind foot is on the ground ) , and for quadrupeds diagonality ( percent of stride interval that a forefoot lags behind ipsilateral hind foot ) . Nevertheless , there might be general principles of organization that underlie the diversity of locomotor styles ( Blickhan and Full , 1993 ) . Following the pioneering program set up by Marey , one important research goal in the study of comparative physiology of movement is ‘to point out the laws which are common for all forms and manifestations of locomotion’ ( Marey , 1874 ) . Although this ambitious goal has not been reached yet , some general principles for terrestrial locomotion have emerged that apply to a wide range of animal species , mainly related to energy saving mechanisms ( Alexander , 1989; Dickinson et al . , 2000 ) on the one hand , and to the neural control of muscle activity patterns ( Lacquaniti et al . , 2013; Grillner , 2018 ) on the other hand . With regard to energy saving mechanisms , the exchange between the gravitational potential energy and the forward kinetic energy due to pendulum-like oscillations of the centre of body mass ( COM ) has been shown to apply to walking of several legged animals ( Cavagna et al . , 1977 ) . With regard to neuromuscular control , four main activity patterns output by spinal motoneurones have been described that are common to different mammals and birds ( Dominici et al . , 2011; Wenger et al . , 2016 ) . They contribute to the different phases of gait cycle , that is limb extension at foot touch-down , body-weight support during stance , limb lift-off , and swing ( Ting et al . , 2015 ) . Also the architecture of the central pattern generators is highly conserved across animal species ( Kiehn , 2016; Grillner , 2018 ) . However , there is a gap in trying to relate the neural command signals to the mechanics of the COM . The COM is a virtual point in the trunk that shifts in space depending on the instantaneous configuration of the body . Its position is determined by the combined motion of the limb segments , as well as by trunk deformations . Therefore , in order to control the position of COM , the central pattern generators for locomotion must coordinate the motion of the limb segments ( Lacquaniti et al . , 1999; Lacquaniti et al . , 2002 ) . Kinematic coordination of limb segments can be described by statistical methods such as principal component analysis ( PCA ) , which projects movements onto a low-dimensional space thereby helping to detect invariant properties of coordination ( Daffertshofer et al . , 2004 ) . Based on this approach , one law of inter-segmental coordination has been described in human locomotion , which involves the planar covariation of the temporal changes of the elevation angles of the lower limbs ( Borghese et al . , 1996 ) . Specifically , limb segment rotations covary so that the three-dimensional ( 3D ) trajectory of the elevation angles lies close to a plane . The findings pertaining to the planar law are very reproducible ( Bianchi et al . , 1998b; Bianchi et al . , 1998a; Ivanenko et al . , 2002; Ivanenko et al . , 2007; Ivanenko et al . , 2008 ) and have been replicated in several laboratories ( e . g . Hicheur et al . , 2006; Noble and Prentice , 2008; Barliya et al . , 2009; Barliya et al . , 2013; Hallemans and Aerts , 2009; Maclellan and McFadyen , 2010; Leurs et al . , 2012; Wang et al . , 2013; Aprigliano et al . , 2016 ) . A planar law applies to both walking and running , as well as to other gait modes ( Grasso et al . , 2000; Ivanenko et al . , 2007 ) . It distinguishes between different developmental stages of human walk ( Cheron et al . , 2001b; Cheron et al . , 2001a; Ivanenko et al . , 2004; Dominici et al . , 2011 ) , as well as between normal and pathological gait ( Grasso et al . , 2004b; Laroche et al . , 2007; Leurs et al . , 2012; Degelaen et al . , 2013; Cappellini et al . , 2016; Ishikawa et al . , 2017; Wallard et al . , 2018 ) . Importantly , it has been shown that the planar covariation in humans is not a trivial consequence of the geometrical and kinematic relationships between different limb segments ( Ivanenko et al . , 2008 ) . Thus , newly walking toddlers lack an adult-like limb segment planar covariation , and children acquire it with walking experience ( Cheron et al . , 2001b; Dominici et al . , 2010 ) . In adults , the planar covariation can be violated in some conditions ( e . g . when stooping and grasping an object during walking ) or can collapse in a simple linear relationship in other conditions ( e . g . during stepping in place , Ivanenko et al . , 2008 ) . Also , spinal cord injured patients often lack the planar covariation ( Grasso et al . , 2004a ) . The functional relevance of the synergic control of segmental motions lies in a reduction of the degrees of angular motion to two , thus matching the degrees of freedom of linear motion of the COM in the sagittal plane . In fact , a significant correlation has been found between the specific orientation of the plane , or equivalently the phase shift between the angular motion at the shank and foot , and the net mechanical power output at the COM at different speeds of walking ( Bianchi et al . , 1998a; Bianchi et al . , 1998b ) . A consistent reduction of the degrees of angular motion has also been observed when additional segments are included in a principal component analysis of motion in the sagittal plane ( Mah et al . , 1994; Borghese et al . , 1996; Daffertshofer et al . , 2004; Wang et al . , 2013; Dewolf et al . , 2018 ) . How general across animal species is the planar law of intersegmental coordination ? Does it happen to be a basic principle of terrestrial locomotion ? One might expect this to be the case , given the potential connection linking the neuromuscular control patterns , the kinematic synergy , and energy saving at the COM . Initial evidence for the application of the planar law beyond the human species comes from the observation of a similar law in bipedal walking of Japanese macaques ( Ogihara et al . , 2012 ) and common quails ( Ogihara et al . , 2014 ) , as well as in quadrupedal walking Rhesus monkeys ( Courtine et al . , 2005 ) , dogs ( Catavitello et al . , 2015 ) and cats ( Shen and Poppele , 1995; Lacquaniti et al . , 1999 ) . However , apart from the above-mentioned studies , the planar law has received little attention so far in the context of comparative studies of animal locomotion . Our aim here was to look into the synergic control of segmental motions during terrestrial walking in a large set of mammals and birds . It is worth stressing that considering the large variety of body size , posture , limb configuration and segment proportions , one might not expect the same inter-segmental coordination across different animal species . Using a computational approach similar to that proposed by Gatesy and Pollard ( 2011 ) , Figure 1A illustrates the examples of permissible combinations of elevation angles assuming a fixed endpoint and constant hip ( or shoulder ) height during stance ( Notice that the results of Figure 1 do not change substantially if one relaxes the constraint of a constant hip ( or shoulder ) height during stance , since the vertical excursion of these proximal joints is relatively small during walking . Thus , the vertical oscillations due to the pendulum-like behavior of the limb during stance are 9 . 6±3 . 8% [mean±SD across all animals] of HL length for hip height , while the corresponding oscillations of shoulder height are 11 . 4±4 . 3% of FL length ) . Notice that the potential range of segment motions varies substantially across animals . For instance , the segment that can rotate the most is the foot for the mandrill while it is the shank for the avocet ( see the relative size and shape of the ellipsoids in Figure 1A ) . Thus , the height of the hip ( or shoulder ) above ground and the limb segment proportions impose very different constraints on the range of angular motion across animals . Furthermore , the same planar covariation law may not apply to the forelimb ( FL ) angles and to the hindlimb ( HL ) angles of the same animal , nor may it apply to different animals . For instance , transfer of FL angles to HL angles in the cheetah results in aberrant trunk deformations , while transfer of HL angles of the camel to HL angles of the flamingo fails to predict a realistic hip height of the ipsilateral and contralateral limbs ( Figure 1B ) . We examined the basic kinematic patterns of limb motion in several animal species , belonging to diverse taxonomic groups of birds and therian mammals . Specifically , we studied 54 different animal species belonging to 2 classes and 18 orders , including 10 species of birds and 44 species of mammals , whose body mass spanned 5 orders of magnitude . Most recordings involved free walking in a natural environment . The animals we analyzed varied widely in size and mass , from the mouse ( typical mass 0 . 03 kg ) to the elephant ( about 4000 kg , Table 1 ) . General gait parameters are reported in Figure 2—figure supplement 2 . The stride duration ranged between 0 . 3 and 3 s , the longest in the hippopotamus ( 3 s ) , and the shortest in the mouse ( 0 . 3 s ) ( Figure 2—figure supplement 2D ) . The mean trunk orientation during walking corresponded to the most erect posture in humans ( 93° relative to the horizontal ) and gibbons ( 76° ) , less vertical posture in birds ( ~10–45° ) , and nearly horizontal trunk in quadrupeds ( 8 ± 7° ) ( Figure 2—figure supplement 2E ) . Animals were free to move spontaneously at their natural speed and each limb was on the ground for more than a half of the gait cycle ( Figure 2—figure supplement 2C ) , as expected for walking gait . On average , the contacts of hindlimbs with the ground were shorter than those of forelimbs by ~10% of the gait cycle in quadrupeds ( 69 ± 5% for HL and 74 ± 4% for FL , n = 35 , p<0 . 001 , paired t-test ) , as Hildebrand ( Hildebrand , 1976 ) also reported for some animals . Touchdown events of the homologous limb pairs ( i . e . left and right HL , or left and right FL ) were almost equally spaced in time , around 50% of the gait cycle ( between left and right limbs ) , so that all recorded gaits were roughly symmetrical ( Hildebrand , 1976; Cartmill et al . , 2002; Abourachid , 2003; Frigon , 2017 ) . Most quadrupedal animals adopted the lateral gait pattern ( i . e . when the HL touchdown is followed by the ipsilateral FL touchdown , tFL < tFLcontr , Figure 2—figure supplement 2B ) , consistent with the literature ( Miller et al . , 1975; Abourachid , 2003; Schmitt , 2003; Righetti et al . , 2015 ) , while primates showed the diagonal sequence ( even though several primate species can also use a lateral sequence ) ( Hildebrand , 1976; Cartmill et al . , 2002; Frigon , 2017 ) . The gait of some quadrupeds can be defined as lateral sequence-diagonal couplets , since footfalls of contralateral limbs were almost synchronous ( e . g . for the hippopotamus tFL = 41% , tFLcontr = 91% ) , so that their gait was almost diagonal even though with a lateral sequence . In sum , a wide range of recorded animals demonstrated significant differences in the stride duration , trunk orientation , limb lengths and bilateral footfall patterns ( Figure 2—figure supplement 2 ) . There were also differences in the range of angular limb segment movements between animals and between HL and FL in quadrupeds , which will be reported in the following section . Figure 2 illustrates the range of angular motion ( ROM ) of HL and FL segment elevation angles . For mammals , the ROMs of thigh , shank , and foot elevation angles were on average 38 ± 22° , 62 ± 16° , 71 ± 23° , respectively , and for the scapula , upper arm , lower arm and hand they were 29 ± 13° , 50 ± 22° , 75 ± 20° , 96 ± 33° , respectively . For birds , the ROM of the thigh segment was smaller than in mammals ( on average by ~27° , p<0 . 00001 , unpaired t-test ) , while it was larger for the shank ( by ~16° , p=0 . 001 ) and it did not differ significantly for the foot ( p=0 . 64 ) ( Figure 2 ) . Figure 2—figure supplement 2F illustrates the ranges of linear limb movements . For all animals , horizontal limb endpoint excursions were significantly larger than the vertical ones , and for some animals they were relatively small ( e . g . for artiodactyls ) while for others ( e . g . for carnivores , empty squares in Figure 2—figure supplement 2F ) they exceeded the limb length ( 1L ) . In general , for quadrupeds , horizontal limb excursions were significantly greater for FL ( 0 . 95L ) than for HL ( 0 . 87L ) . The animals walked at their natural speeds and some differences in the normalized limb endpoint excursions ( Figure 2—figure supplement 2F ) could be related to variations in the walking speed across animals . However , the reported difference between horizontal HL and FL endpoint excursions could not be related to walking speed since we compared them for the same animals . These inter-limb differences were expected , given the corresponding differences in the relative stance phase duration between the limbs ( Figure 2—figure supplement 2C ) . We used serially homologous HL and FL segments and models for comparing the kinematics of the HL and FL , starting from the distal segment: foot-hand , shank-lower arm , and thigh-upper arm . However , the scapula segment also undergoes significant rotations in the sagittal plane ( Figure 2 ) . While PCA can be applied also in four dimensions for FL , using a tri-segmental model makes it easier to compare the kinematic synergies between FL and HL . Therefore , for FL we used two separate tri-segmental models ( Fischer and Blickhan , 2006 ) : FLlow – 'upper arm–lower arm–hand’ and FLupp – ‘scapula–upper arm–lower arm’ ( Figure 3A right panel ) . We found that the planar covariation law of limb segment motions holds for walking in all recorded animal species , despite significant differences in body size , limb segment configuration and gait parameters ( Figure 2 and Figure 2—figure supplement 2 ) . Figure 3 shows the results of PCA applied to the kinematic data of different animals . Figure 3A shows examples of the ensemble-averaged elevation angles ( across strides ) as a function of the normalized gait cycle ( upper panels ) and a corresponding three-dimensional view of these angles ( lower panels ) for one bipedal ( avocet ) and one quadrupedal ( elephant ) animal . The foot at HL and the hand at FL touchdown corresponds to the top of the loops in the lower panels . The trajectories progress in the counter clockwise direction of the loops . The grids represent the best fitting planes , defined by the first two eigenvectors of the PCA . Note the differences in the orientation of the covariation plane at HL between the two animals , and between HL and FL for the elephant . Note also a roughly similar orientation of the covariation plane between the two models of FL ( FLupp and FLlow , Figure 3A ) . Planarity of the data was quantified by computing the percentage of variance accounted for by the third eigenvector ( PV3 ) of the data covariation matrix: the closer is PV3 to 0 , the smaller the deviation from planarity . The results showed that the planarity was obeyed by all species ( PV3 ranged from 0 . 04% to 5 . 3% across all limbs and animals ) ( Figure 3B ) . We examined mainly the intersegmental coordination of the elevation angles , rather than that of the relative joint angles ( so called anatomical angles ) , because the former capture the overall limb configuration in external space . Indeed , the elevation angles identify the orientation of each segment relative to the direction of gravity ( vertical direction ) . Moreover , the time course of the anatomical angles of flexion-extension in human locomotion is more variable across subjects and trials than that of the elevation angles , and the planarity of the anatomical angles trajectories is weaker ( Borghese et al . , 1996; Barliya et al . , 2009 ) . In our recorded animals , when we applied the PCA to the anatomical angles ( hip , knee and ankle for HL , and shoulder , elbow and wrist for FL ) , the planarity indexes ( PV3 ) were: 2 . 7 ± 2 . 3% for HL angles ( ranging from 0 . 02% in guinea fowl to 8 . 0% in porcupine ) and 4 . 0 ± 2 . 5% for FL angles ( ranging from 0 . 7% in ox to 13 . 9% in porcupine ) . Therefore , although also the anatomical angles trajectories tend to be constrained close to one plane , the planar coordination of the elevation angles is stronger and less variable , in agreement with what was previously reported for human walking ( Borghese et al . , 1996; Barliya et al . , 2009 ) . It is interesting to note that the shape of the 3D trajectories generated by the three elevation angular waveforms differed across animals and limbs ( Figure 3—figure supplement 2 ) . All trajectories presented a closed loop , as expected by the cyclic nature of the gait , and were elongated in the direction of PC1 given that PC1 explains the largest fraction of variance ( Figure 3—figure supplement 2A ) . We quantified the relative width of the loop by computing the relative amplitude of PC2 with respect to PC1 ( Figure 3—figure supplement 2B ) . The width of the loop in birds ( 0 . 54 ± 0 . 11 ) was significantly larger than for HL in mammals ( 0 . 26 ± 0 . 06 ) ( p=0 . 00002 , unpaired t-test ) . In addition , there was a difference between HL and FLlow ( 0 . 49 ± 0 . 11 ) and between HL and FLupp ( 0 . 31 ± 0 . 08 ) gait loops in quadrupeds ( p<0 . 005 , paired t-test ) . There were no significant differences between the loops of birds and FLlow of mammals ( p=0 . 25 ) . While planarity of the 3D loops held for all animals ( Figure 3B ) , the orientation of the covariation plane differed , due to different phase relationships between elevation angles . The third eigenvector ( u3 ) of the covariance matrix is orthogonal to the best fitting plane and characterizes its orientation . Figure 3—figure supplement 1 illustrates the direction cosines of the normal to the plane ( i . e . the dot products of u3 with the unit vectors along each of the three axes ) for all animals . One can notice a greater scatter of the u3 components of HL across mammals than the corresponding components in birds or those of FL in mammals . One way to visualize the u3 vector in a 3D space for all animals is to plot it on a sphere as shown in Figure 4 . Note a higher u3 dispersion for HL ( Figure 4A ) than for FL ( Figure 4B ) , consistent with the greater variability of u3 components in HL noticed above . Using the empirical shape criterion ( Fisher et al . , 1993 ) , we distinguished the girdle distribution ( a type of distribution of directions with a concentration about a given plane ) of the eigenvectors in mammals HL from the clustered distribution of birds and of mammals FL . In particular , this criterion revealed that the four groups ( birds HL , mammals HL , mammals FLupp , mammals FLlow ) had concentration parameters significantly different ( p<0 . 0001 ) and belonged to different distribution , respectively a von-Mises Fisher distribution with concentration parameters k = 106 . 9 , a Kent distribution ( girdle-like shape ) with k = 5 . 7 , and two other von Mises distribution with k = 20 . 6 , k = 33 . 7 . High values of the concentration parameters showed that the population mean directions were different across the four groups ( p<0 . 0001 ) . Furthermore , it is worth noting that the vectors were not dispersed randomly on a sphere , but tended to lay on the plane defined by the mean u1 vector across animals ( since u1 was roughly the same; overall , it deviated from the mean u1 by 2 . 4° [spherical standard error] for HL , by 2 . 7° for FLupp and by 2 . 6° for FLlow ) . The lower panels of each box in Figure 4A and B illustrate the rotation of the u3 vectors on this plane for each animal and limb . This plane is perpendicular to the averaged u1 vector across animals , and α defines the angle of rotation of each u3 on the plane . In sum , the full limb behavior in all walking animals can be expressed as two principal components identifying a given covariation plane ( Figure 3 ) . While the orientation of the covariation plane of the FL appears fairly conserved across species ( Figure 4B ) , the covariation plane of HL varies across mammalian species by a rotation ( α-angle ) about a well-defined axis ( Figure 4A ) . We searched for the presence of a phylogenetic signal in the wide scatter of α-values of rotation for HL across animal species , in order to frame the data scatter in an evolutionary context . Although the K index ( Blomberg et al . , 2003 , see Materials and methods ) we used for the presence of a phylogenetic signal in the α-angle for HL was statistically significant , its value was rather low ( K = 0 . 10 , n = 54 , p=0 . 04 ) ( Figure 4—figure supplement 1 ) , suggesting that the pattern of α-angles distribution is hardly dependent on phylogenetic relatedness of the species considered . Such a pattern may occur when close relatives are less similar than distant ones . To search for biomechanical correlates of inter-species differences in the orientation of the covariation plane , we performed a linear regression between the value of α-angle ( Figure 4A , B ) and the ROM of limb segment elevation angles , the phase shift between elevation angles ( i . e . timing of their minima ) , and the ratio between limb segment lengths . The rationale for using these biomechanical parameters was that the animals differed significantly in terms of limb proportions ( Figure 2—figure supplement 1 ) , ROMs ( Figure 2 ) , and temporal sequence of minima of elevation angles ( Figure 5A , bottom panels ) . In particular , the phase shifts between elevation angles waveforms are strongly related to the rotation of the covariation plane ( Bianchi et al . , 1998b; Barliya et al . , 2009 ) , and they can be assessed using the relative timing of the minima ( Bianchi et al . , 1998b; Catavitello et al . , 2015 ) . We found that the values of α-angle were best correlated with the phase shift Δfoot-shank ( r2 = 0 . 5 , p<0 . 0000001 , Figure 5A ) . Other relatively high ( r2≥0 . 3 ) significant correlations were: for HL , the ratios Lshank/Lfoot and Lthigh/Lfoot ( r2 = 0 . 45 and r2 = 0 . 31 , respectively , p<0 . 00004 ) and , for FLupp , ROMs of the upper and lower arms ( r2 = 0 . 40 and r2 = 0 . 32 , respectively , p<0 . 0002 ) , and the phase shift Δupper arm-scapula ( r2 = 0 . 3 , p<0 . 0003 , Figure 5A right panel ) . After controlling for a potential phylogenetic signal in the response ( and , hence , non-independence of the residuals , see Materials and methods ) , we found that the α-angle remained significantly correlated with Δfoot-shank ( r2 = 0 . 37 , p<0 . 00001 ) and Δupper arm-scapula ( r2 = 0 . 31 , p<0 . 00001 ) . For FLlow , all correlations were very weak ( r2 ~0 . 01–0 . 14 ) , and involved much smaller rotations of the covariation plane ( Figure 4B ) . There were also differences in the temporal sequence of minima of elevation angles between the limbs ( Figure 5B ) . Even though the timing of minima for all segments occurred roughly around the stance-to-swing transition ( since the relative stance duration was about 70% cycle , Figure 2—figure supplement 2C ) , the sequence of minima differed for HL and FL . For instance , one can notice that the distal segment ( hand ) of FL was the last to initiate the swing phase in contrast to the distal segment ( foot ) of HL ( Figure 5B ) . In Figures 4 and 5 , we reported the parameters of the inter-segmental coordination in all animal species . To obtain a general template of HL and FL angular motions for each animal species , we averaged limb segment elevation angles across strides . However , some inter-stride variability in the orientation of the covariation plane and timing of the minima of elevation angles exists . Also , there were some limitations of our measurements ( e . g . due to some variability in the walking speed across strides ) . Nevertheless , it is unlikely that the key differences across limbs and species can be accounted for or masked by inter-stride variability . We quantified the inter-stride variability in the animal species in which we recorded more than 15 strides , namely: dog , donkey and human ( Table 1 ) . First , the correlation between the averaged limb segment elevation angles and those of individual strides was high ( on average r = 0 . 98 , range 0 . 7-1 , the data for all segments and animals being pooled together ) , consistent with repeatable kinematic data across steps in animals ( Faber et al . , 2002; Kim et al . , 2008 ) . Second , the sequence of timing of minima of elevation angles showed systematic features as well . For instance , the distal segment ( hand ) of FL always followed the lower arm segment ( Δhand-lower arm being positive in all strides ) , in contrast to the distal ( foot ) segment of HL ( Δfoot-shank varied but was generally negative ) . Finally , the inter-stride and inter-individual variability in the orientation of the covariation plane ( u3 vector ) was relatively small in comparison with the differences across animals ( Figure 4 ) . For instance , the angular standard deviation of the parameter α ( the angle of rotation of u3 ) across strides was 11° in dogs , 12° in donkeys and 6° in humans ( the strides of all animals for each animal species being pooled together ) , and across animals it was 6° in dogs ( n = 6 ) , and 5° in humans ( n = 6 ) , while the differences in α across animal species were much larger ( ~180° , Figure 4 ) . During forward progression , HL and FL segments oscillate back and forth with specific phasing relative to the footfall pattern ( Figure 3A ) . We confirmed the validity of the planar covariation previously reported in humans ( Bianchi et al . , 1998b ) , macaques ( Courtine et al . , 2005; Ogihara et al . , 2012 ) , birds ( quails , Ogihara et al . , 2014 ) , and dogs ( Catavitello et al . , 2015 ) , and extended it to a large set of other animal species . Even though different recording systems were used in these previous studies , the planarity index ( PV3 ~1–3% ) and the orientation of the covariation plane ( u3 vector ) for the bird ( Ogihara et al . , 2014 ) , human ( Bianchi et al . , 1998a ) and dog ( Catavitello et al . , 2015 ) were similar to those reported in the current study ( Figure 3 ) , confirming the reliability of our kinematic recordings and suggesting that each animal adopted its own pattern of the inter-segmental coordination . Since we recorded animals walking at their preferred speeds , some variability in the covariation plane orientation may be caused by variations in speed across strides . However , it is unlikely the relative invariance in the orientation at the forelimbs ( which in theory might parallel changes in speed ) and in the sequence of timing of minima of elevation angles suggests that the key differences across the limbs and species is accounted for or masked by inter-stride variability , consistent with repeatable kinematic data across steps in animals ( Faber et al . , 2002; Kim et al . , 2008 ) . Also , we verified the inter-individual variability in a few species ( dogs and humans , Table 1 ) , and the angular standard deviation in the covariation plane orientation ( ~5–6° ) was much smaller than the differences in α ( the angle of rotation of u3 ) across all animal species ( ~180° , Figure 4 ) , suggesting that the data presented in Figure 4 are representative for each animal species . One should also consider that the effect of speed may be animal- or gait-dependent . For instance , in human bipedal walking there is some rotation ( although relatively small ) of the covariation plane with speed ( Bianchi et al . , 1998a ) , while in human crawling the orientation of the covariation plane does not depend on speed ( MacLellan et al . , 2017 ) . To obtain a general template of limb segment motion of animals walking at their preferred speeds , we analyzed averaged angles ( as in other previous studies , e . g . , Fischer et al . , 2002 ) . Further studies may reveal a nuanced dependence of planar covariation parameters on walking speed or gait in different animals . Specific limb proportions may play an essential role in the kinematics and energetics of walking ( Leurs et al . , 2011 ) , and they have an impact on a locomotor body schema used for controlling step length ( Ivanenko et al . , 2011 ) . For instance , in humans elongation of the shank segment relative to the thigh segment ( either surgically or using specially designed stilts ) affects the amplitude of distal vs . proximal segment oscillations ( Dominici et al . , 2009 ) and the optimal step length and walking speed ( Leurs et al . , 2011 ) . In line of principle , species with different limb proportions are free to employ identical angular movements , but there are biomechanical constraints of the articulated chains on the end effector positions ( Gatesy and Pollard , 2011 ) and on the possibility of transferring angular changes between disproportionate limbs ( Figure 1B , right panel ) . Limb segment proportions varied significantly across species ( Figure 2—figure supplement 1 ) . For instance , the Lshank/Lfoot and Lthigh/Lfoot ratios showed 10-fold differences in our sample of animals ( range 0 . 5–5 ) . Therefore , it is important to stress that the planar covariation holds for animals with very different limb segment configurations . While biomechanics contributes to the planar law of inter-segmental coordination ( e . g . we found that the u3 vector rotation correlated with Lshank/Lfoot and Lthigh/Lfoot ) , the orientation of the covariation plane reflects specific phase relationships in the control of segment motions ( Bianchi et al . , 1998b; Lacquaniti et al . , 2002; Ivanenko et al . , 2008 ) . For instance , birds form a group of animals with a compact orientation of the u3 vector close to the thigh axis ( Figure 4A , see also Ogihara et al . , 2014 ) while primates show rather variable u3 orientation ( Figure 3—figure supplement 1 ) . Also , birds show characteristically wide gait loops , while for other animals the loops are much narrower for HL ( Figure 3—figure supplement 2 ) . Finally , for quadrupedal animals , the orientation of the covariation plane is noticeably different for HL and FL ( Figures 3 and 4 ) . The latter finding represents a particularly interesting phenomenon that may shed further light on the functional difference between the limbs and their control . HL and FL kinematics are characterized by limb-specific differences in the orientation of the covariation plane ( Figures 3 and 4 ) , the width of the gait loop ( the FLlow loop was wider than the HL loop , Figure 3—figure supplement 2 ) and the amplitude ( Figure 2 ) and phase ( Figure 5B ) of angular motion of distal segments . These results confirm previous observations about different orientation of the HL and FL covariation planes in Rhesus monkey ( Courtine et al . , 2005 ) , dog ( Catavitello et al . , 2015 ) and human crawling ( MacLellan et al . , 2017 ) , and point to the differential control of FL and HL segments in a wide range of mammals . The distinctive orientation of FL and HL segments ( the elbow is facing posteriorly and the knee joint anteriorly ) , a stronger push-off function of HL ( e . g . during jumping ) , and the differences in the leading segment ( as assessed by temporal sequence of minima in the elevation angles around the stance-to-swing transition , Figure 5 ) may impose specific phase-relationships of FL and HL segment oscillations . Finally , neurophysiological differences in the neurotransmitter systems of FL versus HL spinal locomotor controllers ( Gerasimenko et al . , 2009 ) , a strong asymmetry of projections from spinal controllers on neurons for FL versus HL areas of the motor cortex ( Zelenin et al . , 2011 ) , and limb-specific features in the organization and coupling between FL and HL spinal controllers ( Shik and Orlovsky , 1965; Miller et al . , 1975; Yamaguchi , 1986 ) point to limb-specific organization of central pattern generators , with propriospinal linkages facilitating the coordination between FL and HL ( Frigon , 2017 ) . The temporal structure of FL and HL muscle activation patterns is limb-specific too , as is the orientation of the covariation plane ( e . g . in the dog , Catavitello et al . , 2015 ) . Furthermore , the orientation of the covariation plane ( normal to the plane , u3 ) varies more for HL than for FL across animals ( Figure 4 ) . This finding is also compatible with larger kinematic changes in HL movements reported in previous studies . For instance , Fischer et al . ( 2002 ) reported variable lift-off configuration of HL with respect to FL , when comparing different gaits of the same animal . Thus , main kinematic adaptations seem to occur in hindlimbs rather than in forelimbs both across animals ( Figures 4 and 5 ) and across gaits ( Fischer et al . , 2002 ) . This implies considerable adaptability or flexibility in the phase relationships of HL segment motion , which will be considered in the following section . We also searched for the presence of a phylogenetic signal in the scatter of α-values of rotation for HL across animal species , and we found only a weak one . Indeed , the α-values of close relatives were not more similar between each other than to the values of distant ones ( Figure 4—figure supplement 1 ) . This suggests that the planar covariation is a feature that has arisen independently several times during evolutionary history ( Ogihara et al . , 2014 ) . A convergent evolution of this kinematic synergy may be due to both adaptation and constraints acting similarly in distantly related species . Adaptation would arise due to the advantage of a kinematic control law lying at the interface between neural commands and environment . Constraints would depend on the inherent biomechanical coupling between different limb segments . The planar covariation law may emerge from the coupling of neural oscillators with limb mechanical oscillators ( Lacquaniti et al . , 2002; Lacquaniti et al . , 2012 ) , by adjusting the phase of unit burst generators for each joint , segment or groups of muscles ( Grillner , 1981; Kiehn , 2016 ) . The basic mechanism of rhythmic movements is a phase control of muscle activity . In particular , myoelectric signal analysis demonstrated a burst-like temporal organization of basic muscle activation patterns shared by many animal species ( Giszter et al . , 2010; Dominici et al . , 2011; Lacquaniti et al . , 2013 ) , consistent with the existence of a rhythm-generating layer or ‘time-keeping function’ of the central pattern generator for locomotion ( Prentice et al . , 1995; McCrea and Rybak , 2008 ) . Because the activation patterns are pulsatile , muscle activations intervene only during limited time epochs at specific phases of the gait cycle to re-excite the intrinsic oscillations of the system when energy is lost ( Lacquaniti et al . , 1999; Lacquaniti et al . , 2012 ) . This represents a fundamental energy-saving principle of control . The dynamic behavior of the musculo-skeletal system can be modeled through a linear combination of these basic muscle patterns , activated sequentially at touch-down , body-weight support , limb lift-off , and swing ( Lacquaniti et al . , 2012 ) . The specific orientation of the planar covariation is related to the timing of basic muscle activation patterns . In humans , changes in the orientation of the covariation plane with walking speed ( Bianchi et al . , 1998b ) or across different gaits ( Ivanenko et al . , 2007 ) are associated with changes in the timing of basic muscle activation patterns ( Ivanenko et al . , 2004; Cappellini et al . , 2006 ) . In dogs , the phase-coupling between the elevation angles differs systematically between HL and FL ( Figure 4 ) , just as the phase-coupling of the muscle activation patterns ( Catavitello et al . , 2015 ) . Thus , although it is often assumed that central pattern generators control patterns of muscle activity , an equally plausible hypothesis is that they control patterns of limb segment motion ( Lacquaniti et al . , 1999; Lacquaniti et al . , 2002 ) , since the phase relationships between them are inherently inter-related . The full limb behavior can be expressed as the two degrees-of-freedom planar motion for each animal , plus the rotation of the plane about a defined axis ( Figure 4 , see also Ivanenko et al . , 2007 ) . An analytical formulation of the law of inter-segmental coordination in human walking was introduced by Barliya et al . ( 2009 ) , using a mathematical model that represents the rotations of the elevation angles in terms of simple harmonic oscillators with appropriate phase shifts between them . This model can be generalized to the locomotion of other animal species . We found the highest correlation between the rotation of the covariation plane ( α-angle ) and Δfoot-shank ( Figure 5 ) . Therefore , the phase shift between foot and shank segments represents an important parametric tuning of the covariation plane rotation to adapt to animal-specific locomotor patterns ( Figure 4A ) , walking speed ( Bianchi et al . , 1998b ) , gait ( Ivanenko et al . , 2007 ) or walking on different support surfaces ( Dominici et al . , 2010 ) . Figure 6 provides schematically the conceptual framework for modelling the foot-shank phase-shift , while approximating the three segment elevation angles with sinusoidal waveforms Barliya et al . ( 2009 ) . Notice that , as predicted , changing the foot-shank phase results in a progressive rotation of the planar covariation . Critically , the rotated planes ( upper panels in Figure 6A ) closely resemble the experimental planes of different animals ( lower panels ) . Thus , by changing the phase of the foot segment waveform relative to the shank segment from 20° to −50° ( corresponding to the same range of Δfoot-shank in Figure 5A ) , the covariation plane rotates similarly to the plane rotation actually observed across animals ( Figure 6A ) . Interestingly , a similar conceptual model can be applied to account for the rotation of the covariation plane across different gaits in humans: walking , hopping , running , air-stepping , obstacle clearance , crouched walking ( Ivanenko et al . , 2007 ) . Figure 6B illustrates a superposition of u3 vectors across animals ( Figure 4A ) and across different human gaits ( indicated by colored confidence cones ) . Note a similar plane of rotation of the u3 vector across animals and across different human gaits . Therefore , limb kinematics of animal locomotion in the sagittal plane can be modeled by two principal components that determine limb segment coordinated movements ( Figures 3 , 4 and 6B ) . The orientation of the first ( u1 ) and second ( u2 ) principal component axes on the covariation plane is illustrated for selected animals in the bottom panels of Figure 6A by blue and red lines , respectively . It has previously been argued that these components may be equivalent to the length and orientation of limb axis , and define an appropriate endpoint motion for different cat postures ( Lacquaniti and Maioli , 1994 ) as well as human gaits ( Ivanenko et al . , 2007 ) . One way to illustrate the functional significance of these principal components is to plot the changes in limb kinematics resulting by a corresponding shift along u1 ( PC1 ) and u2 ( PC2 ) axes ( Lacquaniti and Maioli , 1994 ) . Examples of stick diagrams generated by such shifts are shown in Figure 6C , demonstrating that PC1 mainly reproduces the changes of limb orientation while PC2 is mainly limited to changes in limb length . We showed that the planar covariation law previously established for humans holds for the terrestrial walk of several mammals and birds . This kinematic synergy lies at the interface between two highly conserved phenomena in animal locomotion , the neural command signals output by central pattern generators on the one hand , and the mechanics of the body COM . The kinematic synergy may therefore represent one specific neuromechanical principle of control of the instantaneous position of the COM , thus contributing to net mechanical energy savings . Our study was exploratory ( although extremely laborious , see Materials and methods ) , since our results were mainly obtained under natural conditions rather than the controlled conditions of a laboratory . Nevertheless , the results we found are grounds for further , more systematic investigations into why the principle of planar covariation might be conserved across animal species . The present findings suggest a modular control organization whereby appropriate coordination of the limb segments for each animal can be reduced to two independent components . The orientation of the covariation plane remarkably differed between HL and FL and between animals , especially for HL ( Figure 4A ) . The major changes in HL plane across animals were associated with the second PC related to the limb length covariation ( u2 vector , Figure 6A ) , probably reflecting different distribution of stiffness and phase control of oscillations and thus the relative rotation of limb segments . Specific limb segment phase relationships are likely advantageous for each individual animal species , and might be the result of evolution . For instance , birds master flight techniques , perform take-off and landing manoeuvres , and accordingly they could adopt a specific yielding kinematic synergy ( Figure 3A , Figure 3—figure supplement 2 , Figure 4A ) . In quadrupeds , different biomechanical functions of HL and FL imply limb-specific couplings of neural oscillators with limb mechanical oscillators . In sum , our study provides an integrative view on the dynamic template of limb segment motion across a wide range of animals and prompts further work to understand functional and evolutionary advantages of specific planar covariation patterns adopted by different species . For most species , a few different animals were recorded ( Table 1 ) . The recordings were made in different locations , most of them at Falconara ( Italy ) Zoo , others at Rome or Nemi ( Italy ) Zoo . The videos of the Sika deer were made at Nara city in Japan ( where these animals are allowed to freely walk with humans ) . Human walking was recorded in the laboratory . For mice , videos came from a previously published study ( Movie S1 in Akay et al . , 2014 ) , while for six therian mammals the kinematics was recorded by Fischer et al . ( 2002 ) using cineradiography to study the contribution of limb segment angles to step length and FL and HL movements . In the latter case , published graphs of limb segment elevation angles were scanned , digitized manually and time-interpolated to fit a normalized 100 points time base for the analysis of the inter-segmental coordination . It is important to note that animals were observed walking spontaneously and at their preferred speed . Furthermore , most recordings were performed in natural outside conditions and on the terrain where each animal lives , with no laboratory stress or human handlers’ interaction . No special permission is required in Italy for non-invasive observation of animals outside laboratory settings in behavioral studies like the present one ( Italian law: DL 26/2014 ) . As for the previously published studies , the mouse ( Akay et al . , 2014 ) and six therian mammals ( Fischer et al . , 2002 ) walked on a treadmill , but the operator adjusted the treadmill speed to obtain preferred speeds of the animals . Table 1 reports the animals analyzed , their scientific name , the typical body weight reported from the literature , the speed , the Froude number ( normalized speed , see below ) , and the number of recorded strides . The recordings of animal walking were made using a Fujifilm Camera ( FinePix SL1000 , at 60 Hz , 37 species ) or a Canon camera ( EOS 550D , at 50 Hz , 10 species ) . Cameras were fixed on a tripod to limit vibrations during recordings , and were oriented roughly orthogonal to the direction of animal walk . The distance between the camera and the animals ranged between 4 and 10 m for all recordings , depending on where the animal walked with respect to the observation point of the experimenter . For humans , we recorded both overground ( with the Fujifilm camera ) walking in one subject and treadmill walking in five subjects ( at 5 km/h , using a 9-camera Vicon-612 system , Oxford , UK , sampling rate 100 Hz ) . From video recordings , we identified successful sequences of strides when the gait occurred in the sagittal plane steadily and on a straight path roughly perpendicular to the optical axis of the camera to minimize errors in 2-D kinematic analysis ( Kim et al . , 2008 ) . Only complete strides were analyzed using hindlimb touchdown as the onset . We obtained the kinematics of both the right and left side by recording locomotion in both directions ( relative to the camera ) , and the kinematic data were pooled together because both sides in walking have similar and repeatable locomotion characteristics ( Hildebrand , 1967; Alexander and Jayes , 1983 ) . The contralateral HL and FL endpoints were used only to characterize the interlimb coordination ( diagonality of gait , see below ) . The number of recorded strides varied across animals; on average , we recorded 9 ± 10 ( mean ± SD ) successful strides per animal ( 452 strides total , Table 1 ) . Once we selected the successful strides in videos , the reconstruction was performed using the Tracker software ( v . 4 . 95 ) , a free video analysis and modeling tool built on the Open Source Physics Java framework . The anatomical landmarks of the ipsilateral side ( with respect to the camera ) tracked in the reconstruction were: hip ( HIP ) , knee ( KNE ) , and ankle ( ANK ) joints , base of the external metatarsal ( or tarso-metatarsus in birds ) ( MT ) , endpoint ( end of the distal phalanx ) of the hindlimb ( HEP ) . For quadrupedal animals , we also tracked the following forelimb landmarks: shoulder ( SHO ) , elbow ( ELB ) , and wrist ( WRI ) joints , base of the external metacarpal ( MC ) , and endpoint ( end of the distal phalanx ) of the forelimb ( FEP ) ( see Figure 2—figure supplement 1A , B left panels ) . In addition , we tracked the dorsal border of the scapular spine ( SCA ) , nose ( NOS ) , and tail endpoint ( TEP ) landmarks . To characterize the interlimb coordination ( diagonality of gait ) , the contralateral hindlimb ( C_HEP ) , and forelimb ( C_FEP ) endpoints were also tracked . All anatomical landmarks were manually tracked frame by frame using a wireless touchpad with digital stylus ( Wacom Bamboo Pad CTH-300 ) , and using the skeleton model of each animal ( as derived from the literature ) for guidance ( Figure 2—figure supplement 1A , B left panels ) . In total , 33 , 510 frames were processed by a very experienced person ( author GC ) in 7227 hr of work . These kinematic data were further processed in the context of a multi-segmented bipedal ( Figure 2—figure supplement 1A , left panel ) and quadrupedal ( Figure 2—figure supplement 1B , left panel ) model . The analyses were performed using custom-made algorithms implemented in Matlab . The kinematic data were low-pass filtered using a zero-lag , fourth-order dual-pass Butterworth filter with a cutoff of 10 Hz . Next , we applied a custom model-based algorithm that uses the average segment length over all frames in each trial , and optimizes the locations of joint centres by constraining the changes in the limb segments lengths ( Catavitello et al . , 2015 ) . Since we recorded animal walking under natural conditions and we included animals that are usually difficult to train and work with ( e . g . lion , cougar , tiger ) or have very large sizes ( e . g . ostrich , addax , giraffe , elephant , hippopotamus ) , we could only apply the markerless approach to reconstruct the kinematic data ( Catavitello et al . , 2015 ) . We previously verified that this approach yields reliable results in the assessment of phase relationships between limb segment angles ( Catavitello et al . , 2015 ) . In this previous study , we compared the results of the kinematic analysis of canine locomotion obtained from a video camera with those obtained with a high-performance 3D motion-capture system ( SIMI Motion system , Unterschleissheim , Germany , sampling rate 100 Hz ) , and we found very similar characteristics of the inter-segmental coordination ( Catavitello et al . , 2015 ) . In the present study , we compared the results in five species ( goose , pigeon , guinea fowl , elephant , and cat ) with published data on elevation angles in the same or related species walking at comparable speeds . In the latter case , the data were obtained by means of high-performance 3D motion-capture systems ( Shen and Poppele , 1995; Ren et al . , 2008; Stoessel and Fischer , 2012 ) . We scanned the graphs of limb segment elevation angles published in these reports , digitized them manually and time-interpolated to fit a normalized 100 points time base for the analysis of the inter-segmental coordination . We found a good agreement between our results and those obtained in related species in the previous publications . Thus , on average the root mean square ( RMS ) difference between the angular waveforms of our study and the corresponding ones of the previous studies was 5 . 5 ± 3 . 1° , and the average correlation coefficient between angular waveforms was 0 . 94 ( obtained by pooling all segments and all steps together . Furthermore , the orientation of the covariation plane derived from both sets of studies was almost identical: the angular difference between u3 vectors was on average 0 . 7° for HL and 0 . 5° for FLlow . In addition , in the present study , we compared the results obtained from the video camera recordings of the human subject in the laboratory with those reported in the literature . At matched walking speed ( mean=1 . 6 m/s across all strides ) , the characteristics of the planar covariation of the limb segment elevation angles were identical to those reported by Bianchi et al . ( Bianchi et al . , 1998b ) ( orientation of the covariation plane [see below]: u3t=0 . 04 , u3s= -0 . 72 , u3f =0 . 68 ) . Finally , while all anatomical landmarks were manually tracked by a single person ( see above ) , we verified the inter-rater reliability for a few animals . To this end , we asked another researcher , very experienced in tracking kinematic data but unaware of the details of the present study , to track the videos of the avocet , camel , cheetah , gibbon and Sika deer . The mean RMS difference between the angular waveforms obtained by the two persons was 3 ± 3° and the mean correlation coefficient was 0 . 99 ( all elevation angles and all strides being pooled together ) . The corresponding difference in the orientation of the covariance plane ( u3 ) obtained by the two persons was low: on average 0 . 8° for HL , 2 . 0° for FLupp and 2 . 5° for FLlow . The general model of the HL and FL segments is shown in Figure 2—figure supplement 1A , B ( left panels ) . Whole limb and trunk orientations were defined from the HIP-HEP ( HL ) , SCA-FEP ( FL ) and SCA-HIP ( trunk ) segments . The hindlimb was modeled as the multi-segmented limb from HIP to HEP , consisting of the thigh , shank , foot and toes segments , while the forelimb ( from SCA to FEP ) included the scapula , upper arm , lower arm , hand , and finger segments . Whole limb and limb segment elevation angles relative to the vertical were calculated and analyzed , angles being positive when the distal marker was located anterior to the proximal marker . Mean trunk inclination was defined as the mean angle between the trunk and the horizontal reference ( Figure 2—figure supplement 2E ) . The gait cycle for each limb was defined as the time-interval between two successive maxima of the limb orientation waveform ( Catavitello et al . , 2015 ) . The stance phase ( when the foot was on the ground ) corresponded to the time window between the maximum and the following minimum of the limb orientation waveform . We considered the gait cycle defined by the ipsilateral HL and FL ( facing the camera ) . The contralateral HL and FL endpoints were used only to characterize the interlimb coordination ( diagonality of gait ) . The quadrupedal gait ( inter-limb coupling ) can be characterized by the footfall sequence ( Hildebrand , 1976 ) . To this end , the phase lag was computed as the relative timing ( tFL , tFLcontr ) of the FL cycle onset with respect to HL , and expressed as a percentage of the gait cycle ( Figure 2—figure supplement 2B ) . The lateral gait is determined when the HL touchdown is followed by the ipsilateral FL touchdown , whereas in the diagonal sequence it is followed by the contralateral FL touchdown . Limb endpoint excursion was determined separately for fore-aft and up-down ( relative to the body ) movements ( Figure 2—figure supplement 2F ) . To compare different animals , the calculated values were normalized to hindlimb length ( L ) , defined as the sum of the average lengths of thigh , shank , and foot segments over all frames in each video ( except for humans , in which case we used the thigh + shank length , as it is more commonly accepted in the literature due to the heel contact with the ground ) . To estimate walking speed V , we computed the distance covered by the hip landmark of a given animal during a stride . Since for most animals we could not measure the individual body length in meters , we approximated it using data from the literature in order to convert V from pixels/s to m/s . We also assessed the dimensionless walking speed ( Froude number , Fr ) , which is suitable for the comparison of the speed of locomotion in animals of very different size ( Alexander , 1989 ) . The Froude number is given by Fr= V2g∙L , where g is the acceleration of gravity . The estimated Froude numbers are reported in Table 1 . We used a tri-segmented limb model ( Fischer and Blickhan , 2006 ) and serially homologous HL and FL segments , starting from the distal segment: foot-hand , shank-lower arm , and thigh-upper arm . However , the scapula segment also undergoes significant rotations in the sagittal plane in most mammalian groups during locomotion . Accordingly , the tri-segmented model for HL included thigh ( HIP-KNE ) , shank ( KNE-ANK ) and foot ( ANK-MT ) interconnected segments , while for FL we used two tri-segmental models ( Fischer and Blickhan , 2006 ) : 1 ) FLupp – scapula ( SCA-SHO ) , upper arm ( SHO-ELB ) , and lower arm ( ELB-WRI ) , and 2 ) FLlow – upper arm , lower arm , and hand ( WRI-MC ) . Our study was mainly focused on a general locomotor pattern in various species . Accordingly , the waveforms of the elevation angles of the limb segments were time interpolated over individual gait cycles to fit a normalized 100-point time base , and averaged first across strides and then across animals , in order to obtain a general template of HL and FL angular motion for each animal species . Nevertheless , even though the kinematic data tend to be repeatable across consecutive strides ( Faber et al . , 2002; Kim et al . , 2008 ) , we also report the inter-stride variability for a few animal species , in which we recorded more than 15 strides ( dog , donkey , human , Table 1 ) . The inter-segmental coordination of the elevation angles of HL and FL segments was evaluated in position space using principal component analysis ( PCA ) as previously described ( Borghese et al . , 1996; Bianchi et al . , 1998b; Ivanenko et al . , 2007; Catavitello et al . , 2015 ) . To assess planar covariation of limb segment motion , we computed the covariance matrix of the ensemble of time-varying elevation angles ( after subtraction of their mean values ) . The three eigenvectors u1 , u2 and u3 , rank-ordered on the basis of the corresponding eigenvalues , correspond to the orthogonal directions of maximum variance in the sample scatter . The first two eigenvectors u1 and u2 identify the best-fitting plane of angular covariation . The third eigenvector ( u3 ) is the normal to the plane , and defines the plane orientation in the 3D space of the elevation angles . The planarity of the trajectories was quantified by the percentage of total variation ( PV3 ) accounted for by the third eigenvector ( for ideal planarity , PV3 = 0% ) . Our main analysis was focused on the elevation angles that capture directly the limb configuration in external space . However , we also report the results of the PCA applied to the relative ( anatomical ) angles of the hip , knee and ankle for HL , and shoulder , elbow and wrist for FL . To characterize the distribution and differences in the orientation of the covariation plane ( u3 ) across groups ( birds HL , mammals HL , mammals FLupp , mammals FLlow ) , we used the empirical shape criterion ( Fisher et al . , 1993 ) and the high concentration parameter test ( Tsagris et al . , 2017 ) . In particular , we distinguished the girdle distribution from the clustered distribution of the eigenvectors based on the empirical shape criterion . Briefly , let x1 , y1 , z1 , … , xn , yn , zn be the direction cosines of a sample of points on the unit sphere . The location of these points can be synthetized by their sample mean vector ( x- , y- , z- ) , which is defined as x- , y- , z-= ( ∑xi , ∑yi , ∑zi ) , where the sum is over the number of points n . It is useful to express the mean vector in polar form as x- , y- , z-=R*x-0 , y-0 , z-0 , the scalar product of a unit vector x-0 , y-0 , z-0 with its resultant length R . If the points x1 , y1 , z1 , … , xn , yn , zn are considered as having equal mass , then their center of mass is x- , y- , z- , which has direction x-0 , y-0 , z-0 and distance R from the origin . Then , the mean direction x-0 , y-0 , z-0 defines the location of the sample , and the mean resultant length R-=R/n provides a measure of how concentrated the sample is . If the points are concentrated close together , R will be close to 1 , whereas an increasing scatter results in smaller values of R . We then computed the eigenvalues ( τ1 , τ2 , τ3 ) of the orientation T defined as T=∑xi2∑xiyi∑xizi∑xiyi∑yi2∑yizi∑xizi∑yizi∑zi2 . These eigenvalues give an indication of the general shape of the data set . The empirical shape criterion is based on the assumption that the shape of the distribution will determine its position in the bi-dimensional space created by the variables a= log⁡ ( τ3/τ2 ) and b= log⁡ ( τ2/τ1 ) . For the girdle distribution , the empirical shape value ( defined as γ=a/b ) is less than 1 , while for the clustered distribution it exceeds 1 ( see Fig3 . 15b in Fisher et al . , 1993 ) . The degree of alignment of the samples of u3 vectors was assessed by the concentration parameter k , which is a measure of the concentration of the sample about the mean direction ( the higher is k , the more clustered the data ) , which we used to distinguish the population mean directions ( Tsagris et al . , 2017 ) . We also added an analysis of kinematic traits taking into account a consideration of the data variation in an evolutionary context . In this context , species are not independent data points . Indeed , closely related species derive from a common ancestor and should be weighted more than distant species , since they share common characteristics ( Garland et al . , 2005 ) . We performed a phylogenetic analysis of the orientation of the covariance plane for the HL ( α angle , Figure 4A ) and its correlation with the phase shifts between limb segment movements ( Figure 5A ) . If traits under consideration of closely related species have similar values while the similarity decreases with increasing phylogenetic distance , one can infer the presence of a high phylogenetic signal of the trait . Vice versa , a weak phylogenetic signal corresponds to the situation when close relatives are not more similar on average than distant relatives . The latter may happen in the presence of a convergent evolution or when the traits are randomly distributed across a phylogeny . We reconstructed a taxonomic tree with the data stored in the NCBI Taxonomy Browser ( https://www . ncbi . nlm . nih . gov/Taxonomy/Browser/wwwtax . cgi ) . We used the Brownian motion model ( random walk in continuous time ) to estimate trait evolution: modifications of the trait value through time occur gradually and they are independent of the current state . At the tips of the phylogenetic tree , the expected covariance between trait values of the species is proportional to their common history . The history of each tip , that is species , is computed as the sum of their branch length . Then , a phylogeny can be represented as an n x n phylogenetic variance–covariance matrix , where n is the number of species in the phylogeny . The diagonals of the matrix correspond to the total length of the tree and they represent the species variances , while the off-diagonal elements are computed as the sum of their shared branch lengths and are the covariances between species pairs . We computed the branch lengths with the function ‘compute . brlen’ in the ‘phytools’ R package ( Revell , 2009 ) . We computed Blomberg’s K that measures phylogenetic signal by quantifying the amount of observed trait variance relative to the trait variance expected under Brownian motion ( Blomberg et al . , 2003 ) . K is the ratio of the mean squared error of the tip data in relation to the phylogenetic mean of the data divided by the mean squared error extracted from a generalized least-squares model ( PGLS , Phylogenetic Generalized Least-Squares ) that uses the phylogenetic variance–covariance matrix in its error structure ( Kamilar and Cooper , 2013 ) . K can vary continuously from zero , indicating that there is no phylogenetic signal in the trait ( i . e . the trait has evolved independently of phylogeny ) , to infinity . K = 1 indicates that there is strong phylogenetic signal and the trait has evolved according to the Brownian motion model of evolution , while K > 1 indicates that close relatives are more similar than expected under a Brownian motion model of trait . We performed the evolutionary phylogenetic correlation by fitting a linear module using PGLS . The correlations were computed using the ‘phytools’ and ‘geiger’ R packages . The ancestral states estimation was computed assuming Brownian motion , using the functions ‘fastAnc’ and ‘plotSimmap’ in the ‘phytools’ R package ( Revell , 2009 ) . Statistical analyses were performed using Matlab and R software . Descriptive statistics included the calculation of the mean values and standard deviations . For each species , the parameters were first averaged across gait cycles and then across animals for the same species before subsequent analyses . The general gait parameters ( trunk orientation , stride and stance duration , inter-limb coupling , Froude number and endpoint excursions ) were not computed for animals marked by an asterisk in Table 1 ( since this information was not provided in Fischer et al . ( 2002 ) ) , so we performed statistics on these parameters for a smaller number of animals ( n = 48 ) as compared with the analysis of the inter-segmental coordination ( n = 54 ) . To assess some general kinematic parameters , paired t-tests were used to evaluate differences between HL and FL in quadrupeds , while unpaired t-tests were used to assess differences between the groups of animals when appropriate ( e . g . birds vs . mammals HL or FL ) . Statistical analysis of spherical data was used to characterize the mean orientation of the normal to the covariation plane . In spherical statistics , we distinguished the girdle distribution from the clustered distribution of the eigenvectors with the empirical shape criterion ( Fisher et al . , 1993 ) . Multi-way ANOVA and high concentration parameters test for spherical data were carried out with the R package built for directional statistics ( Tsagris et al . , 2017 ) . A linear regression analysis was used to assess the relationship between limb parameters and rotation of the covariation plane . The correlation coefficients were Z-transformed before the statistical analysis . Reported results are considered statistically significant for p<0 . 05 . We provided Source Data files for figures . They are labeled as Source Data x-figure x . zip ( where x is the figure number ) , contain numeric data and , where necessary , the code to reproduce the figure . See the readme file for each folder .	Animals have evolved very different body shapes and styles of movement that are adapted to their needs in the habitats they live in . For example , mice , lions and many other animals use four limbs to walk , while humans and birds only use two limbs . The styles animals use to walk also differ in terms of how long each foot is on the ground during a single stride , and for four-legged animals , in how long a forefoot lags behind the hindfoot on the same side of the body during the stride . Yet , there are general principles in how walking is organized that are shared between animals of vastly different shapes and sizes . Many animals save energy during walking by swinging the center of their body mass back and forth like a pendulum . Networks of neurons are responsible for controlling how and when animals move , and these networks have similar architectures and patterns of activity in many different mammals and birds . How do signals from the nervous system regulate the position of the center of body mass while an animal walks ? Here , Catavitello et al . addressed this question by analyzing how over 50 different species of birds and mammals walked around in zoo enclosures and other semi-natural or natural environments . The species studied ranged in size from mice weighing around 30 grams to elephants weighing around 4 tonnes . The team also studied human volunteers walking on treadmills . The experiments show that all the species studied coordinate their limbs in the same way , so that the angle to which a particular segment of a limb can bend varies together with the angles that the other limb segments bend . This coordination implies that the movement of the center of body mass is regulated and energy is saved . Along with providing new insight into how walking evolved , these findings may aid research into new approaches to treat walking impairments in humans and other animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2018	A kinematic synergy for terrestrial locomotion shared by mammals and birds
The neuronal DNA-/RNA-binding protein Pur-alpha is a transcription regulator and core factor for mRNA localization . Pur-alpha-deficient mice die after birth with pleiotropic neuronal defects . Here , we report the crystal structure of the DNA-/RNA-binding domain of Pur-alpha in complex with ssDNA . It reveals base-specific recognition and offers a molecular explanation for the effect of point mutations in the 5q31 . 3 microdeletion syndrome . Consistent with the crystal structure , biochemical and NMR data indicate that Pur-alpha binds DNA and RNA in the same way , suggesting binding modes for tri- and hexanucleotide-repeat RNAs in two neurodegenerative RNAopathies . Additionally , structure-based in vitro experiments resolved the molecular mechanism of Pur-alpha's unwindase activity . Complementing in vivo analyses in Drosophila demonstrated the importance of a highly conserved phenylalanine for Pur-alpha's unwinding and neuroprotective function . By uncovering the molecular mechanisms of nucleic-acid binding , this study contributes to understanding the cellular role of Pur-alpha and its implications in neurodegenerative diseases . Purine-rich element-binding protein A ( Pur-alpha ) plays a crucial role in postnatal brain development . Pur-alpha-deficient mice appear normal at birth but develop severe neurological abnormalities after 2 weeks and die shortly after birth ( Hokkanen et al . , 2012; Khalili et al . , 2003 ) . These mice show fewer cells in the brain cortex , hippocampus , and cerebellum as a consequence of decreased proliferation of the precursor cells . Further studies indicated that Pur-alpha co-localizes with Staufen and FMRP and that Pur-alpha ( -/- ) mice display dendritic mislocalization of both proteins ( Johnson et al . , 2006 ) . In support of its important neuronal function , point mutations in the human Pur-alpha gene have been found to cause the so-called 5q31 . 3 microdeletion syndrome , which is characterized by neonatal hypotonia , encephalopathy , and severe developmental delay ( Lalani et al . , 2014; Hunt et al . , 2014; Tanaka et al . , 2015 ) . Pur-alpha is an ubiquitously expressed , multifunctional protein that binds to both DNA and RNA and is known to regulate replication , transcription , and translation ( Johnson et al . , 2013 ) . It has been shown that Pur-alpha binds to single- and double-stranded nucleic acids that contain GGN motifs . Such regions are found at origins of DNA replication and enhancers of TATA-box lacking genes , such as c-myc or the myelin-basic protein , which Pur-alpha regulates . Pur-alpha has also been routinely purified from cytoplasmic kinesin-containing ribonucleoprotein particles ( RNPs ) ( Kanai et al . , 2004; Ohashi et al . , 2000 ) , further supporting its role in mRNA localization and showing that Pur-alpha is a core factor in localizing mRNPs . Besides its ability to bind RNA and DNA , Pur-alpha possesses dsDNA-destabilizing activity in an ATP-independent fashion ( Darbinian et al . , 2001 ) . This function has been suggested as important for DNA replication and transcription regulation . It was postulated that Pur-alpha , being a transcription activator , contacts the purine-rich strand of promoter regions and displaces the pyrimidine-rich strand , which would allow the binding of other proteins and activation of transcription ( Darbinian et al . , 2001; Wortman et al . , 2005 ) . The role of Pur-alpha-dependent unwinding activity in RNA localization and in RNA-based neuropathological disorders is currently unknown . One particularly interesting interaction partner of Pur-alpha is the RNA helicase Rm62 , the Drosophila ortholog of p68 . It is implicated in transcriptional regulation , pre-mRNA splicing , RNA interference , and nucleo-cytoplasmic shuttling ( Qurashi et al . , 2011 ) . Thus , their joint function could be the initial unwinding of short dsRNA regions by Pur-alpha followed by helicase-dependent melting of larger regions for the regulation of RNA processing , translational control , and transport . Nucleic acid-binding of Pur-alpha is mediated by three central PUR repeats ( Graebsch et al . , 2010; Graebsch et al . , 2009 ) , which are N-terminally flanked by unstructured , glycine-rich sequences and C-terminally by glutamine- and glutamate-rich regions ( Figure 1A; Johnson et al . , 2013 ) . In the recently published crystal structure of Pur-alpha each of both PUR repeats I and II consist of a four-stranded antiparallel beta-sheet , followed by a single alpha-helix ( Graebsch et al . , 2009 ) . Repeat I and II fold into an intramolecular dimer that serves as a DNA-/RNA-binding domain . The third repeat leads to intermolecular dimerization ( Figure 1A; Graebsch et al . , 2009 ) . Despite these insights , it remains unclear how Pur-alpha interacts with its nucleic-acid targets to mediate its cellular functions . Furthermore , the mechanistic basis and physiological importance of its unwinding activity remains unresolved . 10 . 7554/eLife . 11297 . 003Figure 1 . Pur-alpha uses similar binding modes for DNA and RNA . ( A ) Schematic representation of the Drosophila Pur-alpha protein , comprising 274 amino acids . Cartoon shows PUR repeat I ( green ) and II ( blue ) , forming the intramolecular DNA-/RNA-binding PUR domain , and PUR repeat III ( grey ) mediating dimerization . The N-terminal unstructured , glycine-rich region and the C-terminal glutamine-/glutamate-rich region are indicated by Gly and Gln/Glu , respectively . Numbers indicate amino-acid positions of domain boundaries . ( B , C ) Radioactive EMSA with Drosophila Pur-alpha repeat I-II . ( B ) Pur-alpha repeat I-II binds to MF0677 ssDNA ( left ) and ssRNA ( right ) with similar affinities . ( C ) Pur-alpha repeat I-II binds to CGG-repeat ssDNA ( left ) and RNA ( right ) also with similar affinity , but less strong than to the MF0677 sequence . Open arrowheads indicate free and filled arrowheads indicate protein-bound DNA/RNA oligonucleotides . ( D ) Fluorescence-polarization experiments with full-length Pur-alpha and nucleic acids . The full-length protein shows a twofold stronger binding to MF0677 ssRNA when compared to MF0677 ssDNA . ( E ) Binding of unlabeled GCGGA ssDNA and ssRNA to 15N-labeled Pur-alpha repeat I-II ( 50 µM ) followed by NMR spectroscopy . ( Left ) Overlay of 1H , 15N HSQC NMR spectra of free ( black ) , DNA-bound ( red , 1:1 ratio ) and RNA-bound ( cyan , 1:1 ratio ) protein , respectively . ( Right ) Close-up on the dashed area with the same color code . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 00310 . 7554/eLife . 11297 . 004Figure 1—figure supplement 1 . Purification and quality control of Drosophila Pur-alpha protein derivatives used in this study . ( A , C–E ) Size exclusion chromatogram ( blue ) of the final purification step with the Superdex 75 10/300 GL column . Peak fractions ( red dash ) were pooled , concentrated and analyzed on SDS–PAGE . ( A ) Pur-alpha repeat I-II ( 17 kDa ) , ( C ) Pur-alpha repeat I-II ( 17 kDa ) in NMR buffer , ( D ) Pur-alpha repeat III ( 10 kDa ) , ( E ) Pur-alpha full length . ( B ) Overlay of CD spectra of Pur-alpha wild-type and mutant protein ( color code as indicated in the figure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 00410 . 7554/eLife . 11297 . 005Figure 1—figure supplement 2 . 1H , 15N HSQC NMR spectra showing NMR titrations of 15N-labeled Pur-alpha repeat I-II ( 50 µM ) with increasing amounts of unlabeled GCGGA ssDNA and RNA , respectively . ( A ) Titration with DNA . Peaks corresponding to the free and DNA-bound protein states ( protein:DNA 1:0 , 1:0 . 5 , 1:1 , 1:2 . 5 and 1:5 ratio ) are represented in blue , cyan , green , orange and red , respectively . ( B ) Titration with RNA . Peaks corresponding to the free and RNA-bound protein states ( protein:RNA 1:0 , 1:0 . 5 , 1:1 , 1:2 . 5 , and 1:5 ratio ) are represented with the same color code as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 005 Pur-alpha has been implicated in two so-called RNA repeat-expansion diseases , which have been the focus of a number of recent studies . The first one contains expansions in the well-studied fmr1 gene . Individuals with 55 to 200 CGG repeats , termed pre-mutation , develop the neurodegenerative Fragile X-associated Tremor/Ataxia Syndrome ( FXTAS ) ( Hagerman et al . , 2001 ) , whereas healthy individuals have less than 54 trinucleotide CGG repeats in their 5’-UTR region ( Oostra and Willemsen , 2009 ) . It is generally accepted that expression of FMR1 mRNA with abnormal trinucleotide-repeat expansions are the main cause of FXTAS . The second Pur-alpha related disease is caused by repeat expansions of G4C2-hexanucleotides in the first intron of the c9orf72 transcript . These repeat expansions are considered as the most common genetic abnormality in amyotrophic lateral sclerosis ( ALS ) and familial frontotemporal lobal degeneration ( FTLD ) ( Stepto et al . , 2014 ) . The diseases associated with both types of repeat expansions are accompanied by the formation of repeat RNA-containing protein inclusions ( Sareen et al . , 2013; Stepto et al . , 2014 , ; Xu et al . , 2013 ) , suggesting sequestration of proteins as potential mechanism of pathology . Pur-alpha is incorporated into the inclusions of both types of disease and associates directly with the repeat RNAs ( Jin et al . , 2007 , ; Xu et al . , 2013; Rossi et al . , 2015 ) . In fly and mouse models , the overexpression of Pur-alpha can overcome repeat-dependent neurodegeneration of both diseases ( Jin et al . , 2007 , ; Xu et al . , 2013 ) , suggesting a direct contribution of Pur-alpha to neuropathology . Expression of 95 CGG repeats in human neuroblastoma-derived SK-N-MC cells not only induced the formation of nuclear inclusions but also impairs the architecture of the nuclear laminar and activates DNA repair-associated histone variants ( Hoem et al . , 2011 ) . The expression of G4C2-repeat expansions cause nuclear trafficking defects , which contribute to neurotoxicity in ALS/FTLD ( Freibaum et al . , 2015; Jovicic et al . , 2015; Zhang et al . , 2015 ) . Recent studies also showed that repeat-associated non-AUG ( RAN ) translation occurs from CGG- as well as from G4C2-repeat RNAs and that the resulting proteins can form cytoplasmic aggregates , potentially contributing to pathology ( Mori et al . , 2013; Todd et al . , 2013 ) . It is likely that a combination of RNA toxicity and RAN-derived protein aggregates contribute to the full manifestation of FXTAS . Here , we used NMR chemical shift titrations together with in vitro-binding assays to demonstrate that the nucleic acid-binding domain of Pur-alpha binds RNA and DNA in the same manner . We present the co-crystal structure of Pur-alpha with a CGG trinucleotide-repeat DNA , providing a detailed structural explanation for nucleotide recognition . Pur-alpha interacts with this single-stranded DNA fragment in a sequence-specific manner with guanines and additional contacts to the phosphordiester backbone . The observed binding mode of Pur-alpha also explains its interaction with G4C2-hexanucleotide repeats . Mutational analyses as well as determination of the complex stoichiometry confirm that the DNA-/RNA-binding domain of Pur-alpha has two nucleic acid-binding sites . The structure also revealed that a highly conserved phenylalanine causes disruption of the normal base stacking and leads to a strong torsion of the DNA strand , which plays a central role in Pur-alpha’s dsDNA-unwinding activity . In vivo analyses of mutant proteins reveal that nucleic-acid binding and unwinding studied in vitro are both essential for Pur-alpha’s function in vivo . This information together with the crystal structure of its C-terminal dimerization domain allows us to propose a mechanism of how full-length Pur-alpha binds and unwinds dsDNA regions . In order to assess if Pur-alpha has a binding preference for ssDNA or ssRNA , we performed electrophoretic mobility shift assays ( EMSA ) with the nucleic acid-binding domain of Drosophila Pur-alpha , consisting of repeats I-II ( PUR repeat I-II; Figure 1A; Figure 1—figure supplement 1A , B ) and radiolabeled DNA or RNA oligonucleotides ( 24 nt ) of identical sequence . The MF0677 sequence was chosen as a physiological Pur-alpha target found upstream of the human c-myc gene ( Haas et al . , 1993; Haas et al . , 1995 ) . In addition , we used a CGG-repeat sequence because Pur-alpha binds to these repeats in the 5’UTR of the FMR1 mRNA upon incorporation into FXTAS inclusions ( Jin et al . , 2007; Sofola et al . , 2007 ) . In these EMSA , the affinity for the physiological Pur-alpha target MF0677 is much higher ( KD ~200 nM ) than for the disease-related CGG-repeat sequence ( KD ~2 µM ) ( Figure 1B , C; KD estimated from EMSA ) . However , the binding affinities for ssDNA and ssRNA of the same sequence showed no major differences . Since full-length Pur-alpha contains a third PUR repeat , which mediates its dimerization , and additional N- and C-terminal sequences ( Figure 1A ) , we also compared DNA and RNA binding of full-length Pur-alpha ( Figure 1—figure supplement 1E ) . For quantification of the nucleic acid-binding affinity , we performed fluorescence-polarization experiments . Full-length Pur-alpha showed a two-fold preference in binding to MF0677 ssRNA ( KD = 0 . 7 µM ) over MF0677 ssDNA ( KD = 1 . 4 µM; Figure 1D ) . Thus , sequences outside PUR repeats I-II seem to moderately affect nucleic-acid binding . For a more comprehensive , residue-resolved comparison of ssDNA and ssRNA binding , we performed NMR chemical shift titration experiments with 15N-labeled Drosophila Pur-alpha repeat I-II ( Figure 1—figure supplement 1C ) and short unlabeled GCGGA ( 5 nt ) DNA and RNA fragments . The 1H , 15N HSQC NMR spectrum of Pur-alpha alone shows well separated cross peaks ( Figure 1E; Figure 1—figure supplement 2A , B ) , indicating that the protein is correctly folded . Addition of either ssDNA or ssRNA resulted in almost identical , well-localized chemical shift perturbations of backbone and sidechain amide protons ( Figure 1E; Figure 1—figure supplement 2A , B ) . Most NMR signals of residues involved in binding disappeared upon addition of DNA/RNA , thus pointing toward an intermediate exchange regime , which is characteristic for binding affinities in the high nanomolar to micromolar range . In summary , the NMR titration experiments indicate identical binding modes of PUR repeat I-II for ssDNA and for ssRNA involving the same residues in both cases . In order to obtain high-resolution structural information of Pur-alpha binding to nucleic acids , we performed co-crystallization experiments of Pur-alpha repeat I-II with either CGG-repeat DNA or RNA . Crystals of Pur-alpha repeat I-II with a GCGGCGG trinucleotide-repeat ssDNA diffracted to a resolution of 2 . 0 Å . The structure was solved by molecular replacement and refined to Rwork and Rfree of 16 . 3% and 21 . 5% , respectively ( Table 1 ) . 10 . 7554/eLife . 11297 . 006Table 1 . Data collection/processing and refinement statistics ( molecular replacement ) for the two crystal structures of Drosophila Pur-alpha repeat I-II/DNA co-complex and Pur-alpha repeat III alone . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 006Pur-alpha repeat I-II/DNAPur-alpha repeat IIIData collection/processing PDB ID Beamline Wavelength ( Å ) Detector Distance ( mm ) Number of images Oscillation range ( ° ) Space group5FGP ESRF ID23-2 0 . 8726 265 . 433 144 2 . 5 P212125FGO ESRF ID14-1 0 . 9334 261 . 345 180 1 . 0 P 1 21 1Cell dimensions a , b , c ( Å ) 81 . 9 , 40 . 2 , 48 . 861 . 5 , 55 . 5 , 67 . 8α , β , γ ( ° ) 90 . 0 , 90 . 0 , 90 . 090 . 0 , 95 . 7 , 90 . 0Resolution ( Å ) 50 . 0-2 . 0 ( 2 . 05-2 . 0 ) 50-2 . 6 ( 2 . 67-2 . 6 ) Rsym or Rmerge12 . 5 ( 79 . 3 ) 11 . 1 ( 68 . 0 ) I / σI 18 . 85 ( 2 . 61 ) 10 . 4 ( 1 . 97 ) Completeness ( % ) 99 . 4 ( 94 . 3 ) 96 . 5 ( 98 . 4 ) Redundancy13 . 1 ( 7 . 6 ) 1 . 9 ( 1 . 9 ) Refinement Resolution ( Å ) 41 . 9-2 . 047 . 7-2 . 6No . reflections11 , 34914 , 001Rwork / Rfree16 . 3 / 21 . 520 . 6 / 28 . 9No . atoms Protein12073144Ligand/ion145-Water126112B-factors Protein24 . 812 . 5Ligand30 . 4-Water35 . 29 . 55R . m . s . deviations Bond lengths ( Å ) 0 . 010 . 01Bond angles ( ° ) 1 . 251 . 36Ramanchandran plot Allowed ( % ) Additionally allowed ( % ) Disallowed ( % ) 96 . 0 3 . 3 0 . 7 93 . 0 6 . 5 0 . 5Values in parentheses are for highest-resolution shell . The DNA-bound protein shows the typical intramolecular dimer with two PUR repeats tightly intertwined with each other , forming a globular PUR domain ( Figure 2A; Video 1; Figure 2—figure supplement 1A; Graebsch et al . , 2009 ) . Each PUR repeat consists of a N-terminal four-stranded antiparallel beta sheet followed by an alpha helix . A superposition of the previously published Pur-alpha repeat I-II apo-structure ( PDB ID 3K44 ) ( Graebsch et al . , 2009 ) with the structure of the protein-DNA co-complex showed only a root-mean-square deviation ( RMSD ) of atomic positions of 1 . 14 Å ( Figure 2—figure supplement 1B ) . When a flexible loop region from residues L107 to K120 was excluded , the RMSD improved to 0 . 83 Å . Thus , no major conformational changes occur in the PUR domain upon nucleic-acid binding , which is consistent with the results obtained from NMR chemical shift titrations . 10 . 7554/eLife . 11297 . 007Figure 2 . Crystal structure of Drosophila Pur-alpha repeat I-II in complex with the GCGGCGG ssDNA reveals that one molecule of Pur-alpha repeat I-II can bind two molecules of ssDNA . ( A ) Cartoon representation of the backbone model of the DNA-/RNA-binding domain formed by PUR repeat I ( green ) and II ( blue ) in complex with two DNA molecules ( pink ) . Important protein residues involved in DNA interactions are depicted in red with side chains . ( B ) Schematic representation of Pur-alpha interaction with DNA molecules 1 and 2 . Both PUR repeats are involved in DNA binding . Pur-alpha mainly binds to guanine bases , but also to one cytosine and the sugar phosphate backbone . ( C–F ) Detail of the protein-DNA interaction sites . ( G ) Nitrocellulose filter ( top ) from binding assays showing the titration of Pur-alpha repeat I-II to a constant amount of MF0677 ssDNA . The measured intensities from the filters were quantified . The graph ( bottom ) shows intensities from one representative binding assay , with the concentration of saturation marked with a dashed line . Three independent filter-binding assays yielded a mean saturation of 1 : 0 . 58 ± 0 . 1 µM ( ssDNA : Pur-alpha repeat I-II ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 00710 . 7554/eLife . 11297 . 008Figure 2—figure supplement 1 . Analysis of the structural model of Drosophila Pur-alpha repeat I-II in complex with DNA . ( A ) Stereo view of the Pur-alpha/DNA structure model showing the protein ( grey ) , the DNA molecule 1 ( magenta ) , and the ( 2Fo-Fc ) electron density map ( blue ) . ( B ) Structure alignment of the protein backbones of Pur-alpha repeat I-II apo-structure ( PDB ID 3K44; depicted in red ) and in complex with ssDNA ( this study; depicted in blue ) reveals no major conformational changes upon DNA binding . RMSD value is indicated in the figure . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 00810 . 7554/eLife . 11297 . 009Figure 2—figure supplement 2 . Within the crystal structure the protein-bound DNA anneals with another symmetry-related DNA molecule . Cartoon representation shows Pur-alpha repeat I-II ( blue ) bound to two DNA molecules ( pink ) and the symmetry-related Pur-alpha-DNA complex ( grey and cyan , respectively ) . The 5’ ends of the symmetry-related DNA molecules ( DNA 1 and 1’ ) are base pairing ( see close-up on the right side ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 00910 . 7554/eLife . 11297 . 010Figure 2—figure supplement 3 . The DNA-binding site consisting of K138 , N140 , and R142 ( KNR II ) on PUR repeat II has its equivalent at the positions K61 , N63 , and R65 on PUR repeat I ( KNR I ) . ( A ) Cartoon representation of Pur-alpha repeat I ( green ) and II ( blue ) bound to two molecules of DNA ( pink ) . Side chains of KNR I are indicated as red sticks . Only K61 interacts directly with the G6’ base of DNA molecule 2 . ( B ) Zoom-in of ( A ) with indicated distance measurements . A close-up of KNR II binding to DNA 1 is shown in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01010 . 7554/eLife . 11297 . 011Figure 2—figure supplement 4 . Amino acid sequence alignment of Pur-alpha . ( A ) Sequence alignment from different species . D . m . , Drosophila melanogaster; H . s . , Homo sapiens; D . r . , Danio rerio; C . e . , Caenorhabditis elegans . ( B ) Sequence alignment between the three PUR repeats of Drosophila Pur-alpha . Color-coding from blue to red reflects the range of sequence conservation from 0 to 100% . Asterisk indicates positions , which have a single , fully conserved residue . Colon indicates conservation between groups of strongly similar properties . Period indicates conservation between groups of weakly similar properties . Secondary structure assignment is based on the crystal structure of Pur-alpha repeat I-II and Pur-alpha repeat III . Red boxes indicate DNA interaction sites seen in the crystal structure . Blue boxes indicate additional residues used in mutational analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01110 . 7554/eLife . 11297 . 012Video 1 . Movie of the crystal structure of Drosophila Pur-alpha repeat I-II in complex with the GCGGCGG ssDNA . Color-coding as in Figure 2A . Movie relates to Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 012 In the crystal structure , the DNA molecule 1 ( DNA 1 ) is clamped between residues of PUR repeat I and II ( Figure 2A , B; Video 1 ) . PUR repeat II binds DNA 1 with the residue K138 of its β-sheet , and residues N140 and R142 of the short linker ( Figure 2B , C ) , whereas PUR repeat I contacts the DNA 1 via residues Q52 , S53 , and K54 in its short linker ( Figure 2B , D ) . Pur-alpha mainly binds to stacking guanine bases , but also to one of the cytosines ( C5 ) and to the sugar phosphate backbone ( Figure 2B ) . Within the crystal lattice the first two bases ( G1 and C2 ) of the 5’-end of DNA 1 are base pairing with the 5’-end of the symmetry related DNA molecule ( DNA 1’; Figure 2—figure supplement 2 ) . The cytosine C5 in the middle of the DNA 1 strand is twisted and does not stack with the neighboring guanines ( Figure 2E ) . Instead , F145 from the β-sheet of PUR repeat II stacks with the neighboring guanine G4 and thereby blocks the space for the cytosine C5 ( Figure 2E , Video 1 ) . In the crystal structure , an additional DNA-binding event was observed for PUR repeat I . The residues Y57 , D59 , K61 , K70 , and R80 of the β-sheet interact with the 3’-end of the second DNA molecule ( DNA 2 ) ( Figure 2B , F; Video 1 ) . This interface is similar but not identical to the DNA 1-binding site on PUR repeat II . The three DNA-contacting amino acids K138 , N140 , and R142 of PUR repeat II are also found in corresponding positions of PUR repeat I ( Figure 2—figure supplement 3 ) . However , in PUR repeat I only K61 but not N63 or R65 contact the DNA 2 molecule . Thus , although there is a conservation of DNA-contacting residues on both PUR repeats , in the crystal structure their modes of binding are not identical . This observation hints toward a potentially asymmetric binding of nucleic acids on both protein surfaces of Pur-alpha I-II . To test if Pur-alpha also interacts with two DNA oligonucleotides in solution , we performed filter-binding assays with Pur-alpha repeat I-II and MF0677 ssDNA ( 24 nt ) . Pur-alpha repeat I-II was titrated at near-stoichiometric concentrations to a constant amount ( 1 µM ) of radiolabeled DNA and blotted onto a nitrocellulose membrane ( Figure 2G ) . Plots of the signal intensities against the protein concentrations yielded a mean saturation at 0 . 58 ± 0 . 1 µM ( n=3 ) of Pur-alpha ( Figure 2G ) . This indicates a stoichiometric ratio of 1:2 ( protein:DNA ) and confirms that like in the crystal structure ( Figure 2A ) Pur-alpha repeat I-II binds two molecules of ssDNA in solution . All amino acids involved in DNA binding within the crystal structure ( Figure 2B ) are conserved ( Figure 2—figure supplement 4A ) . To assess the importance of these contacts in solution , we generated structure-guided mutations and tested their effect on DNA/RNA binding . The binding motif consisting of K138 , N140 , R142 , and F145 on PUR repeat II ( KNR II and F II , respectively ) is also found on PUR repeat I ( K61 , N63 , R65 , and F68; KNR I and F I , respectively ) . Hence , these residues were replaced by alanines and tested for nucleic acid-binding in vitro . For the QSK I – KNR II mutant the residues Q52 , S53 , K54 , were replaced by glycine and the residues K138 , N140 , R142 by alanines , since a pure alanine mutant tended to aggregate . Correct folding of all generated Pur-alpha mutants was verified by circular dichroism ( CD ) spectroscopy ( Figure 1—figure supplement 1B ) . First , radioactive EMSA were performed with CGG-repeat and MF0677 DNA/RNA oligomers ( 24 nt ) . Except for Pur-alpha mutant F I , all other mutants showed decreased binding to DNA and RNA oligonucleotides with both motifs ( Figure 3A–E , G; Figure 3—figure supplement 1A–E ) . In order to quantify these interactions , we performed fluorescence-polarization experiments with fluorescein-labeled MF0677 DNA and different variants of Pur-alpha . The effects observed in EMSA of mutations in Pur-alpha I-II were confirmed by these experiments ( Figure 3H; Figure 3—figure supplement 2 ) . Of note , mutations in PUR repeat I ( KNR I , F I ) had less severe effects on DNA binding than mutations in repeat II ( KNR II , F II ) . 10 . 7554/eLife . 11297 . 013Figure 3 . Mutations in Pur-alpha repeat I-II decrease nucleic-acid binding and dsDNA unwinding . ( A–E ) Radioactive EMSA with wild-type and mutant Pur-alpha repeat I-II . All mutants show a decrease in binding affinity , except for the F I mutant in C . Open arrowheads indicate free and filled arrowheads indicate protein-bound DNA/RNA oligonucleotides . ( F ) Unwinding assays with wild-type and mutant Pur-alpha repeat I-II . Protein was titrated to a dsDNA substrate containing a GGN motif . Pur-alpha repeat I-II is able to separate the DNA strands , whereas mutations in both repeats ( QSK I – KNR II ) ( top ) and the F II mutation ( bottom ) abolish the unwinding activity . ( G ) Summary of the results of all EMSA and unwinding experiments of Pur-alpha derivatives and mutants . Original data are shown in Figure 3A–F and Figure 3—figure supplement 1 . ( H ) Summary of the results of fluorescence-polarization experiments of Pur-alpha derivatives and mutants with MF0677 ssDNA . Original data are shown in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01310 . 7554/eLife . 11297 . 014Figure 3—figure supplement 1 . Drosophila Pur-alpha repeat I-II mutants show decreased binding affinity to DNA and RNA and decreased dsDNA-unwinding activity . ( A–E ) Radioactive EMSA with wild-type or mutant Pur-alpha repeat I-II and with MF0677 ssDNA ( left ) or RNA ( middle ) and CGG-repeat RNA ( right ) . EMSA with CGG-repeat DNA is shown in Figure 3A–E . All mutants show decreased nucleic acid binding , except for the F68 mutant ( F I ) ( D ) , the counterpart to F145 on repeat II ( F II ) . Open arrowheads indicate free and filled arrowheads indicate protein-bound DNA/RNA oligonucleotides . ( F ) Mutations in repeat I ( KNR I ) ( left ) or in the identical motif in repeat II ( KNR II ) ( middle ) decreased the unwinding activity . ( Right ) Decreased unwinding also occurs upon mutation of F I . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01410 . 7554/eLife . 11297 . 015Figure 3—figure supplement 2 . Fluorescence-polarization measurements with wild type or various mutants of Pur-alpha I-II and MF0677 ssDNA . All experiments were performed as triplicates . KD values and standard deviations are given as insets in the plots . Curve fitting was performed as one-site binding . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 015 A large portion of the ssDNA 1 strand in the co-complex is stabilized in its conformation by aromatic stacking of G1 , C2 , G3 , G4 and G6 , G7 ( Figure 2C–F ) . F145 of Pur-alpha shows particularly unusual characteristics by undergoing aromatic stacking with G4 ( Figure 2E ) . This protein-DNA interaction blocks additional DNA-base stacking events and forces the DNA to flip out its cytosine base ( C5 ) , leading to a strong twist of the DNA 1 strand . It was previously reported that Pur-alpha unwinds short stretches of dsDNA in an ATP-independent manner ( Darbinian et al . , 2001 ) . However , the molecular basis of this function has not been understood to date . Since the sequence-specific interactions of Pur-alpha with DNA and the aromatic stacking of DNA with F145 seem incompatible with binding to dsDNA , we wondered which interactions are of foremost importance for the unwinding of dsDNA . Using a previously described unwinding assay ( Darbinian et al . , 2001 ) , we compared ATP-independent unwinding activity of wild-type and mutant Pur-alpha repeat I-II on a dsDNA substrate . When the main binding sites on PUR repeat I and II were mutated ( QSK I – KNR II ) unwinding was abolished ( Figure 3F , G ) , most likely due to impaired DNA binding ( Figure 3D , G; Figure 3—figure supplement 1A ) . In contrast , mutation of F145 ( F II ) abolished the unwinding activity without a complete loss of DNA binding ( Figure 3E–G; Figure 3—figure supplement 1E ) . All other mutations showed reduced DNA binding ( Figure 3A–C , G; Figure 3—figure supplement 1B–D ) and only decreased unwinding ( Figure 3—figure supplement 1F ) . Together these observations suggest that the heterotypic stacking of DNA-bases with F145 in PUR repeat II stabilizes the single-stranded conformation of DNA and enforces a twist of the bases that is important for its unwinding activity . To understand the role of the third repeat of Pur-alpha ( Figure 1A; Figure 1—figure supplement 1D ) for DNA/RNA binding , we determined its crystal structure . Initial datasets were obtained from native crystals at 2 . 7 Å resolution , from which electron-density maps were calculated by molecular replacement with the apo-structures of Pur-alpha from Borrelia and Drosophila as search templates ( PDB-IDs: 3NM7 and 3K44 , respectively ) . The final structure model was obtained in the same way from selenomethionine-derivatized crystals at 2 . 6 Å resolution ( Table 1; Figure 4A; Figure 4—figure supplement 1 ) . The structure consisting of two repeat III molecules shows the same overall fold as repeat I-II with an RMSD of 1 . 5 Å , and only few differences in the amino acid composition of its putative nucleic-acid-binding surface ( Figure 4A; Figure 2—figure supplement 4B ) . 10 . 7554/eLife . 11297 . 016Figure 4 . Crystal structure of PUR repeat III and assessment of its weak nucleic-acid-binding and unwinding activity . ( A ) Crystal structure of Drosophila Pur-alpha repeat III . Two molecules ( one depicted in brown , the other in yellow ) of repeat III form a dimer with intertwined α-helices , very similar to Pur-alpha repeat I-II . ( B ) Radioactive EMSA with PUR repeat III and the MF0677 DNA/RNA ( top ) and the CGG DNA/RNA oligonucleotides ( bottom ) . Repeat III shows only weak binding affinity to each of both sequences , regardless of whether they consist of DNA or RNA . Open arrowheads indicate free and filled arrowheads indicate protein-bound DNA/RNA oligonucleotides . ( C ) Pur-alpha repeat III shows only weak dsDNA-unwinding activity compared to PUR repeat I-II . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01610 . 7554/eLife . 11297 . 017Figure 4—figure supplement 1 . Analysis of the structural model of Drosophila Pur-alpha repeat III . Stereo view of the helical region of chain B ( grey ) , from lysine 254 to proline 235 with ( 2Fo-Fc ) electron-density map ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 017 PUR repeat III was previously suggested to mainly mediate dimerization of Pur-alpha ( Graebsch et al . , 2009 ) . However to date , no binding of PUR repeat III to nucleic acids has been measured . We therefore performed EMSA and observed that Pur-alpha repeat III bound with weaker affinities to CGG repeats and to MF0677 than Pur-alpha repeat I-II ( Figures 3G and 4B ) . Also in fluorescence-polarization experiments , PUR repeat III bound MF0677 ssDNA over 30-times weaker than PUR repeat I-II ( Figure 3H; Figure 3—figure supplement 2 ) . The main DNA/RNA interactions of full-length Pur-alpha might therefore occur via the first two PUR repeats . Although Pur-alpha repeat III does not have a phenylalanine in the corresponding position of F145 of PUR repeat II , it also contains a conserved aromatic residue ( Y219 ) , which could potentially undergo stacking with DNA bases and support dsDNA unwinding ( Figure 2—figure supplement 4B ) . However , in unwinding assays almost no activity was observed for PUR repeat III ( Figures 3G and 4C ) . These observations confirm that PUR repeats I-II mediate the main nucleic-acid-binding and unwinding activities and suggest that repeat III might predominantly mediate dimerization . To assess the physiologic relevance of our in vitro findings , we relied on a previously reported Drosophila model . Overexpression of CGG-repeat RNA in the Drosophila eye induces neuronal degeneration and as a consequence the rough eye phenotype ( compare Figure 5A with 5B; Jin et al . , 2003 ) . Overexpression of Pur-alpha can rescue the eye phenotype in a dose-dependent manner , suggesting that this protein is sequestered into the inclusions ( Jin et al . , 2007 ) . We compared the rescue by wild-type Pur-alpha with DNA-/RNA-binding and unwinding mutants . Whereas the wild-type protein achieved a full rescue ( Figure 5C ) , expression of the QSK I – KNR II mutant failed to ameliorate the rCGG repeat-induced neuronal toxicity ( Figure 5D ) . On the other hand , a previously reported double mutant R80A/R158A ( R I – R II ) that impairs nucleic-acid binding ( Figure 3H; Figure 3—figure supplement 2; Graebsch et al . , 2009 ) was still able to suppress the rCGG repeat-mediated toxicity ( Figure 5E ) . Thus , there might be differences in the binding for ssCGG repeats and the requirements for neuronal rescue . Most interestingly , however , is the observation that also the mutant F II , which still binds DNA/RNA but fails to unwind dsDNA , is unable to rescue neurodegeneration ( Figure 5F ) . Together these observations confirm the physiologic importance of the nucleic-acid-protein contacts observed in the crystal structure . In addition , these findings formally establish that the binding and unwinding of nucleic acids is required to modulate toxicity caused by pathogenic CGG RNA . 10 . 7554/eLife . 11297 . 018Figure 5 . Mutations in Pur-alpha’s nucleic-acid-binding domain abolish rescue of CGG RNA-mediated neurodegeneration . ( A–F ) Scanning electron microscope pictures of the eyes of adult flies . ( A ) Wild-type fly , ( B ) flies expressing ( CGG ) 90-EGFP/+ alone , together with wild-type Pur-alpha ( C ) , with the QSK I – KNR II mutant ( D ) , with the previously published R I – R II mutant ( E ) ( Graebsch et al . , 2009 ) and with the F II mutant ( F ) . Only wild-type Pur-alpha and the R I – R II mutant can rescue the neurodegeneration induced by rCGG repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 018 Pur-alpha repeat I-II shows strong and specific binding to its physiological target MF0677 DNA located upstream of the c-myc gene ( Bergemann et al . , 1992 ) , but much weaker binding to CGG-repeat RNA ( Graebsch et al . , 2009 ) . For this reason , it has been suggested that the binding of Pur-alpha to DNA is stronger than to RNA and , as a consequence , that there might be differences in the binding modes to both nucleic-acid targets . In this study , we directly compared Pur-alpha binding to RNA and DNA oligonucleotides of the same sequence and found no major differences ( Figure 1B–D ) . This suggests that the higher affinity for MF0677 ( KD ~200 nM; Figure 1B ) over CGG repeats ( KD ~2 µM; Figure 1C ) is due to differences in sequence and not the absence of the 2’ OH group in the DNA . This interpretation found further support from NMR titrations with 15N-labeled Pur-alpha repeat I-II and oligonucleotides . The spectra showed similar chemical shift perturbations , regardless of whether it was DNA or RNA , indicating that both nucleic acids are bound in the same way ( Figure 1E ) . Finally , the crystal structure of the Pur-alpha/DNA co-complex showed that a hydroxyl-group on the 2’ position of the pentose ring of the RNA sugar backbone would not cause steric clashes ( Figure 2A , C-F ) . Together , our biochemical , NMR , and X-ray crystallographic insights indicate that Pur-alpha binds DNA and RNA in the same way and thus will interact equally with both types of nucleic acids in the cell . It is also consistent with the previously suggested Pur-alpha-dependent gene regulation by competitive RNA binding ( Tretiakova et al . , 1998 ) . Previous findings implied that the positively charged β-sheets mediate DNA/RNAbinding , whereas the amphipathic helices might contribute to protein-protein interactions ( Graebsch et al . , 2009 ) . The crystal structure of the protein-DNA co-complex confirms that the β-sheets , together with their short linkers , are involved in DNA binding , in contrast to the α-helices that show no interaction ( Figure 2A ) . A comparison of the Pur-alpha repeat I-II apo-structure ( PDB ID 3K44 ) with the co-structure presented here revealed no significant conformational changes ( Figure 2—figure supplement 1B ) . In the crystal structure , Pur-alpha interacts with nucleic acids by clamping them between its two repeats , mostly by interacting with the guanine bases ( Figure 2A , B ) . Only R142 interacts with the cytosine base C2 . K54 and K138 additionally stabilize the DNA binding by interacting with the sugar phosphate backbone of guanine G4 and cytosine C5 , respectively ( Figure 2B–D ) . Binding therefore occurs sequence specifically and confirms the GGN-binding motif postulated before ( Bergemann and Johnson , 1992 ) . Mutation of the interacting residues resulted in a decreased binding affinity ( Figure 3G , H ) and therefore confirmed the interaction sites seen in the crystal structure . Also mutation of the corresponding KNR motif on PUR repeat I ( KNR I ) caused a decrease in affinity ( Figure 3G , H ) . However , in fluorescence-polarization experiments , the mutation of KNR I had a less severe effect on DNA binding ( KD = 1 . 3 µM ) than mutation of KNR II ( KD = 4 . 1 µM; Figure 3H ) . This is consistent with the observation that in the crystal structure all three residues of KNR II make contacts with DNA 1 ( Figure 2B , C ) , whereas in KNR I only a single amino acid binds to DNA 2 ( Figure 2—figure supplement 3 ) . Also , the F I mutation in repeat I had a less severe effect on MF0677 ssDNA binding than the F II mutation ( Figure 3G , H ) . In summary , these observations suggest that the MF0677 ssDNA is bound asymmetrically by PUR repeats I-II . In FXTAS patients , Pur-alpha binds to CGG-repeat expansions that cause the formation of nuclear inclusions and neurodegeneration ( Oostra and Willemsen , 2003 ) . Pur-alpha is also incorporated into inclusion triggered by G4C2-repeat RNA of patients with ALS and FTLD . The nucleic-acid binding of Pur-alpha observed in the crystal structure can explain both binding events , as it makes sequence-specific interactions with a GGC motif found in both repeat RNAs . The structural model of Pur-alpha repeat I-II forming a PUR domain has two nucleic-acid-binding surfaces . PUR repeats I and II share the identical binding motif ( KNR ) , and adopt the same fold , despite moderate sequence identity of about ~30% ( Figure 2—figure supplement 4; Graebsch et al . , 2010; Graebsch et al . , 2009 ) . Consistent with this finding , we observed a stoichiometric ratio of 1:2 for the PUR domain with ssDNA in filter-binding assays ( Figure 2G ) . Both binding events appear at overlapping but non-identical surface regions ( Figure 2A , B ) , which might prefer different GGN-motifs ( GGA , GGG , GGC , GGT ) as has been previously suggested ( Aumiller et al . , 2012 ) . This might also explain why CGG repeats bind less strongly to Pur-alpha than the MF0677 sequence , which mostly consists of GGA and GGT motifs . Pur-alpha has been previously reported to unwind dsDNA in an ATP-independent manner ( Darbinian et al . , 2001; Wortman et al . , 2005 ) . However , so far , it has not been shown how unwinding is achieved on a molecular level and that this function is physiologically relevant . The crystal structure of our Pur-alpha/DNA co-complex offers a mechanistic explanation: phenylalanine in position 145 of PUR repeat II undertakes base stacking with the guanine G4 and thereby blocks the space for the neighboring cytosine C5 ( Figure 2E ) . Thereupon , the cytosine flips out and the 3’-end of the DNA strand becomes distorted . The interaction of K54 and K138 with the phosphate backbone upstream of the cytosine C5 enforces this strong turn ( Figure 2B–D ) . F145 is highly conserved throughout different species ( Figure 2—figure supplement 4A ) and its mutation ( F II ) abolishes unwinding of dsDNA ( Figure 3F , G ) . Phenylalanine 145 has its structural counterpart in PUR repeat I in position F68 . Although F68 is also highly conserved , in the crystal structure the guanine base stacking is not mediated by this residue . Instead , the conserved Y57 in repeat I stacks with G7 ( Figure 2B , F ) . As mentioned before , the two binding sites of Pur-alpha seen in the crystal structure are asymmetric and might account for sequence-specific binding to nucleic acids with different GGN motifs . To assess the physiological importance of the interactions observed in the crystal structure and validated in vitro , we used the previously reported FXTAS fly model ( Jin et al . , 2007; Jin et al . , 2003 ) . Expression of pre-mutation CGG-repeat RNA in Drosophila induces neurodegeneration , which is easily detectable in abnormalities in the facet eye ( compare Figure 5A with 5B ) . While we observed that overexpression of wild-type Pur-alpha rescues the eye phenotype ( Figure 5C ) , the RNA-binding mutant QSK I - KNR II failed to do so ( Figure 5D ) . Surprisingly , a second , previously published RNA-binding mutant ( Pur-alpha R I – R II ) , which showed strongly reduced MF0677 ssDNA binding ( Figure 3H ) , was able to fully rescue the eye phenotype ( Figure 5E ) . This observation indicates that arginine 80 and 158 are not required for the binding to nucleic acids important for neuroprotection . While the neuroprotection by the R I – R II mutant indicates flexibility in nucleic-acid recognition , the loss of rescue by the QSK I - KNR II mutant formally establishes the requirement of nucleic-acid binding for Pur-alpha-dependent neuroprotection . Additionally , the F II mutation of Pur-alpha , which abolishes its dsDNA-unwinding activity , also impairs the neuroprotective function in the fly model ( Figure 5F ) . These findings indicate that unwinding is important for neuroprotection by Pur-alpha . Recently , de novo mutations in Pur-alpha have been found to cause the so-called 5q31 . 3 microdeletion syndrome . This disease is characterized by neonatal hypotonia , encephalopathy , and severe developmental delay ( Lalani et al . , 2014; Hunt et al . , 2014; Tanaka et al . , 2015 ) . Of the reported mutations ( Figure 6—source data 1 ) , two missense mutations ( A89P , K97E ) are of particular interest from a structure-to-function point of view ( Lalani et al . , 2014 ) . Sequence alignment of Pur-alpha from different species shows that the residues A89 and K97 of the human Pur-alpha protein correspond to the residues A72 and R80 of the Drosophila protein , respectively . These residues are highly conserved ( Figure 2—figure supplement 4A ) . In the crystal structure of the protein/DNA co-complex , A72 does not directly interact with the DNA molecule . Instead it forms backbone hydrogen bonds between the β-strands of PUR repeat I to stabilize the nucleic-acid binding β-sheet ( Figure 6A , top ) ( this study and Graebsch et al . , 2009 ) . When A72 and its disease-causing counterpart A98 in the human protein ( Figure 6A , middle ) are substituted by a proline , the backbone interactions that stabilize the β-sheet very likely become disrupted ( Figure 6A , bottom ) and thus the protein misfolds . 10 . 7554/eLife . 11297 . 019Figure 6 . Pur-alpha mutations found in the 5q31 . 3 microdeletion syndrome can be modeled into the crystal structure of Drosophila Pur-alpha repeat I-II ( green ) in complex with DNA ( cyan ) . ( A ) Residue A72 of the Drosophila protein ( top ) corresponds to the residue A89 ( grey ) of the human protein modeled into the Drosophila crystal structure ( middle ) . In both species , the alanines form backbone hydrogen bonds . In the microdeletion syndrome A89 is mutated to proline , which disrupts backbone interactions ( bottom ) . ( B ) Residue R80 of the Drosophila protein ( top ) corresponds to the residue K97 ( grey ) of the human protein , which was modeled into the crystal structure ( middle ) . Both R80 and K97 are positively charged residues . In Drosophila R80 interacts with the guanine G7 . The same interaction is likely to be mediated by K97 . In the microdeletion syndrome , K97 is mutated to a glutamate , which probably impairs nucleic-acid binding due to its negative charges ( bottom ) . A list of all published mutations in human Pur-alpha leading to the 5q31 . 3 microdeletion syndrome is shown in Figure 6—source data 1 . In this table , their predicted effects on the structure and function of Pur-alpha are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 01910 . 7554/eLife . 11297 . 020Figure 6—source data 1 . Mutations in the gene encoding for human Pur-alpha that result in the 5q31 . 3 microdeletion syndrome . Amino acids in brackets indicate the corresponding positions in Drosophila Pur-alpha , based on the protein-sequence alignment shown in Figure 2—figure supplement 4 . “p . ” indicates a point mutation , “” a stop codon , “fs” a frame shift followed by the number of residues produced until the next stop codon , “del” a deletion , and “delins” a combination of deletion and insertion . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 020 The Drosophila equivalent R80 of the disease-associated human K97 directly binds to the guanine base G7 ( Figure 2B , F and 6B , top ) and its mutation results in reduced nucleic-acid binding ( Graebsch et al . , 2009 ) . It is therefore conceivable that a mutation of K97 to glutamate impairs nucleic-acid interaction because of repulsive forces and causes dysfunction of Pur-alpha ( Figure 6B , middle , bottom ) . Although in our fly model the double mutant R80A/R158A ( R I – R II ) was still able to rescue neurodegeneration ( Figure 5E ) , the reported effect of the K97E mutation in the microdeletion syndrome indicates that nucleic-acid binding by this residue is important at least in humans . Additional interesting disease-causing point mutations in human Pur-alpha are I188T and I206F ( Figure 6—source data 1 ) , which likely impair the intramolecular dimerization of PUR repeats I and II ( Hunt et al . , 2014; , Tanaka et al . , 2015 ) . Taken together , the crystal structure of the Pur-alpha/DNA co-complex presented in this study provides a molecular explanation for the effects of missense mutations in the 5q31 . 3 microdeletion syndrome . Wild-type Pur-alpha binds to origins of replication and promoter regions ( Bergemann and Johnson , 1992 , ; Bergemann et al . , 1992 ) and regulates the transcription of more than 20 genes ( White et al . , 2009 ) . Pur-alpha’s ability to unwind dsDNA might therefore play an important role in the initiation of replication and transcription . One recently reported interaction partner of Pur-alpha that might play a role in this context is the RNA helicase Rm62 ( Qurashi et al . , 2011 ) . In the light of the dsDNA-unwinding activity an intriguing speculation is that Pur-alpha also unwinds dsCGG-repeat RNA . This initial unwinding by Pur-alpha could allow interacting helicases to subsequently regulate RNA processing , transport , and translation . Therefore , it will be important to assess Pur-alpha’s role in unwinding of dsRNA and its interaction with Rm62 . Considering that Pur-alpha repeat I-II has two nucleic-acid-binding sites , it is conceivable that each PUR repeat binds to one of the strands of a duplex DNA molecule thereby unwinding short stretches of dsDNA ( Figure 7A–C , top ) . The insertion of Pur-alpha between both DNA strands might be achieved through spontaneous breathing of the dsDNA helix ( Peyrard et al . , 2009 , ; Jose et al . , 2012 ) . Intercalating residues ( phenylalanine , tyrosine ) might cause further separation of the two DNA strands via base stacking with the guanines and thereby causing the strong twist of the DNA strands . The partly melted duplex DNA could then be further unwound by DNA helicases , which are required for initiation of transcription and replication . In the crystal structure , base pairing is observed between the 5’-G1-C2 bases of two symmetry-related DNA molecules ( Figure 2—figure supplement 2 ) , indicating , that a PUR domain would unwind a short stretch of approximately four to six bases . 10 . 7554/eLife . 11297 . 021Figure 7 . Model for unwinding of dsDNA by full-length Pur-alpha . ( A , top ) Electrostatic surface model of Pur-alpha repeat I-II in complex with one ssDNA molecule ( pink ) . Red and blue colorations of the surface indicate negative and positive electrostatic potentials , respectively . ( B , top ) Cartoon shows in addition the structure model of PUR repeat I ( green ) and II ( blue ) . DNA interaction sites , seen in the crystal structure , are shown as red sticks and correspond to the residues highlighted in Figure 2A . ( C , top ) Model showing the most likely overall trajectory of dsDNA ( pink ) when bound to Pur-alpha repeat I-II . The double-strand is locally unwound and the two separated strands bind to the two opposing binding sites on the protein . ( A , B , bottom ) Representation as in ( A , B , top ) , additionally showing the C-terminus connecting to PUR repeat III . PUR repeat III likely arranges at the opposing site of the nucleic-acid-binding region . ( C , bottom ) Schematic drawing of an intermolecular Pur-alpha dimer bound to dsDNA ( pink ) . PUR repeat III ( grey ) mediates dimerization , potentially orienting both nucleic-acid-binding domains ( repeat I , green and II , blue ) to the dsDNA . There both PUR domains could unwind larger regions of the DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 11297 . 021 We also solved the crystal structure of PUR repeat III ( Figure 4A; Figure 4—figure supplement 1 ) and found that it binds only weakly to DNA/RNA ( Figure 3G , H ) and unwinds dsDNA only slightly ( Figure 4C ) . Since in the crystal structure the C-terminal end of PUR repeat I-II is located on the opposite side of its nucleic-acid-binding surface ( Figure 7A , B , bottom ) , it is unlikely that PUR repeat III causes steric clashes interfering with the nucleic-acid binding by PUR repeat I-II . Hence , PUR repeat III might only facilitate dimerization , thereby guiding a second DNA-/RNA-binding domain ( PUR repeat I-II ) to another GGN motif further upstream or downstream on the dsDNA , where additional DNA-unwinding events could take place ( Figure 7C , bottom ) . How this effect of dimeric Pur-alpha is achieved on a molecular level and if unwinding of longer dsDNA fragments requires its joint action with helicases are main questions to be addressed in future . Escherichia coli BL21 ( DE3 ) cells transformed with pGEX-6P-1::Pur-alpha fragments were grown at 37°C in LB medium supplemented with 100 µg/ml ampicillin . For 15N-labeling of protein cells were grown in M9 minimal medium supplemented with 0 . 5 g/l 15NH4Cl . For selenomethionine-substituted protein , cells were grown in M9 minimal medium supplemented with an amino-acid mix of L-alanine , L-arginine , L-aspartic acid , L-cysteine , L-glutamate , L-glycine , L-histidine , L-isoleucine , L-leucine , L-lysine , L-phenylalanine , L-proline , L-serine , L-threonine , L-tyrosine , L-valine , and selenomethionine ( 100 mg/l each ) . After reaching an OD600 of 0 . 8 , cell cultures were cooled down to 18°C and expression was induced by adding 0 . 25 mM IPTG . Cells were harvested after 18 hr of expression . GST-tagged proteins were purified by GST-affinity chromatography ( GE Healthcare , Munich , Germany ) . After protease cleavage , the GST tag was removed by a glutathione-sepharose column . Nucleic acids were removed by using an anion-exchange Q column ( GE Healthcare ) followed by size exclusion chromatography with buffer containing 250 mM NaCl , 20 mM Hepes pH 8 . 0 . For cysteine-containing and for selenomethionine-substituted proteins 2 mM DTT was added to the buffer . For NMR experiments size exclusion chromatography was performed in 50 mM potassium phosphate buffer pH 7 . 0 and 200 mM NaCl ( NMR buffer ) . Absence of nucleic-acid contamination was confirmed by measuring the ratio of absorption at 260/280 nm ( Edelmann et al . , 2014 ) . To confirm proper protein folding of the Pur-alpha mutants CD spectra ( wavelength 190–260 nm ) were recorded with a JASCO-715 spectropolarimeter at 5°C in a 0 . 1-cm cuvette . Proteins were diluted in buffer containing 250 mM NaCl , 20 mM Hepes pH 8 . 0 , and 2 mM DTT to a final protein concentration of 30 µM in 300 µl total volume . Five scans were taken with a speed of 50 nm/min . Crystallization was carried out with freshly prepared selenomethionine-substituted Pur-alpha repeat I-II ( residues 40–185 ) in size exclusion buffer ( 250 mM NaCl , 20 mM Hepes pH 8 , 2 mM DTT ) . The protein was mixed with commercially purchased GCGGCGG ssDNA oligonucleotides , dissolved in Milli-Q H2O at a ratio 1:2 . 2 ( protein:DNA ) . The final protein concentration was 1 . 77 mg/ml . A drop size of 3 µl and a 2:1 mixture of protein-DNA complex and crystallization solution were used for hanging-drop vapor-diffusion at 21°C using 24-well EasyXtal Crystal Support plates ( Qiagen , Hilden , Germany ) . The crystallization solution contained 50 mM MES pH 5 . 2 , 500 mM ( NH4 ) 2SO4 , 1 mM TCEP , and 16% PEG400 . The total reservoir volume was 500 µl . Rod-shaped crystals of 160 x 20 µm size appeared within 4 days . Prior to data collection , crystals were cryoprotected in mother liquor and flash frozen in liquid nitrogen . Native dataset was recorded at 100 K at beamline ID23-2 ( European Synchrotron Radiation Facility [ESRF] Grenoble , France ) . Crystals diffracted up to 2 . 0 Å . Data were integrated and scaled with XDS ( Kabsch , 1993 ) . Structure was solved by molecular replacement with PHASER ( McCoy et al . , 2007 ) using the apo-structure of Drosophila Pur-alpha 40–185 ( PDB ID 3K44 ) as template and model building was manually completed using COOT ( Emsley et al . , 2010 ) . Refinement of the native data was performed with PHENIX ( Adams et al . , 2010 ) using NCS and TLS . The final model was analyzed with SFCHECK ( Vaguine et al . , 1999 ) , PHENIX , and REFMAC ( Murshudov et al . , 1997; , Terwilliger , 2002 ) . Superpositioning of the apo-structure with the DNA-complexed structure of Pur-alpha was performed with the superpose algorithm ( Krissinel and Henrick , 2004 ) of the program COOT . Images and movie of the crystal structure , superimpositions of the co-complex and apo-structure , as well as electrostatic surface potentials were prepared with PyMol ( Version 1 . 7; Schrodinger LLC . ; http://www . pymol . org/ ) . All crystallographic software was used from the SBGRID software bundle ( Morin et al . , 2013 ) . Structural model and dataset is available http://www . rcsb . org ( PDB-ID: 5FGP ) . Selenomethionine-substituted crystals of Pur-alpha repeat III ( residues 188–258 ) were grown at 4°C with a protein concentration of 0 . 5–2 mg/ml . The crystallization solution contained 50 mM MES pH 6 . 5 , 200 mM NaCl , 16% PEG 3350 , and 6% MPD . Plate-shaped crystals of approximately 70 × 70 × 10 µm size appeared within 2–4 days . For cryo-protection , crystals were shortly incubated in reservoir solution containing 30% ethylene glycol in two steps and then flash frozen in liquid nitrogen . Native dataset was recorded at 100 K at beamline ID14-1 [ESRF] . Crystals showed good diffraction up to 2 . 6 Å and belonged to space group P21 ( see Table 1 ) . The data were integrated and scaled with the XDS program package . Phases were obtained by molecular replacement using PHASER together with Borrelia burgdorferi Pur-alpha and Drosophila melanogaster Pur-alpha repeat I-II structures as a search model . Best results were achieved using a truncated version of the search models lacking the loop regions and poly-serine as amino-acid sequence . Parts of the initial model were built automatically with Buccaneer ( Cowtan , 2006 ) and manually completed using COOT . Refinement was performed with PHENIX using NCS with 6 monomers per asymmetric unit . Structural model and dataset is available http://www . rcsb . org ( PDB-ID: 5FGO ) . For RNA-labeling RNase-free buffers , materials , and reagents were used . Ten picomol of chemically synthesized DNA or RNA oligonucleotides were phosphorylated at the 5’-end with 10 pmol γ-32P ATP by T4 polynucleotide kinase ( New England Biolabs , Frankfurt , Germany ) with buffer A in a final volume of 20 µl . Labeling reaction was carried out at 37°C and stopped after 30 min by incubation at 70°C for 10 min . Labeled oligonucleotides were purified by a NucAway™ Spin column ( Ambion , Ulm , Germany ) and stored at -20°C . The protein-nucleic acid complexes were formed in RNase-free binding buffer containing 250 mM NaCl , 20 mM Hepes pH 8 . 0 , 3 mM MgCl2 , 4% glycerol , 2 mM DTT ) . Serial protein dilutions and a constant amount of radiolabeled nucleic acid ( 2 . 5 nM ) were incubated in a total reaction volume of 20 µl for 20 min at 21°C . DNA-binding experiments contained 25 µg/ml Salmon Sperm DNA , and RNA-binding experiments contained 100 µg/ml yeast tRNA competitor . Ten microliter of the reactions were loaded onto 6% TBE polyacrylamide gels . After electrophoresis ( 45 min , 100 V ) , gels were incubated for 15 min in fixing solution ( [v/v] 10% acetic acid , [v/v] 30% methanol ) , dried in a gel dryer ( BioRad , Munich , Germany ) and analyzed with radiograph films in a Protec Optimax developer ( Hohmann , Hannover , Germany ) . Sequences of oligonucleotides were as follows: MF0677 ssDNA/RNA , 5’-GGAGGTGGTGGAGGGAGAGAAAAG-3’; CGG ssDNA/RNA , 5’- ( CGG ) 8–3’ . For fluorescence-polarization measurements , protein-nucleic acid complexes were formed in buffer containing 500 mM NaCl , 20 mM Hepes pH 7 . 5 , 3 mM MgCl2 , 2 mM DTT ) . In comparison to EMSA , higher salt concentrations were used ( 500 mM versus 250 mM ) to allow for binding experiments at higher protein concentrations without aggregation of Pur-alpha . Serial protein dilutions and a constant amount of fluorescein-labeled MF0677 ssDNA or ssRNA ( 100 nM ) were incubated for 20 min at 21°C in a total reaction volume of 40 µl . DNA-binding reactions contained 25 µg/ml Salmon Sperm DNA and RNA-binding reactions contained 100 µg/ml yeast tRNA as competitor . Measurements were performed on an Envision Multilabel reader ( Perkinelmer ) . The excitation and emission wavelengths were 485 nm and 535 nm , respectively . The dissociation constant was calculated by fitting the data with the one-site binding model included in the program origin ( OriginLab ) . The experiment was performed as triplicates . Equation for one-site binding: y=Bmaxx/ ( k1+x ) . y = specific binding , x = ligand concentration , Bmax = maximum specific binding , k1 = equilibrium binding constant . All NMR spectra were recorded in NMR buffer with 5% D2O at 298 K using a Bruker Avance III spectrometer equipped with a TCI cryogenic probe head , at field strengths corresponding to 900 MHz proton Larmor frequency . To study DNA/RNA binding 1H , 15N HSQC NMR spectra were recorded of 15N-labeled protein ( 50 µM ) titrated with nucleic acids with different stoichiometric ratio of protein:nucleic acid ( 1:0 . 25 , 1:0 . 5 , 1:0 . 75 , 1:1 , 1:1 . 25 , 1:1 . 5 , 1:2 . 5 , and 1:5 ) . For every spectrum , 256 increments in the 15N indirect dimension with eight scans and an interscan delay of 1 s were acquired . Spectra were recorded and processed with Topspin 3 . 2 ( Bruker ) and analyzed with CCPNMR analysis ( Vranken et al . , 2005 ) . Unwinding assays were carried out according to reference ( Darbinian et al . , 2001 ) . A dsDNA substrate was prepared by annealing a complementary 18-mer oligonucleotide to a GGN motif of the M13mp18 ssDNA plasmid . The 18-mer was labeled with γ-32P ATP . Protein dilutions were added to a constant amount of dsDNA substrate ( 100 ng ) in binding buffer composed of 150 mM NaCl , 20 mM Hepes pH 8 . 0 . Samples were incubated at 37°C for 1 hr . The unwinding reaction was stopped by adding SDS to a final concentration of ( v/v ) 0 . 3% . Samples were run on 9% native polyacrylamide gels in 1x TBE buffer for 150 min at 200 V . Gels were incubated for 15 min in fixing solution ( [v/v] 10% acetic acid , [v/v] 30% methanol ) , dried and analyzed with radiograph films . The sequence of the 18-mer oligonucleotide was as follows: 5’-TCAGAGCCGCCACCCTCA-3’ . Filter-binding assays were performed as described ( Wong and Lohman , 1993 ) . Protein was titrated to a constant amount of 1 µM MF0677 ssDNA ( thereof 2 . 5 nM radiolabeled ) in a final volume of 80 µl and incubated for 20 min at 21 °C in binding buffer 150 mM NaCl , 20 mM Hepes pH 8 . 0 . Nitrocellulose filter ( Roth , Karlsruhe , Germany ) was presoaked for 10 min in 0 . 4 M KOH followed by intensive washing with Milli-Q H2O . Nitrocellulose and nylon filters ( Roth ) were then equilibrated in binding buffer for 15 min . Both filters ( nitrocellulose , top; nylon filter , bottom ) were placed into a dot-blot apparatus ( BioRad ) . Vacuum was applied and the wells were washed once with 80-µl binding buffer before and after samples were loaded . The nitrocellulose filters were analyzed using a phosphor imager system to measure the retained radiolabeled oligonucleotides on the nitrocellulose filter . Quantification was done using the dot blot analyzer plug-in of the ImageJ 1 . 47v software ( National Institute of Health , USA ) . Transgenic flies expressing rCGG90 repeats were obtained as previously described ( Jin et al . , 2003 ) . The pUAST constructs were generated by cloning cDNA of full-length Drosophila Pur-alpha into the pUAST transformation vectors . The constructs were confirmed by DNA sequencing and then injected in a w1118 strain by standard methods . Fly lines were grown on standard medium with yeast paste added . All crosses were performed at 25°C . For scanning electron microscopy ( SEM ) images , whole flies were dehydrated in ethanol , dried with hexamethyldisilazane ( Sigma-Aldrich , Hamburg and Seezle , Germany ) , and analyzed with an ISI DS-130 LaB6 SEM/STEM microscope .	Some proteins perform several different tasks inside cells . This is the case for a protein called Pur-alpha , which is essential for neurons to work correctly . For example , Pur-alpha can bind to DNA to regulate gene activity . It also binds to RNA molecules , which are copies of a gene , and helps to distribute them within the neuron . In humans , there are several neurodegenerative diseases in which Pur-alpha is involved . One example is the Fragile X-associated Tremor/Ataxia Syndrome ( FXTAS ) , which causes memory and movement problems . Experiments with isolated proteins and double-stranded DNA show that Pur-alpha is able to separate the two DNA strands . But it was not clear how this DNA unwinding occurs , and the biological significance of this activity was unknown . Other questions also remained unanswered: how does Pur-alpha recognize DNA and RNA ? Does the loss of Pur-alpha’s binding to DNA and RNA contribute to neurodegenerative diseases ? To address these questions , Weber et al . obtained Pur-alpha from the fruit fly and crystallized the protein bound to DNA . A technique called X-ray crystallography was then used to determine the three-dimensional structure of the Pur-alpha/DNA complex in fine enough detail to work out the position of individual atoms . Based on this structure , Weber et al . could introduce mutations that alter the DNA- and RNA-binding region of the protein to investigate the binding mechanism . The crystal structure and experiments with normal and mutant Pur-alpha protein revealed how it unwinds double-stranded DNA: binding of Pur-alpha to DNA causes a strong twist of the DNA molecule , which contributes to separating the strands . Further experiments in fruit flies revealed that both the DNA-unwinding activity and the ability of Pur-alpha to bind DNA/RNA are needed for the protein to work correctly in neurons . Because Pur-alpha is involved in a range of different processes inside cells , a future goal is to identify the DNA and RNA sequences it specifically binds to . This information , together with the insights gained from Weber et al . ’s study , should advance our understanding of why Pur-alpha is essential for maintaining neurons .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Structural basis of nucleic-acid recognition and double-strand unwinding by the essential neuronal protein Pur-alpha
The mTOR complex 1 ( mTORC1 ) and endoplasmic reticulum ( ER ) stress pathways are critical regulators of intestinal inflammation and colon cancer growth . Sestrins are stress-inducible proteins , which suppress both mTORC1 and ER stress; however , the role of Sestrins in colon physiology and tumorigenesis has been elusive due to the lack of studies in human tissues or in appropriate animal models . In this study , we show that human SESN2 expression is elevated in the colon of ulcerative colitis patients but is lost upon p53 inactivation during colon carcinogenesis . In mouse colon , Sestrin2 was critical for limiting ER stress and promoting the recovery of epithelial cells after inflammatory injury . During colitis-promoted tumorigenesis , Sestrin2 was shown to be an important mediator of p53’s control over mTORC1 signaling and tumor cell growth . These results highlight Sestrin2 as a novel tumor suppressor , whose downregulation can accelerate both colitis and colon carcinogenesis . Colorectal carcinoma ( CRC ) is an important contributor to cancer mortality and morbidity . The lifetime risk of developing CRC in the US is 4–5% , and approximately one-third of CRC patients die from the disease ( Siegel et al , 2015 ) . Although the pathogenetic mechanisms underlying CRC development are complex and heterogeneous , several critical genes and pathways important in its initiation and progression are well characterized , such as Wnt-APC , Ras-MAPK , p53 and DNA repair pathways ( Fearon , 2011 ) . In addition to these components , mammalian target of rapamycin complex 1 ( mTORC1 ) , a protein kinase that is essential for cell growth ( Hay and Sonenberg , 2004; Zoncu et al , 2011 ) , was recently found to play a key tumorigenic role during CRC development induced by either colitis ( Thiem et al , 2013 ) or a genetic mutation ( Faller et al , 2015; Hardiman et al , 2014 ) . In addition to promoting cancer cell growth , mTORC1 hyperactivation can lead to unrestricted protein synthesis , resulting in the accumulation of unfolded protein , endoplasmic reticulum ( ER ) stress and tissue injury ( Ozcan et al , 2008; Park et al , 2014; Young et al , 2013 ) , which together can contribute to tumor progression ( Wang and Kaufman , 2014 ) . ER stress has been shown to be critically involved in the pathogenesis of colitis and colon inflammation ( Bertolotti et al , 2001; Cao et al , 2013; Kaser et al , 2008 ) , which is an important risk factor of CRC development ( Thorsteinsdottir et al , 2011 ) and a well-characterized tumor promoter ( Grivennikov et al , 2012; Terzic et al , 2010 ) . Mechanisms of how the mTORC1 and ER stress signaling pathways are regulated in the colon , especially during colon injury , inflammation and tumorigenesis , are poorly understood . Sestrins are a family of stress-inducible proteins that are widely conserved throughout animal species ( Lee et al , 2013 ) . Sestrins were originally identified as a target of the tumor suppressor p53 ( Budanov et al , 2002; Velasco-Miguel et al , 1999 ) . Sestrins have two important functions , suppressing reactive oxygen species ( ROS ) ( Budanov et al , 2004 ) and inhibiting mTORC1 ( Budanov and Karin , 2008 ) . The ROS-suppressing effect of Sestrins is dependent , at least partially , on mTORC1 inhibition , which promotes autophagic degradation of dysfunctional mitochondria or an Nrf2 inhibitor Keap1 ( Bae et al , 2013; Lee et al , 2010; Woo et al , 2009 ) . However , Sestrin can also function as an active oxidoreductase that can directly detoxify ROS such as alkylhydroperoxides ( Kim et al , 2015a ) . Sestrins inhibit mTORC1 through the activation of AMP-activated protein kinase ( AMPK ) and the subsequent inactivation of Rheb GTPases ( Budanov and Karin , 2008; Sanli et al , 2012 ) . Independently of AMPK , Sestrins can also inhibit Rag GTPases ( Chantranupong et al , 2014; Kim et al , 2015b; Parmigiani et al , 2014; Peng et al , 2014 ) , which are essential for mTORC1 activity . Sestrin-mediated inhibition of mTORC1 is also critical for limiting protein synthesis upon unfolded protein accumulation ( Bruning et al , 2013; Park et al , 2014 ) or amino acid starvation ( Peng et al , 2014; Wolfson et al , 2015; Ye et al , 2015 ) , thereby suppressing ER stress or nutrient crisis . In light of these important cellular functions , the present study assessed if Sestrin functions as a coordinator of mTORC1 and ER stress signaling pathways in the colon during intestinal inflammation and carcinogenesis . Our data collected from patient samples , mouse models of colitis and colitis-associated cancer , cultured colon cancer cell lines as well as data mining of large-scale transcriptome analyses , concertedly indicate that Sestrin2 , a member of the Sestrin family , is important for proper regulation of mTORC1 and ER stress pathways during colon injury , and thereby functions as a suppressor of colitis and colon cancer development . Increased ER stress and excessive ROS accumulation are hallmarks of colon inflammation and colitis ( Fritz et al , 2011; Zhu and Li , 2012 ) . Sestrins can be induced upon either of these stresses and are critical to dampen their detrimental consequences ( Bruning et al , 2013; Lee et al , 2013; Park et al , 2014 ) . Therefore , to understand the role of Sestrins in colitis , expression of Sestrins was analyzed in tissues isolated from patients with ulcerative colitis ( UC ) . Sestrin1 mRNA ( SESN1 ) was unaltered in UC ( Figure 1A ) ; however , expression of SESN2 ( Figure 1B ) and SESN3 ( Figure 1C ) was significantly increased in the intestine of patients with UC . 10 . 7554/eLife . 12204 . 003Figure 1 . Protective function of Sestrin2 against colon injury . ( A-C ) Upregulation of human SESN2 and SESN3 expression in ulcerative colitis ( UC ) . mRNA expression of human SESN1-3 was analyzed through quantitative RT-PCR of non-inflamed ( Normal ) and inflamed ( UC ) colon tissues from patients with UC ( n=10; mean ± s . e . m . ) . These samples were histologically confirmed and formerly described ( Xue et al , 2013 ) . ( D-M ) Loss of Sestrin2 impairs recovery from DSS-induced colitis in mice . 6-month-old WT and Sesn2-/- mice ( n=4 each ) were treated with 3% DSS in drinking water for 6 days ( arrows ) , followed by 6 days of regular water . Body weight was measured over 12 days ( D; mean ± s . e . m . ) . At the final day of the experiment , mice were sacrificed and colon length was measured ( E ) . The data are shown as the mean ± s . e . m . The colons were isolated and fixed for H&E staining ( F ) , TUNEL staining ( G ) , PCNA staining ( H ) and F4/80 staining ( I ) . The levels of the indicated mRNAs , which are indicative of active inflammation , were quantified by real-time PCR ( J-M; mean ± s . e . m . ) . p<0 . 05 , p<0 . 01 , p<0 . 001 . P values are from Student’s t-test . Scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 00310 . 7554/eLife . 12204 . 004Figure 1—figure supplement 1 . Hypersensitivity of Sesn2-/-/Sesn3-/- mice against DSS-induced colon injury . 1-year-old WT and Sesn2-/-/Sesn3-/- mice ( n=4 each ) were treated with 3% DSS in drinking water for 7 days ( arrows in A ) , followed by 5 days of regular water . Body weight was measured over 12 days ( A ) . At the final day of the experiment , mice were sacrificed and colon length was measured ( B ) . The colons were isolated and fixed for H&E staining ( C ) , TUNEL staining ( D ) , PCNA staining ( E ) and F4/80 staining ( F ) . Data are shown as the mean ± s . e . m . p<0 . 01 . P values are from Student’s t-test . Scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 00410 . 7554/eLife . 12204 . 005Figure 1—figure supplement 2 . Acute colon injury is comparable between WT and Sesn2-/- mice during DSS treatment . 2-month-old WT and Sesn2-/- mice ( n=4 each ) were treated with 3% DSS in drinking water for 7 days ( DSS – 7d only ) . At this time point , mice were sacrificed and colon length was measured ( A ) . The data are shown as the mean ± s . e . m . The colons were isolated and fixed for H&E staining ( B ) . The levels of Tnfa and Xbp1s mRNAs , which are respectively indicative of active inflammation and ER stress , were quantified by real-time PCR ( C , D; mean ± s . e . m . ) . Scale bars , 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 005 To examine whether colitis-induced Sestrin2 and Sestrin3 play a physiological role in maintaining intestinal homeostasis , WT and Sesn2-/-/Sesn3-/- mice were treated with dextran sulfate sodium ( DSS ) in the drinking water to induce colitis . DSS treatment for 7 days led to substantial weight loss in both WT and Sesn2-/-/Sesn3-/- mice ( Figure 1—figure supplement 1A ) . After placing back on regular water , WT mice recovered their body weight ( Figure 1—figure supplement 1A ) . However , Sesn2-/-/Sesn3-/- mice did not show any recovery and continued to lose body weight until the experimental endpoint ( 5 days during the recovery phase; Figure 1—figure supplement 1A ) . Sesn2-/-/Sesn3-/- mice also showed a dramatic decrease in colon length when compared to WT mice ( Figure 1—figure supplement 1B ) , indicative of strongly exacerbated DSS-induced colitis . Histological examination of colon tissue sections also revealed significant epithelial degeneration in Sesn2-/-/Sesn3-/- mice following the 5 days of recovery from the 7-day DSS treatment , while WT mice exhibited substantial regeneration of epithelial structure at the same time point ( Figure 1—figure supplement 1C ) . The increased susceptibility of Sesn2-/-/Sesn3-/- mice to DSS-induced injury ( Figure 1—figure supplement 1A–C ) was recapitulated in Sesn2-/- mice; although both WT and Sesn2-/- mice develop severe colitis with one week of DSS treatment ( Figure 1D and Figure 1—figure supplement 2 ) , WT mice successfully recovered from injury after one additional week of regular water treatment , while Sesn2-/- mice did not ( Figure 1D–F ) . These results demonstrate a critical role for Sestrin2 in restoring intestinal homeostasis after injury . We examined molecular markers for cell death and inflammation in the colons of WT , Sesn2-/- and Sesn2-/-/Sesn3-/- mice after DSS treatment . At 5 days after DSS injury , WT mice displayed a very small number of apoptotic cells ( Figure 1G and Figure 1—figure supplement 1D ) , consistent with the histological observation showing that the colon epithelium had been restored ( Figure 1F and Figure 1—figure supplement 1C ) . However , a significant number of apoptotic cells were observed in the colon epithelium of both Sesn2-/- and Sesn2-/-/Sesn3-/- mice ( Figure 1G and Figure 1—figure supplement 1D ) , consistent with the degenerative phenotypes observed in these mice . Proliferating cell nuclear antigen ( PCNA ) staining of WT colon displayed a normal pattern of cell proliferation; PCNA staining is confined to the base of colon crypts in WT mice ( Figure 1H and Figure 1—figure supplement 1E ) , where epithelial progenitor cells are undergoing homeostatic proliferation that maintains normal turnover of the epithelium . However , the colon epithelium of both Sesn2-/- and Sesn2-/-/Sesn3-/- mice exhibited an increased number of PCNA-positive cells throughout the degenerated epithelium ( Figure 1H and Figure 1—figure supplement 1E ) . This result suggests that , in order to compensate for the apoptotic loss of epithelial cells , colonocytes of both Sesn2-/- and Sesn2-/-/Sesn3-/- mice were undergoing active proliferation . Immunohistochemical staining of macrophage marker F4/80 ( Figure 1I and Figure 1—figure supplement 1F ) , as well as quantitative RT-PCR examination of inflammation markers Tnfa ( Figure 1J ) , Il6 ( Figure 1K ) , Il1b ( Figure 1L ) and Il10 ( Figure 1M ) , show that Sesn2-/- mice had increased the levels of colon inflammation after DSS injury . These data collectively indicate that Sestrin2 deficiency exacerbates DSS-induced colon damage and inflammation . Inflammatory cytokine signaling instigated by bone marrow-derived immune cells , such as macrophages , is known to be important for the progression of colitis as well as colon cancer ( Terzic et al , 2010 ) . We examined whether the expression of Sestrin2 in the bone marrow-derived hematopoietic compartment is important for the protective role of Sestrin2 in colitis . For this purpose , reciprocal bone marrow chimera experiments were performed: WT bone marrow was introduced into lethally irradiated Sesn2-/- mice ( WT→Sn2 ) while Sesn2-/- bone marrow was introduced into lethally irradiated WT mice ( Sn2→WT ) . Both groups of chimeric mice were subjected to DSS administration ( Figure 2A ) . Interestingly , Sn2→WT mice were similar to WT mice and able to recover from DSS-induced injury ( Figure 2A , B ) . However , more than 70% of WT→Sn2 mice were dead at 6 days following DSS treatment ( Figure 2B ) . WT→Sn2 mice , but not Sn2→WT mice , also experienced dramatic weight loss during the recovery phase following DSS treatment ( Figure 2A ) . Histological examination of the surviving mice revealed that , although the epithelial structure of Sn2→WT mouse colon was nearly completely restored at 6 days after the DSS treatment , WT→Sn2 colon epithelium was degenerated and marked by complete loss of epithelial cells and a robust increase in infiltrating immune cells ( Figure 2C ) . Genotyping of spleen tissues in surviving WT→Sn2 and Sn2→WT mice confirmed that the hematopoietic compartment of recipient mice had been completely substituted with bone marrow cells of the donor ( Figure 2D ) . These results indicate that the extra-hematopoietic presence of Sestrin2 , such as in epithelial cells , is critical for the maintenance of intestinal homeostasis during colitis . 10 . 7554/eLife . 12204 . 006Figure 2 . Sestrin2 prevents colitis-associated ER stress in colonic epithelia . ( A-D ) Sestrin2 expression in the extra-hematopoietic compartment is critical for the resolution of DSS-induced colitis . 3-month-old WT and Sesn2-/- mice ( n=7 each ) were subjected to lethal irradiation and injected with bone marrow cells from age-matched Sesn2-/- ( Sn2→WT ) and WT ( WT→Sn2 ) mice , respectively . After 1 month , mice were subjected to DSS administration as indicated in the panel A . Body weight was measured over 15 days ( A ) . Data are shown as mean ± s . e . m . Percent survival was calculated for each day ( B ) . The colons of surviving mice at the final day of experiment were isolated and fixed for H&E staining ( C ) . The spleens of surviving mice were genotyped for WT ( upper band , ~450bp ) and Sesn2-KO ( lower band , ~200bp ) alleles ( D ) . ( E-H ) Loss of Sesn2 aggravates colitis-induced ER stress in colon . Expression or phosphorylation of ER stress signaling markers were analyzed from indicated mice ( described in Figure 1D-M ) through immunoblotting ( E ) , real-time PCR ( F ) or immunohistochemistry ( G , H ) . Data are shown as mean ± s . e . m . p<0 . 05 , p<0 . 01 . P values are from Student’s t-test . Scale bars , 200 μm ( black ) , 100 μm ( white ) . Molecular weight markers are indicated in bp ( D ) or kDa ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 00610 . 7554/eLife . 12204 . 007Figure 2—figure supplement 1 . Induction of ER stress and Sestrin2 upon DSS treatment . 4-month-old WT mice were treated with 3% DSS in drinking water for 7 days , followed by 7 days of regular water ( RW ) . Mice were sacrificed before the treatment ( Con; n=3 ) , right after 7 days of DSS treatment ( DSS 7d only; n=5 ) or after completion of the experiment ( DSS 7d + RW 7d; n=3 ) . Expressions of ER stress signaling markers ( A ) , inflammation markers ( B ) and Sestrin2 ( C , D ) were analyzed from indicated mice through real-time PCR ( A-C ) or immunoblotting ( D ) . Data are shown as mean ± s . e . m . p<0 . 05 , p<0 . 01 , p<0 . 001 . P values are from Student’s t-test between control and indicated groups . Molecular weight markers are indicated in kDa ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 00710 . 7554/eLife . 12204 . 008Figure 2—figure supplement 2 . Increased ER stress in colon epithelia of Sesn2-/- and Sesn2-/-/Sesn3-/- mice during DSS-induced colon injury . Expression or phosphorylation of ER stress signaling markers were analyzed from indicated mice through immunoblotting ( A , B ) , real-time PCR ( C; n=4 ) or immunohistochemistry ( D , E ) . Data are shown as mean ± s . e . m . p<0 . 05 . P values are from Student’s t-test . Scale bars , 100 μm . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 00810 . 7554/eLife . 12204 . 009Figure 2—figure supplement 3 . Uncropped images of blots . Red boxes indicate the cropped regions . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 009 Sestrin2 protects cells and tissues from ER stress and its pathological sequelae such as metabolic abnormalities , tissue inflammation and apoptotic cell death ( Park et al , 2014 ) . DSS treatment was recently found to induce ER stress in colon epithelia ( Cao et al , 2013 ) , and consistent with this report , we observed that DSS treatment can induce prolonged ER stress signaling that is associated with modestly elevated Sestrin2 expression ( Figure 2—figure supplement 1 ) . Therefore , it is possible that Sestrin2 protects colon epithelium by allowing colonocytes to cope with DSS-induced ER stress insults , and thereby promoting colonic recovery after DSS injury . To test this idea , we analyzed the level of ER stress in WT and Sesn2-/- colon epithelium recovering from DSS injury . Immunoblot analyses of colon tissue showed that Sesn2-/- and Sesn2-/-/Sesn3-/- mice have elevated phosphorylation of pancreatic ER kinase ( PERK ) and increased expression of BiP and CHOP ( Figure 2E and Figure 2—figure supplement 2A , B ) , when compared to WT mice . mRNA expression analysis for ER stress target genes , including spliced XBP1 ( XBP1s ) and BiP cofactor ERdj4 , were also robustly upregulated in the colon of Sesn2-/- mice ( Figure 2F and Figure 2—figure supplement 2C ) . Immunostaining of BiP and CHOP also confirmed the presence of prominent ER stress in the damaged colon epithelium of Sesn2-/- mice ( Figure 2G , H and Figure 2—figure supplement 2D , E ) . These results collectively indicate that endogenous Sestrin2 is critical for the suppression of colon ER stress after DSS insults . Our results show that Sestrin2 expression is induced during UC as a protective mechanism against colonic ER stress and epithelial degeneration . Since pre-existing colitis ( Terzic et al , 2010 ) or tumor-elicited inflammation ( Grivennikov et al , 2012 ) is important for the progression of colon cancer , it is possible that Sestrin2 expression is lost or downregulated during colon carcinogenesis . Thus , we examined the Oncomine database , which contains diverse large-scale genomic and transcriptomic analyses of normal and tumor tissues ( Rhodes et al , 2007 ) , to determine if there is a differential expression of Sestrins between normal and tumor tissues . Surprisingly , in virtually all of the colon cancer transcriptome studies available within the database , which were conducted in diverse platforms using different patient tissues , human Sestrin2 ( SESN2 ) mRNA expression was strongly downregulated in colon adenocarcinoma tissues when compared to normal colon controls ( Figure 3A–G ) ( Gaspar et al , 2008; Graudens et al , 2006; Hong et al , 2010; Kaiser et al , 2007; Cancer Genome Atlas Network , 2012; Skrzypczak et al , 2010 ) . The magnitude of SESN2 suppression in tumors is often among the top 1–5% of all suppressed genes ( Figure 3A ) . Considering that these data are collected from a variety of different human samples , the extent of the difference was very strong ( all studies indicate p<10–4 or much lower ) . In contrast , other major cell type markers , such as Villin ( VIL1 , enterocytes ) , DLL1 ( progenitor cells ) , F4/80 ( EMR1 , macrophages ) , Gr-1 ( LY6G5B , leukocytes ) , or β-catenin ( CTNNB , colon epithelia ) , did not show such strong suppression ( Figure 3—figure supplement 1A–E ) , while the stem cell marker LGR5 was rather upregulated in tumor samples ( Figure 3—figure supplement 1A–E ) . These results indicate that the downregulation of SESN2 mRNA in colon cancer is specific and not an indirect consequence of different compositions of cell subtypes between normal and tumor tissues . 10 . 7554/eLife . 12204 . 010Figure 3 . Downregulation of SESN2 in human colon cancer tissues . ( A ) Oncomine analysis of Sestrin-family gene expression in normal and cancer tissues of different types . Gene summary views for SESN2 , SESN1 and SESN3 genes are shown . Cell color is determined by the best gene rank percentile for the analyses within the cell , as described below the table . Thresholds for gene rank , fold change and P value are also described below the table . Reduction of SESN2 in colorectal cancer tissue ( highlighted in the green box ) was one of the most significant alterations . ( B-G ) human SESN2 mRNA expression in normal colon and colon cancer tissues , derived from six independent studies ( B-G; total n=40 , 78 , 58 , 82 , 80 and 165 , respectively ) conducted in different platforms ( Gaspar et al , 2008; Graudens et al , 2006; Hong et al , 2010; Kaiser et al , 2007; Cancer Genome Atlas Network , 2012; Skrzypczak et al , 2010 ) . ( H ) DNA copy number analysis of human SESN2 gene in normal blood , normal colon and colon cancer tissues ( total n=975 ) , derived from TCGA dataset ( Cancer Genome Atlas Network , 2012 ) . Colon cancer staging in F , G and H is according to the TNM staging system from the American Joint Committee on Cancer ( AJCC ) . All data are shown as the mean ± s . e . m . P values between normal and cancer tissues , calculated from Student’s t-test , are all below 10–4 ( B-H ) . For TCGA dataset , P values between normal and cancer tissues are 1 . 6 x 10–30 ( G ) and 3 . 7 x 10–58 ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01010 . 7554/eLife . 12204 . 011Figure 3—figure supplement 1 . Expression of cell-type specific markers in human colon cancer tissues . ( A-E ) Levels of the following mRNAs were analyzed in the studies described in Figure 3 . SESN2 , Sestrin2; VIL1 , Villin – Expressed in Enterocytes; DLL1 , Delta-like 1 – Expressed in Progenitor cells; EMR1 , F4/80 – Expressed in Macrophages; LY6G5B , Gr-1 – Expressed in Leukocytes; CTNNB , β-catenin – Expressed in Colonic Epithelia; LGR5 , Gpr49 – Expressed in Colonic Stem Cells . All data are shown as the mean ± s . e . m . p<0 . 05 , p<0 . 01 , p<0 . 001 . N/D , not determined . P values were calculated from Student’s t-test between normal and cancer tissues . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 011 SESN2 genome copy is also significantly reduced in colon cancer , and decreases further as the colon cancer progresses ( Figure 3H ) . However , the extent of copy number loss was very small ( ~10% ) , suggesting that transcriptional downregulation , rather than the loss of genomic information , is the major mechanism of SESN2 inhibition during colon cancer progression . SESN2 is a transcriptional target of tumor suppressor p53 ( Budanov et al , 2002 ) , which is one of the most frequently mutated genes in colon cancer ( Fearon , 2011 ) . To test whether p53 mutation plays any role in regulating SESN2 expression during colon carcinogenesis , we analyzed the cancer genome atlas ( TCGA ) dataset ( Cancer Genome Atlas Network , 2012 ) by partitioning tumors based on p53 status . TCGA dataset has comprehensive information regarding the genomic status of each tumor , determined by whole genome/exome sequencing . From this database , we were able to classify all the colon tumor samples into the two groups . One group , designated as 'p53-mutated' , has one or more missense or nonsense coding sequence mutations in the TP53 gene . The second group , designated as 'p53-unknown' , is classified as such because it does not reveal any coding sequence mutations in the TP53 gene; however , it is still possible that these tumors contain TP53 mutations in essential non-coding regions ( e . g . promoters , enhancers or introns ) or other genomic or epigenetic alterations that can lead to functional p53 inactivation ( e . g . MDM2 overexpression ) . Expressions of SESN2 and other known p53 target genes were then analyzed in three different groups: normal colons , 'p53-unknown' tumors and 'p53-mutated' tumors ( Figure 4 and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 12204 . 012Figure 4 . Correlation between p53 status and SESN2 expression in human colon cancer . TCGA colon/colon cancer dataset ( total n=165 ) was partitioned according to the p53 status ( Cancer Genome Atlas Network , 2012 ) . Normal colon tissues do not reveal any TP53 mutation ( n=19 in the dataset ) , while ~43% of tumor samples in the gene expression dataset ( n=63 out of total 146 tumor samples ) , designated as 'p53-mutated' , identified missense or nonsense point mutations in the TP53 coding region . Other tumor samples ( n=83 ) are designated as 'p53-unknown' . ( A ) SESN2 expression was analyzed in indicated tissues . Data are shown as the mean ± s . e . m . p<0 . 01 , *p<0 . 001 . P values were calculated from Student’s t-test . ( B , C ) Expression of SESN2 was analyzed in three different probes , and their correlations were visualized by a scatter plot of individual patient tissue samples . Trend lines were approximated through linear regression . Y axis is in a log scale . Normal colon samples are in blue , 'p53-unknown' tumor samples are in gray , and 'p53-mutated' tumor samples are in red . Pearson’s correlation coefficients ( r ) with P values were calculated and presented . p<0 . 0001 . ( D-F ) Expression of SESN2 , CDKN1A , GADD45A and MDM2 was analyzed and their correlations were analyzed as described above for B and C . p<0 . 01 , *p<0 . 001 , *p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01210 . 7554/eLife . 12204 . 013Figure 4—figure supplement 1 . Correlation between p53 status and expression of CDKN1A , GADD45A , MDM2 , BAX , PUMA ( BBC3 ) , p53AIP1 ( TP53AIP1 ) , TSC2 , AMPKβ ( PRKAB1 , PRKAB2 ) and PTEN in human colon cancer . ( A-J ) Gene expression was analyzed in indicated tissues , using the same dataset as in Figure 4 . The data are shown as the mean ± s . e . m . NS , not significant; p<0 . 05 , **p<0 . 001 . P values were calculated from Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 013 The analyses demonstrated that SESN2 expression is significantly reduced in 'p53-mutated' tumors when compared to 'p53-unknown' tumors ( Figure 4A ) , suggesting a role of p53 in controlling SESN2 expression . This reduction was consistently observed in three independent SESN2 probes ( Figure 4A–C ) . Strikingly , there were almost no overlaps of SESN2 expression levels between the normal colon and 'p53-mutated' tumor groups ( Figure 4B , C ) , while SESN2 levels in 'p53-unknown' tumors overlap with both groups ( Figure 4B , C ) . Other known p53 target genes , such as CDKN1A ( p21 ) , GADD45A and MDM2 ( Riley et al , 2008 ) , did not show this strong correlation: for these genes , considerable overlaps were found between the normal colon and 'p53-mutated' tumor groups in individual samples ( Figure 4D–F ) . Nevertheless , expression levels of these genes have a general positive correlation with expression of SESN2 in individual samples ( Figure 4D–F ) , suggesting that SESN2 , CDKN1A , GADD45A and MDM2 are all regulated through the same p53-dependent mechanism . Indeed , all of these genes are differentially expressed between 'p53-unknown' and 'p53-mutated' tumor groups ( Figure 4A and Figure 4—figure supplement 1A–C ) . Interestingly , some p53 target genes , such as apoptosis mediators BAX , PUMA ( BBC3 ) and p53AIP1 ( TP53AIP1 ) ( Riley et al , 2008 ) or mTORC1 regulators TSC2 , AMPKβ ( PRKAB1 and PRKAB2 ) and PTEN ( Feng et al , 2007 ) , did not show differential expression between the 'p53-unknown' and 'p53-mutated' tumor groups ( Figure 4—figure supplement 1D–J ) , suggesting that the effects of p53 mutations on these genes are minimal in the pathological context of colon carcinogenesis . These results collectively highlight SESN2 as a clinically relevant target of p53 during colon carcinogenesis . To more precisely establish the cause-effect relationship between Sestrin2 and colon cancer development , Sesn2-/- mice were further assessed . The analysis of two-year-old WT and Sesn2-/- mice ( n=6 each ) did not reveal any noticeable colon tumors , suggesting that Sesn2 may not be a classical tumor suppressor gene whose homozygous deletion is sufficient to induce spontaneous cancer development . However , given the strong correlation between SESN2 expression and colon cancer tumorigenesis in humans ( Figure 3 ) , we reasoned that endogenous Sestrin2 may play a role in attenuating tumor development and progression . Thus , Sesn2-/- mice were subjected to an established protocol of colitis-associated colon cancer induction; in this model , colon carcinogenesis is initiated by azoxymethane ( AOM ) administration and promoted by repeated colon injury induced by 2 . 5% DSS . However , none of Sesn2-/- mice ( n=7 ) survived the first round of colitis induction ( Figure 5A ) , while WT mice were able to survive the entirety of the treatment . This is consistent with our findings that Sestrin2 has an important physiological role in maintaining epithelial integrity during colon injury ( Figures 1–2 ) . 10 . 7554/eLife . 12204 . 014Figure 5 . Loss of Sesn2 promotes colon tumor growth in mice . ( A ) 2-month-old WT and Sesn2-/- mice were exposed to a standard protocol of azoxymethane ( AOM ) -dextran sulfate sodium ( DSS ) -induced colon carcinogenesis as outlined in this figure panel . However , none of the Sesn2-/- mice ( n=7 ) survived after the first round of administration of DSS , suggesting that they are hypersensitive to DSS-induced colitis . ( B ) WT and Sesn2-/- mice ( n=11 each ) were subjected to a modified protocol of AOM-DSS-induced colon cancer . By reducing the dose of DSS to 1 . 5% , we were able to keep a substantial number of Sesn2-/- mice ( n=9 ) alive until the experimental endpoint . Tumor incubation period was extended to 100 days to compensate for the lower dose of DSS treatment . ( C-G ) After completion of the experiment , colons were examined under a dissection microscope ( C ) , and tumor number ( D ) , average tumor size ( E ) , size of individual tumors ( F ) and total tumor burden ( G ) were analyzed and presented as means from each mouse and as mean ± s . d . of the whole groups . ( H-N ) Sestrin2-deficient tumors exhibit increased proliferation . Colon tumor ( T ) and normal colon ( N ) tissues of indicated mice were subjected to immunohistochemistry of β-catenin ( H ) , PCNA ( I ) and γ-H2AX ( J ) or TUNEL staining ( K ) . PCNA- ( L ) , γ-H2AX- ( M ) and TUNEL-positive ( N ) cells from indicated tissues were quantified and presented as mean ± s . e . m . NS , not significant; p<0 . 05 , p<0 . 01 , p<0 . 001 . P values are from Student’s t-test . Scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 014 To overcome the DSS-induced lethality of Sesn2-/- mice , a lower dose of DSS ( 1 . 5% ) was administered to both control and experimental groups in the colitis-associated cancer model ( Figure 5B ) . This modification enabled both WT and Sesn2-/- groups to survive through three inflammatory and recovery phases , and develop macroscopically visible colon tumors at 100 days after AOM injection ( Figure 5C ) . There was no substantial difference in the number of tumors between WT and Sesn2-/- mice ( Figure 5D ) . However , tumor size ( Figure 5E , F ) and burden ( Figure 5G ) were dramatically increased in Sesn2-/- mice , suggesting that Sestrin2 loss promoted tumor growth and progression in the colitis-associated cancer model . Tumors developed in both WT and Sesn2-/- mice displayed classical characteristics of colon adenomas , such as nuclear β-catenin staining ( Figure 5H ) , increased cell proliferation ( Figure 5I; PCNA staining ) , DNA damage ( Figure 5J; γ-H2AX staining ) and apoptotic cell death ( Figure 5K; TUNEL staining ) . Quantification of PCNA- , γ-H2AX- and TUNEL-positive cells showed that tumor cell proliferation ( Figure 5L ) is significantly increased in tumors isolated from Sesn2-/- mice , while DNA damage ( Figure 5M ) and apoptosis ( Figure 5N ) were not significantly altered by the loss of Sestrin2 . Using the reciprocal bone marrow chimera experiments followed by the AOM-DSS treatments ( Figure 6—figure supplement 1 ) , we assessed whether the expression of Sestrin2 in the hematopoietic or extra-hematopoietic compartment is important for the tumor-suppressive role of Sestrin2 in colon cancer . Compared to the Sn2→WT mice , WT→Sn2 mice exhibited strongly exacerbated tumor growth phenotypes , which was evident in both tumor number and size ( Figure 6A–D ) , indicating that the extra-hematopoietic expression of Sestrin2 , such as in colon epithelia , is critical for colon tumor suppression . 10 . 7554/eLife . 12204 . 015Figure 6 . Sestrin2 loss increases mTORC1 signaling in colon cancer . ( A-D ) 3-month-old WT and Sesn2-/- mice ( n=11 each ) were subjected to lethal irradiation and injected with bone marrow cells from age-matched Sesn2-/- ( Sn2→WT ) and WT ( WT→Sn2 ) mice , respectively . After 1 month , mice were subjected to AOM/DSS administration as indicated in Figure 5B . After completion of the experiment , colons were examined under a dissection microscope ( A ) , and tumor number ( B ) , colon length ( C ) and tumor size ( D ) were analyzed and presented as mean ± s . e . m . ( E , F and J ) Colon tumor ( T ) and normal colon ( N ) tissues of indicated mice were subjected to immunohistochemistry of phospho-Ser235/236-S6 ( E and J ) or phospho-Thr37/46-4E-BP ( F ) . ( G-I , K and L ) Colon tumor ( T ) and normal colon ( NT ) tissues of WT and Sesn2-/- ( S2 ) mice ( G-I ) , as well as WT→Sn2 and Sn2→WT mice ( K and L ) , were subjected to immunoblotting of indicated mTORC1 and mTORC2 signaling markers ( G and K ) . Relative band intensities were quantified through densitometry and presented as mean ± s . e . m ( H , I and L; n=6 in each group ) . p<0 . 05 , **p<0 . 001 . P values are from Student’s t-test . Scale bars , 200 μm . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01510 . 7554/eLife . 12204 . 016Figure 6—figure supplement 1 . Efficiency of bone marrow transplantation . Spleen tissues from S2→WT ( A ) and WT→S2 ( B ) mice described in Figure 6 were genotyped for WT ( upper band , ~450bp ) and Sesn2-KO ( lower band , ~200bp ) alleles . Molecular weight markers are indicated in bp . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01610 . 7554/eLife . 12204 . 017Figure 6—figure supplement 2 . Additional immunohistochemistry results for Figure 6E ( A ) , Figure 6F ( B ) and Figure 6J ( C ) . Colon tumor ( T ) and normal colon ( N ) tissues of indicated mice were subjected to immunohistochemistry of phospho-Ser235/236-S6 ( A and C ) or phospho-Thr37/46-4E-BP ( B ) . Scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01710 . 7554/eLife . 12204 . 018Figure 6—figure supplement 3 . Additional immunoblotting results for Figure 6G ( A ) and Figure 6K ( B ) . Colon tumor ( T ) and normal colon ( NT ) tissues of WT and Sesn2-/- ( S2 ) mice ( A ) , as well as WT→Sn2 and Sn2→WT mice ( B ) , were subjected to immunoblotting of indicated mTORC1 and mTORC2 signaling markers . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 01810 . 7554/eLife . 12204 . 019Figure 6—figure supplement 4 . Uncropped images of blots . Red boxes indicate the cropped regions . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 019 It is possible that epithelial Sestrin2 expression tissue-autonomously suppresses colon tumor growth . Indeed , Sestrin2 is a potent inhibitor of mTORC1 ( Budanov and Karin , 2008; Lee et al , 2013 ) , which is known to be critical for tumorigenic growth in colon cancer cells ( Faller et al , 2015; Hardiman et al , 2014 ) . Therefore , we analyzed mTORC1 downstream target proteins in colon cancer and normal colon tissues from WT and Sesn2-/- mice through immunohistochemistry . Phosphorylation of ribosomal protein S6 , which is mediated by an mTORC1 target p70 S6 kinase ( S6K ) , is dramatically increased in tumors isolated from Sesn2-/- mice ( Figure 6E and Figure 6—figure supplement 2A ) . Consistent with this observation , phosphorylation of eukaryotic translation initiation factor 4E-binding protein ( 4E-BP ) , an additional mTORC1 target , was also prominently increased in Sesn2-/- colon cancer tissues , when compared to adjacent normal colon tissues as well as to WT colon cancer tissues ( Figure 6F and Figure 6—figure supplement 2B ) . These immunohistochemical observations were also confirmed by immunoblotting experiments; S6 and 4E-BP phosphorylation was prominently increased in colon tumors of Sesn2-/- mice , compared to those of WT mice ( Figure 6G–I and Figure 6—figure supplement 3A ) . Tumors from WT→Sn2 mice , but not those from Sn2→WT mice , also exhibited hyperactive mTORC1 signaling , indicated by increased S6 phosphorylation ( Figure 6J–L and Figure 6—figure supplement 2C , 3B ) . However , AKT Ser473 phosphorylation , which is mediated by mTORC2 , a complex distinct from mTORC1 , was not significantly altered by Sestrin2 loss in colon cancer tissues ( Figure 6G ) . Collectively , these data suggest that Sestrin2 loss selectively leads to mTORC1 hyperactivation in colon cancer tissues , which subsequently allowed for prominent tumor overgrowth . Unlike human colon cancer tissues ( Figures 3 , 4 ) , mouse colon cancers induced by the AOM-DSS treatment displayed Sestrin2 expression at a level comparable to that of normal colon tissues ( Figure 6G ) . The expression of Sestrin2 in mouse colon tumor cells may provide an explanation for the relatively low mTORC1 activity in the tumors ( Figure 6E–L ) that do not progress to adenocarcinomas ( Rosenberg et al , 2009 ) . Interestingly , it has been formerly reported that most mouse colon tumors induced by the AOM-DSS treatment do not contain p53 mutations or misregulation ( Nambiar et al , 2004; Takahashi and Wakabayashi , 2004 ) , and this may provide an explanation of why Sestrin2 expression is sustained in this specific mouse model of colon cancer . Indeed , SW480 cells , a human colon cancer cell line that displays very low p53 activity , demonstrated a strong downregulation of Sestrin2 expression , while RKO and HCT116 cells , which have wild-type p53 activity , expressed a relatively high amount of Sestrin2 proteins ( Figure 7A ) . The p53-deficient HCT116 cells exhibited dramatic downregulation of Sestrin2 expression when compared to the parental HCT116 cells ( Figure 7B ) . Expression of Sestrin2 in hyperplastic mouse colon tissues , which have mutations in both Apc and Kras genes ( Feng et al , 2013 ) , was also reduced by the loss of p53 or a dominant-negative mutation of p53 ( Figure 7C ) . Consistent with clinical data on human SESN2 expression described above ( Figure 4 ) , these results further support the notion that p53 is critical for Sestrin2 expression in colon cancer cells . 10 . 7554/eLife . 12204 . 020Figure 7 . p53 controls mTORC1 signaling through Sestrin2 in colon cancer cells . ( A-F ) Whole cell or tissue lysates from the following experiments were subjected to immunoblotting of indicated proteins . ( A ) Human colon cancer cell lines ( RKO , HCT116 and SW480 ) were serum-starved for 24 hr and then treated with 10% FBS ( serum ) for 2 hr . ( B ) p53-knockout ( HCT116 p53-/- ) and control HCT116 cells were serum-starved for 24 hr and then treated with 10% FBS for indicated time . ( C ) CDX2P-CreERT2Apcflox/flox KrasLSL-G12D/+ p53flox/flox ( left three lanes ) , CDX2P-CreERT2Apcflox/flox KrasLSL-G12D/+ p53+/+ ( centre three lanes ) , and CDX2P-CreERT2Apcflox/flox KrasLSL-G12D/+ p53R270H/+ ( right three lanes ) mice ( Feng et al , 2013 ) were daily injected with 100 mg/kg tamoxifen ( i . p . ) for 3 days , and dysplastic colon tissues were harvested after 8 days . ( D ) HCT116 p53-/- cells were infected with GFP- or Sestrin2-expressing lentiviruses . After 24 hr , cells were serum-starved for 24 hr and treated with 10% FBS for indicated time . ( E ) SW480 cells were infected with GFP- or Sestrin2-expressing lentiviruses . After 24 hr , cells were serum-starved for 24 hr and treated with 10% FBS for indicated time . ( F ) RKO cells , stably infected with lentiviruses expressing shRNA targeting luciferase ( sh-Luc ) or the SESN2 ( sh-SESN2 ) gene , were serum-starved for 24 hr and then treated with 10% FBS for indicated time . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 02010 . 7554/eLife . 12204 . 021Figure 7—figure supplement 1 . Sestrin2 inhibits colon cancer cell growth through mTORC1 inhibition . ( A and B ) RKO cells , stably infected with lentiviruses expressing shRNA targeting luciferase ( sh-Luc ) or the SESN2 ( sh-SESN2 ) gene , were subjected to colony-forming assay . In brief , cells were seeded sparsely ( 1000 cells per well for 6-well plates ) and grown for 10 days in normal medium with or without rapamycin ( Rap , 20 nM ) . Colonies were fixed in methanol and stained with 0 . 1% crystal violet ( C581 , Fisher Scientific ) . The plates were imaged under EPSON Perfection V30 scanner ( A ) . Relative colony growth was determined by densitometry of stained images ( B; n=3 from three different plates per group ) . ( C and D ) SW480 cells were infected with lentiviruses expressing GFP or SESN2 . At 24 hr after the infection , the cells were subjected to a colony-forming assay as described above ( n=3 ) . All data are shown as the mean ± s . e . m . p<0 . 05 , *p<0 . 01 . P values are from Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 02110 . 7554/eLife . 12204 . 022Figure 7—figure supplement 2 . Sestrin2-deficiency renders cancer cells less sensitive to chemotherapy . ( A-C ) RKO cells , stably infected with lentiviruses expressing shRNA targeting luciferase ( sh-Luc ) or the SESN2 ( sh-SESN2 ) gene , were subjected to colony-forming assay . In brief , cells were seeded sparsely ( 1000 cells per well for 6-well plates ) , treated with 5-fluorouracil ( 5-FU , 1 μM ) and irinotecan ( CPT-11 , 10 μM ) as indicated for 24 hr , and grown for 14 days in normal medium . Colonies were fixed in methanol and stained with 0 . 1% crystal violet ( C581 , Fisher Scientific ) . The plates were imaged under EPSON Perfection V30 scanner ( A ) . Relative colony growth was determined by densitometry of stained images ( B; n=3 from three different plates per group ) , and the growth suppression effects of 5-FU and CPT-11 were calculated from the densitometry results ( C; n=3 from three different plates per group ) . ( D , E ) RKO cells , stably infected with sh-Luc or sh-SESN2 gene , were treated with indicated concentrations ( μM ) of 5-FU and CPT-11 for 24 hr and subjected to immunoblotting of indicated proteins . ( F ) Current model of how Sestrin2 suppresses colon cancer progression . During colitis , Sestrin2 promotes restoration of colon homeostasis after injury through limiting ER stress . However , during carcinogenesis , p53 is inactivated and Sestrin2 expression is downregulated , which subsequently causes hyperactivation of mTORC1 signaling and promotes colon cancer development and growth . The p53-Sestrin2 axis may also be important for the effect of chemotherapy in attenuating colon cancer growth . All data are shown as the mean ± s . e . m . p<0 . 05 , **p<0 . 01 . P values are from Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 02210 . 7554/eLife . 12204 . 023Figure 7—figure supplement 3 . Uncropped images of blots . Red boxes indicate the cropped regions . Molecular weight markers are indicated in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 12204 . 023 p53 inhibits mTORC1 signaling , and this regulation may be an important contributor to the tumor suppressive activity of p53 ( Agarwal et al , 2016; Feng and Levine , 2010 ) . To understand if Sestrin2 , acting as a p53 target , has an essential role in regulating mTORC1 and colon cancer cell growth , Sestrin2 was overexpressed in p53-deficient HCT116 cells and SW480 cells , which have low SESN2 expression and high mTORC1 signaling ( Figure 7A , B ) . Restoration of Sestrin2 suppressed serum-induced phosphorylation of mTORC1 substrates , S6K and 4E-BP ( Figure 7D , E ) , but not mTORC2 substrate AKT ( Figure 7E ) . Sestrin2 silencing in RKO cells enhanced the mTORC1 signaling in both starved and serum-stimulated cells ( Figure 7F ) , indicating that endogenous Sestrin2 indeed functions to inhibit mTORC1 signaling in human colon cancer cells . As mTORC1 is known to control cell growth , we examined the effect of Sestrin2 on clonogenic growth of RKO and SW480 cells . A colony forming assay of Sestrin2-silenced RKO cells demonstrated an increase in clonogenic growth compared to the control cells ( Figure 7—figure supplement 1A , B ) , suggesting that Sestrin2 is an inhibitor of cancer cell growth . To further investigate if this Sestrin2 activity is dependent on mTORC1 hyperactivation , we treated control and Sestrin2-silenced cells with rapamycin , an inhibitor of mTORC1 . Rapamycin inhibited clonogenic growth in both control and Sestrin2-silenced cells , and interestingly , the growth difference between control and Sestrin2-silenced cells was diminished upon rapamycin treatment ( Figure 7—figure supplement 1A , B ) . These results indicate that Sestrin2 attenuates cancer cell growth primarily through inhibition of a hyperactive mTORC1 . On the other hand , expression of Sestrin2 in SW480 cells substantially inhibited clonogenic growth ( Figure 7—figure supplement 1C , D ) , supporting the idea that Sestrin2 is critical for inhibition of colon cancer cell overgrowth . As a stress-inducible cell growth regulator , Sestrin2 may be also important for the responsiveness of colon cancer cells to chemotherapeutic treatments . To examine this possibility , we treated control and Sestrin2-silenced RKO cells with two representative chemotherapeutic agents , 5-fluorouracil ( 5-FU ) and irinotecan ( CPT-11 ) . Although control cell growth was strongly suppressed by 5-FU or CPT-11 , Sestrin2 silencing rendered cells less susceptible to the chemotherapeutic drug treatments ( Figure 7—figure supplement 2A–C ) , suggesting that Sestrin2 loss may confer chemotherapy resistance to colon cancer cells . We also examined Sestrin2 expression and mTORC1 signaling in RKO cells when treated with 5-FU and CPT-11 . After 5-FU or CPT-11 treatments , Sestrin2 expression was slightly elevated in WT cells ( sh-Luc; Figure 7—figure supplement 2D , E ) . Interestingly , mTORC1 signaling , monitored by p-S6K and p-4E-BP1 , was prominently upregulated in sh-SESN2 cells after 5-FU treatment ( Figure 7—figure supplement 2D ) , suggesting that Sestrin2 suppresses mTORC1 activation after 5-FU treatment . In contrast , CPT-11 treatment reduced mTORC1 signaling , and Sestrin2 silencing led to modest but persistent mTORC1 upregulation ( Figure 7—figure supplement 2E ) . This mTORC1 upregulation as a result of Sestrin2 loss could have conferred chemoresistance to RKO cells against 5-FU and CPT-11 . The current study reveals how the stress-inducible protein Sestrin2 can coordinate ER stress and mTORC1 signaling pathways to maintain epithelial homeostasis and limit colitis and colon cancer development during colon injury ( Figure 7—figure supplement 2F ) . Yet , it is still possible that the mechanisms underlying increased susceptibility to colitis and colon cancer are separate from each other . During colitis stress , Sestrin2 functions to suppress ER stress and to promote regeneration of colon epithelium ( Figures 1 , 2 ) . However , Sestrin2 expression is lost during human colon carcinogenesis ( Figure 3 ) through inactivation of tumor suppressor p53 ( Figures 4 , 7 ) . Because expression of Sestrin2 is important for suppressing hyperactive mTORC1 signaling ( Figure 6 ) and tumor outgrowth ( Figure 5 ) , loss of Sestrin2 expression in human colon cancer serves as a critical tumorigenic mechanism . Indeed , Sestrin2 negatively controlled cell growth in various human colon cancer cell lines , which was dependent on mTORC1 regulation ( Figure 7—figure supplement 1 ) . Furthermore , inactivation of Sestrin2 conferred chemoresistance to colon cancer cells ( Figure 7—figure supplement 2 ) , rendering them difficult to treat with conventional chemotherapeutic methods . In addition to mTORC1 regulation , Sestrin2 is also known to reduce oxidative stress ( Budanov et al , 2004 ) by functioning as an activator of anti-ROS transcription factor Nrf2 ( Bae et al , 2013 ) or as an alkylhydroperoxidase ( Kim et al , 2015a ) . Thus , it is possible that the loss of Sestrin2 can contribute to cellular accumulation of ROS , which can promote DNA damage and genomic mutations that facilitate tumor development ( Sablina et al , 2005 ) . However , analysis of γ-H2AX did not show a significant increase in DNA damage between colon tumors of WT and Sesn2-/- mice ( Figure 5J , M ) . This result suggests that the mTORC1-regulatory function , rather than the ROS-inhibiting function , is the main contributor of Sestrin2’s tumor suppressive activity in colon tissues . Nevertheless , it is still possible that Sestrin2 attenuates tumor growth , at least partially by inhibiting tumor-associated epithelial damage and inflammation ( Figures 1 , 2 ) , which are well-characterized promoters of colon cancer growth ( Grivennikov et al , 2012; Terzic et al , 2010 ) . It is also possible that Sestrin2 exerts tumor suppressive activity additionally through its apoptosis-inducing function , which was recently discovered ( Ding et al , 2015 ) . The mTORC1-suppressing and ROS-reducing activities are shared between all members of the Sestrin family ( Sestrin1-3 ) ( Lee et al , 2013; Nogueira et al , 2008 ) . However , only Sestrin2 was shown to be downregulated in human colon cancer tissues , and expression of Sestrin1 and Sestrin3 was unchanged or slightly upregulated in the colon cancer tissues . Nevertheless , Sestrin1 and Sestrin3 are strongly downregulated in several types of cancer tissues , such as lung cancers and lymphomas ( Figure 3A ) , suggesting that they may be involved in anti-tumorigenic processes in tissues other than the colon . Drosophila Sestrin , which is the only Sestrin homologue expressed in Drosophila , was formerly shown to be a feedback inhibitor of mTORC1 , which can suppress hyperplastic tissue growth provoked by oncogenic mTORC1 hyperactivation ( Lee et al , 2010 ) . As the current study demonstrates Sestrin2 to be an inhibitor of colon cancer growth , future studies on the role of Sestrin1 and Sestrin3 in carcinogenic processes of other tissues may reveal a conserved tumor-suppressive role of Sestrin-family proteins . For immunoblotting , we obtained S6K ( sc-230 ) , PERK ( sc-13073 ) and p-PERK ( sc-32577 ) antibodies from Santa Cruz Biotechnology , Dallas , TX , human Sestrin2 ( 10795-1-AP ) antibody from Proteintech , Chicago , IL , BiP ( 3177 ) , CHOP ( 2895 ) , p-Thr389 S6K ( 9234 ) , p-Ser235/236 S6 ( 2211 ) , S6 ( 2317 ) , p-Thr37/46 4E-BP ( 2855 ) , 4E-BP ( 9452 ) , p-Ser473 AKT ( 9273 ) and AKT ( 4691 ) antibodies from Cell Signaling Technology , Danverse , MA , actin ( JLA20 ) antibody from Developmental Studies Hybridoma Bank ( DSHB , Iowa city , IA ) , and tubulin ( T5168 ) antibody from Sigma , St . Louis , MO . Mouse Sestrin2 antibody was described ( Ro et al , 2014b ) . For immunostaining , we obtained p53 ( sc-6243 ) , PCNA ( sc-7907 ) , β-catenin ( sc-59737 ) , CHOP ( sc-575 ) from Santa Cruz Biotechnology , F4/80 ( mf48000 ) from Invitrogen , Carlsbad , CA , BiP ( 3177 ) , γ-H2AX ( 2577 ) , p-Ser235/236 S6 ( 2211 ) and p-Thr37/46 4E-BP ( 2855 ) from Cell Signaling Technology . Azoxymethane ( AOM ) , dextran sulfate sodium ( DSS ) , rapamycin , 5-fluorouracil ( 5-FU ) and irinotecan ( CPT-11 ) were from Sigma . Human colon cancer cell lines , including RKO , SW480 and HCT116 , were obtained from American Type Culture Collection ( ATCC , Manassas , VA ) and cultured in Dulbecco’s modified Eagle’s medium ( DMEM , Invitrogen ) containing 10% fetal bovine serum ( FBS , Sigma ) , 50 U/ml penicillin and 50 mg/ml streptomycin . The cells were authenticated by Short Tandem Repeat ( STR ) profiling at ATCC , tested negative for mycoplasma infection , and subcultured for less than 6 months prior to initiation of the described experiments . p53-knockout HCT116 cells were obtained from Dr . Bert Vogelstein ( Johns Hopkins University , Baltimore , MD ) . The p53 loss in this cell line was confirmed by western blot . All cultures were maintained in a 37°C incubator with 5% CO2 . The lentiviral constructs for Sestrin2 overexpression and silencing are formerly described ( Budanov and Karin , 2008 ) . Viruses were generated and amplified in the Vector Core facility at the University of Michigan ( UM ) . Cells and tissues were lysed in RIPA buffer ( 50 mM Tris-Cl pH 7 . 4 , 150 mM NaCl , 1% sodium deoxycholate , 1% NP-40; 0 . 1% SDS ) or cell lysis buffer ( 20 mM Tris-Cl pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 2 . 5 mM sodium pyrophosphate , 1 mM β-glycerophosphate , 1 mM Na3VO4 , 1% Triton-X-100 ) containing protease inhibitor cocktail ( Roche , Indianapolis , IA ) , and processed as formerly described ( Ro et al , 2014a ) . Protein samples were boiled in SDS sample buffer for 5 min , separated by SDS-PAGE , transferred to PVDF membranes and probed with primary antibodies ( 1:200 for Santa Cruz antibodies , and 1:1000 for all other antibodies ) . After incubation with secondary antibodies conjugated with HRP ( Bio-rad; 1:2000 ) , chemiluminescence was detected using LAS4000 ( GE , Fairfield , CT ) systems or X-ray films . Immunoblot images were quantified by densitometry , and protein expressions were presented as relative band intensities . Uncropped images of immunoblots are provided in Figure 2—figure supplement 3 , Figure 6—figure supplement 4 and Figure 7—figure supplement 3 . Total RNA was extracted from tissues or cells using Trizol reagent ( Invitrogen ) , and cDNA was made using MMLV-RT ( Promega , Madison , WI ) and random hexamers ( Invitrogen ) . Quantitative PCR was performed in a Real-Time PCR detection system ( Applied Biosystems , Foster city , CA ) with iQTM SYBR Green Supermix ( Bio-rad , Hercules , CA ) and relevant primers . Relative mRNA expression was calculated from the comparative threshold cycle ( Ct ) values relative to β-Actin . Primers for Sestrins , inflammation markers ( Tnfa , Il6 , Il1b and Il10 ) , ER stress markers ( Xbp1s , ERdj4 , Gadd34 , Ero1α , Edem1 , and Pdi ) and β-Actin were formerly described ( Park et al , 2010; Park et al , 2014; Ro et al , 2014a ) . mRNA expression data and genome copy data from various studies were retrieved from Oncomine database ( Rhodes et al , 2007 ) . Exome sequencing information regarding the status of p53 was retrieved from Supplmentary Table 2 of TCGA colon cancer paper ( Cancer Genome Atlas Network , 2012 ) . TCGA data , retrieved from Oncomine , were manually partitioned into 'p53-mutated' and 'p53-unknown' groups according to the p53 gene status . When multiple probes for a same gene were found from the database , probes whose average values between normal and cancer groups ( or between normal and 'p53-mutated' groups ) are close to zero in log scale were selected for further analysis . Bar graphs were plotted in a linear scale and control values were normalized to one . Scatter plots were presented in a logarithmic scale using raw data . Correlation and linear regression analyses were performed in Graphpad Prism 6 . WT , Sesn2-/- and Sesn2-/-/Sesn3-/- mice ( Lee et al , 2012 ) and CDX2P-CreERT2Apcflox/floxmice ( Feng et al , 2013 ) were used for this study . These mice are on a C57BL/6 background . Mice were maintained in filter-topped cages and given free access to autoclaved regular chow diet at the UM according to the NIH and institutional guidelines . All animal studies were ethically approved ( protocol approval numbers: PRO00005712 and PRO00004019 ) and overseen by the University Committee on Use and Care of Animals ( UCUCA ) at the UM . For colitis induction , mice received water with 3% DSS for 6–7 days ( inflammatory phase ) . Then the mice were placed on regular drinking water for 5–7 days ( recovery phase ) as formerly described ( Xue et al , 2013 ) . For tumor induction , mice were treated with AOM ( 10 mg/kg body weight ) . At 5 days following the AOM injection , mice received water with 1 . 5% DSS for 7 days ( inflammatory phase ) . Then , the mice were placed on regular drinking water for 14 days ( recovery phase ) . The mice were subjected to two more inflammatory and recovery cycles for tumor induction as in Figure 5A , B . Reciprocal bone marrow chimera experiments were done as described in our recent paper ( Anderson et al , 2013 ) , and the mice were subjected to DSS or AOM/DSS treatment at 1 month after the bone marrow transplantation to allow for complete substitution of the hematopoietic compartment . The mice in the same experiments are from an age-matched , co-housed cohort , and animal numbers were determined according to our previous studies ( Anderson et al , 2013; Xue et al , 2013 ) . A dissecting microscope ( 4x magnification ) was used to assess the tumor number and size . Tumor size was defined as the mean of the two largest diameters measured with digital calipers . Tumor volume was derived from tumor size . Consistent with their histological appearance , a spherical shape was assumed for colon polyps , thus tumor volume = 4/3πr3 , where r = radius . Tumor burden/load is defined as the total polyp volume per animal , which is the product of polyp number and polyp volume . Colons were removed , flushed with PBS , fixed in 4% paraformaldehyde at 4°C overnight and paraffin-embedded for histological analyses . Antigen retrieval was performed in 10 mM sodium citrate at 95°C for 15 min . For immunostaining of PCNA , β-catenin , F4/80 , BiP , γ-H2AX , p-Ser235/236 S6 and p-Thr37/46 4E-BP , colon sections were incubated with corresponding primary antibodies ( 1:50 , 1:50 , 1:100 , 1:200 , 1:50 , 1:200 and 1:100 , respectively ) , followed by incubation with biotin-conjugated secondary antibodies ( 1:200 ) and streptavidin-HRP ( 1:300 ) . The HRP activity was visualized with diaminobenzidine staining . Haematoxylin counterstaining was applied to visualize nuclei . For immunostaining of CHOP , colon sections were incubated with primary antibody ( 1:50 ) , followed by incubation with Alexa 594-conjugated secondary antibody . DAPI counterstaining was applied to visualize nuclei . TdT-mediated dUTP nick end labeling ( TUNEL ) assay was performed using In Situ Cell Death Detection Kit TMR-Red ( 1215792910 , Roche ) . The samples were analysed under an epifluorescence-equipped light microscope ( Meiji MT6300 ) .	An organ that is inflamed has an increased risk of developing cancer . Inflammation can be elicited in various ways; and intestinal inflammation and colon cancer development are often associated with a protein complex – called mTORC1 – being overactive in the tissue . This protein complex has been studied in other contexts and is known to instruct cells to produce more proteins . However , when too much protein is made too quickly , cells cannot carry out their routine quality checks . This , in turn , can lead to unfolded proteins accumulating in the cell , which is stressful and damaging , and can cause inflammation . Increased production of proteins and other biomolecules can also allow the uncontrolled growth of cancer cells . Other recently discovered proteins – called sestrins – can counteract the cancer-promoting effects of overactive mTORC1 . Sestrins achieve this via several mechanisms , but as yet almost nobody had studied the role of these proteins in intestinal inflammation and colon cancer . Ro , Xue et al . deleted the genes for two members of the sestrin family , called Sestrin2 and Sestrin3 , in mice and showed that their colons were more prone to inflammation . Additional analysis showed that people with ulcerative colitis – a condition in which the colon is chronically inflamed – have elevated levels of Sestrin2 , whereas very low levels of Sestrin2 could be detected in tissue samples from patients with colon cancers . These data suggested that Sestrin2 might be trying to protect cells from injury and acts as a barrier to cancer formation . Ro , Xue et al . then used biochemical techniques in human cancer cells grown in the laboratory to show that Sestrin2 inhibits mTORC1 , making these cells grow less . Colon cancer cells with little or no Sestrin2 were also more resistant to chemotherapy than control cells with normal levels of Sestrin2 . Lastly , a type of colon cancer that is associated with inflammation grew faster in mice that lacked the gene for Sestrin2 . Taken together these findings represent evidence that Sestrin2 acts as a tumor suppressor in the colon . Future experiments might investigate how losing Sestrin2 makes these cells more resistant to chemotherapy and whether sestrins act as tumor suppressors in other tissues as well .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2016	Tumor suppressive role of sestrin2 during colitis and colon carcinogenesis
C4 photosynthesis represents a most remarkable case of convergent evolution of a complex trait , which includes the reprogramming of the expression patterns of thousands of genes . Anatomical , physiological , and phylogenetic and analyses as well as computational modeling indicate that the establishment of a photorespiratory carbon pump ( termed C2 photosynthesis ) is a prerequisite for the evolution of C4 . However , a mechanistic model explaining the tight connection between the evolution of C4 and C2 photosynthesis is currently lacking . Here we address this question through comparative transcriptomic and biochemical analyses of closely related C3 , C3–C4 , and C4 species , combined with Flux Balance Analysis constrained through a mechanistic model of carbon fixation . We show that C2 photosynthesis creates a misbalance in nitrogen metabolism between bundle sheath and mesophyll cells . Rebalancing nitrogen metabolism requires anaplerotic reactions that resemble at least parts of a basic C4 cycle . Our findings thus show how C2 photosynthesis represents a pre-adaptation for the C4 system , where the evolution of the C2 system establishes important C4 components as a side effect . The dual-specific enzyme ribulose 1 , 5-bisphosphate carboxylase/oxygenase ( Rubisco ) catalyzes two opposing reactions—the carboxylation and the oxygenation of ribulose 1 , 5-bisphosphate . The former reaction yields 3-phosphoglycerate ( 3-PGA ) , whereas the latter produces 2-phosphoglycolate ( 2-PG ) . 3-PGA is reduced to carbohydrates in the Calvin–Benson cycle and incorporated into biomass . However , 2-PG is toxic , which requires its removal by a metabolic repair pathway called photorespiration ( Anderson , 1971; Bowes et al . , 1971; Ogren , 1984; Leegood et al . , 1995 ) . In the photorespiratory cycle , 2-PG is regenerated to 3-PGA , but it involves the release of formerly assimilated CO2 and NH3 , entails energy costs for the plants and reduces the efficiency of photosynthesis by up to 30% ( Ehleringer et al . , 1991; Bauwe et al . , 2010; Raines , 2011; Fernie et al . , 2013 ) . Eight core enzymes are required for photorespiration , which in higher plants are located in the chloroplast , the peroxisome , and the mitochondrion ( Bauwe et al . , 2010; Figure 1A ) . The pathway rescues ¾ of the carbon , which would otherwise be lost through the oxygenase activity of Rubisco ( Peterhansel et al . , 2010; Fernie et al . , 2013 ) . Ammonia refixation in the chloroplast by the combined activities of glutamine synthase ( GS ) and glutamine oxoglutarate aminotransferase ( GOGAT ) is an integral part of photorespiration . 10 . 7554/eLife . 02478 . 003Figure 1 . The genus Flaveria as a model organism to study C4 evolution . Schematic view of the photorespiratory pathway ( A ) , the NADP-ME type C4 pathway as it can be found in C4 Flaveria species ( B ) and the C2 photosynthesis pathway ( C ) . ( D ) Phylogeny and physiological properties of selected Flaveria species . The phylogeny was redrawn according to McKown et al . ( 2005 ) , CO2 compensation points are taken from Ku et al . ( 1991 ) , incorporation of 14CO2 is from Moore et al . ( 1987 ) and the ratios of GLDP B ( expressed in all chlorenchyma cells ) and GLDP A ( expressed in bundle sheath cells only ) are from Schulze et al . ( 2013 ) . ( Abbreviations: AGT: serine glyoxylate aminotransferase; AlaAT: alanine aminotransferase; AspAT: aspartate aminotransferase; GDC: glycine decarboxylase complex; GGT: glutamate , glyoxylate-aminotransferase; GLYK: D-glycerate 3-kinase; GOX: glycolate oxidase; HPR: hydroxypyruvate reductase; MDH: malate dehydrogenase; NADP-ME: NADP dependent malic enzyme; PEPC: phosphoenolpyruvate carboxylase; PGLP: 2-phosphoglycerate phosphatase; PPDK pyruvate , phosphate-dikinase; RUBISCO: Ribulose-1 , 5-bisphosphat-carboxylase/-oxygenase; SHM: serine hydroxymethyltransferase; 2-OG: oxoglutarate; 2-PG 2-phosphoglycolate; 3-PGA: 3-phosphoglycerate; Gln: glutamine; Glu: glutamate; OAA: oxaloacetate; PEP: phosphoenolpyruvate; TP: triosephosphate ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 003 In hot and dry environments and under low atmospheric CO2 conditions , when the oxygenation activity of Rubisco is increased , the high rate of photorespiration becomes unfavorable for the plants ( Sage , 2001 , 2013 ) . C4 plants possess a mechanism that minimizes the oxygenase function of Rubisco and thereby reduces photorespiration and decreases the loss of carbon . C4 photosynthesis is based on a division of labor between two different cell types , mesophyll and bundle sheath cells , which are organized in a wreath-like structure called ‘Kranz Anatomy’ ( Haberlandt , 1904; Dengler and Nelson , 1999 ) . Atmospheric CO2 is initially fixed in the mesophyll by phosphoenolpyruvate carboxylase ( PEPC ) , and the resulting four-carbon compound is transported to the bundle sheath cells and decarboxylated by NADP/NAD malic enzyme or phosphoenolpyruvate carboxykinase ( Hatch et al . , 1975 ) . Thereby CO2 is concentrated at the site of the Rubisco in the bundle sheath cells ( Hatch , 1987 ) , outcompeting the molecular oxygen . As a consequence , photorespiration is drastically reduced as compared to C3 plants , and C4 plants are characterized by a high photosynthetic efficiency ( Figure 1B ) . C4 plants have evolved multiple times independently from C3 ancestors . The evolution of C4 photosynthesis occurred at least 62 times in 19 different families of the angiosperms ( Sage et al . , 2011 ) , implying a low evolutionary barrier towards expression of this trait . The analysis of recent intermediate species ( Bauwe and Kolukisaoglu , 2003; Sage , 2004; Bauwe , 2011; Sage et al . , 2012 , 2013; Schulze et al . , 2013 ) indicates that establishing a photorespiratory CO2 pump was an early and important step in the evolution towards C4 photosynthesis ( Figure 1C ) . Since the two-carbon compound glycine serves as a transport metabolite , this photorespiratory CO2 concentrating mechanism is also termed C2 photosynthesis . Computational modeling of the evolutionary trajectory from C3 to C4 photosynthesis indicated C2 photosynthesis represented an evolutionary intermediate state ( Heckmann et al . , 2013; Williams et al . , 2013 ) as well suggesting that C2 photosynthesis is a prerequisite for the evolution of C4 . However , it remained unclear if the evolution of C2 photosynthesis fosters the evolution of C4 photosynthesis beyond providing a selection pressure to reallocate Rubisco to the bundle sheath . In the present study , we have used the genus Flaveria as a model system for investigating the transition from C2 to C4 photosynthesis . To this end , we study a phylogenetic framework consisting of C3 , C3–C4 intermediate , and C4 species ( Powell , 1978; Edwards and Ku , 1987; Ku et al . , 1991 ) of this genus which rather recently evolved C4 ( Christin et al . , 2011 ) , focusing on genes encoding photorespiratory enzymes and other components of C2 photosynthesis . The genus Flaveria contains three main phylogenetic groups , of which the first diverging group includes all C3 Flaveria . Clade B contains seven C3–C4 intermediate species and the C4-like species F . brownii . All C4 Flaveria species belong to clade A , which also contains several C4-like species and the C3–C4 intermediate F . ramosissima ( McKown et al . , 2005; Figure 1D ) . We hypothesized that the analysis of species in the genus Flaveria combined with in silico modeling elucidates the evolutionary changes accompanying and following the establishment of the C2 pathway . To this end we simulated the metabolism of C2 plants by coupling a mechanistic model of C3–C4 intermediate photosynthesis ( von Caemmerer , 2000; Heckmann et al . , 2013 ) with a detailed modified stoichiometric model of C4 photosynthesis ( Dal'Molin et al . , 2010 ) , and investigated the evolution of C4 photosynthesis and photorespiration by following the changes in mRNA and protein abundance along the evolutionary path . RNA and protein amounts of the majority of the photorespiratory enzymes were reduced in C4 as compared to C3 species . In contrast , photorespiratory mRNA and protein amounts did not decrease in the C3–C4 intermediate species but were mostly equal or even higher than in the C3 species , demonstrating that the establishment of the photorespiratory CO2 pump in the genus Flaveria relies on coordinated changes in the expression of all core photorespiratory enzymes . Metabolic modeling in combination with comparisons of transcript abundances in the different Flaveria species strongly indicates that introduction of C2 photosynthesis has a direct impact on the nitrogen metabolism of the leaf . Its implementation necessitates the parallel establishment of components of the C4 cycle to cope with these changes in refixation of photorespiratory nitrogen . Based on these results , we predict a mechanistic interaction between C4 and C2 photosynthesis . To study the evolution of the expression of photorespiratory and C4 cycle genes during the transition from C3 to C4 photosynthesis in the genus Flaveria , nine species reflecting the evolutionary trajectory taken were selected , including two C3 ( F . robusta and F . pringlei ) , two C4 ( F . bidentis and F . trinervia ) , and five C3–C4 intermediate species ( Figure 1D ) . According to their CO2 compensation points and the percentage of carbon initially fixed into malate and aspartate , F . chloraefolia and F . pubescens were earlier classified as type I C3–C4 intermediates . F . anomala and F . ramosissima belong to the type II C3–C4 intermediates and F . brownii is classified as a C4-like species ( Edwards and Ku , 1987; Moore et al . , 1987; Cheng et al . , 1988; Ku et al . , 1991 ) . Type I C3–C4 intermediates are defined as solely relying on the photorespiratory CO2 concentration cycle whereas a basal C4 cycle activity is present in type II C3–C4 intermediates species . C4-like species exhibit much higher C4 cycle activities but lack complete bundle sheath compartmentation of Rubisco activity ( Edwards and Ku , 1987 ) . Four independent experiments with plants grown during different seasons were performed to identify differences between the species that are dependent on their different modes of photosynthesis and independent of environmental influences . For each experiment the plants were seeded concurrently and grown side-by-side under greenhouse conditions . The second and fourth visible leaves from the top of all nine species were harvested at noon on the same day for transcript and protein analysis . Plants for experiment one were harvested in September 2009 , for experiment two in June 2010 , for experiment three in October 2010 and for experiment four in April 2011 . The amounts of the core photorespiratory and C4 enzymes were assessed by immunoblotting using specific antibodies raised against synthetic peptides or recombinant proteins . The abundances of the corresponding RNAs as well of C4 cycle associated transcripts were quantified by total transcriptome sequencing . The transcriptomes of the different Flaveria species were sequenced by Illumina technology following standard procedures . In total , close to 200 Gb of raw sequence data were produced . After filtering of low quality reads 30 to 58 million reads per species and experiment were quantified ( Figure 2—source data 1 ) . In a cross species approach , we mapped the sequences onto the minimal set of Arabidopsis thaliana coding sequences using the BLAST-like alignment tool BLAT ( Kent , 2002 ) as described previously ( Gowik et al . , 2011 ) ( Figure 2—source data 2 , data available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . q827h ) . We were able to align approx . 50% of our reads to the Arabidopsis transcripts . This is lower as compared to a similar approach using 454 sequencing ( Gowik et al . , 2011 ) and likely due to the shorter read length of the Illumina compared to the 454 reads . To overcome the low mapping efficiency , the leaf transcriptomes of Flaveria species were assembled de novo based on 454 ( Gowik et al . , 2011 ) and Illumina reads ( this study ) . Among the contigs from F . robusta , we identified full-length transcripts for all photorespiratory and C4 genes in the focus of the present study and used these for further read mapping and detailed analysis . To evaluate the variation between the four independent experiments , we performed hierarchical sample clustering and a principal component analysis of the transcript profiles derived from read mapping on the minimal set of Arabidopsis coding sequences . Hierarchical sample clustering using Pearson correlation and average linkage clustering shows that the transcript profiles of all Flaveria species were quite similar in all four experiments since the samples cluster strictly species-wise ( Figure 2A ) . The transcriptome patterns are influenced by the photosynthesis type and the phylogenetic relationships of the different species . The two C4 species , both belonging to clade A , cluster together as do the two C3 species that belong to the basal Flaveria species . Within the C3–C4 intermediates the two more advanced intermediates F . ramosissima and F . anomala cluster together , the only pattern which contradicts phylogenetic proximity since F . ramosissima belongs to clade A and F . anomala belongs to clade B . The last cluster consists out of the C3–C4 intermediates F . chloraefolia and F . pubescens , and the C4-like species F . brownii . 10 . 7554/eLife . 02478 . 004Figure 2 . Variation of transcript profiles of the individual Flaveria species between the four experiments . ( A ) Hierarchical sample clustering of all expressed transcripts . The tree was calculated with the MEV program using the HCL module with Pearson correlation and the average linkage method . ( B ) Principal component analysis of transcript levels . The first three components explain 27% of the total variance . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 00410 . 7554/eLife . 02478 . 005Figure 2—source data 1 . Results of the Illumina sequencing and cross species read mapping . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 00510 . 7554/eLife . 02478 . 006Figure 2—source data 2 . Quantitative information for all reads mapped in a cross species approach onto the reference transcriptome from Arabidopsis thaliana . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 006 Principle component analysis supports the results of the hierarchical clustering . The samples are mainly separated by photosynthesis type and phylogenetic relationships with the two intermediate species from different phylogenetic trajectories again forming a tight cluster ( Figure 2B ) . The first three components , shown in Figure 2B , explain only 27% of the total variance . This is in good accordance with earlier results where it was shown that about 16% of all analyzed genes showed photosynthesis type related expression changes when the transcriptomes of the C4 species F . trinervia and F . bidentis , the C3 species F . robusta and F . pringlei and the C3–C4 intermediate species F . ramosissima were compared ( Gowik et al . , 2011 ) . Photorespiratory genes are expressed in all species and photorespiratory proteins are detected in all species . To visualize the differences in transcript and protein abundance heat maps were plotted ( Figure 3 ) . The transcription of all photorespiratory genes except the transport proteins DIT1 and DIT2 and one isoform of GLDH was downregulated in the C4 species F . bidentis and F . trinervia compared to the C3 species F . pringlei und F . robusta ( Figure 3A , Figure 3—source data 1 ) . Both dicarboxylate transporters play an important role in generating the transfer acids in the C4 pathway of NADP-ME plants such as F . trinervia and F . bidentis ( Renne et al . , 2003; Gowik et al . , 2011; Kinoshita et al . , 2011 ) . This may explain why their expression pattern is more similar to the C4 genes than to the other photorespiratory genes . 10 . 7554/eLife . 02478 . 007Figure 3 . Abundance of photorespiratory transcripts and proteins in leaves of individual Flaveria species . Normalized transcript ( A ) and protein ( B ) amounts are plotted as heat maps . Transcript amounts were determined by Illumina sequencing of the leaf transcriptomes and read mapping on selected F . robusta full length transcript sequences . Protein amounts were determined by protein gel blots . See Figure 3—source data 1 for absolute transcript levels , Figure 3—source data 2 for protein quantification and Figure 3—figure supplements 1 and 2 for immunoblots . Fp: F . pringlei ( C3 ) ; Fro: F . robusta ( C3 ) ; Fc: F . chloraefolia ( C3–C4 ) ; Fpu: F . pubescens ( C3–C4 ) ; Fa: F . anomala ( C3–C4 ) ; Fra: F . ramosissima ( C3–C4 ) ; Fbr: F . brownii ( C4-like ) ; Fb: F . bidentis ( C4 ) ; Ft: F . trinervia ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 00710 . 7554/eLife . 02478 . 008Figure 3—source data 1 . Transcript abundance of photorespiratory genes determined by read mapping on F . robusta full length transcript sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 00810 . 7554/eLife . 02478 . 009Figure 3—source data 2 . Quantification of photorespiratory proteins by protein gel blot . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 00910 . 7554/eLife . 02478 . 010Figure 3—figure supplement 1 . Results of the protein analyses . 30 µg of total protein was electrophoresed on a polyacrylamide-SDS gel and stained with coomassie blue as control of total protein concentrations . Fp: F . pringlei ( C3 ) ; Fro: F . robusta ( C3 ) ; Fc: F . chloraefolia ( C3–C4 ) ; Fpu: F . pubescens ( C3–C4 ) ; Fa: F . anomala ( C3–C4 ) ; Fra: F . ramosissima ( C3–C4 ) ; Fbr: F . brownii ( C4-like ) ; Fb: F . bidentis ( C4 ) ; Ft: F . trinervia ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01010 . 7554/eLife . 02478 . 011Figure 3—figure supplement 2 . Results of the protein analyses . Immunoblot results with antibodies against photorespiratory proteins ( C ) Immunoblot results with antibodies against C4 proteins . Fp: F . pringlei ( C3 ) ; Fro: F . robusta ( C3 ) ; Fc: F . chloraefolia ( C3–C4 ) ; Fpu: F . pubescens ( C3–C4 ) ; Fa: F . anomala ( C3–C4 ) ; Fra: F . ramosissima ( C3–C4 ) ; Fbr: F . brownii ( C4-like ) ; Fb: F . bidentis ( C4 ) ; Ft: F . trinervia ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 011 The amounts of photorespiratory transcripts did not decrease gradually from C3 to C4 but the expression levels in the C3–C4 intermediate species F . chloraefolia , F . pubescens , F . anomala and F . ramosissima were mostly equal or higher than in the C3 species . An exception are the transcripts of one GLDP , one GLDH and one SHM isoform which are drastically down-regulated also in the C3–C4 intermediate species . It was shown earlier that the down-regulation of this GLDP isoform is tightly associated with the establishment of the C2 pathway in Flaveria ( Schulze et al . , 2013 ) . The down-regulation of the GLDH and SHM isogenes might have similar reasons since both enzymes are also involved in glycine decarboxylation . Only the C4-like species F . brownii is intermediate with respect to photorespiratory transcripts . 19 of 27 transcripts are reduced compared to the C3–C4 intermediate and C3 species but have higher levels than the true C4 species F . bidentis and F . trinervia . Exceptions are the components of the glycine decarboxylase complex as the respective transcripts levels are equal to these in the C3 and C3–C4 intermediate species ( Figure 3A ) . The expression patterns described above were not only found for the genes encoding the core enzymes of photorespiration but also for the genes responsible for recycling of ammonia set free during photorespiration , GS/GOGAT . Also the genes of recently discovered transporters associated with photorespiration , PLGG1 and BOU ( Eisenhut et al . , 2013; Pick et al . , 2013 ) , behave accordingly . To test whether transcript abundance reflects protein abundance , amounts of core photorespiratory proteins in the leaves of all nine species were quantified by protein gel blots . To this end we generated antibodies against conserved peptides from Flaveria GLDP , GLDT , GLDL , SHM , HPR , PGLP and GLYK proteins . Total leaf proteins were extracted from plant material harvested together with the material used for RNA isolation and equal amounts of protein were separated via SDS gel-electrophoresis prior to blotting ( Figure 3—figure supplement 1 ) . The changes of protein amounts essentially reflected the changes of the amounts of the corresponding transcripts ( Figure 3B , Figure 3—figure supplement 2 , Figure 3—source data 2 ) . The amounts of core photorespiratory proteins in the C3–C4 intermediates were equal to the amounts in the C3 species . A clear reduction of these proteins can be observed only for the true C4 species and the C4–like species . F . brownii exhibits intermediate amounts of most photorespiratory proteins . This indicates that the regulation of photorespiratory genes mainly occurs on the transcriptional level and that our approach to analyze the photorespiratory activity by comparative transcriptomics is reasonable . While the overall patterns remain similar between all independent experiments , individual proteins and transcripts vary between the four experiments . This likely reflects adjustments of photorespiratory gene expression to the different light and temperature conditions in our green house in the different seasons of the year . We conclude that the four experiments support the establishment of a photorespiratory C2 cycle early during C4 evolution in Flaveria and that this C2 cycle was maintained until Rubisco activity was constricted to the bundle sheath cells in the true C4 Flaveria species . While the principal physiological differences between C3 and C4 leaves are widely understood , knowledge about the metabolic reconfiguration required to implement a functional C2 pathway into a C3 leaf is incomplete . In particular , moving glycine from mesophyll to bundle sheath cells ( Hylton et al . , 1988; Morgan et al . , 1993 ) does not only translocate carbon , it also transports one nitrogen atom per two carbon atoms . Evidently , implementing the C2 carbon pump requires balancing of metabolic routes to maintain homeostasis of both carbon and nitrogen metabolism ( Monson and Rawsthorne , 2000 ) . How this can be achieved is non-intuitive and it thus requires a systematic analysis by metabolic modeling . To this end , we simulated the leaf metabolism of a C2 plant using an integrated model . We coupled a mechanistic model of C3–C4 intermediate photosynthesis ( von Caemmerer , 2000; Heckmann et al . , 2013 ) with a modified genome-scale stoichiometric model of C4 photosynthesis that was designed to describe the entire metabolic interactions of mesophyll and bundle sheath cells in C4 leaves ( Dal'Molin et al . , 2010 ) . We used the mechanistic model to predict constraints for the stoichiometric model . It provided values for net CO2 uptake , Rubisco carboxylation as well as oxygenation in mesophyll and bundle sheath , CO2 leakage from the bundle sheath , PEPC activity in the mesophyll , activity of NADP-ME in the bundle sheath , plasmodesmatal flux of glycine and serine , and decarboxylation by the GDC . Given specific activities of the C2 and C4 cycles in the mechanistic model , we used flux balance analysis ( FBA ) to predict detailed flux distributions that follow biologically realistic optimality criteria ( Varma and Palsson , 1994 ) . We employed a maximization of leaf biomass production , followed by a minimization of the sum of absolute fluxes including transport processes . In the minimization of total flux , we allocated higher weights to plasmodesmatal fluxes in order to account for the trade-off between CO2 leakage and diffusion of metabolites between the cells . This framework allows us to investigate the most parsimonious implementation of C2 and C4 cycles , given a hypothesis about which metabolites are suitable for plasmodesmatal transport . The first outcome of simulating the photorespiratory CO2 pump was that the establishment of the C2 pathway has indeed a direct impact on the nitrogen metabolism of the leaf . It transports two molecules of glycine from the mesophyll to the bundle sheath , where one molecule each of serine , CO2 , and ammonium are produced . CO2 is fixed by bundle sheath Rubisco and serine is transferred back to the mesophyll , where it is used for the regeneration of phosphoglycerate and photorespiratory glycine . This results in a net transport of CO2 but also ammonia from the mesophyll to the bundle sheath . To create a noticeable CO2 enrichment in the bundle sheath , the C2 cycle must run with an appreciable capacity; indeed , the mechanistic model of C3–C4 intermediate photosynthesis predicted an oxygenation rate of Rubisco of about one third of its carboxylation rate . Running at such rates , the C2 cycle will create a massive nitrogen imbalance between mesophyll and bundle sheath cells , as was also predicted earlier by Monson and Rawsthorne ( 2000 ) . Within the stoichiometric model , the free diffusion of ammonia between the two cell types was not allowed , since ammonia is toxic and known to effectively uncouple electrochemical gradients ( Krogmann et al . , 1959 ) . Thus , ammonia must be refixed in the bundle sheath cells and transferred back to the mesophyll in the form of amino acids . According to the intergrated model , ammonia is fixed by glutamine synthetase and glutamine oxoglutarate aminotransferase ( GS/GOGAT ) in the bundle sheath cells ( Figure 4 ) . Consistent with this prediction , we found that GS/GOGAT transcripts were upregulated in the C3–C4 intermediate species ( Figure 3 ) . 10 . 7554/eLife . 02478 . 012Figure 4 . Flux Balance Analysis of the C2 photosynthetic pathway . Predicted fluxes if ( A ) major amino acids and the corresponding oxoacids and dicarbonic acids are allowed to freely diffuse between cells , ( B ) the α-ketoglutarate and glutamate transfer between mesophyll and bundle sheath was constrained ( C ) additionally the transfer of alanine and pyruvate between mesophyll and bundle sheath was constrained ( D ) transfer of all nitrogen containing compounds except for glycine and serine , which are used by the C2 cycle were constrained . Fluxes are given in µmol s−1 m−2 . Values in brackets show minimum and maximum of flux resulting from flux variability analysis . Flux of dissolved gasses , sucrose , inorganic compounds and processes that carry flux below 1 µmol s−1 m−2 are not shown . The sums of absolute fluxes over the plasmodesmata for the different variants were ( A ) : 17 . 8 µmol s−1 m−2; ( B ) : 18 . 4 µmol s−1 m−2; ( C ) : 19 . 0 µmol s−1 m−2; ( D ) : 22 . 1 µmol s−1 m−2 . See Figure 4—source data 1 for plasmodesmatal fluxes . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01210 . 7554/eLife . 02478 . 013Figure 4—source data 1 . Fluxes over plasmodesmata depending on the weight on plasmodesmatal fluxes including flux variability analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 013 Estimating whether a certain metabolite is suitable for maintaining a diffusional gradient between mesophyll and bundle sheath is an unsolved problem . The impact on regulatory mechanisms and homeostasis of the C3 leaf may render some metabolites unsuitable to serve as transport metabolites . We address this problem by modeling multiple scenarios that assume different transport metabolites . If major amino acids and the corresponding oxoacids and dicarbonic acids are allowed to freely diffuse between cells in an integrated model representing a C2 cycle , glutamate is predicted to be transferred to the mesophyll , where it is deaminated by GGT , regenerating the photorespiratory glycine . The resulting 2-oxoglutarate is transferred back to the bundle sheath cells ( Figure 4A ) . The model preference for glutamate/2-oxoglutarate reflects the minimization of total flux in the FBA model , as this effectively minimizes the number of active enzymatic reactions and holds the plasmodesmatal flux for ammonia balance at one acceptor and one transport metabolite . To elucidate if alternative solutions exist that contain more steps but retain the same biomass output , the 2-oxoglutarate transfer between mesophyll and bundle sheath was constrained to prevent the glutamate/2-oxoglutarate exchange . The integrated model then predicts an alanine/pyruvate shuttle ( Figure 4B ) . The glutamate produced by GS/GOGAT activity in the bundle sheath cells is used by alanine aminotransferase ( Ala-AT ) to aminate pyruvate . The resulting alanine is transferred to the mesophyll and trans-aminated by Ala-AT resulting in pyruvate and glutamate . The glutamate is used to regenerate photorespiratory glycine and pyruvate is transferred back to the bundle sheath . If alanine and pyruvate transfer are also constrained , the model predicts an aspartate/malate shuttle ( Figure 4C ) . This includes the oxidation of malate in the bundle sheath . The resulting oxaloacetate ( OAA ) is aminated by aspartate aminotransferase ( Asp-AT ) and aspartate moves to the mesophyll . Here aspartate is trans-aminated by Asp-AT and malate is regenerated by reduction of the resulting OAA and transferred to the bundle sheath . In all these scenarios , further increasing the weights on plasmodesmatal flux leads to transporter metabolites with increased N carrying capacity such as asparagine ( Figure 4—source data 1 ) . In a restrictive scenario , all nitrogen containing compounds were excluded from plasmodesmatal transport , except for glycine and serine , which are used by the C2 cycle itself . In this case , the model predicts that bundle sheath derived ammonia is transferred from glutamate to phosphohydroxy-pyruvate by phosphoserine aminotransferase to yield phosphoserine; phosphoserine is then converted to serine by phosphoserine phosphatase . Finally , the serine moves to the mesophyll . This variant includes the transfer of 3-phosphoglycerate from the mesophyll to the bundle sheath , where it is converted to phosphohydroxy pyruvate by 3-phosphoglycerate dehydrogenase ( Figure 4D ) . In C3 plants , basal activities of the typical C4 cycle enzymes are present ( Aubry et al . , 2011 ) . When our integrated model is parameterized to include an active C4 cycle , it predicts that a contingent of the bundle sheath ammonia will be transferred to the mesophyll cells by the C4 cycle as a biomass neutral alternative to the 2-OG/Glu shuttle or as the unique solution when additional weight on plasmodesmatal fluxes is applied ( Figure 5—source data 1 ) . In this solution malate is decarboxylated in the bundle sheath cells . CO2 is refixed by Rubisco , and the resulting pyruvate is aminated by Ala-AT . Alanine moves to the mesophyll cells , where ammonia is fed into the photorespiratory cycle by Ala-AT and GGT . The resulting pyruvate is converted back to malate by PPDK , PEPC , and NADPH-dependent MDH ( Figure 5A ) . Flux variability analysis shows that only marginal variability in the fluxes of the shuttle is possible ( Figure 5—source data 1 ) . According to our model predictions , the cycle is active even at low PEPC activities , such as those measured in C3 Flaveria species ( Gowik et al . , 2011; Heckmann et al . , 2013 ) . When the C4 cycle runs with low capacity , according the model , the surplus of bundle sheath ammonia is transferred back to the mesophyll by the glutamate/2-oxoglutarate shuttle . Once the capacity of the C4 cycle gradually increases , the recirculation of nitrogen is shifted from the glutamate/2-oxoglutarate shuttle towards the C4 cycle ( Figure 5B ) . The predicted biomass production increases linearly with C4 cycle activity ( Figure 5C ) . Thus , our model predicts a strong interaction between C2 and C4 photosynthesis . 10 . 7554/eLife . 02478 . 014Figure 5 . Mechanistic interaction between C2 and C4 cycle . ( A ) Predicted fluxes when the model is parameterized to include activity of the C4 cycle enzymes . Fluxes are given in µmol s−1 m−2 . Values in brackets show minimum and maximum of flux resulting from flux variability analysis . The sum of absolute flux over plasmodesmata was 21 . 9 µmol s−1 m−2 . Flux of dissolved gasses , sucrose , inorganic compounds and processes that carry flux below 1 µmol s−1 m−2 are not shown . See Figure 5—source data 1 for plasmodesmatal fluxes . ( B ) Predicted activities of Ala-AT in mesophyll ( black line ) and bundle sheath ( gray line ) cells and predicted transfer of α-ketoglutarate from mesophyll to bundle sheath cells ( black dashed line ) and glutamate from bundle sheath to mesophyll cells ( gray dashed line ) at low C4 cycle activities . ( C ) Changes in biomass production with varying ( low ) activity of the C4 cycle in a C2 plant . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01410 . 7554/eLife . 02478 . 015Figure 5—source data 1 . Fluxes over plasmodesmata depending on the weight on plasmodesmatal fluxes including flux variability analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 015 When the C2 cycle is running with high capacity , our integrated modeling approach predicts the necessity of auxiliary metabolite fluxes between mesophyll and bundle sheath cells to prevent a massive nitrogen imbalance . Among those auxiliary fluxes were the pyruvate/alanine and the malate/aspartate exchanges . The metabolites used in these shuttles also serve as transport metabolites in C4 photosynthesis . Furthermore , the model highlights the possibility that a low capacity C4 cycle balances part of the C2 cycle ammonia production . Therefore we analyzed in detail the expression of C4 cycle related genes in our dataset . True C4 Flaverias , such as F . bidentis or F . trinervia , are believed to use a NADP-ME type C4 cycle ( Moore et al . , 1984; Ku et al . , 1991; Meister et al . , 1996; Gowik et al . , 2011 ) . All genes associated with this type of C4 photosynthesis are gradually upregulated in the analyzed C3–C4 intermediate species in line with their degree of ‘C4-ness’ . This is true for the typical C4 enzymes like PEPC , PPDK , MDH , NADP-ME , Ala-AT and a plastidic aspartate aminotransferase ( Asp-AT ) , as well as for several C4 associated transporters , such as the pyruvate transporter BASS2 , the H+/Na+ exchanger NHD , the PEP translocator CUE1 and the putative malate and aspartate transporters DIT1 and DIT2 ( Weber and von Caemmerer , 2010; Brautigam et al . , 2011; Furumoto et al . , 2011; Gowik et al . , 2011 ) . The regulators of the C4 enzymes ( like PEPC kinase or the PPDK regulatory protein ) and enzymes with auxiliary functions of C4 enzymes ( like pyrophosphatases or adenosinmonophosphatases ) show a similar pattern ( Figure 6 , Figure 6—source data 1 ) . To corroborate the results of the transcript abundance measurements , selected C4 cycle enzymes ( PEPC , PPDK , and NADP-ME ) were measured by immunoblotting . The protein abundance correlates well with the transcript abundance ( Figure 6 , Figure 6—source data 2 ) . 10 . 7554/eLife . 02478 . 016Figure 6 . Abundance of C4 related transcripts and proteins in leaves of individual Flaveria species . Normalized transcript ( A ) and protein ( B ) levels are plotted as heat maps . Transcript amounts were determined by Illumina sequencing of the leaf transcriptomes and read mapping on selected F . robusta full length transcript sequences . Protein amounts were determined by protein gel blots . See Figure 6—source data 2 for absolute transcript level , Figure 6—source data 2 for protein quantification and Figure 3—figure supplement 1 for immunoblots . ( C ) Mean values of transcript levels from all four experiments were clustered by hierarchical using the HCL module of MEV program with Pearson correlation and the average linkage method . The relative transcript abundance for PEPC , PPDK , NADP-ME and Ala-AT ( mean values from all four experiments ) are plotted for all nine species . Fp: F . pringlei ( C3 ) ; Fro: F . robusta ( C3 ) ; Fc: F . chloraefolia ( C3–C4 ) ; Fpu: F . pubescens ( C3–C4 ) ; Fa: F . anomala ( C3–C4 ) ; Fra: F . ramosissima ( C3–C4 ) ; Fbr: F . brownii ( C4–like ) ; Fb: F . bidentis ( C4 ) ; Ft: F . trinervia ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01610 . 7554/eLife . 02478 . 017Figure 6—source data 1 . Transcript abundance of C4 cycle genes determined by read mapping on F . robusta full length transcript sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01710 . 7554/eLife . 02478 . 018Figure 6—source data 2 . Quantification of C4 proteins by protein gel blots . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 01810 . 7554/eLife . 02478 . 019Figure 6—figure supplement 1 . Results of the protein analyses . Immunoblot results with antibodies against C4 proteins . Fp: F . pringlei ( C3 ) ; Fro: F . robusta ( C3 ) ; Fc: F . chloraefolia ( C3–C4 ) ; Fpu: F . pubescens ( C3–C4 ) ; Fa: F . anomala ( C3–C4 ) ; Fra: F . ramosissima ( C3–C4 ) ; Fbr: F . brownii ( C4-like ) ; Fb: F . bidentis ( C4 ) ; Ft: F . trinervia ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02478 . 019 The expression changes of C4 cycle genes do not all follow the same quantitative pattern ( Figure 6C ) . Although all of these genes gradually increase in expression when plants gain C4 properties , as judged , for example , by the percentage of 14CO2 directly fixed into C4 acids ( Vogan and Sage , 2011 ) , the quantitative changes in gene expression are quite different . PEPC and PPDK transcript amounts increase slowly in the C3–C4 intermediates F . chloraefolia , F . pubescens , and F . anomala , more steeply in the advanced C3–C4 intermediate F . ramosissima and the C4-like species F . brownii before reaching the highest transcript abundances in the true C4 species ( Figure 6C ) . In contrast , NADP-ME and Ala-AT gene expression already increase in expression in the more C3-like intermediate species . Their expression rises more linearly in the further advanced intermediates and plateaus in the C4-like and C4 species . If one uses the different Flaveria species as evolutionary proxies as suggested by the results of Heckmann et al . ( 2013 ) , these results suggest that NADP-ME and Ala-AT are strongly upregulated earlier in evolution than other C4 core enzymes like PEPC or PPDK . F . chloraefolia is classified as a type I C3–C4 intermediate species , and no enhanced C4 cycle activity should be present in this species based on the classification . We detected upregulation of all NADP-ME type associated C4 genes , with some of the genes showing comparable small increases in expression ( Figure 6 ) . This is in line with the results of 14CO2 uptake studies that indicate about 14% of CO2 is directly incorporated into C4 acids in F . chloraefolia , whereas only 6% goes into C4 acids directly in the C3 species F . pringlei ( Moore et al . , 1987 ) . We think therefore that a basal C4 cycle activity is present in F . chloraefolia and its classification as type I C3–C4 intermediate is questionable . A gene encoding a mitochondrial NAD dependent malate dehydrogenase as well as several cytosolic and especially one mitochondrial Asp-AT were upregulated exclusively in the C3–C4 intermediate species and the C4-like F . brownii ( Figure 6 ) . Often , high activities of these genes are associated with the NAD-ME or PEP-CK type of C4 photosynthesis . NAD dependent malic enzyme and PEP carboxykinase genes were only very lowly expressed in all analyzed Flaverias , and no obvious differences between the C3–C4 intermediates and the other species could be found ( Figure 2—source data 2 , data available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . q827h ) . We found no transcriptomic evidence that ammonia is recirculated by the phosphoserine pathway predicted by the model that restricts the free diffusion of all amino acids except serine and glycine . The amounts of transcripts for all three enzymes of this pathway , i . e . , phosphoserine aminotransferase , phosphoserine phosphatase , and 3-phosphoglycerate dehydrogenase , were found to be very low in all analyzed Flaveria species ( Figure 2—source data 2 , data available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . q827h ) ( Mallmann et al . , 2014 ) . Taken together , these data imply that the anaplerotic ammonia shuttle , required to maintain the nitrogen homeostasis in mesophyll and bundles sheath cells of plants performing C2 photosynthesis , is active in all analyzed C3–C4 Flaveria species , as predicted by the computer simulations . Furthermore , it appears that even the most C3-like C3–C4 intermediate species analyzed within the present study , F . chloraefolia , exhibits low level C4 cycle activity . This activity is again in accordance with the in silico model , which predicts the C4 cycle to be a highly efficient ammonia recirculation pathway . The expression of photorespiratory genes , including all genes encoding the core enzymes of the pathway , most of the transporters , and the enzymes involved in ammonia refixation , is not downregulated in the analyzed intermediate species; the transcript and protein amounts remain constant or in some cases are even higher compared to C3 species . A significant drop in photorespiratory gene expression is only observed in the C4-like species F . brownii and is decreased further in the C4 species . Together with earlier results ( Schulze et al . , 2013 ) , this indicates that indeed a C2 photosynthetic cycle is active in all these C3-C4 intermediate Flaveria species and that a reduction in photorespiratory transcripts and proteins only occurs once the amounts of Rubisco have been reduced in the mesophyll as was described for the C4-like species F . brownii ( Bauwe , 1984; Holaday et al . , 1988 ) . Rubisco reduction in the mesophyll is thus a late step of C4 evolution , which in the Flaveria series appears to not occur gradually but rather abruptly towards the end of the evolutionary trajectory . It is followed by a strong increase of C4 cycle activity , as can be deduced from the upregulation of PEPC and PPDK genes in the real C4 species ( Figure 6C ) , when the primary CO2 fixation is completely taken over by PEPC . In the intermediate species C2 and C4 cycles operate in parallel leading to similar or higher photorespiratory gene expression compared with the C3 species . The model of the C2 cycle and the underlying metabolism proposes GS/GOGAT , Ala-AT , and Asp-AT to be involved in balancing the amino groups during C2 cycle operation ( Figure 4 ) . The transcriptome data from the C3-C4 intermediate Flaveria species largely support the results of our integrated model for the C2 pathway ( Figure 6 ) . In these species we found an upregulation of genes involved in the three most likely mechanisms for the recovery of ammonia predicted by the model . GS/GOGAT , which catalyzes the primary refixation of ammonia in the bundle sheath cells , is important for all three versions of ammonia shuttles ( Figure 4 ) and is upregulated in the intermediate species . Transcripts for the glutamate/2-oxoglutarate shuttle , the alanine/pyruvate shuttle , and the aspartate/malate shuttle are enriched in all C3–C4 intermediates compared to the C3 and C4 Flaverias . For the alanine/pyruvate shuttle , Ala-AT is needed in the bundle sheath and the mesophyll cells . Ala-AT is upregulated already in the least advanced C3–C4 intermediates F . chloraefolia and F . pubescens , but also in all the other C3–C4 intermediates . Ala-AT transcripts are also highly abundant in the true C4 Flaverias since Ala-AT is directly involved in the C4 cycle when alanine is used as transport metabolite . We found several Asp-AT and two MDH genes upregulated in the C3–C4 intermediate species ( Figure 6 ) . The chloroplast-located MDH and Asp-AT genes are involved in the C4 cycle of C4 Flaverias , in which malate and aspartate are used concurrently as C4 transport metabolites ( Meister et al . , 1996 ) . Two further Asp-AT genes and another MDH gene were found to be upregulated exclusively in the C3–C4 intermediates , including the C4-like species F . brownii . The most likely reason for upregulation of these genes is their involvement in the recirculation of photorespiratory ammonia by a malate/aspartate shuttle . The pathways of ammonia recirculation between mesophyll and bundle sheath foreshadow the establishment of a true C4 cycle ( Figure 4 ) . All variants described above need the establishment of inter- and intra-cellular transport capacities for amino acids and small organic acids , which are also needed for a functional C4 cycle ( Weber and von Caemmerer , 2010 ) . The existence of an aspartate/malate and an alanine/pyruvate shuttle anticipates important components of a functional C4 pathway . Our transcript data imply that both of these shuttles are active in C3–C4 intermediate Flaverias . Only a few additions would be required to convert these pathways of ammonia recirculation into a C4-like CO2 concentration mechanism , that is , malate would have to be decarboxylated in the bundle sheath cells and pyruvate would have to be converted to malate in the mesophyll . Our transcript data implies that this conversion of the photorespiratory ammonia recirculation pathway into a C4-like CO2 concentrating pump must have been an early event in C4 evolution of Flaveria since already in the least advanced intermediates such as F . chloraefolia and F . pubescens , NADP-ME transcripts are elevated and their amounts increase in parallel with Ala-AT and Asp-AT transcript levels . To extend the pathways of ammonia recirculation into a rudimentary C4 cycle , a capacity to regenerate malate from pyruvate in the mesophyll is required . As deduced from the transcriptome data , the enzymatic functions required are also already enhanced in the least advanced C3–C4 intermediates , since we observe a low but consistent upregulation of PEPC and PPDK genes in these species compared to the C3 Flaverias . Measurements of radiolabeled CO2 incorporation support the view that a rudimentary C4 cycle is already operating in intermediate Flaveria species ( Rumpho et al . , 1984; Monson et al . , 1986; Moore et al . , 1987; Chastain and Chollet , 1989 ) . F . chloraefolia as well as F . pubescens incorporate a higher percentage of 14CO2 into the C4 compounds malate and aspartate ( 11 . 3% and 24 . 9% ) than the C3 species F . pringlei and F . cronquistii ( 4 . 1% and 7 . 7% ) ( Vogan and Sage , 2011 ) . Thus even the least advanced intermediates analyzed in this study run already a low-level C4 cycle , which assists in recycling the ammonia liberated by GDC in the bundle sheath cells . The question arises whether amino group transfer initially exclusively happened via amino acid/oxoacid pairs or whether the enzymatic content of C3 plants immediately supported a shuttle that also involved decarboxylation and carboxylation reactions . C3 plants have considerable capacity for the decarboxylation of four-carbon organic acids in their bundle sheath cell ( Hibberd and Quick , 2002; Brown et al . , 2010 ) and measurements of total leaf NAD-ME and NADP-ME activity in C3 plants repeatedly demonstrated basal activities for various C3 species ( Wheeler et al . , 2005; Aubry et al . , 2011; Maier et al . , 2011 ) . C3 plants also accumulate high amounts of organic C4 acids like malate or fumarate during the day ( Zell et al . , 2010 ) , which are produced by PEPC , the only enzyme capable of producing C4 acids de novo . It is tempting to hypothesize that plants use a malate/alanine shuttle to recycle parts of the ammonia liberated by glycine decarboxylation from the very beginning of the C2 cycle . If the C4 cycle is superimposed onto a C2 cycle operating in a C3–C4 intermediate plant , the C2 photosynthesis model predicts a mechanistic interaction between the C2 and C4 cycles ( Figure 5 ) . When the C4 cycle is running , the photorespiratory ammonia is recirculated from the bundle sheath to the mesophyll cells by moving malate from the mesophyll to the bundle sheath and transferring alanine back to the mesophyll . This malate/alanine cycling leads to a net transport of ammonia from the bundle sheath into the mesophyll cells . In contrast to the other mechanisms of ammonia recirculation described above , the C4 cycle does not only lead to a net transport of ammonia from the bundle sheath to the mesophyll but additionally also to a net transport of CO2 in the opposite direction . Thus CO2 is transferred from the mesophyll to the bundle sheath without increasing the number of transport processes between the cells . By elevating the CO2 concentration in the bundle sheath cells the C4 cycle acts cooperatively with the C2 cycle . The bundle sheath Rubisco would work under a more elevated CO2 concentration and thus operate more effectively compared to a pure C2 plant , leading to an increased biomass production . The C4 cycle thus has a dual beneficial effect: an efficient nitrogen shuttle is combined with a CO2 concentrating pump . To investigate the possible interaction with regard to biomass , a C4 cycle at the enzyme capacities of C3 plants was allowed and tested for biomass changes ( Figure 5 C ) . When the C4 cycle is running with PEPC activities comparable to those found in C3 Flaveria species , the model already predicts a gain in biomass production compared to the C2 cycle on its own . Under these conditions , the bulk of photorespiratory ammonia is recycled through a rudimentary C4 cycle limited by the C4 cycle flux capacity . The model predicts that biomass production will be further enhanced with higher activity of the C4 cycle . Consequently , there is permanent positive selection on enhancing the activity of the currently rate limiting enzyme once a C4 cycle is running . The evolutionary scenario described above is in good agreement with the Flaveria transcriptome data . We observe gradual increases in the amounts of C4 transcript with increasing ‘C4-ness’ of the C3–C4 intermediates until the most advanced species F . brownii . The abundance of NADP-ME and Ala-AT transcripts increases faster than the transcript abundance of the other core C4 genes like PEPC , PPDK , MDH or Asp-AT . This implies that these evolutionary changes were driven by selection on high bundle sheath decarboxylation capacity , consistent with the idea that the C4 cycle began as an auxiliary pathway to the C2 cycle to recirculate photorespiratory ammonia . Hence , in this early phase , the main purpose of the C4 cycle was to provide the ammonia acceptor pyruvate . The C2 model and its evolutionary implications are consistent with the properties of the C3–C4 intermediate Flaveria species including F . brownii , which possess mesophyll Rubisco activity and consequently the C2 photosynthetic pathway . The next iteration during C4 evolution in Flaveria must have been the restriction of Rubisco activity to the bundle sheath , making the C2 cycle obsolete , as observed for the true C4 Flaveria species . The establishment of a photorespiratory CO2 pump , termed C2 photosynthesis , is thought to be an important step in C4 evolution . Recent work has shown how C3 Flaverias were preconditioned for the evolution of the C2 pathway and how the C2 cycle was implemented on the molecular level ( Sage et al . , 2013; Schulze et al . , 2013 ) . Together with the present work , this gives us a detailed picture of what happened in the early and intermediate stages during C4 evolution in Flaveria . We have argued that the establishment of the C2 cycle requires the implementation of at least components of the C4 pathway , if not the whole pathway . This fact might be a partial explanation for the polyphyletic evolution of C4 photosynthesis . Only the C2 cycle has to evolve to set a system on a slippery slope towards C4 photosynthesis . Nature seems to confirm this idea . So far , 66 independent origins of C4 photosynthesis could be identified . In contrast , there are only seven known groups with independent origins of C2 plants and no direct ancestry to C4 species ( Sage et al . , 2012 ) . If one assumes that all recent C4 lineages evolved via C2 intermediates , which appears likely ( Sage et al . , 2012; Heckmann et al . , 2013; Williams et al . , 2013 ) , this would mean that the C2 pathway evolved 73 times independently and that over 90% of these C2 plant containing lineages proceeded to C4 photosynthesis . This indicates that the C2 photosynthetic pathway must indeed be a strong enabler of C4 photosynthesis . It will be highly enlightening to analyze these C2 groups without ancestry to C4 species , like Moricandia , Steinchisma or Mollugo , to find out in how far they differ from groups that evolved the C4 pathway and why C4 evolution may have been hampered in these groups . The close evolutionary interconnection of the C2 and the C4 pathway could be seen as an example of metabolic exaptation ( Barve and Wagner , 2013 ) . Exaptation or pre-adaptation was defined as an adaptation involving the co-option of traits that originally evolved for a different purpose ( Gould and Vrba , 1982 ) . While both C2 and C4 act as carbon shuttles to the bundle sheath cells , the two systems achieve this goal through different biochemical processes . In particular , the amino acid shuttle in the C2 system evolved to transport nitrogen , and its later use in C4 photosynthesis to shuttle carbon thus represents a molecular exaptation . Our findings therefore corroborate the general idea that the evolution of complex traits may be accelerated through exaptations ( Darwin , 1872; Gould and Vrba , 1982; Barve and Wagner , 2013 ) . We do not know if the scenario on the early and intermediate stages of evolution described above is limited to the genus Flaveria or if it is valid for C4 evolution in general . Our prediction of the C2 pathway being a strong facilitator of C4 evolution should apply to all C4 origins , as the integrated model is not specific to Flaveria . F . pringlei , F . robusta , F . chloraefolia , F . pubescens , F . anomala , F . ramosissima , F . brownii , F . bidentis and F . trinervia plants were grown in the green house at University of Duesseldorf side-by-side and harvested at four different points of time over the year . The plants were grown in 17-cm pots on soil ( C-400 with Cocopor [Stender Erden , Schermbeck Germany] fertilized with 3 g/l Osmocote exact standard 3 to 4 M [Marysville , USA] ) with additional light for 16 hr per day until 50 to 60 cm height and before the onset of flowering . Plants for experiment one were harvested in September , for experiment two in June , for experiment three in October and for experiment four in April . The plant material was immediately frozen in liquid nitrogen , stored at −80°C and used for the following analyses . Total RNA was isolated from the second and fourth leaves according to ( Westhoff et al . , 1991 ) followed by a DNAse treatment . After phenol/chloroform extraction and precipitation with NaAc and isopropyl alcohol the RNA was dissolved in H2O . The RNA quality was tested with the Agilent 2100 bioanalyzer . 1 µg of total RNA was used for cDNA library generation , which was accomplished with the TruSeq RNA Sample Preparation Kit ( Illumina Inc . , San Diego , USA ) via the Low-Throughput Protocol ( TruSeq RNA Sample Preparation Guide , Illumina Proprietary Catalog # RS-930-2001 , Part # 15008136 Rev . A , November 2010 ) . Clusters were generated with the TruSeq SR Cluster Kit v2 according to the Reagent Preparation Guide with the Illumina cBot device . The single read sequencing was performed with the Illumina HiSeq2000 . Sequences of transcripts from genes involved in photorespiraton , C4 photosynthesis and refixation and recirculation of photorespiratory ammonia were identified among de novo assembled transcripts of F . robusta . De novo assembly was performed with either CLC Genomics Workbench ( CLC-Bio , Aaarhus , Denmark ) or the Velvet/Oases software package ( Schulz et al . , 2012 ) using F . robusta 454 ( Gowik et al . , 2011 ) and Illumina reads ( this study ) . After quality control and processing , Illumina reads were aligned to the F . robusta transcript sequences with the CLC Genomics Workbench using standard parameters . Read mapping against a minimal set of coding sequences ( Brautigam et al . , 2011 ) of the TAIR 9 release of the Arabidopsis thaliana genome ( http://www . Arabidopsis . org/ ) was performed using BLAT ( Kent , 2002 ) as described in ( Gowik et al . , 2011 ) . The MEV software package ( http://www . tm4 . org/mev . html ) was used for plotting heat maps , hierarchical clustering and principal component analysis . Total proteins were isolated from plant material harvested together with the material for RNA isolation according to Shen et al . ( 2007 ) and quantified using the RC-DC protocol ( Bio-Rad Laboratories , Hercules , USA ) . 30 µg of total protein was electrophoresed on polyacrylamide-SDS gels ( Schägger and von Jagow , 1987 ) and electrophoretically transferred to nitrocellulose membranes ( Protran BA85 , 0 . 45 μm; Schleicher & Schuell , Dassel , Germany ) for 1 hr with 0 . 8 mA per cm2 . Specific primary antibodies were raised against conserved Flaveria peptides ( Agrisera Vännäs , Sweden ) . For the detection of specific proteins the nitrocellulose membranes were incubated with the primary antibodies and a Horseradish peroxidase-conjugated secondary antibody ( Sigma-Aldrich , St . Louis , USA ) . An enhanced chemiluminescent Horseradish peroxide substrate was added and signals were recorded using a Fuji LAS-4000 mini CCD camera system . The signals were quantified with the Multi Gage analysis software ( Fujifilm , Tokyo , Japan ) . As loading control a gel was stained for 45 min with 0 . 25% Coomassie blue , 50% methanol , 7% acetic acid , and destained in 50% methanol , 7% acetic acid . In order to model the metabolic integration of C2 and C4 cycle in the context of leaf metabolism , we conducted Flux Balance Analysis ( FBA ) based on a genome-scale metabolic reconstruction of C4 metabolism , C4GEM ( Dal'Molin et al . , 2010 ) . This reconstruction contains a complex biomass reaction including carbohydrates , cell wall components , amino acids and nucleotides ( Dal'Molin et al . , 2010 ) . FBA is a powerful tool to understand the adaptation of metabolism on a genomic scale . Since metabolite concentrations are not modeled explicitly , fluxes related to carbon concentration mechanisms ( CCMs ) cannot be captured by this constraint-based approach alone . To account for this issue , we coupled the FBA model with a mechanistic model of C3–C4 photosynthesis ( von Caemmerer , 2000; Heckmann et al . , 2013 ) . C4GEM representing NADP-ME types was provided by the authors and FBA was conducted using this model:Maximize cTv subject to Sv=0 . vmin , i≤vi≤vmax , iwhere c is the vector of coefficients in the objective function , here the leaf biomass production . v is the vector of fluxes through the network reactions , S is the stoichiometric matrix of the metabolic network , and vmin and vmax represent constraints on the respective fluxes . In order to test hypotheses concerning nitrogen metabolism in C3–C4 intermediate plants , S had to be modified . The plasmosdesmatal transport reactions in the original C4GEM model include malate , pyruvate , 3-phosphoglycerate , trioses , phosphates , sucrose , aspartate , alanine , phosphoenolpyruvate , CO2 , and O2 . Reactions were added to S in order to include transport of serine , glycine , glutamate , glutamine , asparagine , threonine , 2-oxoglutarate and water over the mesophyll/bundle sheath interface . Furthermore , the lack of photosystem II in the bundle sheath of certain C4 plants does not hold in our scenario ( Nakamura et al . , 2013 ) and we added a reaction for linear electron transport to the bundle sheath . C4GEM does not contain a reaction for a plastidal NADP-dependent malate dehydrogenase in the bundle sheath; we added this reaction to S . In addition to the stoichiometric matrix S , the constraints used in C4GEM were modified: The original constraint on leaf sucrose production was changed to result in an output ratio of sucrose to amino acids of about 5 ( Riens et al . , 1991 ) . Fixed constraints on production of starch and fatty acids are not appropriate in the coupled framework . Since we are not aware of data that explains how these fluxes scale with net CO2 assimilation rate , the constraints were removed from the model . Reactions belonging to the GS/GOGAT system were assumed to be irreversible . Nitrogen is available in the form of nitrate as opposed to NH3 in the original model . Since there is no evidence suggesting mesophyll specificity of PEPC in intermediate Flaveria species , we unconstrained PEPC flux in the bundle sheath . To couple the genome-scale FBA model with the mechanistic model of carbon fixation , the following reactions were constrained using the values predicted by the mechanistic model: net CO2 uptake , Rubisco carboxylation and oxygenation in mesophyll and bundle sheath , CO2 leakage from the bundle sheath , PEPC activity in the mesophyll , activity of NADP-ME in the bundle sheath , plasmodesmatal flux of glycine and serine and decarboxylation by the GDC complex . The lower bound on glycine diffusion ( Vmin , Gly ) , serine diffusion ( Vmin , Ser ) , and GDC reaction ( Vmin , GDC ) can be obtained from the rate of Rubisco oxygenation in the mesophyll ( Vom ) and the fraction of photorespiratory CO2 in the bundle sheath derived from mesophyll oxygenations ( ξ ) :vmin , Gly=ξVom , vmin , Ser=0 . 5ξVom , vmin , GDC=0 . 5ξVom The mechanistic model was parameterized to the C3 state as given in Heckmann et al . ( 2013 ) , with the exception of the parameter ξ , which was set to a value of 0 . 98 ( i . e . , the majority of GDC activity was restricted to the bundle sheath . Derivation from transcriptome data is given in Heckmann et al . ( 2013 ) ) . These constraints on the reactions of the photorespiratory pump are necessary to adequately predict C2 photosynthesis because of the inability of FBA alone to model CCMs ( see discussion above ) . In the FBA part of the model , a minimization of total flux ( MTF ) analysis was conducted in order to narrow down the space of optimal solutions:Minimize ∑i=1nwi\|vi\| subject to: Sv=0 . vmin , i≤vi≤vmax , icTv=cTvFBAwhere vFBA is the flux distribution of the FBA optimization described above . w denotes a vector of weights , where plasmodesmatal flux received a higher weighting factor ( 1 . 1 for plasmodesmatal exchange , 1 for the remaining reactions ) . This method implements a simple minimization of protein costs for a given optimal biomass production . The higher weights on plasmodesmatal fluxes account for the trade-off between CO2 containment in the bundle sheath and metabolite diffusion between the cells . Since this trade-off is difficult to quantify , we conducted a sensitivity analysis by varying the weight on plasmodesmatal transport reactions . In order to investigate the possible range that fluxes can take while yielding an optimal solution , flux variability analysis was conducted: For each vi: Maximize or Minimizevi . Subject to: Sv=0 . vmin , i≤vi≤vmax , icTv=cTvFBA∑i=1nwi\|vi\|=sopt Where sopt is the minimum for the weighted sum of absolute flux found in the MTF optimization . All simulations were conducted in the R environment for statistical computing ( R Core Team , 2013 ) using the sybil library ( Gelius-Dietrich et al . , 2013 ) . The read data have been submitted to the National Center for Biotechnology Information Short Read Archive under accession numbers SRP036880 ( F . bidentis ) , SRP036881 ( F . anomala ) , SRP036883 ( F . brownii ) , SRP036884 ( F . chloraefolia ) , SRP036885 ( F . pringlei ) , SRP037526 ( F . pubescens ) , SRP037527 ( F . ramosissima ) , SRP037528 ( F . robusta ) and SRP037529 ( F . trinervia ) .	Environmental pressures sometimes cause different organisms to independently evolve the same traits . A dramatic example of this phenomenon , which is called convergent evolution , can be seen in the modes used by plants to convert carbon dioxide from the air into starch during photosynthesis . Early plants existed in an environment with high levels of carbon dioxide in the air . Over time , carbon dioxide levels decreased , so plants evolved more efficient types of photosynthesis to cope . A very efficient type of photosynthesis , called C4 photosynthesis essentially represents a carbon dioxide concentration mechanism . It has evolved at least 62 times independently in 19 different families of flowering plants . Scientists have shown that a less advanced , low-efficiency version of photosynthetic carbon dioxide concentration , called C2 photosynthesis , is a stepping-stone to C4 photosynthesis . It is also known that the evolution of C4 photosynthesis required changes to the expression patterns of thousands of genes , but the exact mechanism that leads from C2 photosynthesis to C4 photosynthesis is not clear . To explore this in greater detail , Mallmann , Heckmann et al . studied plants from the genus Flaveria , which belongs to the same family as sunflowers and asters . Under identical greenhouse conditions , plants that use three different photosynthetic pathways—C3 photosynthesis , C4 photosynthesis , or an intermediate between the two—were grown and their gene expression patterns were compared . Computer simulations were used to model the metabolism of plants that relied on C2 photosynthesis . Based on the modeling , it appears that C2 photosynthesis shifts the balance of nitrogen metabolism between two types of cell that are critical to photosynthesis . To rebalance the nitrogen , several genes are expressed to trigger an ammonia recycling mechanism . The same genes are turned on during C4 photosynthesis , and this recycling mechanism include parts of the C4 process . The findings of Mallmann , Heckmann et al . suggest that the initial steps in C4 photosynthesis evolved to prevent nitrogen imbalance . Over time , this mechanism was co-opted to become part of a more efficient form of photosynthesis , which may explain why so many different plants evolved from C2 to C4 photosynthesis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "evolutionary", "biology" ]	2014	The role of photorespiration during the evolution of C4 photosynthesis in the genus Flaveria
Protein output from synonymous codons is thought to be equivalent if appropriate tRNAs are sufficiently abundant . Here we show that mRNAs encoding iterated lysine codons , AAA or AAG , differentially impact protein synthesis: insertion of iterated AAA codons into an ORF diminishes protein expression more than insertion of synonymous AAG codons . Kinetic studies in E . coli reveal that differential protein production results from pausing on consecutive AAA-lysines followed by ribosome sliding on homopolymeric A sequence . Translation in a cell-free expression system demonstrates that diminished output from AAA-codon-containing reporters results from premature translation termination on out of frame stop codons following ribosome sliding . In eukaryotes , these premature termination events target the mRNAs for Nonsense-Mediated-Decay ( NMD ) . The finding that ribosomes slide on homopolymeric A sequences explains bioinformatic analyses indicating that consecutive AAA codons are under-represented in gene-coding sequences . Ribosome ‘sliding’ represents an unexpected type of ribosome movement possible during translation . Messenger RNA ( mRNA ) transcripts can contain errors that result in the production of incorrect protein products . Both bacterial and eukaryotic cells have evolved mechanisms to deal with such errors which involve ( 1 ) proteolytic degradation of the aberrant protein product , ( 2 ) mRNA decay and ( 3 ) ribosome rescue ( Shoemaker and Green , 2012 ) . One such mRNA surveillance pathway in eukaryotes targets mRNAs that lack stop codons ( Non-Stop-Decay or NSD ) . In these cases , actively translating ribosomes are thought to read into the 3′ terminal poly ( A ) sequence of the mRNA triggering ribosome pausing as poly ( lysine ) is translated , followed by the recruitment of ubiquitin ligases , mRNA decay and ribosome recycling factors ( review Klauer and van Hoof , 2012 ) . Given the substantial amount of premature ( or alternative ) polyadenylation that has been documented in eukaryotes ( Ozsolak et al . , 2010 ) , it seems that such an mRNA surveillance pathway might have considerable biological significance . Similarly , in bacteria , while no ‘NSD-like’ response has been characterized , it is known that poly ( A ) sequences are added to mRNAs in the process of being degraded ( review Dreyfus and Régnier , 2002 ) , and so ribosomes on these mRNAs may encounter similar challenges . The utilization in bacteria and eukaryotes of 3′ poly ( A ) tails as non-coding elements may reflect a common solution to the challenges for the ribosome in translating such sequences . Most studies investigating how NSD works have been conducted in yeast using reporter constructs . Early studies in Saccharomyces cerevisiae revealed that mRNAs lacking stop-codons are targeted for decay both in a reaction dependent on the exosome-associated factor Ski7 ( van Hoof et al . , 2002 ) and in a more canonical degradation reaction involving decapping and 5′ to 3′ exonucleolytic degradation ( Frischmeyer et al . , 2002 ) . Other factors involved in NSD have since been discovered; these include Dom34 and Hbs1 which facilitate ribosome rescue during NSD ( Izawa et al . , 2012; Tsuboi et al . , 2012 ) , Ltn1 and Not4 which ubiquitinate the protein products on non-stop mRNAs ( Dimitrova et al . , 2009; Bengtson and Joazeiro , 2010 ) , and a number of other factors genetically identified as critical for poly ( basic ) -mediated stalling ( Kuroha et al . , 2010; Brandman et al . , 2012; Chiabudini et al . , 2014 ) . Although many players in NSD have been identified and their functions defined , there remain critical gaps in our understanding . In this manuscript , we focus on what must be the earliest events in NSD , the translation of poly ( lysine ) sequences by the ribosome . NSD is widely thought to be triggered by unfavorable electrostatic interactions that occur in the ribosomal exit tunnel when ribosomes translate the poly ( lysine ) sequences encoded by poly ( A ) . Indeed , biochemical studies in rabbit reticulocyte lysate with proteins interrupted by iterated poly ( lysine ) and poly ( arginine ) sequences indicate that positively charged residues do slow translation and produce transiently arrested species ( Lu and Deutsch , 2008 ) . Other examples of peptide-mediated stalling have also been documented in bacterial and eukaryotic systems . In some cases , such as the tnaC gene , secM , or ermCL in bacteria , the peptide stalling motif is several amino acids in length and appears to specifically engage the contours of the exit tunnel to elicit stalling ( Gong and Yanofsky , 2002; Nakatogawa and Ito , 2002; Vazquez-Laslop et al . , 2008; Seidelt et al . , 2009; Bhushan et al . , 2011; Ito and Chiba , 2013; Arenz et al . , 2014 ) . Poly ( proline ) sequences have recently been shown to cause stalling during translation in bacteria and eukaryotes in the absence of specialized bypass factors , EFP and eIF5A , respectively ( Doerfel et al . , 2013; Gutierrez et al . , 2013; Ude et al . , 2013 ) . In this case , proline is thought to adopt a conformation that interferes with the ribosome active site geometry . Here we take a high-resolution biochemical look at the molecular events that occur when the ribosome translates poly ( lysine ) peptides . We find that insertion of consecutive AAA lysine codons into reporters has a stronger negative impact on protein expression than insertion of an equivalent number of AAG lysine codons in both eukaryotes and bacteria . Kinetic and toeprinting studies in an in vitro reconstituted Escherichia coli translation system reveal that differential protein output is the downstream consequence of ribosome pausing followed by an unanticipated ribosome movement on successive AAA codons that we refer to as ‘sliding’ . When sliding occurs in the middle of genuine ORFs in a cell , frame is lost and ribosomes encounter out of frame stop codons that result in canonical ( stop-codon mediated ) termination . In eukaryotes , such premature termination events target the mRNA for non-sense mediated decay ( NMD ) . The finding that the ribosome can robustly slide on poly ( A ) sequences explains bioinformatic analyses revealing that consecutive AAA codons are under-represented in ORFs in all genomes ( unpublished data ) and helps to rationalize the widespread usage of poly ( A ) sequence as a regulatory rather than a coding feature . To begin investigating the translation of poly ( lysine ) -encoding sequences , we created a series of mCherry- and luciferase-based reporter constructs ( Figure 1A ) containing no insert , glutamic acid ( GAA ) repeats , or consecutive lysine residues encoded by various combinations of AAA and AAG codons . These reporters were introduced into S . cerevisiae and E . coli cells and the protein products visualized by either luminescence or fluorescence , respectively ( Figure 1B ) . The insertion of twelve consecutive negatively charged glutamic acid residues ( GAA ) had no negative impact on production of the reporter protein ( Figure 1B ) . By contrast , the addition of consecutive lysine residues generally resulted in overall less protein production ( Figure 1B ) , consistent with previous studies of poly ( lysine ) -containing reporters ( Ito-Harashima et al . , 2007; Lu and Deutsch , 2008; Chiabudini et al . , 2012 ) . Interestingly , we find that protein output from the poly ( lysine ) -containing reporters is codon dependent in both bacteria and yeast; reporters containing iterated AAG lysine codons generate more protein than those with an equivalent number of synonymous AAA codons ( Figure 1B ) . The relative differences in expression of AAG- vs AAA-encoded poly ( lysine ) -containing reporters in E . coli and S . cerevisiae are comparable ( 4 ± 0 . 3-fold more in E . coli and 3 ± 1-fold more in S . cerevisiae from reporters with AAG12 vs AAA12 ) . 10 . 7554/eLife . 05534 . 003Figure 1 . Protein production is differentially diminished by iterated lysine codons ( AAA vs AAG ) in E . coli and S . cerevisiae . ( A ) Schematics of the mCherry ( top ) and luciferase ( middle , and bottom ) reporters used in this study . The mCherry reporter contains an N-terminal thioredoxin ( Thrdx ) domain , 3HA-tag , sequence of interest ( black section ) , followed by the C-terminal mCherry sequence . The top luciferase reporter includes a 2HA tag followed by sequences of interest ( used for study in Figure 1B ) . The second luciferase reporter ( used in Figure 6 ) has sequences of interest at the N-terminal end of Renilla . Firefly is used in this construct as an internal control in the second luciferase reporter . ( B ) Relative amounts of protein expressed from reporters expressed in E . coli ( mCherry , red ) and S . cerevisiae ( luciferase , green ) . Error bars results from for the standard error of at least three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 003 One potential explanation for the codon-dependent expression of poly ( lysine ) -containing proteins could be that the ribosome more rapidly incorporates lysine on AAG than AAA codons . In E . coli a single tRNA with a UUU anti-codon decodes both lysine codons ( Chan and Lowe , 2009 ) , making E . coli an excellent system for studying differences in the production of poly ( lysine ) peptides . We measured the rate of lysine incorporation using a previously described reconstituted E . coli translation system ( Youngman et al . , 2004; Gromadski et al . , 2006; Zaher and Green , 2009 ) on a series of lysine-encoding simple mRNAs including: AUG-AAA-UUC-AAG-UAA ( MKFK-Stop ) , AUG-UUC-AAA ( MFK ) , AUG- ( AAA or AAG ) 5-UAA ( MK ( A or G ) 5-Stop ) . Only Lys-tRNALys was included during the translation of MKFK-Stop and MK5-Stop mRNAs while both Lys-tRNALys and Phe-tRNAPhe were present when MFK was translated . Electrophoretic TLC ( eTLC ) readily resolved the reaction products allowing for analysis of intermediate and complete peptide products ( Figure 2A ) . The quantitated data were modeled in Mathematica using the kinetic scheme displayed in Figure 2B ( see ‘Material and methods’ ) . These experiments reveal that addition of a single lysine in a heteropolymeric sequence is rapid and independent of whether lysine is the first or second amino acid incorporated ( Figure 2C , rate constants for formation of MK and MFK peptides are 12 s−1 and 7 s−1 , respectively ) ; these rates are similar to those typically measured for peptide bond formation in this in vitro system ( Gromadski et al . , 2006 ) . For messages containing iterated lysine codons , the rate constant for translating the first lysine codon is similarly fast ( k1 , obs from 2–19 s−1 , Figure 2C ) for AAA and AAG codons . However , subsequent lysines in an iterated sequence are added with considerably slower kinetics on both AAA ( k2 , obs = 0 . 0005 and k3 , obs = 0 . 0003 s−1 ) and AAG codons ( k2 , obs = 0 . 009 and k3 , obs = 0 . 015 s−1 ) ( Figure 2C ) . We note that the rate of second lysine addition during the translation of MK5-STOP messages are somewhat slower on AAA relative to AAG codons , potentially partially explaining the decreased overall protein output on these mRNAs . More importantly , however , these data show that the reactivity of the second Lys-tRNALys on iterated lysine containing messages ( such as MK5-Stop ) is substantially reduced ( at least 130-fold ) on both lysine codon-containing mRNAs relative to normal elongation rates . Interestingly , the addition of a second lysine to messages with fewer sequential lysine codons ( such as MK2F-STOP ) does not exhibit such a striking kinetic defect ( k2 , obs is not largely affected , data not shown ) . These data suggest that the identity of the message ( i . e . a long poly ( A ) sequence ) plays a critical role in the observed slowing of elongation . Toeprinting assays performed using the E . coli PURE cell-free translation system are consistent with these observations; E . coli ribosomes stall when the second lysine codon in iterated ( AAA ) - and ( AAG ) -codon containing sequences is positioned in the A site ( Figure 2—figure supplement 1 ) . Together , these results reveal that translating consecutive lysines in a poly ( lysine ) peptide sequence , either on iterated AAA or AAG codons , can lead to substantial kinetic delays in vitro . 10 . 7554/eLife . 05534 . 004Figure 2 . Kinetic defect observed on addition of second and third lysine residues in iterated lysine stretch . ( A ) Example eTLC displaying the E . coli translation products of a AUG- ( AAA ) 5-UAA message . The ± poles of the electrophoretic TLC are indicated . MK4 and MK5 products ( and those with greater numbers of lysine ) are difficult to resolve in this system but the other products are easily visualized . ( B ) Kinetic scheme for rate constants of sequential lysine additions to peptide chain . ( C ) Bar graph displaying rate constants for the addition of individual lysines to a variety of messages: MKFK-Stop ( gray ) , MKA5-Stop ( blue ) , MKG5-Stop ( black ) , and MFK ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 00410 . 7554/eLife . 05534 . 005Figure 2—figure supplement 1 . Ribosomes stall while adding a second lysine . Toeprinting assays were performed with constructs containing 1–12 consecutive lysines inserted ( either AAG and AAA codons ) . Assays were performed in the presence and absence of thiostrepton to mark ribosomes on the initiating AUG codon . Sequences on which toeprints appear are highlighted in red . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 00510 . 7554/eLife . 05534 . 006Figure 2—figure supplement 2 . Modeling of rate constants in Mathematica . ( A ) Kinetic scheme used to model the rate constants of sequential lysine additions to the peptide chain ( same as Figure 2A ) . We also attempted to model with peptidyl-tRNA drop-off rates included . Inserting peptidyl-tRNA drop-off into our model decreases the quality of fits , and returns rates of drop-off small enough that they are inconsequential relative to the time scale of the reaction . ( B ) The top panel displays the differential equations used to solve for each rate constant . The bottom panels display the mathematical solutions for the differential equations . These solutions were used to perform modeling and fit the data . The fits were performed both iteratively ( e . g . , we solved for k1 by fitting the plots measuring the disappearance of M , then input that value into the equation describing the appearance of MK to solve for k2 ) and by letting all of the values float for each data set . In both cases , the rate constants modeled were essentially the same , indicating that the first lysine is added quickly ( k1 ) , and subsequent lysines ( k2 , k3 ) are added more slowly . ( C ) An example fit in Mathematica showing time course for the formation and depletion of MK on a message with AAG codons . This time course , for example , was used to model the k2 value . ( D ) R2 values for the fits for the appearance and disappearance of each species used to model rate constants . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 006 As we explored the kinetics of lysine incorporation , we evaluated the ability of the ribosome to translate a variety of MK ( A or G ) 2 di-lysine messages ( Figure 3A ) . Unexpectedly , we found that messages containing iterated AAA codons generate extended peptides longer than the designed coding sequence ( Figure 3A ) . When E . coli initiation complexes ( programmed with fMet-tRNAfMet ) are reacted with Lys-tRNALys on messages containing two consecutive lysine codons followed by a variety of non-lysine codons ( Phe ( UUC ) , Val ( GUC ) , or Stop ( UAA ) ) , only MKK peptide should be synthesized . However , we see the formation of a majority population of extended peptide product containing at least four lysines on all messages with two consecutive AAA codons ( Figure 3B , lanes 2-4 ) . In contrast , equivalent messages with two AAG codons predominantly form the expected MKK product ( Figure 3B , compare lane 3 vs 5 ) . We also find that a mixed sequence of lysine codons ( AAA-AAG ) can form some extended peptide ( Figure 3—figure supplement 1 ) . These data suggest that 5 As in a row are sufficient to promote the addition of extra lysines in vitro . We note that the identity of the codon that follows the di-lysine sequence is not relevant to the observed amount of extended peptide product ( Figure 3B , Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 05534 . 007Figure 3 . E . coli ribosomes add extra lysines on messages containing two sequential AAA , but not AAG , lysine codons . ( A ) Illustration of the ribosome on the entire MKA2-Stop message . ( B ) eTLCs showing the peptide products resulting from translation of indicated messages with Lys-tRNAlys ( but no other tRNAs or release factors ) present . ( C ) eTLC displaying the peptide products resulting from the translation of indicated messages in the presence of Lys-tRNALys alone , or in the presence of Lys-tRNALys + factors ( either RF1 or Phe-tRNAPhe ) necessary for messages to be fully translated . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 00710 . 7554/eLife . 05534 . 008Figure 3—figure supplement 1 . E . coli ribosomes add extra lysines to peptides translated on messages containing sequential AAA-AAG lysine codons . TLC showing all of the peptide products resulting from translation of MKA2V-Stop , MKA3-Stop , MKA4-STOP , and MKAKGF-Stop messages with Lys-tRNAlys ( but no other tRNAs or release factors ) present . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 00810 . 7554/eLife . 05534 . 009Figure 3—figure supplement 2 . Quantification of the percentage of translated peptide containing more lysine residues than expected . Translation reactions were run in the presence of either Lys-tRNAlys only , or Lys-tRNAlys and other factors ( Phe-tRNAPhe or RF1 ) . All errors bars represent the standard error from at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 00910 . 7554/eLife . 05534 . 010Figure 3—figure supplement 3 . T7 transcribed messages visualized on 15% denaturing PAGE gel . ( A ) In vitro transcribed mRNAs used in our in vitro studies run as distinct , single bands on high-resolution denaturing PAGE gel . The RNA is visualized with methylene blue stain . ( B ) The mRNAs encoding consecutive AAA codons result in discrete length toeprint signatures , yielding specific bands corresponding to the full-length message on our toeprints . We also performed RACE experiments on in vitro T7 transcribed mCherry reporter mRNAs containing A18-36 sequences and found that with high frequency , our RNAs contained the expected number of As . Importantly , in both the cell-free system and in vivo , T7 RNA polymerase is responsible for transcribing the mRNAs relevant to the output . Together , these data provide strong evidence that the mRNAs utilized throughout this study are accurately transcribed by T7 RNA polymerase . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 010 The production of peptide products containing more than the encoded number of lysines is surprising , especially given that there are no nearby upstream or downstream in-frame or out-of-frame lysine codons in these mRNAs ( Figure 3A ) . We speculate that these extended peptides result from the ribosome repeatedly moving backwards by at least three nucleotides to position an AAA Lys codon in the A site , and then subsequent standard peptide bond formation . Toeprinting assays performed on iterated AAA- and AAG-containing mRNAs provide further support for such irregular movement of ribosomes specifically on iterated AAA codons ( Figure 2—figure supplement 1 ) ; the toeprint on the iterated AAA sequence is diffuse relative to the discrete toeprint seen on iterated AAG sequence . In the course of performing our experiments we carefully considered reports suggesting that T7 RNA polymerase could promiscuously add extra adenosines to poly ( A ) messages ( Tsuchihashi and Brown , 1992; Ratinier et al . , 2008 ) ; no experiment that we performed revealed any evidence for such heterogeneity in our mRNA products ( Figure 3—figure supplement 3 ) . Unlike better studied −1 and +1 frameshifting events , these data suggest that ribosomes on iterated AAA sequences are making unexpected and large excursions from their initial frame; we refer to this process as ‘ribosome sliding’ . The observation of ribosome sliding on iterated AAA codons is surprising given that the ribosome must somewhat regularly translate mRNA sequences in vivo that contain two consecutive AAA codons . While three or more AAA codons in a row are selected against in gene coding sequences , there are thousands of examples of two consecutive AAA codons in S . cerevisiae and E . coli genes ( see further details in bioinformatic analysis below , Table 1 ) . In the experiments described in Figure 3A , ribosome initiation complexes formed on the specified MKA2-Stop and MKA2F-Stop messages ( Figure 3B ) were only supplied with Lys-tRNALys and essential elongation factors; the subsequent substrates normally present in vivo after the formation of MKK peptide were left out . To determine if ribosome sliding occurs in more typical circumstances , we performed elongation reactions on the same mRNAs , but where both Lys-tRNALys and the relevant other downstream substrates ( release factor 1 ( RF1 ) or Phe-tRNAPhe ) were added to the ribosome initiation complexes . The result is clear; in this latter case , the anticipated MKKF or MKK peptide products are predominantly generated ( Figure 3C , Figure 3—figure supplement 2 ) . These data suggest that ribosome sliding on iterated AAA sequences occurs more slowly than the normal rate of peptidyl transfer with Phe-tRNAPhe or RF1-catalyzed peptide release , respectively . Moreover , these results readily explain how the ribosome can normally translate ( at least two ) sequential AAA codons in vivo without sliding . When there are more than two AAA codons in a row , each lysine after the first is added slowly ( Figure 2B ) , raising the possibility that sliding may become relevant on such messages . 10 . 7554/eLife . 05534 . 011Table 1 . Bioinformatic analyses of poly ( lysine ) sequencesDOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 011OrganismSequenceOccurancesFraction observedFraction expectedEnrichmentE . coliAAG-AAG2440 . 080 . 081 . 01AAG-AAA9020 . 290 . 201 . 45AAA-AAG5440 . 180 . 200 . 87AAA-AAA14160 . 460 . 520 . 88AAG-AAG-AAG90 . 070 . 023 . 37AAG-AAG-AAA200 . 160 . 062 . 89AAG-AAA-AAG210 . 170 . 063 . 03AAA-AAG-AAG40 . 030 . 060 . 58AAG-AAA-AAA360 . 290 . 142 . 00AAA-AAG-AAA290 . 230 . 141 . 61AAA-AAA-AAG40 . 030 . 140 . 22AAA-AAA-AAA10 . 010 . 380 . 02AAG-AAG-AAA-AAG10 . 250 . 0216 . 07AAG-AAG-AAA-AAA10 . 250 . 046 . 20AAA-AAG-AAA-AAA20 . 500 . 104 . 78S . cerevisiaeAAG-AAG38450 . 210 . 141 . 45AAG-AAA51830 . 280 . 241 . 20AAA-AAG45050 . 240 . 241 . 04AAA-AAA48580 . 260 . 390 . 69AAG-AAG-AAG2610 . 160 . 052 . 87AAG-AAG-AAA2340 . 140 . 091 . 57AAG-AAA-AAG2240 . 130 . 091 . 51AAA-AAG-AAG1890 . 110 . 091 . 27AAG-AAA-AAA2110 . 130 . 150 . 87AAA-AAG-AAA2610 . 160 . 151 . 07AAA-AAA-AAG1170 . 070 . 150 . 48AAA-AAA-AAA1710 . 100 . 240 . 43AAG-AAG-AAG-AAG240 . 100 . 024 . 88AAA-AAG-AAG-AAG280 . 120 . 033 . 48AAG-AAA-AAG-AAG230 . 100 . 032 . 86AAG-AAG-AAA-AAG190 . 080 . 032 . 36AAG-AAG-AAG-AAA270 . 110 . 033 . 35AAG-AAG-AAA-AAA130 . 050 . 060 . 99AAG-AAA-AAG-AAA190 . 080 . 061 . 44AAA-AAG-AAG-AAA110 . 050 . 060 . 83AAA-AAG-AAA-AAG170 . 070 . 061 . 29AAG-AAA-AAA-AAG50 . 020 . 060 . 38AAA-AAA-AAG-AAG90 . 040 . 060 . 68AAG-AAA-AAA-AAA90 . 040 . 090 . 42AAA-AAG-AAA-AAA140 . 060 . 090 . 65AAA-AAA-AAG-AAA60 . 030 . 090 . 28AAA-AAA-AAA-AAG50 . 020 . 090 . 23AAA-AAA-AAA-AAA90 . 040 . 150 . 25The prevalence precise sequences encoding 2–3 consecutive lysine residues in E . coli and S . cerevisiae are displayed . The raw number of ‘occurrences’ are listed for each sequence . The enrichment values listed reflect the fraction observed/fraction expected . The initial in vivo observation that protein production is more severely impacted by iterated AAA than AAG codons ( Figure 1 ) was recapitulated using the PURExpress E . coli cell-free translation system ( NEB ) ( Figure 4A ) . This system contains all factors required for normal translation , but lacks cellular factors involved in the degradation of RNA or proteins that might obscure interesting effects on translation . When the mCherry reporters ( described in Figure 1A ) were expressed in this system , we find that iterated AAA-containing reporters produce less protein than their iterated AAG-containing counterparts ( Figure 4A , lanes 3 vs 4 ) . Additionally , we note the appearance of a truncated protein product generated from the iterated AAA-containing reporter ( Figure 4A , lane 3 ) . This band is slightly larger than the size of protein produced when a stop-codon is positioned at the insertion site ( Figure 4A , lanes 2–3 ) . 10 . 7554/eLife . 05534 . 012Figure 4 . Ribosomes ‘slide’ into new frame on poly ( A ) -containing messages in the PURE in vitro translation system . ( A ) Expression of mCherry reporters ( Figure 1A ) in the E . coli PURE cell-free translation system ( NEB ) . The truncated band generated from the ( AAA ) 12 reporter is boxed in red . The expected sizes of the full-length , STOP protein and truncated reporter are 42 kDa , 15 kDa , and 17 kDa , respectively . ( B ) Expression of mCherry reporters in the PURE system and subsequent treatment of peptide products with RNase A . Only the positive control ( with a truncated mRNA species ) yielded a peptidyl-tRNA product that shifted in mobility upon RNase A treatment . ( C ) Expression of mCherry reporters ( Figure 1A ) in the PURE in vitro translation system in the presence and absence of RFs ( RFs = RF1 , RF2 , and RF3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 01210 . 7554/eLife . 05534 . 013Figure 4—figure supplement 1 . Truncated product release is independent of RF3 in the PURExpress cell-free translation system . mCherry reporters ( Figure 1A: no insert , AAG12 , AAA12 ) were expressed in the PURExpress cell-free translation system lacking release factors ( RFs ) ( light gray ) . RFs were added back to the reactions individually ( RF1 in green , RF2 in purple ) , and in combination ( RF1/3 in red and Rf2/3 dark gray ) . The plot displays the fraction of protein in the truncated band ( 100% × ( radioactivity in truncated band ) / ( radioactivity in truncated + full-length bands ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 01310 . 7554/eLife . 05534 . 014Figure 4—figure supplement 2 . Western blot ( α-HA ) of mCherry reporters ( Figure 1A ) expressed in E . coli . The full-length peptide product is noted with the solid arrow , and the truncated band is highlighted with the dotted arrow . WT = no insert . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 014 To ask whether the truncated band is the typical product of a stalled ribosome , a peptidyl-tRNA , we subjected the products of our PURE reactions to RNase A treatment ( Figure 4B ) . As a positive control , we observed that peptidyl-tRNA product generated from a non-stop mRNA ( Figure 4B , lanes 9–10 ) does indeed change in mobility when treated with RNase A ( see uppermost band resolve into smaller peptide products from this inefficiently translated mRNA ) . By contrast , the truncated band generated from the ( AAA ) 12-containing reporter does not shift in mobility on a gel following RNase A treatment ( Figure 4A , lanes 5–6 ) . We closely examined our reporter sequence and found that there are several out of frame stop-codons following the ( AAA ) 12 insert ( Supplementary file 1 ) . We next showed that the truncated band is generated by RF-mediated peptide release , likely on a canonical stop codon reached following ribosome sliding on poly ( A ) sequence ( Figure 4A , lanes 7–8 ) . Further experiments indicate that both RF1 and RF2 can promote release of this product and that the release reaction is independent of RF3 ( Figure 4—figure supplement 1 ) . The formation of truncated product from our ( AAA ) 12 reporters is a signature that reports on ribosome sliding on iterated AAA sequences . We note that the truncated band is also observed when the mCherry reporter is expressed in E . coli ( and a western is performed with an α-HA antibody ) ( Figure 4—figure supplement 2 ) . Together , these data provide evidence that ribosome slipping on iterated AAA sequences occurs both in a fully reconstituted translation system and in E . coli . To determine the minimum number of consecutive lysine or adenosine residues necessary for ribosomes to robustly slide on the iterated AAA-containing reporters , we expressed reporter constructs containing 3 , 6 , 9 or 12 lysines ( encoded by AAA ) in the PURExpress E . coli cell-free translation system ( Figure 5 ) . Truncated product ( which we have determined to be a signature of ribosome sliding ) was generated with as few as three consecutive lysines . We next asked whether the number of lysines residues or the number of consecutive adenosine nucleotides determines the extent of ribosome sliding . In this case , reporters were created containing a three lysine ( K3 ) insert encoded by 9 , 10 , 11 , or 13 As in a row ( Figure 5 ) . We find that an A11 repeat results in the robust formation of truncated product ( Figure 5 , Figure 5—figure supplement 1 ) while little product is seen with A9 or A10 sequences , though each sequence encodes the same number of consecutive lysines . 10 . 7554/eLife . 05534 . 015Figure 5 . Position and length of poly ( A ) stretch contributes to ribosome ‘sliding’ in the PURE in vitro translation system . Expression of mCherry reporters containing poly ( A ) inserts of various lengths in the presence ( + ) and absence ( − ) of RFs . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 01510 . 7554/eLife . 05534 . 016Figure 5—figure supplement 1 . Quantification of the efficiency of ribosome sliding on mCherry reporters expressed in the PURExpress system . mCherry reporters ( Figure 1A: no insert , and various A stretches ) were expressed in the PURExpress cell-free translation system ( Figure 5 ) . The plot reports the percent of truncated peptide product expressed relative to total peptide product for each reporter ( 100% × ( radioactivity in truncated band ) / ( radioactivity in truncated + full-length bands ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 016 In eukaryotic systems , NMD is a quality control system that recognizes mRNAs containing premature termination codons ( PTC ) and targets them for degradation . Upf1 is a key protein in NMD and upf1Δ cells stabilize PTC-containing transcripts . Previous studies established that when ribosomes frameshift during translation , these mRNAs are typically targeted for decay by NMD because the ribosomes generally encounter an out of frame premature termination codon ( Belew et al . , 2011 , 2014 ) . We proposed that if the ribosome slides on iterated AAA-containing mRNAs in yeast , as it does in the bacterial system , then iterated AAA-containing mRNAs should be targeted by NMD . We addressed this possibility by measuring the levels of ( AAA ) 12 , ( AAG ) 12 , and ( AAGAAGAAA ) 4-containing mRNAs in two different yeast-expressed reporter systems ( Figure 1A ) in wild-type and upf1Δ cells . First , as a control , we measured the mRNA levels of luciferase reporters containing no insert , an engineered premature stop codon ( positive control ) , and a stem-loop known to trigger an alternative mRNA quality control pathway , no-go decay ( negative control ) ( Doma and Parker , 2006 ) . We find that the levels of mRNA for PTC and stem-loop containing reporters are lowered ( PTC = 2 fold , stem-loop = 21 fold ) relative to reporters with no insert in wild-type yeast cells . Moreover , as expected , the level of PTC , but not stem-loop-containing , mRNA is recovered when the reporters are expressed in upf1Δ cells ( Figure 6A ) . When this same experiment was performed with a luciferase reporter containing an ( AAA ) 12 sequence , we find that reporter mRNA levels are substantially reduced in wild-type cells ( >50-fold down ) , and that these levels are partially recovered in a upf1Δ strain ( Figure 6A ) . These results suggest that the ( AAA ) 12-containing reporter is indeed a target of NMD in vivo . 10 . 7554/eLife . 05534 . 017Figure 6 . Deletion of Upf1p results in recovery of mRNA levels for poly ( A ) reporters in yeast . Luciferase ( A ) and mCherry ( B ) reporters ( Figure 1A ) were expressed in wild-type and upf1Δ S . cerevisiae , and the levels of reporter RNA were quantified by qRT-PCR . Various insertions including 12 lysines ( ( AAA ) 12 , ( AAG ) 12 , ( AAG2AAA ) 4 ) , stem-loop , or premature termination codon ( PTC ) in the coding sequence are specified on the x-axes . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 01710 . 7554/eLife . 05534 . 018Figure 6—figure supplement 1 . mRNA half-life of reporter containing iterated AAA codons is Upf1 dependent . Representative experiments measuring the amount of mCherry reporter mRNA in wild-type BY4741 ( black ) and upf1Δ ( blue ) cells as a function of time following transcriptional shut-off for reporters containing ( A ) ( AAG ) 12 , ( B ) ( AAA ) 12 , and ( C ) ( AAG2AAA ) 4 inserts . ( D ) The measured half-lives for decay of mCherry reporter mRNA in wild-type ( BY4741 ) and upf1Δ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 01810 . 7554/eLife . 05534 . 019Figure 6—figure supplement 2 . eRF1:3 does not prematurely terminate translation on coding sequences in poly ( lysine ) messages . MKA5-STOP message was translated with Lys-tRNALys present by S . cerevisiae ribosomes in a previously described yeast in vitro reconstituted system ( Shoemaker , et al . , 2010 ) . The reaction was allowed to proceed for 10 min; aliquots of the reaction were quenched at various time points with KOH to hydrolyze the peptidyl-tRNA bond and allow for the visualization of discrete peptide products ( right panel ) . After 10 min , eukaryotic release factors eRf1:eRF3 were added; time points quenched with formic acid; these lanes allow for visualization of peptides released from peptidyl-tRNA ( shown in the left panel ) . Normally , eRF1:eRF3 should only catalyze the release of peptide products from ribosomes on stop codons . The release reaction was allowed to proceed for 5 min ( left panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 019 To more directly compare our S . cerevisiae and E . coli results , we performed experiments instead using the related mCherry reporters ( Figure 1A ) with no insert , or a variety of lysine inserts ( ( AAG ) 12 , ( AAA ) 12 , and ( AAGAAGAAA ) 4 ) . In addition to measuring the absolute levels of reporter mRNAs in wild-type and upf1Δ cells ( Figure 6 ) , we asked whether the rates of mRNA decay for these reporters are impacted in the upf1Δ knock-out background ( Figure 6B and Figure 6—figure supplement 1 ) . We chose to include a mixed AAA/AAG reporter in addition to the simpler AAA and AAG repeat reporters because this sequence is commonly used to report on the NSD phenomenon ( Dimitrova et al . , 2009; Chiabudini et al . , 2012 , 2014 ) . Indeed , a recent study with an ( AAGAAGAAA ) 4-containing reporter argued that a truncated product generated by such a construct resulted from an unusual release factor-dependent termination event on a sense ( lysine ) codon ( Chiabudini et al . , 2014 ) . In an attempt to recapitulate these results , we directly looked for evidence of eRF1:eRF3-mediated termination activity on iterated lysine mRNAs in vitro using a yeast reconstituted translation system ( Shoemaker et al . , 2010 ) ; we see no evidence that such an event can occur ( Figure 6—figure supplement 2 ) . We propose that an alternative explanation for the published data could be that the ribosome slides out of frame on the ( AAGAAGAAA ) 4 sequence , resulting in premature termination on a previously out-of-frame stop codon , akin to what we observe in the PURE E . coli cell-free translation system ( Figure 4C ) . This possibility seemed particularly likely given that we observed sliding activity on a AUG-AAA-AAG-UUC-STOP sequence in our in vitro reconstituted E . coli system ( Figure 3—figure supplement 1 ) . In wild type and upf1Δ cells , we find that the level of the ( AAG ) 12 containing reporter mRNA is unchanged relative to the mCherry reporter with no insert ( Figure 6B ) . In contrast , the levels of ( AAGAAGAAA ) 4 and ( AAA ) 12 reporter mRNAs are significantly reduced compared to the control ( no insert ) reporter ( 15-fold and 30-fold , respectively ) . These observations are consistent with the low levels of protein expressed in vivo from these reporters relative to sequences containing no insert or ( AAG ) 12 ( Figure 1B ) . As with the luciferase reporters , the level of mCherry mRNA containing an ( AAA ) 12 insert is partially recovered by the deletion of UPF1 ( Figure 6B ) . Strikingly , when the ( AAGAAGAAA ) 4-containing reporters are expressed in upf1Δ cells , the mRNA levels are nearly fully recovered . The mRNA half-lives for these reporters are similarly recovered in the upf1Δ cells ( Figure 6—figure supplement 1 ) . Thus both the ( AAGAAGAAA ) 4 and ( AAA ) 12 reporter mRNAs are targeted by NMD in yeast cells ( Figure 6B ) . These results are consistent with a model invoking ribosome sliding followed by recognition of out-of-frame premature termination codons . We performed bioinformatic analyses of fully annotated ORFs to evaluate the codon usage in sequences of consecutive lysines found in the E . coli and S . cerevisiae transcriptomes . In both organisms , AAA codons are found more commonly than AAG codons ( 62% AAA vs 38% AAG in yeast , and 72% AAA vs 28% AAG in bacteria ) ; however , consecutive AAA codons are under-represented relative to their overall codon usage ( Table 1 ) . This is highlighted by the observation that the longer the stretch of lysines , the lower the likelihood of the motif being comprised solely of AAA codons ( Table 1 ) . Such an underrepresentation of AAA codons becomes pronounced in runs of 3 or 4 lysine codons in both organisms . In E . coli , only a single AAA-AAA-AAA sequence is present , which is 50-fold less common than expected based on the frequency of AAA codons; in contrast , ( AAG ) 3 sequences are found 3 . 3-fold more often than expected . In S . cerevisiae , the trends are similar; there are 2 . 3 and 4-fold fewer ( AAA ) 3 and ( AAA ) 4 sequences , respectively , than expected . Conversely , ( AAG ) 3 and ( AAG ) 4 sequences are threefold to fivefold more abundant than expected . These data together argue that evolution has selected against the use of long runs of A to encode sequential lysines within ORFs . Although many of the major players in NSD have been identified , a high-resolution mechanistic understanding of how translation of poly ( A ) sequences triggers NSD has been missing . Here , we provide mechanistic insight into what initially happens when the ribosome encounters poly ( A ) sequence . First , we find that the expression of proteins containing poly ( lysine ) stretches is codon-dependent in both bacteria and eukaryotes , with reporters containing iterated AAA codons consistently producing less protein than those with equivalent AAG codons ( Figures 1 , 4 ) . This differential protein output is not the result of imprecise RNA polymerase action ( Figure 3—figure supplement 3 ) nor likely of disparities in the rate of adding lysine codons ( Figure 2 ) ; lysines are slowly incorporated on iterated AAA and AAG codons . Instead , the codon-dependent disparity primarily stems from an unusual sliding event that occurs when ribosomes encounter consecutive AAA codons ( Figures 3 , 4 ) . Our observation that ribosomes can slide in multiple frames on iterated AAA sequences provides a rationale for consecutive AAA codons being substantially under-represented in open reading frames in most genomes ( see Bioinformatic discussion below , Table 1 and ( unpublished data ) . Our biochemical data in E . coli lead us to propose a model ( Figure 7 ) for what happens to the ribosome during the translation of homopolymeric A sequences . On these messages , the first lysine is added quickly ( k1 , obs ) while subsequent lysines are added more slowly , causing the ribosome to pause . We note that the rate constants measured in the in vitro assay reflect all of the processes that can occur each time a new lysine moiety is added to the growing polypeptide chain ( Lys-tRNALys binding , peptidyl-transfer , translocation , peptidyl-tRNA drop-off , 70S complex instability , etc ) . We suspect it to be unlikely that ribosome pausing is caused solely by dramatically large defects in peptidyl-transfer , but instead may result from ribosomes that become effectively inactivated ( e . g . as a result of complex instability on homopolymeric A messages , etc ) . Whatever the cause for an initial ribosome pausing event on iterated AAA sequences , the ribosome can either slide or perform another round of peptide bond formation . If the ribosome slides such that another AAA codon is positioned in the A site , the next step will also be slow , while if sliding somehow positions a non-lysine codon in the A site , recovery from slow elongation may occur . In our in vitro system translating di-lysine messages , we are able to observe sliding when consecutive AAA-codons are present because we force a strong pause after MKK formation by leaving out downstream factors required for translation to proceed ( Figure 3 ) . Our data suggest that ribosome sliding on iterated AAA sequences is the major difference between the translation of poly ( AAA ) - and poly ( AAG ) -containing messages that results in substantially different protein outputs . While each sequential addition of lysine in an iterated AAG sequence may be slow , the ribosome maintains frame and ultimately is able to produce full-length protein . By contrast , with repeated AAA sequences , the ribosome can eventually escape the homopolymeric A sequence through repeated sliding events , often emerging out-of-frame from the A stretch , and thus unable to produce full-length protein . 10 . 7554/eLife . 05534 . 020Figure 7 . Model for events during ribosome sliding . In this model translation is paused following the addition of the first lysine . The ribosome can than either slide or perform another round of peptide bond formation . If an AAA codon is positioned in the A site after sliding , the next step will also be slow , while if sliding results in a non-lysine codon in the A site , recovery from slow elongation may occur . DOI: http://dx . doi . org/10 . 7554/eLife . 05534 . 020 Ribosome sliding on poly ( A ) is distinct from traditional programmed ribosomal movements such as +1 ( Farabaugh and Björk , 1999; Taliaferro and Farabaugh , 2007 ) and −1 frame-shifts ( Dinman et al . , 1991; Plant et al . , 2003; Caliskan et al . , 2014; Chen et al . , 2014; Kim et al . , 2014 ) . During a programmed frame-shifting ( PRF ) event , specific signals direct elongating ribosomes to shift reading frame by one base in the 5′ ( −1 ) or 3′ ( +1 ) direction ( Dinman , 2012 ) . −1 PRFs signals are typically characterized by a ‘slippery’ sequence ( X XXY YYZ ) that is modulated by the presence of a downstream secondary structure , most commonly a pseudoknot ( Plant et al . , 2003; Jacobs et al . , 2007; Caliskan et al . , 2014; Chen et al . , 2014; Kim et al . , 2014 ) . The secondary structure impairs the normal movement of the ribosome during translocation , and promotes the frame-shift event in an EF-G dependent manner ( Caliskan et al . , 2014; Chen et al . , 2014 ) . +1 PRFs signals are more diverse than −1 PRFs , but still generally depend on a slippery sequence and a downstream element ( e . g . , secondary structure or rare codon ) that causes the ribosome to pause ( Dinman , 2012 ) . Iterated A stretches are inherently slippery and contain a built-in translation pause ( adding consecutive lysines is slow—Figure 2 ) , however the poly ( A ) sequences that we have studied lack significant secondary structure downstream that might contribute to limiting unregulated ribosome sliding . As such , when ribosomes slide on iterated AAA codons , forward and backward movements may be permitted . The scale of the movements undergone during a ribosome sliding event may be more similar to those documented in translational bypassing on the gene product 60 of bacteriophage T4 which is synthesized from a discontinuous reading frame ( Samatova et al . , 2014 ) . Importantly , however , in contrast to this specific concerted large-scale movement ( 50 nucleotides ) which results in the production of a single peptide product , ribosome sliding is different in that no single outcome appears to be encoded by the event . The inability of the ribosome to translate a discrete product on homopolymeric A sequences likely explains the bioinformatic analyses demonstrating that poly ( A ) sequences are strongly selected against in coding sequences containing iterated lysines ( Table 1 ) . Consistent with this idea , in E . coli we find that the minimum length ( 11 ) of a homopolymeric A sequence needed to trigger ribosome sliding in the PURE cell-free translation system ( Figure 5 , and Figure 5—figure supplement 1 ) correlates with the length of lysine stretch at which homopolymeric sequences are selected against in mRNA coding regions ( Table 1 ) . There are multiple reports in the literature indicating that frame-shifted ribosomes can trigger NMD ( Belew et al . , 2011 , 2014 ) . We find that mRNA levels for reporters containing ( AAA ) 12 or ( AAGAAGAAA ) 4 , but not ( AAG ) 12 sequences , are reduced in a Upf1-dependent manner . These data are consistent with the idea that sliding on homopolymeric A stretches can eventually lead to ribosomes reaching out-of-frame premature termination codons ( Figure 6 ) . A recent report in the literature argued that translation of poly ( lysine ) stretches led to an unusual termination event on a sense codon ( AAA or AAG ) mediated by eRF3 ( presumably in concert with its binding partner eRF1 ) ( Chiabudini et al . , 2014 ) . These observations bring to mind premature termination events on sense codons documented in E . coli ( Zaher and Green , 2009 ) ; this quality control system was proposed to increase the fidelity of translation by minimizing frame-shifting and eliminating errors made during tRNA selection . We note that the premature termination event that we previously documented in E . coli was highly dependent on RF3 , while the termination event documented in E . coli in this manuscript at homopolymeric A sequences is RF3-independent ( Figure 4—figure supplement 1 ) . Given the clear evidence that we provide for ribosome sliding in the E . coli system and the inability to observe eRF1:eRF3-mediated peptide release on homopolymeric A programmed yeast ribosome complexes in vitro ( Figure 6—figure supplement 2 ) , we suggest that the most likely explanation for the eRF3-dependent truncated product generated in yeast cells on ( AAGAAGAAA ) 4–encoding reporters in Chiabudini et al . is the result of ribosome sliding and canonical recognition of downstream premature stop codons . We note that there are multiple out-of-frame stop codons following the ( AAGAAGAAA ) 4-repeat that could account for the observed products in Chiabudini et al . ( Chiabudini et al . , 2014 ) . We were intrigued by the observation that the ( AAGAAGAAA ) 4 reporter mRNA levels are more efficiently recovered than those of the ( AAA ) 12 reporter mRNA in a UPF1-deletion strain . We speculate that the more modest sliding within the ( AAGAAGAAA ) -repeats might be distinguished from the sliding on ( AAA ) -repeats in an important way . Sliding within homopolymeric AAA sequence most typically results in another nearby AAA codon being poised in the A site , and another inefficient elongation event with Lys-tRNALys . Ribosomes that eventually exit the poly ( A ) sequence to reach heteropolymeric sequence and an out-of-frame downstream premature stop codon will trigger NMD; ribosomes that struggle to get past the very long stretch of iterated lysine codons will instead trigger NSD . As such , the mRNA levels for the ( AAA ) 12 reporter are partially recovered by a UPF1 deletion and partially recovered by a DOM34 deletion ( data not shown ) . By contrast , on the ( AAGAAGAAA ) -repeat reporters , sliding has the potential to quickly place the ribosomes in a more productive frame for efficient elongation ( one frame will result in Arg-Arg-Lys ( RRK ) repeats while the other frame will result in Glu-Glu-Lys ( EEK ) repeats ) . While we might predict that the poly ( basic ) RRKRRKRRKRR peptide will also be slowly translated , a ribosome that slides into the frame encoding the EEKEEKEEKEE peptide should be able to resume efficient elongation . As such , fewer ribosomes may trigger NSD and , instead , a majority of ribosomes will reach downstream premature stop codons that trigger NMD . These ideas can easily be understood in the context of the model in Figure 7 where differences in the elongation rates ( e . g . slow for iterated lysine residues but fast for incorporation of other amino acids ) will impact the relative contribution of ribosome sliding to overall outcome . NSD was originally identified by following the degradation of transcripts lacking termination codons ( Frischmeyer et al . , 2002; van Hoof et al . , 2002 ) . These studies led to the idea that NSD is triggered when the ribosome stalls while translating a poly ( basic ) lysine sequence . NSD is commonly studied using reporters in yeast that contain poly ( basic ) inserts; common lysine and arginine inserts that have been investigated include ( AAA ) 12 , ( AAG ) 12 , ( AAG-AAG-AAA ) 4 , and ( CGG- ( CGA ) 2-CGG- ( CGC ) 2 ) 2 ( Ito-Harashima et al . , 2007; Dimitrova et al . , 2009; Bengtson and Joazeiro , 2010; Brandman et al . , 2012; Chiabudini et al . , 2012 , 2014 ) . Consistent with our findings , previous studies reported differences in protein output in yeast when these different sequences are translated ( Ito-Harashima et al . , 2007; Dimitrova et al . , 2009 ) ; iterated AAA codons are more detrimental to overall expression than iterated AAG codons . Despite these differences , because the mRNA and protein levels for all of these are broadly sensitive to known NSD factors ( Ltn1 , Dom34 , Ski7 ) , poly ( basic ) sequences have been treated equally . Our results demonstrating that ribosomes can slide on consecutive AAA codons suggest that there may be important distinctions to be made in considering these reporters and that there may be substantial mechanistic overlaps in these systems . Even though cells rarely maintain homopolymeric A sequences in ORFs , there are some situations where the ribosome likely must deal with homopolymeric A stretches in both bacteria and eukaryotes . In bacteria , mRNAs are typically polyadenylated as part of the normal decay process ( Dreyfus and Régnier , 2002 ) . For example , ribosome sliding might provide an escape for ribosomes already engaged on these mRNAs ( a form of ribosome rescue ) . In eukaryotes , virtually all mRNAs in the cell are polyadenylated , but usually a stop codon is found at the end of the encoded ORF . However , there is abundant recent evidence indicating that a significant portion of yeast ( 14% ) and human ( 9% ) genes contain at least one alternative polyadenylation site within their coding sequence ( Ozsolak et al . , 2010 ) . It has even been suggested that premature polyadenylation may become up-regulated in cancerous cells ( Berg et al . , 2012 ) . In cases where premature polyadenylation takes place within the ORF , the ribosome will surely encounter a homopolymeric A sequence , likely triggering so called Non-Stop-Decay ( NSD ) . In light of the results presented here , we would suggest that the triggering of NSD ( and associated mRNA decay , proteolysis and ribosome recycling ) occurs following the slow translation of iterated lysines and ribosome sliding events . The ubiquity of premature polyadenylation suggests that NSD broadly serves as an important pathway for regulating gene expression . The observation of synonymous AAG to AAA changes in iterated lysine stretches in genes upregulated in cancer provides support for the significance of this mechanism of gene regulation ( unpublished data ) . The widespread use of polyadenylation for non-coding purposes in mRNA transcripts may find its origins in the inability of the decoding machine , the ribosome , to carefully control the behavior of these sequences . The Thrdx-HA-mCherry ( Figure 1A , Supplementary file 1 ) no insert reporter expressed in E . coli and the PURExpress cell-free translation system was created using Gateway cloning to include the 2HA-mCherry sequence in the pBAD-DEST49 vector . The vectors containing inserts ( Thrdx-HA-insert-mCherry: ( AAA ) 12 , ( AAA ) 6 , ( AAG ) 12 , ( AAGAAGAAA ) 4 , ( GAA ) 12 , TAA ( STOP ) , ( A ) 9-13 , etc ) were subsequently derived from this clone . To create the mCherry reporter expressed in yeast ( Figure 1A ) , the Thrdx-HA-mCherry and Thrdx-HA-insert-mCherry sequences were amplified out of the pBAD-DEST49 vectors and cloned into the p-ENTR/D-TOPO vector . The vector was then reacted with lr-clonease II to move the sequences into the pYES-DEST52 plasmid . The dual luciferase reporter described in Figure 1A was based on the dual luciferase plasmid from Takacs et al . ( 2011 ) . In this reporter , Renilla and Firefly luciferase are under the control of ADH and GPD promoters , respectively . We inserted sequences of interest into the N-terminus of Renilla luciferase . The single Renilla luciferase reporter described in Figure 1A was cloned into pYES2 ( with a Gal promoter ) using the Gateway cloning system . Thrdx-HA-mCherry and Thrdx-HA-insert-mCherry constructs were expressed in 6 ml E . coli grown in LB-Ampicillin . The cells were grown to an OD of 0 . 4–0 . 6 , induced with 25 μl of 5 g/10 ml arabinose , then harvested 2 hr post-induction . In yeast , the Thrdx-HA-mCherry constructs were expressed in wild-type and upf1Δ S . cerevisiae ( BY4741 ) grown in 5 ml of–URA/+galactose media to an OD of 0 . 6 . The single luciferase reporters were transformed into yeast and grown in–URA/+galactose media , and harvested at an OD of 0 . 6 . Proteins production was analyzed via fluorescence , luminescence ( Figure 1 ) or western blot analysis ( Figure 4—figure supplement 2 ) . 70S initiation complexes ( ICs ) were prepared using E . coli ribosomes programmed with various mRNAs and f-[35S]-Met-tRNAMet in the P site . mRNAs were generated by transcription with T7 polymerase and ICs were formed , pelleted , and resuspended as previously described ( Youngman et al . , 2004 ) on our messages of interest . Translation assays were initiated when equal volumes of ternary complex ( 10–20 μM charged tRNA , 12 μM EFG , 60 μM EfTu ) were added to 0 . 2 nM 70S initiation complexes . Assays were performed in 219-Tris buffer ( 50 mM Tris pH 7 . 5 , 70 mM NH4Cl , 30 mM KCl , 7 mM MgCl2 , 5 mM βME ) . The limited addition of iterated lysines on a MKA5-STOP message was also observed in polymix buffer ( 50 mM K2HPO4 pH 7 . 5 , 95 mM KCl , 5 mM NH4Cl , 5 mM Mg ( OAc ) 2 , 0 . 5 mM CaCl2 , 8 mM putrescine , 1 mM spermidine , 1 mM DTT ) . To measure the rates of amino acid incorporation , the reactions are quenched with 500 mM KOH ( final concentration ) at discrete time points ( 0 s–30 min ) either by hand or on a quench-flow apparatus . For assays including release factors for the duration of the reaction ( Figure 3C ) , RF1 and additional GTP were added prior to the initiation of translation ( final concentrations 1 μM and 200 μM , respectively ) . The time-points were diluted 1:10 in nuclease free water and the reactants , intermediates and products visualized by electrophoretic TLC , as previously described ( Zaher and Green , 2009 ) . The reactants , products and intermediates were visualized by phosophorimaging and quantified with ImageQuant . The kinetic fits were modeled using Mathematica ( details in Figure 2—figure supplement 2 ) . The Thrdx-HA-mCherry and Thrdx-HA-insert-mCherry reporters were expressed in the PURExpress in vitro translation system ( NEB , Ipswitch , MA ) from PCR products . The peptidyl-tRNA construct was generated by creating a truncated mRNA lacking a stop codon directly after the Thrdx-HA sequence . The PURExpress reactions were initiated by mixing 1 μl of PCR product ( 29–22 ng/μl ) , 2 μl of solution A , 1 . 5 μl of solution B , and 0 . 6 μl of 35S-methionine . The reactions were run for 45–60 min at 37°C . Following translation , the products were immediately heat-denatured and loaded on a 4–12% Bis-Tris gel at 4°C in XT-MES buffer . For the experiments in which the PURExpress reaction products were treated with RNase A ( Figures 4B ) , 0 . 5–1 μg of RNase A ( Ambion , Grand Island , NY ) was added to each reaction and solutions were incubated on ice for an additional 30 min before being denatured and loaded on a gel . The peptide products of the PURExpress reactions were visualized by Phosphoimager and quantified with ImageQuant ( Figure 3—figure supplement 2 , and Figure 5—figure supplement 1 ) . DNA templates were PCR amplified from plasmids ( PCR-Blunt II-TOPO vector ) encoding MEA ( insert ) EAEDYKDD sequences . The PURExpress cell-free transcription-translation system ( NEB , Ipswich , MA ) was used for in vitro protein synthesis . Reactions were run for 30 min at 37°C by mixing 0 . 2-pmol of DNA template , 2 . 5 μl of Solution A and 1 μl of Solution B along with either 0 . 5 μl of DMSO ( 5% ) or thiostrepton ( 0 . 5 mm in 5% DMSO ) . 1 pmol of 32PATP-labeled NV1 primer was added , and reverse transcription was performed with AMV as previously described ( Vazquez-Laslop et al . , 2008; Tanner et al . , 2009 ) . Reactions were phenol and chloroform extracted , ethanol precipitated and visualized on a 6% denaturing PAGE gel . Sequencing lanes were generated from plasmids using the Sequenase 2 . 0 DNA sequencing kit ( Affymetrix , Santa Clara , CA ) . All bands were visualized by PhosphorImager . Reporter mRNA levels were quantified by qRT-PCR using the iQ5 iCycler system ( Bio-Rad , Hercules , CA ) and iQ SYBR Green Supermix ( Bio-Rad , Hercules , CA ) . To measure the rate of mRNA decay in yeast for our mCherry reporters , we grew wild-type and upf1Δ cells expressing reporters in–ura/galactose media at 30°C to an OD600 of 0 . 4 . Cells were washed three times with–ura media lacking sugar , then re-suspended in -ura/glucose media; the transcription of the reporter is shut-off by glucose . Samples were collected at discrete time points ( 0–90 min ) , and mRNA levels were analyzed by qRT PCR . E . coli K-12 substrain MG1655 complete genome , 4140 ORFs ( data source: GenBank:U00096 . 3; http://www . ncbi . nlm . nih . gov/nuccore/U00096 . 3 ) and S . cerevisiae 5887 verified ORFs ( data source: http://downloads . yeastgenome . org/sequence/S288C_reference/orf_protein/ ) have been used for extraction of lysine codon numbers and analyses of consecutive codons shown in Table 1 . Expected values for consecutive variants of lysine AAA and AAG codons were calculated based on observed values for a single AAA and AAG codons and their probabilities to be found in such arrangments . Observed values were calculated based on data from genomic distribution and total numbers of variants for two , three or four consecutive lys codons , respectively .	Genes provide the instructions to assemble proteins from smaller molecules called amino acids . When a gene is ‘switched on’ , the DNA that makes up the gene is copied into messenger ribonucleic acid ( or mRNA ) molecules , composed of building blocks called nucleotides . There are four types of nucleotides in mRNA molecules—commonly referred to as A , C , G , and U—and a set of three nucleotides is called a codon . A molecular machine called a ribosome moves along an mRNA molecule translating the codons into protein . Each codon instructs the ribosome to add a particular amino acid to the chain of amino acids that will make up the protein . Some codons do not specify an amino acid but instead mark the point on the mRNA that the ribosome should stop and release the new protein . Most mRNAs have nucleotides beyond the ‘stop’ codon and these often contain a long stretch of A nucleotides , one after the other , which is known as the poly ( A ) tail . Some mRNA copies may contain poly ( A ) tails before a stop codon , which can lead to the production of alternate and potentially harmful proteins . Cells have developed ways to identify and dispose of these mRNAs and their protein products . For example , in yeast and other eukaryotes , if an mRNA is missing a stop codon , the ribosomes will continue to translate along the mRNA into the poly ( A ) tail where they stall and are eventually removed . When this happens , the mRNA and protein are rapidly destroyed . However , it is not clear how this works . Koutmou et al . studied the translation of a series of artificial mRNAs that contained different numbers of A nucleotides in codons of either AAA or AAG . Both of these codons specify the same amino acid , and should therefore be translated equivalently . The experiments show that the ribosomes read the AAA and AAG codons differently . When consecutive AAA codons are found in the mRNA , the level of protein production is significantly lower than when the mRNA contains iterated AAGs instead . Koutmou et al . found that when ribosomes encounter consecutive AAA codons they undergo an unusual ‘sliding’ movement and are unable to accurately produce proteins . When a cell detects this abnormal sliding behavior , it rapidly triggers the destruction of the mRNA molecule . In contrast , when ribosomes encounter consecutive AAG codons , they slow down but do not slide , and therefore produce a correct protein . Koutmou et al . 's findings also provide an explanation for why there are relatively few AAA codons within the regions of genes that encode proteins . The prevalence of alternative forms of mRNAs with poly ( A ) sequences before their stop codons suggests that ribosome sliding may contribute to an important pathway to control the activity of genes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	Ribosomes slide on lysine-encoding homopolymeric A stretches
Linear Ubiquitin chain Assembly Complex ( LUBAC ) is an E3 ligase complex that generates linear ubiquitin chains and is important for tumour necrosis factor ( TNF ) signaling activation . Mice lacking Sharpin , a critical subunit of LUBAC , spontaneously develop inflammatory lesions in the skin and other organs . Here we show that TNF receptor 1 ( TNFR1 ) -associated death domain ( TRADD ) -dependent TNFR1 signaling in epidermal keratinocytes drives skin inflammation in Sharpin-deficient mice . Epidermis-restricted ablation of Fas-associated protein with death domain ( FADD ) combined with receptor-interacting protein kinase 3 ( RIPK3 ) deficiency fully prevented skin inflammation , while single RIPK3 deficiency only delayed and partly ameliorated lesion development in Sharpin-deficient mice , showing that inflammation is primarily driven by TRADD- and FADD-dependent keratinocyte apoptosis while necroptosis plays a minor role . At the cellular level , Sharpin deficiency sensitized primary murine keratinocytes , human keratinocytes , and mouse embryonic fibroblasts to TNF-induced apoptosis . Depletion of FADD or TRADD in Sharpin-deficient HaCaT cells suppressed TNF-induced apoptosis , indicating the importance of FADD and TRADD in Sharpin-dependent anti-apoptosis signaling in keratinocytes . Tumor necrosis factor receptor 1 ( TNFR1 ) -mediated signaling is regulated by multiple ubiquitination events involving linear ( Met1 ) , Lys63 and Lys48 ubiquitination of several target proteins ( Wertz and Dixit , 2008; Verhelst et al . , 2011; Schmukle and Walczak , 2012 ) . Upon TNF stimulation , a receptor proximal signaling complex ( termed complex I ) consisting of the TNFR1-associated death domain ( TRADD ) , receptor-interacting protein kinase 1 ( RIPK1 ) , TNF receptor-associated factor 2 ( TRAF2 ) , and cellular inhibitor of apoptosis proteins 1 and 2 ( cIAP1/2 ) is recruited to the intracellular domain of TNFR1 ( Haas et al . , 2009 ) . Ubiquitination of proteins in complex I , including RIPK1 , cIAP1/2 , and TRAF2 , leads to the recruitment of additional signaling components that facilitate activation of nuclear factor-κB ( NF-κB ) and mitogen-activated protein ( MAP ) kinase signaling cascades ( Wu et al . , 2005; Varfolomeev et al . , 2008; Haas et al . , 2009 ) . It has been shown that the transforming growth factor beta-activated kinase 1 ( TAK1 ) /TAK1-binding protein 2 ( TAB2 ) complex is recruited into complex I through the interaction between the TAB2-Npl4 zinc finger ( NZF ) and Lys63-linked ubiquitin chains ( Kanayama et al . , 2004; Kulathu et al . , 2009; Sato et al . , 2009 ) . cIAP-mediated ubiquitination of RIPK1 and cIAPs themselves was shown to result in the recruitment of the E3 ligase complex Linear Ubiquitin chain Assembly Complex ( LUBAC ) ( Haas et al . , 2009 ) into complex I . LUBAC consists of a catalytic protein , HOIL-1L interacting protein ( HOIP ) /ring finger protein 31 ( Rnf31 ) , and two other critical subunits , Sharpin/Shank-interacting protein-like 1 ( SIPL1 ) and HOIL-1L/RanBP-type and C3HC4-type zinc finger containing 1 ( Rbck1 ) ( Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) . Using biochemical and cell biological approaches , LUBAC has been shown to specifically generate linear ubiquitin chains , linked via Met1 , and these chain types are important for pathway activation ( Kirisako et al . , 2006 ) . To date , LUBAC is the only E3 ligase complex identified that catalyzes linear ubiquitin chain generation . HOIP belongs to a RING-in-between-RING ( RBR ) -type of E3 ligase family and constitutes the catalytic center in LUBAC ( Kirisako et al . , 2006; Stieglitz et al . , 2013 ) . Interestingly , HOIP requires binding to either Sharpin or HOIL-1L for its catalytic action ( Kirisako et al . , 2006; Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011; Stieglitz et al . , 2012 ) . It was shown that the LUBAC-mediated ubiquitination of NEMO in the IκB kinase ( IKK ) complex is critical for the NF-κB signaling pathway . Sharpin or HOIL-1L deficiency partially suppress TNFR1-induced NF-κB activation , suggesting that these components show some degree of functional redundancy in regulating NF-κB signaling ( Haas et al . , 2009; Tokunaga et al . , 2009; Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) . Earlier studies identified Sharpin as the gene mutated in the chronic proliferative dermatitis mice ( Sharpincpdm/cpdm ) , which spontaneously develop severe chronic inflammation primarily in the skin but also in other tissues such as the gut , lung , liver , and esophagus ( Gijbels et al . , 1996; Seymour et al . , 2007 ) . The pathogenesis of multi-organ chronic inflammation in Sharpincpdm/cpdm mice depends on TNF , as double Sharpincpdm/cpdm;Tnf−/− mice did not develop signs of inflammatory skin and liver disease ( Gerlach et al . , 2011 ) . These results showed that Sharpin has an essential function in preventing TNF-induced chronic inflammation . However , the molecular mechanisms that are controlled by Sharpin to prevent TNF-induced inflammatory disease remain poorly understood . Here we show that the skin inflammation in Sharpincpdm/cpdm mice is triggered by TNFR1-mediated TRADD- and FADD-dependent apoptosis of keratinocytes . Previous studies showed that TNF is required for the development of multi-organ inflammation in Sharpincpdm/cpdm mice ( Gerlach et al . , 2011 ) . To address whether this function of TNF is mediated by TNFR1 , we crossed Sharpincpdm/cpdm mice with Tnfrsf1a−/− animals . Double deficient Sharpincpdm/cpdm;Tnfrsf1a−/− mice did not develop skin inflammation , demonstrating that TNF-induced TNFR1 signaling is essential for the pathogenesis of inflammatory skin lesions in Sharpincpdm/cpdm mice ( Figure 1 ) . We then tried to identify the cellular target of the pathogenic TNFR1 signaling in Sharpincpdm/cpdm mice . We have recently shown that TNFR1 signaling in NF-κB-deficient epidermal keratinocytes drives psoriasis-like skin inflammation in mice ( Kumari et al . , 2013 ) , identifying keratinocytes as an important cellular target of pathogenic TNF signaling in skin inflammation . To address whether TNFR1 signaling in epidermal keratinocytes drives the skin inflammation in Sharpincpdm/cpdm mice , we crossed Sharpincpdm/cpdm mice with K14Cre-Tnfrsf1afl/fl ( TNFR1E-KO ) mice that lack TNFR1 specifically in keratinocytes ( Figure 1A ) . These Sharpincpdm/cpdm;TNFR1E-KO mice did not develop any macroscopic signs of skin inflammation ( Figure 1B ) . In addition , histological analysis of Sharpincpdm/cpdm;TNFR1E-KO mice skin revealed a normal epidermis without keratinocyte death ( cleaved caspase-3 staining in Figure 1C ) , skin inflammation ( F4/80 staining in Figure 1C ) , or epidermal hyperplasia ( H&E , Keratin 6 , Keratin 10 , and Loricrin staining in Figure 1C and quantification in Figure 1D ) , similar to Sharpincpdm/cpdm;Tnfrsf1a −/− animals ( Figure 1C , D ) . These results demonstrate that TNFR1 signaling in epidermal keratinocytes is essential for the pathogenesis of skin inflammation in Sharpincpdm/cpdm mice . 10 . 7554/eLife . 03422 . 003Figure 1 . Tumor necrosis factor receptor 1 ( TNFR1 ) signaling in keratinocytes triggers chronic proliferative dermatitis phenotype in Sharpincpdm/cpdm mice . ( A ) Flow cytometric analysis of TNFR1 expression on the isolated keratinocytes from mice with the indicated genotypes . ( B and C ) Macroscopic pictures , Hematoxylin and Eosin staining ( H&E ) , Keratin 6 , 14 , 10 and Loricrin as well as cleaved caspase-3 and F4/80 staining of the skin sections from 14-week-old littermate mice of the indicated genotypes . The scale bars are 100 μm . ( D ) Microscopic quantification of the epidermal thickness from 12–18-week-old mice of the indicated genotypes and their littermate controls ( Ctr ) , which consisted of the following genotypes: Sharpincpdm/wt;Tnfrsf1a−/− , Sharpinwt/wt;Tnfrsf1a−/− , Sharpincpdm/wt;Tnfrsf1afl/fl , Sharpincpdm/wt;TNFR1E-KO , and Sharpinwt/wt;TNFR1E-KO . The Sharpincpdm/cpdm group consisted of Sharpincpdm/cpdm;Tnfrsf1afl/fl and Sharpincpdm/cpdm;Tnfrsf1afl/wt mice that were littermates of the Sharpincpdm/cpdm;TNFR1E-KO mice . The Sharpincpdm/cpdm;Tnfrsf1a−/− mice were derived from a different line and shown here is the picture and the staining from the age-matched mice . Bars represent mean values ± SEM . Statistical significance was determined using the Student's t test ( *p ≤ 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 003 Having established keratinocyte-intrinsic TNFR1 signaling as a key spatial event triggering skin inflammation in Sharpincpdm/cpdm mice , we sought to investigate the cell death mechanisms by which epithelial TNFR1 induces the inflammatory response . We and others have shown that increased numbers of keratinocytes in the epidermis of Sharpincpdm/cpdm mice undergo apoptosis , as indicated by the presence of cleaved caspase-3 ( Ikeda et al . , 2011; Liang and Sundberg , 2011 ) ( also see Figure 1C ) . In addition , it was suggested that Sharpin deficiency sensitizes primary keratinocytes to both TNF-induced caspase-dependent apoptosis and RIP-kinase-dependent necroptosis ( Gerlach et al . , 2011 ) . We therefore used genetic mouse models to address the role of FADD/caspase-8-dependent apoptosis and RIPK3-dependent necroptosis in Sharpincpdm/cpdm mice . To address the role of RIPK3-dependent necroptosis , we generated mice lacking both Sharpin and RIPK3 by crossing Sharpincpdm/cpdm with Ripk3−/− animals ( Figure 2 ) . Double deficient Sharpincpdm/cpdm;Ripk3−/− mice developed skin lesions similar to those of Sharpincpdm/cpdm mice , demonstrating that RIPK3 deficiency did not prevent the development of skin inflammation ( Figure 2A , B ) . However , the initiation of the skin phenotype was delayed in Sharpincpdm/cpdm;Ripk3−/− animals , which started to show lesions after the age of 10 weeks but showed a large variability in onset and severity with some mice showing only mild lesions even at the age of 19 weeks ( Figure 2—figure supplement 1 ) . Sharpincpdm/cpdm mice also showed variability with lesion onset between 8 and 11 weeks , but all mice showed severe lesions by the age of 12–14 weeks . Quantification of epidermal thickness revealed that RIPK3 deficiency mildly ameliorated the severity of skin lesions ( Figure 2C ) . These results showed that , although RIPK3-dependent necroptosis contributes to accelerating the onset and exacerbating the severity of the phenotype , it is not essential for the pathogenesis of the inflammatory skin lesions in Sharpincpdm/cpdm mice . 10 . 7554/eLife . 03422 . 004Figure 2 . Fas-associated protein with death domain ( FADD ) deficiency in keratinocytes prevents skin inflammation in Sharpincpdm/cpdm mice . ( A and B ) Macroscopic pictures , Hematoxylin and Eosin staining ( H&E ) , Keratin 6 , 14 , 10 and Loricrin as well as cleaved caspase-3 and F4/80 staining of skin sections from 14-week-old mice of the indicated genotypes . The scale bars are 100 μm . ( C ) Microscopic quantification of the epidermal thickness from 12–18-week-old mice of the indicated genotypes and their littermate controls ( Ctr ) , which consisted of the following genotypes: Sharpinwt/wt;Tnfrsf1afl/fl , Sharpincpdm/wt;Faddfl/fl;Ripk3−/− , Sharpinwt/wt;Faddfl/fl;Ripk3−/− , Sharpincpdm/wt;FADDE-KO;Ripk3−/− and Sharpinwt/wt;FADDE-KO;Ripk3−/− . The Sharpincpdm/cpdm group consisted of Sharpincpdm/cpdm;Tnfrsf1afl/fl and Sharpincpdm/cpdm;Tnfrsf1afl/wt mice that were littermates of the Sharpincpdm/cpdm;TNFR1E-KO mice . Sharpincpdm/cpdm;Ripk3−/− mice were derived from the same breeding line as Sharpincpdmcpdm;FADDE-KO;Ripk3−/− mice and consisted of the genotype Sharpincpdm/cpdm;Faddfl/fl;Ripk3−/− . Bars represent mean values ± SEM . Statistical significance was determined using the Student's t test ( p ≤ 0 . 001 , p ≤ 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 00410 . 7554/eLife . 03422 . 005Figure 2—figure supplement 1 . Variability among Sharpincpdm/cpdm;Ripk3−/− mice at a similar age . The gross appearance of severe skin lesions in two Sharpincpdm/cpdm;Ripk3−/− mice at the age of 19 weeks ( left ) and a Sharpincpdm/cpdm;Ripk3−/− mouse at the same age with only a very mild phenotype ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 005 Next , we sought to address whether FADD/caspase-8-dependent apoptosis of Sharpin-deficient keratinocytes induces TNF-dependent skin inflammation in Sharpincpdm/cpdm mice . Since deficiency of caspase-8 or FADD alone in epidermal keratinocytes triggers a RIPK3-dependent skin inflammation ( Bonnet et al . , 2011; Weinlich et al . , 2013 ) , we could not directly investigate the role of FADD or caspase-8 in the TNF-induced death of Sharpin-deficient keratinocytes in vivo . However , taking advantage of the fact that concomitant deletion of RIPK3 fully prevents skin lesion formation in mice with keratinocyte-restricted FADD knockout ( FADDE-KO ) ( Bonnet et al . , 2011 ) , we generated Sharpincpdm/cpdm;Ripk3−/− mice that also lacked FADD specifically in keratinocytes ( Figure 2 ) . These Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice did not develop any macroscopic ( Figure 2A ) or histological ( Figure 2B ) skin lesions up to the age of 4 months , as they showed normal keratinocyte proliferation and differentiation without any signs of inflammation or epidermal hyperplasia ( Figure 2B , C ) . Moreover , apoptosis of keratinocytes observed in Sharpincpdm/cpdm mice was completely prevented in Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice , as shown by the absence of cleaved caspase-3 positive cells ( compare Figure 2B with Figure 1C ) . Taken together , these results showed that combined inhibition of FADD/caspase-8-dependent apoptosis and RIPK3-dependent necroptosis prevented keratinocyte death and the development of skin lesions in Sharpincpdm/cpdm mice , providing in vivo genetic evidence that skin inflammation is triggered by TNFR1-induced death of Sharpin-deficient keratinocytes . Our findings strongly suggest that FADD-dependent apoptosis of Sharpin-deficient keratinocytes triggers skin inflammation . However , since the role of FADD can only be addressed in a RIPK3-deficient background , it remains possible that FADD-dependent apoptosis and RIPK3-dependent necroptosis might share a redundant function in inducing the cell death of Sharpin-deficient keratinocytes and triggering skin inflammation . To directly address the role of TNFR1-induced apoptosis in Sharpincpdm/cpdm mice , we employed mice carrying conditional alleles for TRADD , an adapter molecule that is important for the induction of inflammatory and apoptotic signaling downstream of TNFR1 ( Chen et al . , 2008; Michallet et al . , 2008 ) . It has been shown that TRADD deficiency partially inhibits TNFR1-induced activation of NF-κB and MAP kinase pathways and fully prevents TNFR1-induced apoptosis in mouse embryonic fibroblasts ( MEFs ) in vitro and in hepatocytes in vivo ( Ermolaeva et al . , 2008 ) . To examine the role of TRADD in TNFR1-induced apoptosis and necroptosis , we analyzed the response of wild type and TRADD-deficient primary MEFs to TNF stimulation in the presence of cycloheximide ( CHX ) , caspase inhibitor ( Z-VAD-FMK ) , and RIPK1 inhibitor ( Necrostatin-1 ) ( Figure 3A ) . As expected , TRADD-deficient MEFs were resistant to apoptosis induced by TNF and CHX . However , in contrast to earlier studies ( Pobezinskaya et al . , 2008 ) , we found that TRADD-deficient MEFs were sensitive to necroptosis induced by TNF , CHX , and Z-VAD-FMK . These results demonstrated that TRADD deficiency specifically blocks TNFR1-induced apoptosis ( Figure 3A ) . We therefore generated Sharpincpdm/cpdm mice lacking TRADD specifically in keratinocytes by crossing Sharpincpdm/cpdm with K14Cre-Traddfl/fl mice ( Sharpincpdm/cpdm;TRADDE-KO ) ( Figure 3B–D ) . Indeed , keratinocyte-restricted TRADD deficiency prevented skin lesion development in Sharpincpdm/cpdm mice , as shown by macroscopic and histological analysis ( Figure 3B–D ) . Collectively , our results show that TNFR1-induced TRADD- and FADD-dependent apoptosis of Sharpin-deficient keratinocytes triggers the chronic proliferative dermatitis phenotype in Sharpincpdm/cpdm mice . 10 . 7554/eLife . 03422 . 006Figure 3 . Tumor necrosis factor receptor 1-associated death domain ( TRADD ) deficiency in keratinocytes prevents skin inflammation in Sharpincpdm/cpdm mice . ( A ) The percentage viability of wild type ( WT ) mouse embryonic fibroblasts ( MEFs ) ( n = 3 ) and TRADD-deficient MEFs ( n = 3 ) upon tumor necrosis factor ( TNF ) , cycloheximide ( CHX ) , caspase inhibitor ( zVAD ) , and Necrostatin-1 ( Nec ) treatment alone or in combination for 20 hr and measurement by WST-1 assay . Bars represent average cell viability ( ± SD ) of three independent experiments . ( B and C ) Macroscopic gross appearance of the WT ( +/+ ) and littermate mice of the indicated genotype at the age of 12 weeks ( B ) and ( H&E ) , Keratin 6 , 14 , 10 and Loricrin as well as cleaved caspase-3 and F4/80 staining of the skin sections from 12-week-old mice of the indicated genotypes ( C ) . Scale bars in ( C ) are 100 μm . ( D ) Microscopic quantification of the epidermal thickness from 12-week-old littermate mice of Sharpincpdm/cpdm , Sharpinwt/wt;TRADDE-KO , Sharpincpdm/cpdm;TRADDE-KO and age-matched WT ( +/+ ) is shown . Bars represent mean values ± SD . Statistical significance was determined using the one-way ANOVA test ( *p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 006 In addition to dermatitis , Sharpincpdm/cpdm mice develop splenomegaly and inflammation in other organs such as liver and lung ( Figure 4A , B ) . We found that systemic deficiency of TNFR1 prevented the development of liver inflammation in Sharpincpdm/cpdm mice ( Figure 4A ) , consistent with earlier results showing that TNF deficiency also inhibited liver inflammation in these animals ( Gerlach et al . , 2011 ) . We also found that TNFR1 deficiency prevented lung inflammation ( Figure 4A ) and the development of splenomegaly ( Figure 4B ) in Sharpincpdm/cpdm mice . TNFR1 deficiency also corrected the splenic structure defects in Sharpincpdm/cpdm mice ( Figure 4A ) , in contrast to the report by Gerlach et al . ( 2011 ) that TNF deficiency could not rescue the splenic structural abnormalities of Sharpincpdm/cpdm mice . This contradiction likely stems from the fact that Gerlach et al . compared the spleens of Sharpincpdm/cpdm mice with wild type mice . However , considering that TNFR1- and TNF-deficient animals have altered splenic structures characterized by lack of B cell lymphoid follicles and marginal zone abnormalities ( Pasparakis et al . , 1996a , 1996b , 2000 ) , TNFR1 or TNF deficiency cannot restore the spleen structure of Sharpincpdm/cpdm mice to that of a wild type mouse . Keratinocyte-restricted TNFR1 deficiency could not rescue the extracutaneous pathologies in Sharpincpdm/cpdm mice , suggesting that these develop independently from the skin lesions ( Figure 4A , B ) . These results demonstrate that TNFR1 signaling in non-epidermal cells triggers splenomegaly and extracutaneous inflammation in Sharpincpdm/cpdm mice . We also analyzed the liver , lung , and spleen of Sharpincpdm/cpdm;Ripk3−/− and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice . Interestingly , we observed partial normalization of liver and lung inflammation ( Figure 5A ) , splenomegaly and the splenic structure in Sharpincpdm/cpdm;Ripk3−/− mice ( Figure 5A , B ) , suggesting that RIPK3-mediated necroptosis contributes to the extracutaneous inflammatory pathologies . Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice showed very similar histology of the liver , lung , and spleen to Sharpincpdm/cpdm;Ripk3−/− mice , indicating that the extracutaneous phenotype in Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice is mainly attributed to RIPK3 signaling and not to epidermal FADD signaling ( Figure 5A ) . In addition , we performed cleaved caspase-3 staining and TUNEL staining on liver , lung , and spleen tissue sections obtained from Sharpincpdm/cpdm , Sharpincpdm/cpdm;Ripk3−/− , Sharpincpdm/cpdm;Tnfrsf1a−/− , Sharpincpdm/cpdm;TNFR1E-KO , and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice and from their respective controls Sharpincpdm/wt , Sharpincpdm/wt;Ripk3−/− , Sharpincpdm/wt;Tnfrsf1a−/− , Sharpincpdm/cpdm;TNFR1E-KO , and Sharpincpdm/wt;FADDE-KO;Ripk3−/− mice ( n = 2–4 mice from each genotype ) ( Figure 5—figure supplements 1–3 ) . We observed increased active caspase-3 and TUNEL positive cells in the lung and liver of Sharpincpdm/cpdm mice compared with Sharpincpdm/wt mice , whereas the spleen did not seem to have an increased number of cleaved caspase-3 positive cells compared with the control mice . The numbers of cleaved caspase-3 positive and TUNEL positive cells were reduced in Sharpincpdm/cpdm;Ripk3−/− , Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− and Sharpincpdm/cpdm;Tnfrsf1a−/− mice , whereas Sharpincpdm/cpdm;TNFR1E-KO mice showed similar cleaved caspase-3 and TUNEL positive cells to Sharpincpdm/cpdm mice . These findings essentially showed that , in the mice showing no or less inflammation , the numbers of both cleaved caspase-3 and TUNEL positive cells were reduced . Furthermore , we observed that liver and lung inflammation as well as splenomegaly and the defect of splenic structure were not altered by keratinocyte-specific depletion of TRADD in Sharpincpdm/cpdm mice ( Figure 6A , B ) , providing further support for the notion that the development of extracutaneous organ inflammation in Sharpincpdm/cpdm mice occurs independently from the skin lesions . 10 . 7554/eLife . 03422 . 007Figure 4 . Tumor necrosis factor receptor 1 ( TNFR1 ) deficiency in Sharpincpdm/cpdm mice rescues the inflammation of lung , liver and splenomegaly but not epidermal keratinocyte-restricted knockout of TNFR1 . ( A and B ) H&E staining of liver , lung and spleen , and macroscopic pictures of spleen from mice with the indicated genotypes as well as measurement of spleen weight from 12–18-week-old Sharpincpdm/cpdm , Sharpincpdm/cpdm;Tnfrsf1a−/− , and Sharpincpdm/cpdm;TNFR1E-KO mice and their littermate controls ( Ctr ) , which consisted of the following genotypes: Sharpincpdm/wt;Tnfrsf1a−/− , Sharpinwt/wt;Tnfrsf1a−/− , Sharpincpdm/wt;Tnfrsf1afl/fl , Sharpincpdm/wt;TNFR1E-KO , and Sharpinwt/wt;TNFR1E-KO . The Sharpincpdm/cpdm group consisted of Sharpincpdm/cpdm;Tnfrsf1afl/fl and Sharpincpdm/cpdm;Tnfrsf1afl/wt mice that were littermates of the Sharpincpdm/cpdm;TNFR1E-KO mice . Scale bars in ( A ) are 100 μm . Results are expressed as mean values ± SEM . Statistical significance was determined using unpaired Student's t test ( two-tailed ) ( p ≤ 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 00710 . 7554/eLife . 03422 . 008Figure 5 . Receptor-interacting protein kinase 3 ( Ripk3−/− ) and epidermis-specific Fas-associated protein with death domain ( FADD ) together with Ripk3−/− ( FADDE-KO;Ripk3−/− ) in Sharpincpdm/cpdm mice partially rescues the inflammation of the lung , liver , and splenomegaly . ( A and B ) H&E staining of liver , lung and spleen , and macroscopic pictures of spleen from mice with the indicated genotypes as well as measurement of spleen weight from 12–18-week-old Sharpincpdm/cpdm , Sharpincpdm/cpdm;Ripk3−/− and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice and their littermate controls ( Ctr ) , which consisted of the following genotypes: Sharpinwt/wt;Tnfrsf1afl/fl , Sharpincpdm/wt;Faddfl/fl;Ripk3−/− , Sharpinwt/wt;Faddfl/fl;Ripk3−/− , Sharpincpdm/wt;FADDE-KO;Ripk3−/− , and Sharpinwt/wt;FADDE-KO;Ripk3−/− . The Sharpincpdm/cpdm group consisted of Sharpincpdm/cpdm;Tnfrsf1afl/fl and Sharpincpdm/cpdm;Tnfrsf1afl/wt mice that were littermates of the Sharpincpdm/cpdm;TNFR1E-KO mice . The Sharpincpdm/cpdm;Ripk3−/− mice were derived from the same breeding line as Sharpincpdmcpdm;FADDE-KO;Ripk3−/− mice and consisted of the genotype Sharpincpdm/cpdm;Faddfl/fl;Ripk3−/− . Scale bars in ( A ) are 100 μm . Results are expressed as mean values ± SEM . Statistical significance was determined using unpaired Student's t test ( two-tailed ) ( p ≤ 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 00810 . 7554/eLife . 03422 . 009Figure 5—figure supplement 1 . Cell death in the spleen of Sharpincpdm/cpdm , Sharpincpdm/cpdm; Tnfrsf1a−/− , Sharpincpdm/cpdm;TNFR1E-KO , Sharpincpdm/cpdm;Ripk3−/− and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice . ( A and B ) Cleaved caspase-3 ( A ) and TUNEL ( B ) staining on the spleen tissue sections from mice with the indicated genotypes between 12–18 weeks . The control mice compared were from the same breeding but not always littermates . Representative images are shown . The scale bars are 100 μm . Nuclei in ( B ) were visualized using DAPI . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 00910 . 7554/eLife . 03422 . 010Figure 5—figure supplement 2 . Cell death in the liver of Sharpincpdm/cpdm , Sharpincpdm/cpdm; Tnfrsf1a−/− , Sharpincpdm/cpdm;TNFR1E-KO , Sharpincpdm/cpdm;Ripk3−/− and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice . ( A and B ) Cleaved caspase-3 ( A ) and TUNEL ( B ) staining of liver tissue sections from mice with the indicated genotypes between 12–18 weeks . The control mice compared were from the same breeding but not always littermates . Representative images are shown . The scale bars are 100 μm . Nuclei in ( B ) were visualized using DAPI . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 01010 . 7554/eLife . 03422 . 011Figure 5—figure supplement 3 . Cell death in the lung of Sharpincpdm/cpdm , Sharpincpdm/cpdm; Tnfrsf1a−/− , Sharpincpdm/cpdm;TNFR1E-KO , Sharpincpdm/cpdm;Ripk3−/− and Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− mice . ( A and B ) Cleaved caspase-3 ( A ) and TUNEL ( B ) staining of lung tissue sections from mice with the indicated genotypes between 12–18 weeks . The control mice compared were from the same breeding but not always littermates . Representative images are shown . The scale bars are 100 μm . Nuclei in ( B ) were visualized using DAPI . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 01110 . 7554/eLife . 03422 . 012Figure 6 . Epidermal keratinocyte-restricted knockout of tumor necrosis factor receptor 1-associated death domain ( TRADD ) ( TRADDE-KO ) in Sharpincpdm/cpdm mice has a minor effect on the inflammation of the lung , liver , and splenomegaly . ( A and B ) H&E staining of liver , lung , and spleen and macroscopic pictures of spleen as well as measurement of spleen weight from mice with the indicated genotypes derived from wild type ( +/+ ) , Sharpincpdm/cpdm , TRADDE-KO or Sharpincpdm/cpdm;TRADDE-KO mice at the age of 12 weeks . The mice used here are the littermates . Scale bars in ( A ) are 100 μm . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( p ≤ 0 . 001 and **p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 012 Our in vivo genetic experiments showed that TNFR1-induced apoptosis of Sharpincpdm/cpdm keratinocytes causes skin inflammation in Sharpincpdm/cpdm mice . We therefore investigated the molecular mechanisms by which Sharpin prevents TNFR1-induced apoptosis . We first examined the induction of apoptosis in MEFs derived from Sharpincpdm/cpdm mice stimulated with TNF alone or TNF + CHX using various assays . We observed increased sensitivity to TNF + CHX-induced apoptosis in Sharpincpdm/cpdm MEFs compared with wild type MEFs , as determined by immunocytochemical detection of cleaved caspase-3 ( Figure 7A ) , detection of annexin V positive cells by fluorescence-activated cell sorting ( FACS ) analysis ( Figure 7B ) , analysis of the cleavage of caspase-3 and Poly ( ADP-ribose ) polymerase ( PARP ) by immunoblotting ( Figure 7C ) ( Ikeda et al . , 2011 ) , and luminescent-based caspase-8 activity assay ( Figure 7D ) . Sharpincpdm/cpdm MEFs also showed increased cleavage of caspase-3 and PARP ( Figure 7E ) and caspase-8 activity ( Figure 7F ) in response to stimulation with TNF alone although , in the absence of CHX , the response was considerably weaker . To further analyze the cellular functions of Sharpin in anti-apoptotic signaling , we used a human keratinocyte cell line , HaCaT cells ( Boukamp et al . , 1988 ) . By using lentiviral-based shRNA knockdown , Sharpin was stably depleted in HaCaT cells ( Figure 8A ) . Sharpin-deficient HaCaT and control shRNA-introduced HaCaT cells were treated with TNF alone or TNF + CHX , and the induction of apoptosis was assessed by FACS analysis of annexin V positive cells ( Figure 8B ) and measurement of caspase-8 activity ( Figure 8C ) . Similar to Sharpincpdm/cpdm MEFs , Sharpin knockdown sensitized HaCaT cells to apoptosis induced by TNF or TNF + CHX . Interestingly , treatment with Necrostatin-1 , a RIPK1 inhibitor , suppressed caspase-8 activation in Sharpin-deficient HaCaT cells ( Figure 8D ) , suggesting an involvement of RIPK1 in the induction of TNF-induced death of Sharpin-deficient keratinocytes as suggested recently by Berger et al . ( Berger et al . , 2014 ) . Our results collectively suggest that Sharpin plays a critical role in protecting against TNF-induced apoptosis . To distinguish between the possibilities that the anti-apoptotic protection afforded by Sharpin is a LUBAC-independent function of Sharpin , we assessed the involvement of the catalytic LUBAC component HOIP in the regulation of apoptosis . HOIP expression was stably knocked down in HaCaT cells using shRNA ( Figure 8—figure supplement 1A ) , and the cells were examined for apoptosis induced by TNF alone or TNF + CHX using FACS analysis ( Figure 8—figure supplement 1B ) and a caspase-8 activity assay ( Figure 8—figure supplement 1C and D ) . As expected and consistent with a LUBAC-dependent Sharpin function , HOIP deficiency sensitized HaCaT cells to apoptosis induced by TNF alone or by TNF + CHX , akin to that observed in Sharpincpdm/cpdm MEFs or Sharpin-deficient HaCaT cells . There results provide strong support that a LUBAC-dependent Sharpin function plays a role in the regulation of apoptosis in keratinocytes . 10 . 7554/eLife . 03422 . 013Figure 7 . Sharpin regulates tumor necrosis factor ( TNF ) -induced apoptosis signaling cascade in mouse embryonic fibroblasts ( MEFs ) . ( A–D ) TNF- and cycloheximide ( CHX ) -induced apoptosis in wild type ( +/+ ) or Sharpincpdm/cpdm MEFs . Apoptosis in MEFs stimulated with TNF ( 10 ng/ml ) and CHX ( 1 μg/ml ) for 4 hr was examined by immunofluorescent staining using α-cleaved caspase-3 antibody with Alexa488 conjugated secondary antibody ( A ) , by fluorescence-activated cell sorting ( FACS ) analysis using annexin V staining ( B ) , by immunoblotting using α-cleaved caspase-3 and α-Poly ( ADP-ribose ) polymerase ( PARP ) antibodies ( C ) , or by caspase-8 activity assay measured using a luminol-based assay . Scale bars in ( A ) are 100 μm . ( E and F ) TNF-induced apoptosis in MEFs analyzed by immunoblotting ( E ) or by caspase-8 activity assay ( F ) as in ( C and D ) . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( p ≤ 0 . 05 , p ≤ 0 . 01 , p ≤ 0 . 001 , *p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 01310 . 7554/eLife . 03422 . 014Figure 8 . Sharpin regulates tumor necrosis factor ( TNF ) -induced apoptosis signaling cascade in HaCaT cells . ( A ) Immunoblotting of stable knockdown Sharpin in HaCaT cells using α-Sharpin antibody . Control shRNA ( Ctr ) was used for the control knockdown . α-Vinculin antibody was used for the loading control . ( B ) Fluorescence-activated cell sorting ( FACS ) analysis of annexin V staining in parental , Ctr , and Sharpin knockdown HaCaT cells , stimulated with TNF ( 100 ng/ml ) for 16 hr or TNF with cycloheximide ( CHX ) ( 1 μg/ml ) for 6 hr . ( C ) Caspase-8 activity measurement using a luminol-based assay upon stimulation with TNF alone , or TNF and CHX for the indicated time in Ctr and Sharpin knockdown HaCaT cells . ( D ) Caspase-8 activity measurement upon stimulation with TNF , TNF + CHX with or without Necrostatin-1 ( Nec ) ( 30 μmol ) for 6 hr in Ctr and Sharpin knockdown HaCaT cells . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 01410 . 7554/eLife . 03422 . 015Figure 8—figure supplement 1 . A LUBAC component , HOIL-1L interacting protein ( HOIP ) , plays a role in tumor necrosis factor ( TNF ) -induced apoptosis in HaCaT cells . ( A–D ) Immunoblot of HOIP ( A ) after stable knocked down of HOIP in HaCaT cells by shRNA . α-Tubulin antibody was used for the loading control . Fluorescence-activated cell sorting ( FACS ) analysis of annexin V positive cells after stimulation of control ( Ctr ) and HOIP knockdown HaCaT cells with TNF alone or TNF + cycloheximide ( CHX ) ( B ) and caspase-8 activity assay ( B and D ) after stimulation of Ctr and HOIP knockdown HaCaT cells with TNF alone ( C ) or TNF + CHX ( D ) . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( p ≤ 0 . 01 , **p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 015 As keratinocyte apoptosis and skin inflammation in Sharpincpdm/cpdm mice were suppressed by epidermal-specific deletion of FADD or TRADD , we sought to investigate how the lack of FADD , TRADD , and RIPK3 proteins impact on TNF-induced cell viability in Sharpin-deficient cells . To address this , primary keratinocytes were isolated from newborn pups , Sharpincpdm/cpdm , Sharpincpdm/cpdm;Ripk3−/− , Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− , and their control littermates , Sharpincpdm/wt or Sharpinwt/wt Sharpincpdm/wt;Ripk3−/− and Sharpincpdm/wt;FADDE-KO;Ripk3−/− , respectively . Cells were treated with increasing concentrations of TNF alone or TNF + CHX for 24 hr and their viability was analyzed using the WST-1 assay ( Figure 9A ) . Although TNF + CHX treatment strongly induced the death of Sharpin-deficient keratinocytes , TNF treatment alone had a very small effect in reducing the viability of Sharpin-deficient keratinocytes by about 10% compared with controls . Interestingly , the combined lack of FADD and RIPK3 in Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− keratinocytes fully rescued the increased sensitivity of Sharpin-deficient keratinocytes to TNF + CHX ( Figure 9A ) . However , keratinocytes obtained from Sharpincpdm/cpdm;Faddfl/fl;Ripk3−/− showed a similar response to TNF + CHX as Sharpincpdm/cpdm keratinocytes , demonstrating that RIPK3 deficiency does not prevent the death of Sharpin-deficient keratinocytes . Therefore , Sharpin deficiency primarily sensitizes keratinocytes to FADD-mediated apoptosis and not to RIPK3-mediated necroptosis . To further examine a direct role of FADD in Sharpin-deficient cells without involvement of RIPK3 , we generated HaCaT cells in which Sharpin and FADD were both stably knocked down by shRNA ( Figure 9B ) . Upon treatment with TNF alone or TNF + CHX , HaCaT cells lacking both Sharpin and FADD showed reduced caspase-8 activity compared with Sharpin-deficient HaCaT cells ( Figure 9C , D ) . Similar to the keratinocytes , we generated FADD-deficient Sharpincpdm/cpdm MEFs and analyzed the effect of FADD deficiency on apoptosis induced by TNF alone and by TNF + CHX in Sharpincpdm/cpdm MEFs ( Figure 9—figure supplement 1A ) and observed that FADD deficiency significantly suppressed the annexin V positive cells and caspase-8 activity in Sharpincpdm/cpdm MEFs ( Figure 9—figure supplement 1B and C ) . To address an involvement of TRADD in TNF-induced sensitivity of Sharpin-deficient HaCaT cells , we used HaCaT cells which were knockdown for Sharpin and TRADD expression . In comparison to caspase-8 activity induced by TNF alone or TNF + CHX in Sharpin-deficient HaCaT cells , TRADD deficiency significantly suppressed caspase-8 activation ( Figure 9E–G ) . These data collectively suggest that regulation of Sharpin-dependent anti-apoptosis signaling depends on FADD and TRADD in a cell-intrinsic manner . 10 . 7554/eLife . 03422 . 016Figure 9 . Fas-associated protein with death domain ( FADD ) - and tumor necrosis factor receptor 1-associated death domain ( TRADD ) -dependent enhanced sensitivity of Sharpin-deficient keratinocytes to tumor necrosis factor ( TNF ) -induced apoptosis . ( A ) Percentage viability of primary keratinocytes isolated from Sharpincpdm/cpdm , Sharpincpdm/cpdm;Ripk3−/− , Sharpincpdm/cpdm;FADDE-KO;Ripk3−/− and Sharpincpdm/cpdm;TNFR1E-KO , and their respective control pups ( n = 2 ) upon treatment with increasing TNF concentration ( 20 , 50 and 100 ng/ml ) in the presence or absence of cycloheximide ( CHX ) ( 1 μg/ml ) for 24 hr . Viability of TNF-treated cells was normalized over their untreated control cells , and viability of TNF + CHX-treated cells was normalized over their CHX-treated control cells . The result shown here is representative of two independent experiments . The percentage viability was assessed using the WST-1 assay . Bars represent mean values ± SEM . Statistical significance was determined using the Student's t test ( p ≤ 0 . 01 , p ≤ 0 . 05 ) . ( B–G ) Sharpin with FADD ( B ) or TRADD ( E ) was stably knocked down in HaCaT cells . Knockdown efficiency was analyzed by immunoblotting using α-Sharpin , α-FADD , and α-TRADD antibodies . Caspase-8 activity measurement in Sharpin knockdown , FADD knockdown , and double knockdown of Sharpin and FADD HaCaT cells ( C and D ) as well as Sharpin knockdown , TRADD knockdown , and double knockdown of Sharpin and TRADD HaCaT cells ( F and G ) upon treatment with TNF alone ( C and F ) or TNF + CHX ( D and G ) for the indicated time . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( *p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 01610 . 7554/eLife . 03422 . 017Figure 9—figure supplement 1 . Fas-associated protein with death domain ( FADD ) plays an important role in the Sharpin-dependent apoptosis signaling . ( A ) Immunoblot of FADD after stable knocked down of FADD in wild type ( +/+ ) mouse embryonic fibroblasts ( MEFs ) and Sharpincpdm/cpdm MEFs by shRNA . α-Vinculin antibody was used for the loading control . ( B ) Fluorescence-activated cell sorting ( FACS ) analysis of annexin V positive cells after stimulation with tumor necrosis factor ( TNF ) + cycloheximide ( CHX ) for 4 hr in +/+ shCtr MEFs , +/+ shFADD MEFs , Sharpincpdm/cpdm shCtr MEFs , and Sharpincpdm/cpdm shFADD MEFs . ( C ) Caspase-8 activity measurement upon stimulation with TNF with or without CHX for the indicated time in +/+ shCtr MEFs , +/+ shFADD MEFs , Sharpincpdm/cpdm shCtr MEFs and Sharpincpdm/cpdm shFADD MEFs . Results are expressed as mean values ± SD . Statistical significance was determined using ANOVA test ( *p ≤ 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03422 . 017 Ubiquitination regulates a wide variety of biological functions by generating ubiquitin chains with different linkages on substrates ( Ikeda and Dikic , 2008; Komander and Rape , 2012 ) . One of the atypical linkage types is the Met-1/linearly-linked ubiquitin chain which is specifically generated by the LUBAC E3 ligase complex ( Kirisako et al . , 2006 ) . Sharpin is a critical component of the LUBAC complex ( Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) and its deficiency in mice leads to severe TNF-dependent inflammation in multiple organs including the skin ( Seymour et al . , 2007 ) , suggesting that Sharpin has an important role in preventing inflammation . However , the mechanisms by which Sharpin prevents TNF-mediated inflammation in the skin and other organs have remained elusive . We show here that TNFR1 signaling in keratinocytes is essential for the pathogenesis of skin inflammation in Sharpincpdm/cpdm mice . In addition , we provide genetic evidence that TNFR1-mediated , TRADD- and FADD-dependent apoptosis of Sharpin-deficient keratinocytes induces skin inflammation in these mice . RIPK3 deficiency only mildly delayed and ameliorated the severity of skin lesions in Sharpincpdm/cpdm mice , showing that RIPK3-dependent necroptosis plays a minor role in driving skin inflammation in this model , which was also shown by Rickard et al . ( Rickard et al . , 2014 ) . These results demonstrate that Sharpin deficiency triggers skin inflammation by sensitizing keratinocytes to TNF-induced apoptosis . Furthermore , we found that inflammation in other organs including lung and liver as well as splenomegaly and altered splenic structure observed in Sharpincpdm/cpdm mice also depend on TNFR1 signaling and occur independently of the skin lesions . Together , these findings identified a cell-intrinsic function of Sharpin in inhibiting TNFR1-induced apoptosis that is essential for the maintenance of tissue homeostasis and the prevention of multi-organ inflammation . In this study we have shown that Sharpin and HOIP , components of LUBAC , negatively regulate apoptotic pathways in keratinocytes . Our data suggest that Sharpin and HOIP act in an active ligase complex and raise the intriguing possibility that linear ubiquitination of an unknown target ( s ) inhibits TNF-induced apoptosis in keratinocytes . Ubiquitin signals have been implicated in the regulation of the death-inducing signaling complex ( DISC ) . For example , cullin 3-based ubiquitination of caspase-8 by both Lys48- and Lys63-linked polyubiquitin chains brings caspase-8 into p62-containing aggregates leading to its activation and to commitment to apoptotic cell death ( Jin et al . , 2009 ) . Furthermore , it was shown that FADD ubiquitination induced by Makorin Ring Finger Protein 1 ( MKRN1 ) E3 ligase leads to the proteasome-dependent degradation of FADD , and MKRN1 depletion in breast cancer cells accelerates TNF-related apoptosis-inducing ligand ( TRAIL ) -induced DISC formation and apoptosis ( Lee et al . , 2012 ) . Since we did not detect significant changes in total FADD protein levels in Sharpincpdm/cpdm MEFs in comparison to the wild type cells ( Figure 9—figure supplement 1A ) , it suggests that protein stability is not controlled by Sharpin . It is also important to elucidate spatial and temporal regulation of DISC and TNFR complex II component ubiquitination and to identify the E3 ligases that mediate these ubiquitinations for the better understanding of biological functions of TNF signaling . Lastly , our study sheds new light on the specific functions of Sharpin as an integral component of the LUBAC E3 ligase complex . For example , both HOIL-1L and Sharpin are expressed in keratinocytes and have redundant roles in activating the NF-κB signaling pathway in MEFs ( Haas et al . , 2009; Gerlach et al . , 2011 ) . Yet , Sharpin-deficient mice develop inflammation in the skin whereas HOIL-1L knockout mice showed no obvious dermatitis phenotypes ( Tokunaga et al . , 2009 ) . Interestingly , previous studies have shown that the analysis of complex formation of Sharpin , HOIL-1L , and HOIP suggested that some populations of Sharpin and HOIL-1L may exist in different molecular complexes from HOIP ( Gerlach et al . , 2011; Tokunaga et al . , 2011 ) . This raises the question whether Sharpin and HOIL-1L could individually regulate biological functions depending on the cell types , tissues , and pathogenic conditions . Precise examination of the effects of tissue-specific depletion of the LUBAC components in mice is needed to better understand the functional roles of each of the LUBAC components in vivo . Here we report the specific function of Sharpin in the regulation of apoptosis and skin inflammation , which are mediated through FADD and TRADD . Together , further studies of the regulatory mechanisms controlling inflammation in Sharpincpdm/cpdm mice will be important for the better understanding of the unique functions of Sharpin in vivo . The following mouse lines were used: Sharpincpdm/cpdm C57BL/KaLawRij ( Seymour et al . , 2007; Ikeda et al . , 2011 ) , Traddfl/fl , Faddfl/fl ( Mc Guire et al . , 2010 ) , K14-Cre ( Pasparakis et al . , 2002 ) , Tnfrsf1a−/− ( Pfeffer et al . , 1993 ) , Tnfrsf1afl/fl ( Van Hauwermeiren et al . , 2013 ) , and Ripk3−/− ( Newton et al . , 2004 ) . All animal procedures were conducted in accordance with European , national , and institutional guidelines and protocols and were approved by local government authorities . pLKO . 1-shRNA-control ( AACAAGATGAAGAGCACCAACTCGAGTTGGTGCTCTTCATCTTGTT ) , pLKO . 1-shRNA-mouse FADD ( CCACACTTGGAGCCCAATAAACTCGAGTTTATTGGGCTCCAAGTGTGG ) , pLKO . 1-shRNA-human HOIP ( TGCTCCTTTGGCTTCATATATCTCGAGATATATGAAGCCAAAGGAGCA ) , pLKO . 1-shRNA-human Sharpin ( GTGTTCTCAGAGCTCGGTTTCCTCGAGGAAACCGAGCTCTGAGAACAC ) , pLKO . 1-shRNA-human FADD ( CATGGAACTCAGACGCATCTACTCGAGTAGATGCGTCTGAGTTCCATG ) , and pLKO . 1-shRNA-human TRADD ( CTGAAACTCCACTTGGCCTATCTCGAGATAGGCCAAGTGGAGTTTCAG ) were generated by a standard subcloning method . The following antibodies were purchased and used according to the manufacturers’ recommendations: anti-cleaved caspase-3 antibody ( Asp175 ) ( clone 5A1E; Cell Signaling Technology , Danvers , MA ) , anti-Vinculin antibody ( Sigma , St Louis , MO ) , anti-PARP antibody ( #9542; Cell Signaling Technology ) , anti-FADD antibody ( clone 1F7; ENZO Life Sciences , Farmingdale , NY ) , anti-HOIP antibody ( Aviva Systems Biology , San Diego , CA ) , anti-TRADD antibody ( Santa Cruz Biotechnology , Santa Cruz , CA ) . Anti-Sharpin antibody has been described previously ( Tokunaga et al . , 2009; Ikeda et al . , 2011 ) . Human Embryonic Kidney ( HEK ) 293T cells ( ATCC , Boulevard Manassas , VA ) , immortalized mouse embryonic fibroblasts ( MEFs ) , and HaCaT cells ( a kind gift from Máté Borsos ) were maintained at 37°C in 5% CO2 condition in DMEM ( Sigma ) supplemented with 10% fetal calf serum ( Life Technologies , Carlsbad , CA ) and 100 U/ml penicillin and streptomycin ( Invitrogen , Carlsbad , CA ) . Murine TNF was purchased from PeproTech ( #315-01A , Rocky Hill , NJ ) . Tissue samples from 12–18-week-old mice were fixed in 3 . 8–10% buffered formalin ( skin ) or 4% PFA ( liver , lung and spleen ) and subjected to histological analysis by H&E staining , TUNEL or immunohistochemical analysis . Slides were scanned using Mirax Slide Scanner ( Carl Zeiss , Germany ) . The following antibodies were used: K14 , K6 , K10 and Loricrin ( Covance , Prinston , NJ ) , F4/80 ( clone A3-1 , AbD Serotec , homemade ) , active caspase-3 ( Cell Signaling Technologies ) , and TRADD ( H-278 , Santa Cruz ) . Secondary antibodies were coupled to Biotin ( Dako , Germany ) ; signal was amplified by avidin-biotin-HRP detection system ( ABC VectorLab Elite Kit , Burlingame , CA ) and detected by peroxidase substrate ( VectorLab NovaRed ) . Sections were counterstained with hematoxylin for nuclei visualization . TUNEL staining was performed using TUNEL staining kit form Promega ( Madison , WI ) , according to the manufacturer’s instructions . Flow cytometric ( FACS ) analysis was performed on keratinocytes isolated from newborn pups as described previously ( Tscharntke et al . , 2007 ) and incubated with APC-conjugated anti-TNFR1 antibody ( Biolegend , San Diego , CA ) in PBS-BSA buffer , fixed in 4% PFA , followed by acquisition and analysis using FACS Calibur with accompanying software CellQuest ( BD Bioscience , San Jose , CA ) . For immunoblotting , the method is described elsewhere ( Ikeda et al . , 2011 ) . Briefly , MEFs ( 0 . 05 × 106 ) were plated on 24-well plates . After 24 hr of subculturing , cells were treated with cycloheximide ( CHX ) ( 1 μM ) ( #C7698; Sigma ) or TNF ( 10 ng/ml ) ( PeproTech ) for the indicated times . Depending on the experimental set-ups , retroviral infection was combined . After the treatment , cells were harvested for SDS-PAGE followed by Western blot analysis . For the FACS analysis ( Canto , BD Bioscience , San Jose , CA ) , the percentage of apoptotic cells was quantified by coupled annexin V antibody ( #556419; BD Bioscience ) staining and PI uptake . For WST-1 assay on MEFs , 9000 MEFs from wild type and TRADD-deficient mice were seeded in 96-well plates and treated with CHX ( 10 μg/ml ) , Necrostatin-1 ( 25 μM ) , and Z-VAD-FMK ( 20 μM ) alone or in combination for 20 hr followed by incubation with WST-1 reagent ( Roche , Indianapolis , IN ) and measurement as per the manufacturer’s instructions . For WST-1 assay on keratinocytes , keratinocytes were isolated as described previously ( Kumari et al . , 2013 ) and seeded 20 , 000 cells/well from the indicated genotypes in 96-well plates and stimulated with TNF ( 20 , 50 and 100 ng/ml ) in the presence or absence of CHX ( 1 μg/ml ) for 24 hr followed by incubation with WST-1 reagent ( Roche ) and measurement as per the manufacturer’s instructions . For the caspase-8 activation assay , 50 , 000 cells/well in a 96-well plate were plated and treated with mTNF , CHX , or Necrostatin-1 for the indicated time . Lysates were used for the determination of caspase-8 activity in a luminescent signal-dependent manner following the manufacturer’s protocol ( Promega , caspase-Glo 8 Assay Systems ) . Cells were fixed in 4% paraformaldehyde , permeabilized in 0 . 1%Triton/PBS , blocked in NGS/BSA/0 . 05%-Triton in PBS and incubated with α-cleaved caspase-3 antibody ( Cell Signaling Technologies ) followed by incubation with Alexa488 coupled anti-rabbit antibody ( Invitrogen ) . Nuclei were stained by Vectashield mounting media with DAPI ( VectorLabs ) . Lentiviral production and infection was performed according to the Addgene’s pLKO . 1 protocol with a minor modification . Briefly , pLKO . 1 vector with a packaging and envelope plasmids were transfected into HEK293T cells using Gene Juice ( Novagen ) . After 36 hr of transfection , released lentivirus particles were filtered and used for infection of target MEFs or HaCaT cells using polybrene ( 4 μg/ml ) . After 48 hr of infection , cells were selected with puromycin ( 2 μg/ml ) . Expression of infected protein was monitored by western blotting . Statistical significance was determined using ANOVA ( one-way or two-way ) and unpaired or paired Student's t test ( two-tailed ) by Prism 6 software ( Graph Pad ) or Microsoft Excel ( p ≤ 0 . 05 , p ≤ 0 . 01 , p ≤ 0 . 001 , ***p ≤ 0 . 0001 ) .	In response to an injury or an infection , areas of the body can become inflamed as the immune system attempts to repair the damage and/or destroy any microbes or toxins that have entered the body . At the level of individual cells inflammation can involve cells being programmed to die in one of two ways: apoptosis and necroptosis . Apoptosis is a highly controlled process during which the contents of the cell are safely destroyed in order to prevent damage to surrounding cells . Necroptosis , on the other hand , is not controlled: the cell bursts and releases its contents into the surroundings . Inflammation is activated by a protein called TNF , which is controlled by a complex that includes a protein called Sharpin . Mice that lack the Sharpin protein develop inflammation on the skin and other organs , even in the absence of injury or infection . However , it is not clear how the Sharpin protein controls TNF to prevent inflammation . Kumari et al . have found that inflammation in mice lacking Sharpin depends on TNF interacting with another protein called TRADD . The experiments also show that the inflammation is mainly driven by apoptosis , with necroptosis having only a minor role . Further experiments carried out in mammal cells showed that TRADD and another protein ( called FADD ) work with Sharpin to prevent apoptosis . At the molecular level , Sharpin is known to induce a special type of protein modification ( called linear ubiquitination ) with two partner proteins , so the next challenge is to work out exactly how Sharpin uses this process to prevent apoptosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2014	Sharpin prevents skin inflammation by inhibiting TNFR1-induced keratinocyte apoptosis
Synaptic membrane-remodeling events such as endocytosis require force-generating actin assembly . The endocytic machinery that regulates these actin and membrane dynamics localizes at high concentrations to large areas of the presynaptic membrane , but actin assembly and productive endocytosis are far more restricted in space and time . Here we describe a mechanism whereby autoinhibition clamps the presynaptic endocytic machinery to limit actin assembly to discrete functional events . We found that collective interactions between the Drosophila endocytic proteins Nwk/FCHSD2 , Dap160/intersectin , and WASp relieve Nwk autoinhibition and promote robust membrane-coupled actin assembly in vitro . Using automated particle tracking to quantify synaptic actin dynamics in vivo , we discovered that Nwk-Dap160 interactions constrain spurious assembly of WASp-dependent actin structures . These interactions also promote synaptic endocytosis , suggesting that autoinhibition both clamps and primes the synaptic endocytic machinery , thereby constraining actin assembly to drive productive membrane remodeling in response to physiological cues . At neuronal presynaptic terminals , actin assembly affects many physiological processes including synapse morphogenesis , traffic of numerous vesicular cargoes , and synaptic vesicle endocytosis , organization , and mobility ( Dillon and Goda , 2005; Nelson et al . , 2013; Papandréou and Leterrier , 2018 ) . However , the molecular mechanisms that control F-actin dynamics in space and time at presynaptic membranes are largely unknown . Presynaptic terminals maintain constitutively high local concentrations of actin-associated endocytic regulatory proteins at synaptic membranes ( Reshetniak et al . , 2020; Wilhelm et al . , 2014 ) , yet only a small fraction of this protein pool is likely to be active at any point in time ( in response to vesicle release ) and space ( at <100-nm-diameter endocytic sites ) , suggesting that the endocytic machinery is held in an inactive state at synaptic membranes . However , we do not know the mechanisms that maintain this machinery in an inactive state at the membrane , or how it is activated when and where it is needed . One plausible mechanism to restrict membrane-cytoskeleton remodeling and endocytic activity to specific locations and times may lie in autoinhibition , which is a property of multiple endocytic proteins ( Gerth et al . , 2017; Kim et al . , 2000; Rao et al . , 2010; Stanishneva-Konovalova et al . , 2016 ) . One example is the F-BAR-SH3 protein Nervous Wreck ( Nwk ) , which regulates synaptic membrane traffic at the Drosophila neuromuscular junction ( NMJ ) ( Coyle et al . , 2004; O'Connor-Giles et al . , 2008; Rodal et al . , 2008; Rodal et al . , 2011; Ukken et al . , 2016 ) and whose mammalian homolog FCHSD2 regulates endocytosis and endocytic traffic in mammalian cells ( Almeida-Souza et al . , 2018; Xiao and Schmid , 2020; Xiao et al . , 2018 ) . Nwk/FCHSD2 proteins couple two activities: membrane remodeling and WASp-dependent actin polymerization ( Almeida-Souza et al . , 2018; Rodal et al . , 2008; Stanishneva-Konovalova et al . , 2016 ) . Intramolecular autoinhibitory interactions between the Nwk F-BAR and its two SH3 domains mutually inhibit both Nwk membrane binding and activation of WASp ( Stanishneva-Konovalova et al . , 2016 ) . Unlike other F-BAR-SH3 proteins , which are completely released from autoinhibition upon membrane binding ( Guerrier et al . , 2009; Meinecke et al . , 2013; Rao et al . , 2010 ) , the SH3b domain of Nwk continues to restrict SH3a-mediated WASp activation even after Nwk binds membranes ( Stanishneva-Konovalova et al . , 2016 ) . This suggests that autoinhibition allows Nwk-WASp to remain inactive even after recruitment to the membrane , thus keeping the endocytic machinery in a primed but inactive state . We hypothesized that additional binding partners of NwkSH3b may be required to fully activate membrane remodeling at discrete times and locations at the synapse . An excellent candidate for release of Nwk autoinhibition at synapses is the endocytic adaptor intersectin ( Dap160 in Drosophila ) . Intersectin interacts with numerous endocytic proteins to regulate endocytosis in mammalian cells ( Henne et al . , 2010; Okamoto et al . , 1999; Praefcke et al . , 2004; Pucharcos et al . , 2000; Schmid et al . , 2006; Sengar et al . , 1999; Teckchandani et al . , 2012 ) and has been implicated in several steps of the synaptic vesicle cycle ( Evergren et al . , 2007; Gerth et al . , 2017; Jäpel et al . , 2020; Pechstein et al . , 2010; Pechstein et al . , 2015 ) . Of particular note , intersectin recruits the Nwk homolog FCHSD2 to sites of endocytosis ( Almeida-Souza et al . , 2018 ) , though it is not yet known how this affects FCHSD2 autoinhibition . In Drosophila , Dap160 interacts with WASp , Nwk , and other membrane-remodeling proteins via its four SH3 domains ( SH3AD ) , and regulates the levels and localization of many of these proteins , including Nwk ( Koh et al . , 2004; Marie et al . , 2004; Roos and Kelly , 1998 ) . Further , dap160 mutant phenotypes overlap with those of Nwk and WASp mutants , including impaired synaptic vesicle cycling and synaptic overgrowth ( Coyle et al . , 2004; Khuong et al . , 2010; Koh et al . , 2004; Marie et al . , 2004; Winther et al . , 2013 ) . Finally , intersectin and Dap160 shift localization from synaptic vesicle pools to the plasma membrane in response to synaptic activity ( Evergren et al . , 2007; Gerth et al . , 2017; Winther et al . , 2015 ) , suggesting that Dap160 may provide the spatiotemporal link between salient physiological triggers and Nwk/WASp activation . The high concentration and broad membrane distribution of inactive endocytic proteins ( Reshetniak et al . , 2020; Wilhelm et al . , 2014 ) make it difficult to characterize the molecular dynamics of synaptic endocytosis ( in contrast to non-neuronal cells; Kaksonen and Roux , 2018 ) . To overcome this barrier , we quantified discrete actin assembly events at the Drosophila NMJ as a proxy for productive endocytosis , as actin assembly is both a primary target of the endocytic apparatus under investigation and is required for synaptic vesicle endocytosis in all forms , including at the Drosophila NMJ ( Kononenko et al . , 2014; Wang et al . , 2010; Wu et al . , 2016 ) . This synapse is an ideal system to investigate the molecular dynamics of the endocytic machinery due to its large size , ease of genetic manipulation , and accessibility to live and super-resolution imaging . Here we combine in vitro biochemical approaches with quantitative imaging at the NMJ to define the interactions among Dap160 , Nwk , and WASp that relieve autoinhibition . These interactions drive robust membrane-associated actin assembly in vitro , regulate the frequency and dynamics of synaptic actin structures in vivo , and are functionally required for normal endocytosis at the NMJ . While the importance of actin in synaptic endocytosis is clear ( Kononenko et al . , 2014; Wang et al . , 2010; Wu et al . , 2016 ) , until now there has been no quantitative analysis of individual actin-dependent membrane-remodeling events at synapses . To better understand presynaptic F-actin dynamics and to identify sites where the cytoskeleton- and membrane-remodeling machinery is active , we quantified individual F-actin assembly events by spinning disc confocal microscopy of NMJs presynaptically expressing fluorescent actin probes . To control for developmental variation , all experiments were performed on late third-instar larvae ( ~96–120 hr after egg laying ) on muscle 6/7 NMJs at abdominal segments 3–4 , since the development and physiology of these synapses are well characterized ( Harris and Littleton , 2015 ) . To control for variation in size between neurons , we normalized patch frequencies by the synapse area measured and presented data per 10 µm2 , which is approximately the size of a synaptic bouton in this system . We performed these experiments under resting conditions , where vesicle release is spontaneous at a rate of ~5–6 vesicles/10 µm2/min ( Akbergenova et al . , 2018; Melom et al . , 2013 ) , presumably requiring a similar rate of compensatory endocytosis ( Sabeva et al . , 2017 ) . We first compared the dynamics of three actin markers: GFP::actin , GFP-tagged moesin F-actin-binding domain ( GMA ) , and Lifeact::Ruby . The predominant structures labeled by these markers were transient patches at the presynaptic membrane ( Video 1 , Figure 1A , Figure 1—figure supplement 1A ) , as has been previously observed ( Nunes et al . , 2006; Pawson et al . , 2008; Piccioli and Littleton , 2014 ) . We then quantified individual actin patch dynamics using automated particle tracking and quantification ( Berro and Pollard , 2014; Tinevez et al . , 2017 ) , which captured on the order of 30–50% of visible actin structures ( see 'Materials and methods' , Figure 1—figure supplement 2 , and Figure 6—figure supplement 1 for more details on optimization and validation of actin particle analysis ) . We first imaged at 0 . 25 Hz and measured an average of 1 . 2 GMA patches/10 μm2/min , exhibiting a mean duration of 48 . 0 s ± 45 . 6 s , with an average relative amplitude of 68 ± 32% ( ( Imax-Imin ) /Imean ) ( Figure 1B–D ) . Quantification of GFP::actin and Lifeact::Ruby showed very similar dynamics to GMA , suggesting that these measurements robustly reflect the underlying actin dynamics and not the specific properties of a particular probe . We did note a high percentage of patches in the minimum duration bin , suggesting the existence of even briefer patches . To address this , we also performed imaging at 1 Hz , which could not capture the entire lifetime distribution due to photobleaching but was able to identify a larger population of short-duration patches ( Figure 1—figure supplement 1B ) with an average duration of ~16 +/- 20 s . Given this range of measurements at different sampling frequencies and the efficiency of our automated detection , we estimate that patch frequency is between 2 . 8 and 10 . 3 events/10 µm2/min ( see 'Materials and methods' for calculations ) , on par with the expected frequency of endocytic events , and with a similar albeit broader distribution of durations compared to yeast ( 15 s; Berro and Pollard , 2014 ) and mammalian cells ( ~40 s; Taylor et al . , 2011 ) . We next examined the molecular determinants of synaptic actin patch assembly . Patches strongly co-labeled with Arp3::GFP ( Pearson coefficient 0 . 70 ) , significantly higher than the active zone marker Bruchpilot ( BRP ) , which served as a punctate and membrane-localized negative control ( Figure 1E–F ) . These data suggest that actin patches are predominantly composed of branched F-actin , similar to sites of endocytosis in other cell types ( Akamatsu et al . , 2020; Collins et al . , 2011 ) . To test whether synaptic actin patches require Arp2/3 activation , we analyzed patch dynamics in larvae lacking the Arp2/3 activator WASp . We compared a genomic mutant ( Figure 1G–I ) , likely hypomorphic due to maternal contribution ( Ben-Yaacov et al . , 2001 ) , to presynaptic depletion in neurons expressing WASp RNAi ( Figure 1—figure supplement 1 and C–F ) . Using both approaches allows us to distinguish neuron-autonomous from non-autonomous effects of WASP , which is present both pre- and postsynaptically ( Coyle et al . , 2004 ) . Both genomic and RNAi manipulations significantly reduced the number of actin patches , while genomic mutants also skewed the distribution of patch durations toward both shorter and longer events ( Figure 1I ) . These differences could reflect variable loss of function between the RNAi and mutant , or identify separable presynaptic autonomous ( patch frequency ) vs non-autonomous ( patch duration ) effects of WASp . Overall , these data clearly indicate that WASp is autonomously required in neurons to initiate assembly of presynaptic actin patches , similar to its involvement in endocytosis in yeast , mammalian non-neuronal cells , and in the NMJ ( Hussain et al . , 2001; Kessels and Qualmann , 2004; Khuong et al . , 2010; Madania et al . , 1999 ) . We next examined the synaptic distribution of two likely WASp regulators , Nwk and Dap160 . By conventional and super-resolution microscopy of neurons in diverse organisms , these and other presynaptic membrane-remodeling proteins localize to a broad membrane domain surrounding active zones , termed the periactive zone ( PAZ ) ( Coyle et al . , 2004; Denker et al . , 2011; Gerth et al . , 2017; Koh et al . , 2004; Marie et al . , 2004; Sone et al . , 2000 ) . Consistent with these prior descriptions , we observed by structured illumination microscopy ( SIM ) that the PAZ proteins Nwk and Dap160 localize to a membrane-proximal mesh that surrounds active zones , which were labeled with BRP ( Figure 2A ) . We observed similar results by live imaging of an endogenously tagged Nwk protein by SIM , which revealed most proteins to be close to the plasma membrane ( Figure 2B ) . We then compared the localization of PAZ proteins to F-actin patches at the NMJ . As expected , actin patches were much sparser than the endocytic machinery , and GMA-labeled patches only partially overlapped with each of the presynaptic WASp , Nwk , and Dap160 ( Figure 2C–F; Pearson’s coefficients of 0 . 38 , 0 . 38 , and 0 . 36 , respectively ) . These data confirm that , in sharp contrast to the actin regulatory machinery , which localizes broadly across the PAZ , actin assembly itself is much sparser both spatially and temporally at the NMJ . This raises the question of how PAZ machinery might itself be locally regulated to promote the formation of productive synaptic actin assemblies . The hypothesis that PAZ protein-mediated actin assembly might be locally activated is particularly interesting given that we and others have previously shown that autoinhibition of both Nwk and its mammalian homolog FCHSD2 suppresses both WASp activation and membrane binding ( see Figure 3A for summary model; Almeida-Souza et al . , 2018; Rodal et al . , 2008; Stanishneva-Konovalova et al . , 2016 ) . These results suggest that transient or localized relief of autoinhibition could explain how the PAZ controls actin assembly . To determine if and how the candidate activator Dap160 might relieve Nwk autoinhibition , we first mapped their specific interaction domains using glutathione-S-transferase ( GST ) pulldown assays and found that purified Dap160 SH3C-containing protein fragments ( SH3C , SH3CD , or SH3ABCD ) directly interact with NwkSH3b , while SH3D alone does not ( Figure 3B , Figure 3—figure supplement 1; see Figure 3—figure supplement 2A for details of constructs used ) . Unexpectedly , Dap160 SH3C , SH3D , and SH3CD domain fragments also , each , interact with the isolated Nwk F-BAR domain ( Figure 3—figure supplement 1B ) . We next determined how Dap160 interacts with NwkF-BAR compared to a Nwk fragment containing the F-BAR and both SH3 domains . Dap160-NwkF-BAR interactions were progressively eliminated by increasing salt , suggesting they are mediated by electrostatic interactions . By contrast , Dap160SH3CD-Nwk interactions were maintained ( Figure 3B , Figure 3—figure supplement 1C ) , suggesting that the SH3-SH3 interaction is mediated primarily by hydrophobic interactions , consistent with their mammalian homologs ( Almeida-Souza et al . , 2018; see summary of interactions in Figure 3C ) . Finally , we found that truncation of Dap160SH3CD decreased the levels of Nwk in synaptic boutons similarly to Dap160 knockdown ( Figure 3D , Figure 3—figure supplement 2B–C ) . Dap160ΔSH3CD also exhibited reduced colocalization with Nwk compared to wild-type Dap160 ( Figure 3E , Figure 3—figure supplement 2C ) , further supporting an in vivo requirement for this interaction . Notably , truncation of Dap160SH3D did not exhibit a phenotype in these assays despite lower levels of expression ( Figure 3—figure supplement 2B ) , suggesting that additional factors absent from our in vitro assays may collaborate to regulate Nwk in vivo . We have previously shown that Nwk only weakly activates WASp-dependent actin assembly in vitro , due to Nwk autoinhibition ( Stanishneva-Konovalova et al . , 2016 ) . To test whether Dap160SH3CD might relieve Nwk autoinhibition , we performed pyrene-actin assembly assays ( Figure 4 ) . At moderate Nwk-Dap160 concentrations ( 500 nM and 2 µm , respectively ) , Nwk and Dap160SH3CD significantly enhanced the rate of WASp-Arp2/3-mediated actin assembly compared to Nwk plus WASp alone ( Figure 4A ) . This effect is through Nwk , as Dap160SH3CD had no effect on WASp-Arp2/3 in the absence of Nwk . Further , Dap160 enhancement of Nwk-WASp actin assembly required the Dap160SH3D domain , further showing that the specific Dap160SH3D-NwkF-BAR interaction relieves functional Nwk autoinhibition in vitro . Thus , multiple Nwk-Dap160 interactions work together to relieve autoinhibition of Nwk . To generate salient physiological force , actin assembly must be coupled to membranes , and negatively charged lipids are an important ligand for both Nwk and WASp . Thus , we next tested whether addition of PI ( 4 , 5 ) P2-rich liposomes modified actin assembly by Nwk , Dap160 , and WASp ( Figure 4B ) . Indeed , PI ( 4 , 5 ) P2-containing liposomes synergistically activated WASp-mediated actin assembly in concert with Dap160 and Nwk . By contrast , neither Nwk , PI ( 4 , 5 ) P2 , nor Nwk + PI ( 4 , 5 ) P2 on their own were sufficient to activate WASp above baseline ( Figure 4B ) . Since PI ( 4 , 5 ) P2 is also insufficient to robustly activate either WASp or Nwk under these conditions ( Stanishneva-Konovalova et al . , 2016 ) , our data suggest that WASp activation reflects coordinated relief of Nwk autoinhibition by both Dap160 and membranes . To further explore the coupling between lipid association and actin assembly , we conducted F-actin assembly assays in a droplet assay , in which protein-containing aqueous droplets are surrounded by a lipid interface , with lipid head groups facing the aqueous phase ( Figure 4C ) . In this assay , we found that coordinated interactions among Nwk , Dap160 , and WASp directed actin assembly to the lipid interface . By contrast , substitution of Nwk lacking its autoinhibitory and Dap160-interacting SH3b domain ( NwkΔSH3b ) caused actin to assemble as free-floating asters ( Figure 4C ) . We have previously found that expression of a similarly deregulated fragment ( Nwk1-631 ) at the NMJ led to diffuse actin filament assembly throughout the synapse ( Stanishneva-Konovalova et al . , 2016 ) . Together , these data suggest that NwkSH3b has a dual role in maintaining autoinhibition via Nwk-F-BAR interactions and permitting actin assembly at specific synaptic locations via Dap160-mediated activation . Our actin assembly data suggest that membrane recruitment is a critical regulator of the Nwk-Dap160-WASp complex ( Figure 4B–C ) . To test whether Nwk-Dap160 interactions directly regulate membrane recruitment , we performed liposome cosedimentation assays . We found that Dap160SH3CD enhanced Nwk membrane binding in a dose-dependent fashion ( Figure 5A ) . This effect depended on membrane charge , as Dap160SH3CD significantly enhanced Nwk membrane binding at both 5 and 10% , but not at 2 . 5% PI ( 4 , 5 ) P2 ( Figure 5B ) . Only at 10% PI ( 4 , 5 ) P2 did Dap160SH3CD promote Nwk membrane binding to the same extent as the completely uninhibited NwkFBAR domain alone , suggesting that membrane charge and intermolecular interactions with Dap160 together tune Nwk membrane recruitment . Indeed , this effect required the full Dap160SH3CD-NwkSH3b interaction: Dap160SH3C alone was unable to promote membrane binding by Nwk , and Dap160SH3CD did not enhance membrane binding of Nwk lacking its Dap160-interacting SH3b domain ( Figure 5—figure supplement 1A ) . These data further support the hypothesis that Dap160SH3CD relieves NwkSH3b-mediated autoinhibition . As we found that Dap160SH3CD is insufficient to fully activate membrane binding by Nwk at intermediate phosphoinositide concentrations ( Figure 5A ) , we asked whether WASp could further enhance Nwk membrane recruitment . Indeed , the addition of Dap160SH3CD and WASp together enhanced Nwk membrane association to the level of the isolated F-BAR domain ( Figure 5C ) . Moreover , coordinated binding of all three components resulted in significantly enriched membrane association of both WASp and Dap160 ( Figure 5C ) . We directly observed the coordinated recruitment of Nwk and Dap160 in the presence of WASp using fluorescently labeled proteins on GUVs ( Figure 5D ) . Consistent with the direct Dap160-NwkSH3b interaction , we found that deletion of the NwkSH3b domain abolished both the Dap160SH3CD-dependent increase and the coordinated recruitment of WASp and Dap160 ( Figure 5—figure supplement 1A ) . Notably , addition of Dap160 and WASp did not change the nature of membrane deformations generated by Nwk ( scalloped and pinched membranes; Becalska et al . , 2013 ) , suggesting that Dap160 and WASp together potentiate rather than alter the inherent activity of Nwk ( Figure 5—figure supplement 1D ) . These data indicate that Dap160-Nwk SH3-mediated interactions potentiate Nwk association with membranes in vitro . Finally , to test whether Dap160 promotes Nwk membrane association in vivo , we examined the dynamics of Nwk at the synapse in the presence and absence of Dap160 . Knockdown of Dap160 by RNAi ( Figure 5E , Figure 3—figure supplement 2D ) led to a striking loss of endogenously tagged Nwk::GFP from synaptic membranes ( note strong peripheral labeling in control bouton cross-sections; Figure 5E ) . Further , Dap160 knockdown significantly increased the rate of recovery of Nwk::GFP after photobleaching , consistent with a shift in localization from membrane-bound to cytosolic ( Figure 5F ) . These data suggest that the Dap160SH3CD-Nwk interaction promotes Nwk membrane association in vivo . Taken together , our data indicate that multiple coordinated interactions between Nwk , WASp , Dap160SH3CD , and membranes are required to relieve Nwk autoinhibition , allowing for tight control of membrane-coupled actin assembly in the PAZ . To determine how these mechanisms direct WASp-mediated actin assembly at the synapse , we measured actin dynamics in nwk ( Figure 6A–C , Video 2 ) and dap160 domain ( Figure 6D–F ) mutant NMJs . We predicted two possible but non-exclusive functions based on the dual roles that we found for the Nwk-Dap160-WASp module in vitro: if Nwk and Dap160 are primary activators of WASp , then loss-of-function mutants are likely to diminish patch frequency , duration , or intensity . Importantly , multiple WASp activators exist in the synaptic endocytic machinery ( e . g . , Cip4 and Snx9; Almeida-Souza et al . , 2018; Gallop et al . , 2013; Nahm et al . , 2010; Ukken et al . , 2016 ) , and therefore , these could make significant contributions to WASp activation in addition to Nwk . Conversely , if an important function of autoinhibition is to ‘clamp’ actin assembly at the synapse , we expected that loss of Nwk and/or Dap160 would lead to spurious actin assembly events by these other WASp regulators . We found that both nwk and Dap160ΔSH3CD mutants significantly increased patch frequency ( Figure 6B , E , Figure 6—figure supplement 1 ) , supporting a clamp function for these proteins . We did not detect a difference in the distribution of patch lifetimes , suggesting that it is the frequency of events , and not their duration per se , that changes ( Figure 6C , F ) . We also analyzed actin dynamics using a complementary approach in which we measured the normalized intensity variation ( coefficient of variation , CoV ) over time across the entire NMJ . Interestingly , the magnitude of variation was significantly higher in nwk mutants ( Figure 6—figure supplement 2A–B ) , but the area of the NMJ that was highly variant was similar between genotypes , suggesting that actin assembly is more dynamic in time in these mutants , rather than more extensive in space ( Figure 6—figure supplement 2C ) . We validated this analysis for its sensitivity in detecting changes in event frequency by analyzing synthetic data ( Figure 6—figure supplement 2D , see 'Materials and methods' for details ) . The modeled data suggest that the difference in CoV that we measured between Control and nwk is consistent with a 43% increase in patch frequency , which is slightly higher than our measurement by particle tracking ( 28%; Figure 6A ) . This complementary analysis does not rely on particle tracking and makes no assumptions about the nature of actin dynamics , and is consistent with our particle-based metrics . Thus , we conclude that these phenotypes are robust to the method of analysis used . We next investigated the physiological function of actin patches in vivo . Considering that patch morphology , frequency , and duration resembled endocytic dynamics , we first compared actin patches with the endocytic adaptors Clc and AP2α . Like other endocytic proteins , both presynaptically expressed Clc::GFP and endogenously tagged AP2α::GFP were primarily enriched at the plasma membrane relative to the cytoplasm ( Figure 7—figure supplement 1A ) and covered a large area fraction of the membrane , similar to other endocytic proteins ( Figure 2 ) . In addition to diffuse signal , both probes localized to short- and long-lived puncta , a subset of which dynamically colocalized with actin patches ( Figure 7A–C , Figure 7—figure supplement 1B , Video 3 ) . A significant proportion of endogenously labeled AP2 at the NMJ is likely associated with the closely apposed postsynaptic membrane , which accounts for its slightly lower correlation coefficient with Lifeact::Ruby . Considering that the rates of exo/endocytosis at this synapse at rest are relatively low ( see above ) , these observations suggest that like other PAZ endocytic proteins , a large pool of membrane-localized clathrin coat and adaptor proteins are not actively engaged in endocytosis . Despite these caveats , we found that actin significantly colocalized with both Clc ( Figure 7C ) and AP2 ( Figure 7—figure supplement 1C ) , consistent with a role in endocytosis for these actin-enriched sites . To more rigorously and functionally test the hypothesis that actin patches are endocytic , we acutely disrupted endocytic dynamics using the temperature-sensitive dominant-negative dynamin/shiTS1 allele . When imaged under restrictive conditions , shi disruption decreased the frequency of actin patch dynamics ( Figure 7D–E ) . Together , these data suggest that a significant fraction of presynaptic actin patches are associated with endocytosis . We next tested the physiological requirement of the Nwk and Dap160SH3CD interaction . As both Nwk and Dap160 are implicated in the endocytic trafficking of synaptic growth-promoting bone morphogenetic protein ( BMP ) receptors ( O'Connor-Giles et al . , 2008; Rodal et al . , 2008 ) , we tested whether the Dap160-Nwk interaction was required for normal synaptic growth , which we assayed by counting satellite boutons , a hallmark phenotype of both null mutants . Surprisingly , we found that both Dap160ΔSH3D and Dap160ΔSH3CD truncations rescued satellite bouton numbers to wild-type levels ( Figure 7—figure supplement 2 ) . These data indicate that synaptic vesicle and growth factor endocytosis are mechanistically separable , and suggest that actin dynamics phenotypes in the Dap160ΔSH3CD mutant are not associated with synaptic growth regulation . We next examined synaptic vesicle endocytosis and recycling by FM dye uptake . nwk1/2 null mutants led to a 34% decrease in FM4-64fx uptake compared to controls ( Figure 7F–G ) , an intermediate phenotype compared to dominant negative dynamin in shiTS1 mutants ( 72% decrease ) . dap160 null mutants have been previously shown to exhibit an endocytosis defect ( Koh et al . , 2004; Marie et al . , 2004 ) , so we next tested whether the interaction between Dap160 and Nwk is required to support normal endocytosis . Indeed , we found that expression of Dap160ΔSH3CD in dap160 null mutants also significantly diminished FM dye uptake to a similar extent as loss of nwk ( 27% reduction; Figure 7H–I ) . By contrast , loss of the Dap160SH3D domain alone caused no defects in FM uptake , consistent with the lack of effect of this mutation on Nwk accumulation and localization ( Figure 3D–E ) , suggesting that this interaction , though required in vitro , may be compensated by additional factors in vivo . Both nwk and Dap160ΔSH3CD mutants unloaded FM dye to the same extent as controls , suggesting that diminished endocytosis is a direct phenotype , and not secondary to exocytic deficits ( Figure 7—figure supplement 3 ) . Importantly , these data indicate that spurious actin assembly events in nwk and dap160 mutants are likely to be unproductive for normal endocytosis . Overall , our data support the hypothesis that normal synaptic actin patches represent active endocytic events and indicate that Dap160-Nwk regulation of actin patch dynamics is functionally required for synaptic vesicle endocytosis . Here we provide the first quantitative analysis of the composition and dynamics of individual presynaptic F-actin structures . Numerous studies have examined actin dynamics at the level of entire synapses or qualitatively described dynamics of discrete actin structures ( Bloom et al . , 2003; Colicos et al . , 2001; Nunes et al . , 2006; Piccioli and Littleton , 2014; Sankaranarayanan et al . , 2003; Zhao et al . , 2013 ) , and identified diverse roles for actin , including synaptic vesicle endocytosis ( Holt et al . , 2003; Kononenko et al . , 2014; Soykan et al . , 2017; Watanabe et al . , 2013; Wu et al . , 2016; Zhao et al . , 2013 ) , synaptic vesicle organization and mobilization ( Guzman et al . , 2019; Lee et al . , 2012; Marra et al . , 2012; Owe et al . , 2009; Sakaba and Neher , 2003; Wolf et al . , 2015 ) , active zone organization and function ( Pilo Boyl et al . , 2007; Morales et al . , 2000; Wagh et al . , 2015; Waites et al . , 2011; Wang et al . , 1999 ) , and receptor-mediated endocytosis ( Kim et al . , 2019; Rodal et al . , 2008 ) . Bulk analyses , which do not separate individual dynamic actin structures in space and time , are limited in their ability to discern how the regulation and dynamics of actin contribute to these distinct functions . We leveraged our ability to extract data describing individual structures to find that synaptic actin predominantly assembled into discrete Arp2/3-associated patches , and identified points of control over their dynamics . Specifically , we found that loss of endocytic proteins differentially affected the frequency and kinetics of individual actin patches , which correlate with functional deficits in endocytosis . The link between the actin structures that we observed and endocytic events is supported by several lines of evidence: the morphology and duration of synaptic actin patches are similar to WASp/Arp2/3-dependent endocytic actin dynamics in yeast ( 16 s; Berro and Pollard , 2014 ) and somewhat briefer than in cultured mammalian cells ( ~40 s; Grassart et al . , 2014; Merrifield et al . , 2004; Taylor et al . , 2011 ) . Given the measured time constant for endocytosis ( ~14 s; Poskanzer et al . , 2006 ) and clathrin dependence of vesicle cycling in this synapse ( Heerssen et al . , 2008 ) , these values support the hypothesis that synaptic actin patches are likely sites of clathrin-mediated endocytosis . Further , the frequency of patch assembly , which we measured in resting synapses , approaches the rate of spontaneous synaptic vesicle release at this synapse ( Figure 1—figure supplement 1B , Figure 1—figure supplement 2A ) ( ~5–6/10 μm2/min; Melom et al . , 2013; Akbergenova et al . , 2018 ) . Further , actin patches colocalize partially with endocytic adaptors , and their assembly is sensitive to disruption of endocytosis ( Figure 7 ) . Finally , we found that the same endocytic proteins and protein interactions that regulate endocytosis at this synapse also alter the dynamics of actin patches . Technical challenges due to the high density of endocytic proteins and synaptic vesicle cargoes , and the difficulty of conducting sparse single vesicle measurements at this synapse ( compared to neurons in culture; Chanaday and Kavalali , 2018; Peng et al . , 2012 ) , make it difficult to directly link the dynamics of actin structures to specific membrane or cargo internalization events . However , the frequency of the events captured by our approach makes it unlikely that they represent rare F-actin-dependent events at this synapse , such as those that control macropinocytosis or new bouton growth ( Khuong et al . , 2010; Kim et al . , 2019; Piccioli and Littleton , 2014 ) , and more likely that they represent bona fide endocytic events . Thus , while we do not rule out other biological functions for a subset of patches , together our data indicate that a significant and measurable fraction of synaptic actin patches are associated with endocytosis . Many endocytic proteins accumulate across broad membrane domains at the Drosophila NMJ and other synapses ( Gerth et al . , 2017; Guichet et al . , 2002; Koh et al . , 2007; Roos and Kelly , 1998; Verstreken et al . , 2008 , Verstreken et al . , 2003 ) . Our data indicate that much of this membrane-remodeling machinery is likely held in an inactive state at the presynaptic membrane: Nwk and Dap160 accumulate in a micron-scale membrane domain ( Figure 2 ) , and their loss increases the frequency of short-lived actin patches ( Figure 6 ) . These data suggest that these PAZ proteins are held in a partially autoinhibited state at the membrane in vivo , consistent with our prior in vitro observations ( Stanishneva-Konovalova et al . , 2016 ) . This is further consistent with the broad distribution of Clc and AP2 to the membrane ( Figure 7 , Figure 7—figure supplement 1 ) . Given the comparatively low rate of endocytosis expected at rest at this synapse , this suggests that most Clc and AP2 puncta at the synapse are either not stabilized to form productive endocytic sites ( Aguet et al . , 2013 ) or associated with some non-endocytic functions ( Gimber et al . , 2015 ) . The fact that loss of Nwk increases the frequency of patches while decreasing FM uptake suggests that the actin structures assembled in the nwk mutant are unproductive for synaptic vesicle endocytosis . These spurious patches could reflect non-specific actin assembly , perhaps due to unmasking of the Nwk ligand PI ( 4 , 5 ) P2 at the membrane and/or inappropriate activation of alternative WASp-dependent actin assembly pathways . Indeed , additional WASp activators such as Snx9 and Cip4/Toca-1 may play accessory roles in endocytic actin assembly ( Almeida-Souza et al . , 2018; Gallop et al . , 2013 ) , consistent with our finding that loss of presynaptic WASp leads to a decrease in the total number of patches ( Figure 1G–I , Figure 1—figure supplement 1C–E ) . Our data indicate that at the synapse , where endocytic machinery accumulates at high concentrations ( Wilhelm et al . , 2014 ) and recruitment appears uncoupled from activation , these layers of autoregulation are critical to constrain actin assembly generally . Our findings on autoinhibition and clamping connect two prevailing models of the organization and function of the synaptic endocytosis machinery—preassembly and delivery . In the first model , preassembly of clathrin and accessory proteins is hypothesized to ensure fast endocytosis in response to synaptic vesicle fusion ( Hua et al . , 2011; Mueller et al . , 2004; Wienisch and Klingauf , 2006 ) . Here , Nwk autoinhibition provides a mechanism to assemble an inactive , yet poised endocytic apparatus . In the second model , endocytic machinery associates with the synaptic vesicle pool , providing a ready source or buffer of proteins that can be released to the plasma membrane upon calcium signaling or vesicle fusion ( Bai et al . , 2010; Denker et al . , 2011; Gerth et al . , 2017; Winther et al . , 2015 ) . Because Dap160/intersectin can shuttle between the synaptic vesicle pool and the plasma membrane , is itself subject to autoregulation ( Gerth et al . , 2017 ) , and can regulate other endocytic proteins ( e . g . , dynamin , Nwk ) , it could serve as a single activator that couples the preassembly and delivery models . Our in vitro data show that beyond functioning as a clamp , Nwk and Dap160 collaboratively activate WASp to promote robust actin assembly . Together with the defects we observed in vivo for actin dynamics and FM dye uptake , these data suggest that Dap160-Nwk-WASp interactions could serve as a coincidence detection mechanism to relieve autoinhibition of Nwk and promote productive actin assembly with other WASp activators . Coincidence detection has been demonstrated in several systems to control membrane-associated actin assembly ( Case et al . , 2019; Sun et al . , 2017 ) , suggesting that amplification of WASp membrane binding could drive robust actin patch assembly at synapses . Similarly , in human cells , the interaction between FCHSD2 , intersectin , and WASp promotes actin assembly and endocytic maturation ( Almeida-Souza et al . , 2018 ) or initiation ( Xiao et al . , 2018 ) . The Dap160-Nwk module could act by directing and/or organizing actin assembly specifically at endocytic events , akin to the membrane-directed actin assembly we observed in vitro ( Figure 5D ) , and/or ensure that it is sufficiently robust for productive membrane remodeling ( Akamatsu et al . , 2020 ) . Direct support for these models will require new analytical or imaging approaches to directly visualize the coupling of membranes and actin to the endocytic machinery , in order to distinguish spurious ( due to unclamping ) vs bona fide but underpowered endocytic actin assembly events . Our data suggest that the endocytic machinery can be deployed as clamped , primed , or activated complexes at different locations at the synapse . The next critical step will be to determine the mechanisms that control switching between these states . Many potential mechanisms that link calcium-dependent exocytosis and endocytosis could activate actin assembly . These include direct effects of calcium on the endocytic machinery ( Maritzen and Haucke , 2018 ) , the accumulation of synaptic vesicle cargoes ( Cousin , 2017 ) , stoichiometry-dependent changes in protein interactions or activities ( Case et al . , 2019 ) , changes in membrane mechanics ( Anantharam et al . , 2010; Dai et al . , 1997; Roux et al . , 2010 ) , and changes in membrane charge/mode of membrane binding ( Kelley et al . , 2015 ) . One intriguing possibility is that these mechanisms might enable an endocytic PAZ to rapidly switch between different modes of endocytosis ( e . g . , ultrafast , conventional , or bulk ) in response to a wide range of synaptic activity patterns ( Gan and Watanabe , 2018 ) . These endocytic regulatory mechanisms could also be locally poised to regulate , respond , or adapt to the specific release properties of nearby active zones ( Akbergenova et al . , 2018; Melom et al . , 2013; Dickman et al . , 2006 ) , and serve as novel points of control over synaptic plasticity and homeostasis . Spinning disc confocal time series were acquired at 15 stacks/min ( Figure 1 ) , 60 stacks/min ( Figure 6 , Figure 7E ) , or 2 . 2 stacks/min ( Figure 7A ) . A maximum intensity projection was made of each time point , videos were registered using the FIJI plugin StackReg , and analyzed by Patchtracker , based on Trackmate ( Berro and Pollard , 2014 ) as follows . First , we qualitatively evaluated the optimal intensity threshold for patch detection by identifying the maximum threshold intensity at which all obvious patch structures were detected in the first frame of videos . This process was performed independently by three independent observers over multiple datasets . The threshold for patch detection was normalized to the mean probe intensity in the presynaptic area ( threshold = Probe Mean * 0 . 32 ) . All other settings for patch detection and tracking were default: estimated patch diameter = 0 . 6 μm , median filter = false , subpixel detection = true , linking max distance = 0 . 5 μm , gap-closing distance = 0 . 5 μm , gap-closing frame gap = 0 . For 0 . 25 Hz imaging experiments , patches between 16 and 356 s could be detected . For 1 Hz imaging experiments , patches between 4 and 139 s could be detected . Because this analysis rejects a significant number of detected patches due to tracking defects or tracking path overlap , we estimated the true patch frequency as follows . We combined detections from 0 . 25 Hz and 1 Hz imaging experiments by averaging the frequencies over the shared detection range ( 20–150 s ) and adding the lower and higher duration patches that were specific to each imaging regime ( 4–16 s for 1 Hz and 150–360 s for 0 . 25 Hz ) . Then we ‘corrected’ for rejected tracks and considered the lower bound of the estimate to be the actual , corrected merged frequency of detection ( 2 . 8 patches/10 μm/min ) and the upper bound to include every rejected track ( 10 . 3 patches/10 μm/min ) . We further validated our patch dynamics analysis by measuring patch frequencies at a wide range of patch intensity thresholds and track linking distances . For both 0 . 25 Hz WASp ( Figure 1—figure supplement 2 ) and 1 Hz Nwk ( Figure 6—figure supplement 2 ) datasets , we found measurements of control patch frequencies to be robust to these parameters and in strong agreement with the estimates described above across the entire parameter space tested ( 1 . 1–8 . 4 for 0 . 25 Hz imaging , 1 . 2–7 . 9 for 1 Hz imaging ) . Further , our phenotypic analyses ( decreased patch frequency in WASp mutants and increased frequency in Nwk mutants ) were also both highly robust to tracking parameters . Actin dynamics were also analyzed by measuring intensity variation over time over the entire NMJ , that is , without thresholding or particle tracking . We measured this by extracting the intensity value for each pixel over time and calculating the CoV ( Std Dev/Mean ) for each pixel . We estimated the percentage of ‘highly variant’ pixels by thresholding these values using Li ( Li and Tam , 1998 ) and Moments ( Tsai , 1995 ) algorithms . While these two algorithms gave different estimates of the fraction of NMJs covered by highly variant pixels , both indicated the same relationship between genotypes . To validate this approach , we created synthetic data using a custom FIJI script , with a spatial and temporal scale that matched our in vivo imaging , and in which we varied parameters expected to impact this metric ( signal intensity , noise level , fraction of dynamic pixels , dynamics frequency , dynamics duration , dynamics amplitude ) , and subjected the synthetic data to the same CoV over time analysis . For intensity and colocalization , the presynaptic region was masked in 3D using a presynaptically enriched label: either HRP ( Figure 3—figure supplement 2D ) , Nwk ( Figure 2A , Figure 2C ) , Dap160 ( Figure 2E , Figure 3D–E , Figure 3—figure supplement 2B–C ) , or Lifeact::Ruby ( Figure 1E , Figure 7C . Figure 7—figure supplement 1C ) . For mask generation , images were subjected to a gaussian blur filter and thresholded by intensity . Blur radius and the specific threshold algorithms used were empirically optimized for each experiment to consistently and accurately reflect the presynaptic area in control and mutant groups ( and the same settings were used for all NMJs within any given experiment ) . Signal intensities were measured in 3D using a FIJI script , and colocalization analysis was performed in 3D on Airyscan or SIM reconstructed image stacks using the Coloc2 plugin for ImageJ ( https://imagej . net/Coloc_2 ) . For all images , background was subtracted using the rolling ball method with a radius of 50 pixels . Coprecipitation with GST-tagged proteins was conducted as described previously ( Kelley et al . , 2015 ) . Concentrations of GST fusions on beads were normalized using empty beads and bead volume was restricted to two-thirds of the total reaction volume . GST fusions were incubated by agitation with His-tagged target proteins at room temperature for 1 hr in binding buffer ( 20 mM Tris , pH 8 . 0 , 20 mM KCl , 0 . 5 mM DTT ) . For salt sensitivity experiments , the indicated concentrations of NaCl were used in place of KCl in the binding buffer . Beads were then pelleted and washed once with buffer after removing the supernatant . Pellets and supernatants were then boiled in Laemmli sample buffer and fractionated by SDS-PAGE , followed by Coomassie staining or immunoblotting as noted in figure legends , followed by imaging and analysis on a LICOR Odyssey device . Lipid cosedimentation assays were conducted as described previously ( Becalska et al . , 2013 ) . In brief , liposomes were swelled from dried lipid films in 20 mM 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) , pH 7 . 5 , and 100 mM NaCl . Specific lipid compositions are indicated in the figure legends . Proteins were then mixed with 1 mg/ml liposomes , incubated for 30 min at room temperature , and then pelleted for 20 min at 18 , 000 ×g at 4°C . Pellets and supernatants were then denatured in Laemmli sample buffer and fractionated by SDS-PAGE , followed by Coomassie staining , and imaging and analysis conducted on a LICOR Odyssey device . GUVs were generated by gentle hydration . Briefly , 10 µl of 10 mg/ml lipids dissolved in 19:1 chloroform:methanol were dried under vacuum , and then swelled in 300 µl of 5 mM HEPES 300 mM sucrose , pH 7 . 5 , overnight at 70°C . GUVs were imaged on a Marianas spinning disc confocal system ( see above ) . 3 µl GUVs were diluted into 5 mM HEPES 150 mM KCl , pH7 . 5 , incubated with protein as noted in figures , and imaged using a x100/NA 1 . 4 objective in multiwell slides ( Lab-Tek ) precoated with 1 mg/ml bovine serum albumin . After 30 min of incubation , 1% agarose in 5 mM HEPES 150 mM KCl , pH 7 . 5 , was added ( final agarose concentration , 0 . 5% ) to limit GUV mobility . Images were analyzed by line tracing intensity profiles across a medial optical section of GUVs . Lipids ( 97 . 5% DPHPC [1 , 2-diphytanoyl-sn-glycero-3-phosphocholine] [Avanti Polar Lipids] and 2 . 5% DPHPC:PI ( 4 , 5 ) P2 ) were mixed in chloroform , dried under vacuum , and rehydrated to 23 mM ( 20 mg/ml ) in decane . The indicated proteins were added to the lipid mix at 1:50 vol ratio and pipetted vigorously until cloudy before imaging by spinning disc confocal microscopy . Rabbit muscle actin [5% ( mol/mol ) pyrene-labeled] was gel-filtered , prespun at 90 , 000 xg , exchanged from Ca2+ to Mg2+ , and assembled at a final concentration of 2 . 5 µM as described previously ( Moseley et al . , 2006 ) . Proteins were preincubated with 74 µg/ml liposomes or control buffer for 30 min before actin assembly reactions . Assembly was monitored with a spectrofluorometer ( Photon Technology International ) using an excitation wavelength of 365 nm and an emission wavelength of 407 nm . Rates were calculated from slopes of curves in the linear range , and curves were plotted using GraphPad Prism software . Graphs were prepared and statistical analyses performed using Graphpad Prism software . For normally distributed data , comparisons were made using either t-test or analysis of variance with posthoc Bonferroni’s multiple comparisons test . For non-normally distributed data , comparisons were made using either Mann-Whitney U test or Kruskal-Wallis test with posthoc Dunn’s test . No specific power analyses were performed; sample sizes were chosen based on established protocols and statistical analyses for significance , as detailed for all experiments here and in Supplementary file 1 . Comparison of patch-duration distributions was performed using a Kolmogorov-Smirnoff test . Please see Supplementary file 1 for each statistical test performed for each experiment presented in this study . All data are shown as the mean ± sem . Statistical significance denoted in all graphs p<0 . 05 , p<0 . 01 , **p<0 . 001 .	Neurons constantly talk to each other by sending chemical signals across the tiny gap , or ‘synapse’ , that separates two cells . While inside the emitting cell , these molecules are safely packaged into small , membrane-bound vessels . Upon the right signal , the vesicles fuse with the external membrane of the neuron and spill their contents outside , for the receiving cell to take up and decode . The emitting cell must then replenish its vesicle supply at the synapse through a recycling mechanism known as endocytosis . To do so , it uses dynamically assembling rod-like ‘actin’ filaments , which work in concert with many other proteins to pull in patches of membrane as new vesicles . The proteins that control endocytosis and actin assembly abound at neuronal synapses , and , when mutated , are linked to many neurological diseases . Unlike other cell types , neurons appear to ‘pre-deploy’ these actin-assembly proteins to synaptic membranes , but to keep them inactive under normal conditions . How neurons control the way this machinery is recruited and activated remains unknown . To investigate this question , Del Signore et al . conducted two sets of studies . First , they exposed actin to several different purified proteins in initial ‘test tube’ experiments . This revealed that , depending on the conditions , a group of endocytosis proteins could prevent or promote actin assembly: assembly occurred only if the proteins were associated with membranes . Next , Del Signore et al . mutated these proteins in fruit fly larvae , and performed live cell microscopy to determine their impact on actin assembly and endocytosis . Consistent with the test tube findings , endocytosis mutants had more actin assembly overall , implying that the proteins were required to prevent random actin assembly . However , the same mutants had reduced levels of endocytosis , suggesting that the proteins were also necessary for productive actin assembly . Together , these experiments suggest that , much like a mousetrap holds itself poised ready to spring , some endocytic proteins play a dual role to restrain actin assembly when and where it is not needed , and to promote it at sites of endocytosis . These results shed new light on how neurons might build and maintain effective , working synapses . Del Signore et al . hope that this knowledge may help to better understand and combat neurological diseases , such as Alzheimer’s , which are linked to impaired membrane traffic and cell signalling .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2021	An autoinhibitory clamp of actin assembly constrains and directs synaptic endocytosis
T cell self-tolerance is thought to involve peripheral tolerance and negative selection , involving apoptosis of autoreactive thymocytes . However , evidence supporting an essential role for negative selection is limited . Loss of Bim , a Bcl-2 BH3-only protein essential for thymocyte apoptosis , rarely results in autoimmunity on the C57BL/6 background . Mice with T cell-specific over-expression of Bcl-2 , that blocks multiple BH3-only proteins , are also largely normal . The nuclear receptor Nur77 , also implicated in negative selection , might function redundantly to promote apoptosis by associating with Bcl-2 and exposing its potentially pro-apoptotic BH3 domain . Here , we report that T cell-specific expression of a Bcl2 BH3 mutant transgene results in enhanced rescue of thymocytes from negative selection . Concomitantly , Treg development is increased . However , aged BH3 mutant mice progressively accumulate activated , autoreactive T cells , culminating in development of multi-organ autoimmunity and lethality . These data provide strong evidence that negative selection is crucial for establishing T cell tolerance . Maintenance of T cell self-tolerance is essential for the prevention of autoimmune diseases . A series of mechanisms are thought to be involved in establishing T cell tolerance , the first of which is deletion of potentially self-reactive T cells in the thymus . During T cell development , autoreactive T cells bind self-peptide presented on major histocompatibility complex ( MHC ) molecules with high affinity and receive strong T cell receptor ( TCR ) signals that trigger apoptosis and clonal deletion . Some of these cells can also differentiate down an alternative pathway to a harmless lineage , such as T regulatory cells ( Treg ) , or also CD8αα and invariant NK T cells ( Stritesky et al . , 2012 ) . The importance of this process , termed negative selection , in establishing self-tolerance and preventing autoimmunity was best demonstrated in studies of mice deficient in the autoimmune regulator ( Aire ) gene . These mice develop a multi-organ autoimmune disease as the result of defective deletion of auto-reactive T cells specific to tissue-restricted antigens ( TRA ) , which would normally be expressed by medullary thymic epithelial cells under the control of AIRE ( Anderson et al . , 2002 ) . Humans with loss of AIRE function develop similar disease pathology ( Finnish-German APECED Consortium . , 1997; Nagamine et al . , 1997 ) . However , additional models that implicate a breakdown in negative selection as the driving force behind the development of autoimmune disease are lacking ( von Boehmer and Melchers , 2010 ) . This might suggest that under most circumstances , subsequent tolerizing mechanisms in the peripheral compartments such as anergy or Treg-mediated suppression are sufficient to compensate for defective thymic deletion . Even in the case of AIRE-deficiency , impaired tolerization of CD4+ T cells in the periphery by extra-thymic AIRE expressing cells likely contributes to the development of autoimmunity ( Gardner et al . , 2013 ) . This has cast doubt on the relative importance of negative selection for maintaining immunological self-tolerance . Apoptosis of thymocytes during negative selection is dependent on the mitochondrial apoptosis pathway regulated by the Bcl-2 family of proteins ( Tischner et al . , 2010 ) . In particular , the pro-apoptotic , Bcl-2 homology domain 3 ( BH3 ) only protein , Bim , which is upregulated in response to strong TCR signals , is essential for this process . Bim and other BH3-only proteins are thought to promote apoptosis by binding and inhibiting anti-apoptotic Bcl-2 proteins ( such as Bcl-2 , Bcl-XL and Mcl-1 ) that directly suppress downstream effectors Bax and Bak , which are responsible for disrupting mitochondrial membrane integrity ( Youle and Strasser , 2008 ) . Bim-deficiency has been shown to cause defective thymocyte apoptosis in several TCR transgenic ( HY , OTI , OTII ) models of negative selection ( Bouillet et al . , 2002; Moran et al . , 2011; Zhan et al . , 2011; Suen and Baldwin , 2012 ) . However , this apoptotic defect does not lead to emergence of autoreactive mature T cells except in the case of negative selection to TRA ( Hu et al . , 2009; Kovalovsky et al . , 2010; Suen and Baldwin , 2012 ) . Interestingly , Bim−/− mice on a mixed 129/Sv X C57BL/6 background develop a systemic lupus erythematosus ( SLE ) -like autoimmune disease , indicative of a breakdown in tolerance ( Bouillet et al . , 1999 ) . However , loss of Bim in other immune compartments likely contributes to the disease , especially B cells , which are important mediators of SLE pathology . This phenotype was greatly ameliorated on the autoimmune-resistant C57BL/6 background and is distinct from the multi-organ , T cell dominant disease found in AIRE-deficient mice ( Bouillet et al . , 2001; Labi et al . , 2014 ) . Recently , a study showed that additional loss of Puma ( Bbc3 ) , another BH3-only protein , enhanced the Bim−/− thymic deletion defect and led to the development of immune pathology more similar to that found in AIRE-deficient mice ( Gray et al . , 2012 ) . T cells from these mice were able to transfer the disease , lending support to it being T cell-driven . However , the role of non-T cells in these mice cannot be excluded due to germline deletion of Puma and Bim . In addition , T-cell specific over-expression of Bcl-2 , which can inhibit both Bim and Puma ( Chen et al . , 2005 ) , does not lead to autoimmunity ( Sentman et al . , 1991; Linette et al . , 1995 ) , suggesting that a defect in T cells alone may be insufficient to cause disease . Alternatively , redundant negative selection pathways involving Bim , Puma and the Nur77 family members ( See below ) may not allow Bcl-2 over-expression to block all pathways leading to negative selection . In addition to Bim , the Nur77 family of orphan steroid receptors , which includes Nur77 , Nor-1 and Nurr1 , has also been implicated in apoptosis accompanying negative selection . Nur77 expression , like that of Bim , is induced by strong TCR signals that result in negative selection . Low expression of both proteins has been correlated with defective clonal deletion in Non-obese Diabetic ( NOD ) mice ( Sohn et al . , 2003; Liston et al . , 2004 ) . Furthermore , T cell-specific over-expression of Nur77 , or Nor-1 , results in massive apoptosis of thymocytes ( Cheng et al . , 1997b ) . Expression of a dominant negative Nur77 protein that can block all family members results in inhibition of apoptosis in the F5 and HY TCR transgenic models of negative selection ( Calnan et al . , 1995; Zhou et al . , 1996 ) . Deficiency in Nur77 alone ( Nr4a1−/− ) results in moderate inhibition of apoptosis accompanying negative selection in the OTII and BDC2 . 5 TCR transgenic models ( Fassett et al . , 2012 ) . Interestingly , the Nur77 family has also been shown to transcriptionally regulate Foxp3 expression and deletion of all three Nur77 family members ( Nr4a3−/− X Cd4-Cre+Nr4a1fl/flNr4a2fl/fl ) results in impaired Treg cell production and development of a quick onset , severe autoimmune disease resembling that of Foxp3-deficient mice ( Sekiya et al . , 2013 ) . Therefore , Nur77 could contribute to the maintenance of immunological tolerance both by facilitating deletion of autoreactive thymocytes and ensuring production of Treg cells . The mechanism by which Nur77 and its family members promote thymocyte apoptosis has long been under investigation . We and others showed that strong TCR signals induce the translocation of Nur77 and Nor-1 from the nucleus to the mitochondria in thymocytes , a phenomenon first reported to occur in cancer cells in response to apoptotic stimuli ( Lin et al . , 2004; Thompson and Winoto , 2008; Wang et al . , 2009 ) . At the mitochondria , Nur77 and Nor-1 bind Bcl-2 and cause exposure of Bcl-2's BH3 domain , the ‘death’ domain of the Bcl-2 family of proteins . Bcl-2 BH3 exposure correlates with thymocyte apoptosis in TCR-stimulated thymocytes and HY and F5 TCR transgenic thymocytes undergoing negative selection ( Thompson and Winoto , 2008 ) . BH3-exposed Bcl-2 might act similarly to BH3-only protein Bim to bind and inhibit other anti-apoptotic Bcl-2 proteins , such as Bcl-XL , as was demonstrated in cancer cells ( Kolluri et al . , 2008 ) . It is thus possible that Nur77 converts Bcl-2 to a pro-apoptotic effector during negative selection , preventing over-expression of Bcl-2 from completely blocking negative selection . Here , we generated a T cell-specific Bcl2 BH3 mutant transgenic mouse , in which Bcl-2's purported pro-apoptotic BH3 function should be abolished by changing its conserved BH3 residues , GDD , to alanines ( Cheng et al . , 1997a ) . We found that over-expression of both wild-type and BH3 mutant Bcl-2 efficiently rescued thymocyte apoptosis in two TCR transgenic models of negative selection . However , the BH3 mutant transgene more effectively blocked TCR-induced thymocyte apoptosis in vitro and better rescued high affinity TCR clones from deletion in polyclonal systems in vivo . Interestingly , BH3 mutant transgenic mice , in contrast to the reported wild-type Bcl2 transgenic mouse phenotype , developed multi-organ autoimmune pathology and died around one year-of-age . Thus , we provide strong evidence that a breakdown in thymocyte apoptosis during negative selection is indeed sufficient to cause autoimmune disease . To investigate the role of the Bcl-2 BH3 domain in thymocyte apoptosis , we created T cell-specific Bcl2 BH3 mutant transgenic mice . Three amino acid residues critical for BH3 domain pro-apoptotic function were mutated to alanine ( Figure 1A ) and the BH3 mutated human Bcl2 transgene ( referred to as BH3 ) was expressed under the control of the Cd4 regulatory elements ( Adlam and Siu , 2003; Xue et al . , 2010 ) . The ‘BH3’ transgenic mice ( BH3 Tg ) were generated on the C57BL/6 background and two founder lines were chosen for analysis , BH3 A and BH3 B . For evaluation , the BH3 Tg mice were compared to the T cell-specific wild-type human Bcl-2 strain , LckPr-Bcl2 , also on the C57BL/6 background , referred to here as ‘Bcl-2 Tg’ ( Sentman et al . , 1991 ) . 10 . 7554/eLife . 03468 . 003Figure 1 . Generation of Bcl2 BH3 domain mutant transgenic mice . ( A ) Diagram of the mutation made to the BH3 domain of human Bcl2 to abolish BH3 domain function in the Bcl2 BH3 mutant transgenic ( BH3 Tg ) mice . Amino acids 101–103 of human Bcl2 were mutated from Glycine-Aspartate-Aspartate to Alanine-Alanine-Alanine to abolish BH3 domain function ( Cheng et al . , 1997a ) . ( B ) Flow cytometric analysis of intracellular Bcl-2 expression in BH3 Tg , line A ( BH3 A Tg ) vs LckPr-Bcl2 ( Bcl-2 Tg ) and wild-type ( WT ) thymocyte and mature T cell populations . DN populations were gated based on CD25 and CD44 expression . ( C ) Western blot analysis comparing human vs mouse Bcl-2 expression in the thymus , lymph nodes and spleen of BH3 A Tg vs Bcl-2 Tg and WT mice . ( D ) Flow cytometric analysis of CD4 vs CD8 T cell populations in the thymus and lymph nodes . ( E ) DP , CD4 SP and CD8 SP thymocyte cell numbers in 6-week-old BH3 Tg and Bcl-2 Tg mice compared to WT littermate controls . ( F ) Lymph node mature CD4 and CD8 T cell numbers in 6-week-old BH3 Tg and Bcl-2 Tg mice compared to WT littermate controls . The transgenic mice in ( B–D ) were age-matched within 1 week and compared to a littermate non-transgenic WT control . All mice were between 6 and 10-weeks-old . ( B–F ) are representative of or compiled from at least three independent experiments . Statistics here and in the following figures were calculated by Student's t-test: *p < 0 . 001 , p < 0 . 01 , *p < 0 . 05 , n . s . not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 003 By intracellular staining with an antibody that detects both mouse and human Bcl-2 , we observed that Bcl-2 expression was significantly increased in BH3 Tg thymocytes compared to wild-type controls ( Figure 1B ) . Although the transgene is expressed by the Cd4 promoter , increased Bcl-2 expression was detected in DN ( CD4− CD8− double negative ) thymocytes as early as CD25+ CD44− DN3 ( Figure 1B ) . This was very similar to the expression pattern of the Lck-driven wild-type Bcl2 transgene . Additionally , the level of Bcl-2 upregulation in BH3 Tg vs Bcl-2 Tg thymocytes was very similar in DN , DP and CD4 single positive ( CD4 SP ) thymocytes . Only slightly higher expression of the BH3 transgene was noted in DP and CD4 SP cells . As expected , BH3 transgene expression was downregulated in CD8 single positive ( CD8 SP ) thymocytes due to loss of CD4 expression ( Figure 1B ) . Lymph node CD8 T cells further downregulated BH3 transgene expression , while both BH3 Tg and Bcl-2 Tg CD4 T cells maintained similarly high levels ( Figure 1B ) . Interestingly , we observed a compensatory downregulation of endogenous mouse Bcl-2 in the thymus , lymph nodes and spleen of both BH3 and Bcl-2 Tg mice ( Figure 1C ) . Thus , we do not have to consider a contribution of endogenous mouse Bcl-2 to the phenotype of these mice . Due to close similarity in transgene expression level and pattern in the two BH3 Tg lines , BH3 A and BH3 B , we used these mice interchangeably throughout our study ( Data not shown ) . In both BH3 Tg lines , we observed a decreased proportion of DP thymocytes and an increased proportion of SP thymocytes compared to wild-type littermate controls ( Figure 1D ) . However , only the absolute SP cell numbers were increased , especially for CD4 SP cells , while the DP cell number was comparable to wild-type controls ( Figure 1E ) . This is largely similar to the thymocyte distribution found in the wild-type Bcl-2 Tg mice ( Sentman et al . , 1991 ) . As expected , the peripheral mature T cell compartment reflects the increased SP thymocyte numbers in both the BH3 Tg and Bcl-2 Tg mice ( Figure 1F ) . Relative to wild-type mice , the CD4 to CD8 T cell ratio is therefore somewhat skewed in favor of CD4 T cells in the BH3 Tg mice and CD8 T cells in the Bcl-2 Tg mice . Hence , differential expression in the CD8 compartment is one caveat to comparison of BH3 and Bcl-2 transgenic mice . However , similar expression levels in the CD4 compartment should allow fair comparison of DP , CD4 SP and CD4 T cell phenotypes . A defect in thymocyte apoptosis accompanying negative selection could potentially contribute to increased SP thymocyte and mature T cell numbers in the BH3 and Bcl-2 Tg mice . However , while Bcl2 transgene expression was previously shown to safeguard thymocytes against a variety of apoptotic stimuli , thymocytes were inefficiently protected from apoptosis during negative selection ( Sentman et al . , 1991; Strasser et al . , 1991; Bouillet et al . , 2002 ) . To test whether BH3 Tg thymocytes might be better protected , we first assessed thymocyte apoptosis in response to strong TCR stimuli that mimic negative selection in vitro . In response to anti-CD3 and anti-CD28 antibody stimulation , BH3 transgene expression substantially decreased apoptosis of thymocytes compared to wild-type controls ( Figure 2A ) . In addition , BH3 Tg thymocytes showed significantly increased resistance to apoptosis compared to Bcl-2 Tg thymocytes . 10 . 7554/eLife . 03468 . 004Figure 2 . BH3 transgene expression efficiently blocks TCR-mediated thymocyte apoptosis in vitro and in F5 and HY TCR transgenic models of negative selection . ( A ) Thymocytes were stimulated for 18 hr with anti-CD3 and anti-CD28 ( 20 μg/ml ) and analyzed for the percentage of Annexin V+ cells by flow cytometry . The percentage of Annexin V+ thymocytes in samples left untreated for 18 hr ( i . e . background ) was subtracted . ( B ) Total live thymocyte number in thymii collected 48 hr after injection with the F5 TCR-specific peptide , NP , or PBS . ( C ) CD4 vs CD8 flow cytometric analysis of F5 TCR+ ( vβ11+ ) thymocytes from mice treated as in ( B ) . Data in ( B and C ) are representative of or pooled from at least five independent experiments per transgenic genotype . Mice were between 6 and 9-weeks-old and F5 control mice were littermates to BH3/F5 and Bcl-2/F5 mice . ( D ) Percentage of Cleaved Caspase 3+ thymocytes by flow cytometric analysis treated with 100 ng/ml NP peptide for the indicated times . The percentage of Cleaved Caspase 3+ thymocytes treated with an irrelevant peptide ( i . e . background ) was subtracted at each time point . Efficient activation of thymocytes by the NP peptide was assessed by flow cytometric analysis of CD69 surface expression . Data are representative of three independent experiments . ( E ) Percentage of Cleaved Caspase 3+ thymocytes by flow cytometric analysis treated with 100 ng/ml of the HY TCR-specific peptide , Smcy , for the indicated times . Background was subtracted as in ( D ) . Proper thymocyte activation was confirmed by analysis of CD69 expression . Data are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 004 To test whether the BH3 transgene is also better able to block apoptosis during negative selection in vivo , we examined apoptosis of F5 TCR transgenic thymocytes in BH3 Tg mice crossed to F5 mice ( BH3/F5 ) ( Mamalaki et al . , 1992 , 1993 ) . We also examined Bcl-2/F5 mice for comparison . In response to injection with the F5 TCR-specific peptide , NP , F5 DP thymocytes undergo massive apoptosis ( Mamalaki et al . , 1992 ) . Strikingly , however , expression of either the BH3 or Bcl2 transgene almost completely inhibited thymocyte death , with thymocyte numbers only marginally decreased compared to PBS-injected control mice after 48 hr of treatment ( Figure 2B ) . Analysis of thymocyte proportions indicated a significant rescue of the DP thymocyte population ( Figure 2C ) . We sought to increase the sensitivity of the assay by using a previously described ex vivo thymic slice approach ( Dzhagalov et al . , 2013 ) . This assay allows assessment of apoptosis over a more acute timeline , free of possible cytotoxic effects inherent in the NP injection model ( Mamalaki et al . , 1992 ) . In this assay , transgenic thymic lobes were embedded in agarose , cut into 500 μm slices by vibratome and cultured in transwells . The slices were then briefly exposed to the relevant or an irrelevant control peptide and thymocyte apoptosis was assessed by flow cytometry at 12-hr intervals . Here , again , we saw near complete and equivalent protection of BH3 and Bcl-2 Tg thymocytes from apoptosis in response to NP peptide treatment over a 36-hr time course ( Figure 2D ) . Thymocytes of all genotypes were similarly activated as measured by upregulation of CD69 ( Figure 2D ) . We also performed an analogous thymic slice assay comparing apoptosis of thymocytes from female BH3 or Bcl-2 Tg mice crossed to HY TCR transgenic mice ( BH3/HY or Bcl-2/HY , respectively ) in response to the HY TCR-specific peptide , Smcy . Remarkably , just as in the F5 model , thymocytes expressing either transgene were significantly protected from apoptosis ( Figure 2E ) . Hence , we conclude that Bcl-2 over-expression very efficiently inhibits thymocyte apoptosis accompanying negative selection in these models and this potent blockade could not be enhanced by expression of a BH3 mutant form of Bcl-2 . We next sought to test whether over-expression of Bcl-2 is sufficient to block negative selection in a normal , polyclonal system and if in this context , the BH3 transgene might exhibit enhanced anti-apoptotic activity . To investigate this , we employed the Nur77GFP BAC transgenic reporter mouse system , in which GFP is inserted into the start site of the Nr4a1 ( Nur77 ) transgene ( Moran et al . , 2011 ) . Studies of these mice have clearly demonstrated that the level of Nur77 expression in T cells directly correlates with the intensity of the TCR signals received ( Moran et al . , 2011 ) . During thymocyte selection , cells receiving strong TCR signals are destined to die by negative selection while cells receiving very weak to no signal die by neglect . Hence , the post-selection thymocyte pool is largely comprised of cells expressing an intermediate level of GFP , reflecting their receipt of moderate , positively selecting TCR signals . In Bim−/− X Nur77GFP mice , however , a significant number of post-selection thymocytes express high levels of GFP , indicating that high affinity TCR clones are rescued from negative selection in the absence of Bim ( Stritesky et al . , 2013 ) . This rescue was also reflected by an increase in the total number of post-selection DP and SP thymocytes . When we assessed GFP expression of post-selection thymocytes in BH3/Nur77GFP and Bcl-2/Nur77GFP mice , we observed a strikingly similar phenotype . A large population of GFP-high post-selection DP thymocytes was rescued with the expression of either transgene ( Figure 3A ) . As in Bim−/−/Nur77GFP mice , the proportion of post-selection DP thymocytes was consequently increased compared to wild-type mice , reflecting the rescue of these cells from deletion ( Figure 3A ) . 10 . 7554/eLife . 03468 . 005Figure 3 . BH3 transgene expression rescues high affinity TCR clones in a polyclonal system . ( A ) Flow cytometric analysis of GFP expression in pre-selection ( CD69− TCRβ− ) vs post-selection ( CD69+ TCRβ+ ) DP thymocytes from Nur77GFP mice expressing the BH3 and Bcl2 transgenes . The WT/Nur77GFP post-selection DP histogram was used to set the GFP low ( lo ) , intermediate ( int ) , and high ( hi ) gates . The left end of the ‘int’ gate was set at the base of the GFP− Ctrl histogram and right end was set at the mean GFP fluorescence intensity of the WT/Nur77GFP histogram . The ‘lo’ and ‘hi’ gates were extended from the left and right ends of the ‘int’ gate , respectively . ( B ) Percentage of DP thymocytes that were post-selection and GFP lo , int and hi as defined in ( A ) . ( C and D ) Flow cytometric analysis of GFP expression in CD4 SP ( C ) and CD8 SP ( D ) with the gates from post-selection DP cells in A applied . The GFP lo , int and hi percentages were quantified . All mice were 6-weeks-old and WT/Nur77GFP and GFP− Ctrl mice were littermates to the BH3/Nur77GFP and Bcl-2/Nur77GFP mice . Data were pooled from 10 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 00510 . 7554/eLife . 03468 . 006Figure 3—figure supplement 1 . Nur77 GFP expression in SP thymocytes is regulated by TCR signaling . Flow cytometric analysis of Nur77 GFP expression over time in CD4 SP and CD8 SP thymocytes removed from MHC by culturing . Histograms are representative of duplicate samples and two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 006 To quantify this effect , we defined GFP-high ( GFPhi ) cells as those expressing GFP at a level greater than the mean fluorescence intensity ( M . F . I . ) of wild-type post-selection DP cells; this definition allowed for the most consistent gating across independent experiments ( Figure 3A ) . The percent increase in GFPhi cells over wild-type is indicative of the proportion of cells rescued from negative selection by transgene expression . Interestingly , a significantly greater increase in the percentage of GFPhi post-selection DP cells was observed in BH3/Nur77GFP vs Bcl-2/Nur77GFP mice , suggesting that the BH3 transgene might indeed have an enhanced ability over wild-type Bcl2 to rescue high affinity TCR clones from apoptosis ( Figure 3B ) . No increase was observed in GFP-intermediate ( GFPint ) cells , indicating that the GFPhi population encompasses all rescued cells and that BH3 and Bcl2 transgene expression , similar to Bim-deficiency , only rescues high affinity clones and therefore does not seem to enhance positive selection ( Figure 3A , B; Stritesky et al . , 2013 ) . BH3 and Bcl2 transgene expression also rescued GFPhi CD4 and CD8 SP thymocytes from apoptosis . This likely indicates rescue of high affinity clones from deletion in response to self-antigens presented in the medulla and possibly rescue of DP cells that have differentiated into SP thymocytes . Confirming what has been shown in Nur77GFP mice ( Moran et al . , 2011 ) , GFP expression in BH3 and Bcl-2 Nur77GFP double transgenic mice was also TCR-dependent , as BH3/Nur77GFP and Bcl-2/Nur77GFP SP thymocytes removed from MHC in culture down-regulated GFP similarly to WT/Nur77GFP cells ( Figure 3—figure supplement 1 ) . A greater percentage of GFPhi CD4 SP cells were rescued in BH3 Tg vs Bcl-2 Tg mice ( Figure 3C ) . Consistent with the BH3 transgene being downregulated in CD8 SP cells , BH3 expression rescued a smaller percentage of GFPhi CD8 SP thymocytes compared to CD4 SP ( Figure 3D vs 3C ) . Collectively , these data demonstrate that Bcl-2 over-expression causes a significant defect in thymocyte negative selection even in a normal , polyclonal system and in this context , the BH3 mutant protein rescues significantly more cells from death than wild-type Bcl-2 . Endogenous superantigen negative selection is another polyclonal model of clonal deletion whereby specific Vβ TCR-expressing T cells are deleted through interaction of their specific Vβ chain with endogenously-expressed superantigen bound to class II MHC ( usually I–E ) on antigen presenting cells ( Scherer et al . , 1993 ) . Whether Bcl-2 over-expression can inhibit negative selection against endogenous superantigens is controversial . In one study , Bcl2 transgene expression did not block superantigen negative selection ( Sentman et al . , 1991 ) while another group clearly saw a rescue ( Siegel et al . , 1992 ) . Yet another study reported that cells were rescued from superantigen-mediated deletion in the Bcl2 transgenic thymus , but not in the transgenic lymph nodes ( Strasser et al . , 1991 ) . To assess endogenous superantigen negative selection in BH3 Tg mice , we crossed them to the class II MHC I-E expressing CBA/J strain of mice . CBA/J mice express superantigens that delete T cells expressing Vβ3 , 5 , 6 , 7 , 8 . 1 , 9 and 11 , but not Vβ8 . 2 ( Scherer et al . , 1993 ) . We examined the proportions of Vβ5 , Vβ6 , Vβ8 . 1/8 . 2 and Vβ11 expressing DP and SP thymocytes in BH3 Tg X CBA/J mice and their wild-type littermate controls as well as in Bcl-2 Tg X CBA/J mice . As shown in Figure 4 , the proportions of Vβ5 , Vβ6 and Vβ11 , but not Vβ8 . 1/8 . 2 , expressing cells decreased between the DP and SP stage in non-transgenic mice , indicative of superantigen negative selection . In both BH3 and Bcl-2 Tg mice , however , increased proportions of Vβ5 , Vβ6 and Vβ11 expressing SP thymocytes were observed , suggesting that expression of either transgene can block superantigen-mediated deletion . The BH3 transgene rescued CD8 SP thymocytes less efficiently than the wild-type Bcl2 transgene , reflective of the relative levels of transgenic protein in the CD8 compartment . CD4 SP thymocytes were rescued to a similar extent by both transgenes , however modestly better by the BH3 transgene for Vβ5+ and Vβ6+ cells . Interestingly , we observed a significant increase in Vβ5+ and Vβ6+ ( but not Vβ11+ ) DP thymocytes in the BH3 Tg mice that was not seen in the Bcl-2 Tg mice . The pattern of rescue in this model parallels that observed in the Nur77GFP polyclonal system . Together , these data are consistent with the notion that overexpression of both wild type and BH3 mutant Bcl2 transgenes leads to an enhanced rescue of thymocytes from negative selection , with the BH3 mutant providing a more complete rescue . 10 . 7554/eLife . 03468 . 007Figure 4 . Superantigen negative selection is inhibited in BH3 Tg mice . Flow cytometric analysis of TCR vβ chain expression in DP , CD4 SP and CD8 SP thymocyte subsets from BH3 Tg and Bcl-2 Tg mice ( closed shapes , black ) compared to littermate wild-type controls ( open shapes , grey ) on a C57BL/6 X CBA/J genetic background . All mice were 8-weeks-old . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 007 Our results demonstrate that T cell-specific over-expression of wild-type Bcl-2 clearly results in a rescue of cells receiving strong TCR signals , which would normally undergo apoptosis , yet aged Bcl-2 Tg mice were not reported to develop autoimmune pathology ( Linette et al . , 1995 ) . Hence , we sought to investigate the fate of the rescued , high affinity T cell clones in the Bcl-2 Tg and BH3 Tg mice . We observed that the GFP mean fluorescence intensity was higher in lymph node T cells from both BH3/Nur77GFP and Bcl-2/Nur77GFP mice ( Figure 5A ) . GFP expression was also dependent on TCR signaling , as removal of BH3/Nur77GFP and Bcl-2/Nur77GFP mature T cells from MHC in culture resulted in downregulation of GFP at a rate similar to that observed for WT/Nur77GFP cells ( Figure 5—figure supplement 1 ) . Hence , the increased GFP expression might be leftover from strong TCR signals received in the thymus and/or due to tonic TCR stimulation in the periphery . Interestingly , GFP levels were higher in Bcl-2 Tg vs BH3 Tg T cells , even for CD4+ T cells . The reason for this is unclear , but it may be due to distinct T cell repertoires in these two lines of mice that lead to differences in T cell signaling in the periphery . Treg cells have self-reactive TCRs and express a high level of GFP , however Treg cells did not account for the majority of the GFP-high CD4 T cells ( data not shown ) . This indicates that conventional T cells with higher than normal TCR affinity are permitted to migrate into the periphery . Since presumably autoreactive T cells are found in the periphery , mechanisms must exist to keep these cells in check . One mechanism might be the production of more Treg cells to mediate suppression of the autoreactive cells ( Stritesky et al . , 2012 ) . Interestingly , we observed that GFP-high CD4 SP thymocytes rescued from negative selection express GFP at a very similar level to Treg cells ( Figure 5B ) . Since high TCR affinity contributes to Treg development ( Stritesky et al . , 2012 ) , we hypothesized that more thymocytes might be directed down this pathway in BH3 and Bcl-2 Tg mice . Indeed , we found greater Foxp3+ Treg cell proportions and numbers in the both the thymus and lymph nodes of BH3 and Bcl-2 Tg mice ( Figure 5C ) . BH3 Tg mice had a more significant increase in peripheral Treg numbers compared to Bcl-2 Tg mice ( 3-fold vs 2-fold ) , mirroring the higher mature CD4 T cell numbers in these mice . Most of the increase in total Treg number was caused by an increase in CD25− , and not CD25+ cells . CD25− Treg cells are thought to be derived from diverted , potentially autoreactive thymocytes and are functional , but have reduced suppressive capacity compared to CD25+ Treg cells ( Zhan et al . , 2011 ) . Despite the increase in total Treg number in BH3 and Bcl-2 Tg mice , the ratio of CD25+ Treg cells to total CD4 T cells was similar to wild-type mice ( Figure 5C ) . Thus , some of the rescued high affinity clones appear to be diverted down the CD25− Treg pathway , but not the CD25+ Treg pathway . Intriguingly , consistent with this , we observed that CD25− Treg cells express higher levels of GFP than CD25+ Treg cells , at a similar level to the highest GFP-expressing BH3 Tg clones ( Figure 5D ) . Another mechanism employed to maintain peripheral tolerance is the induction of T cell anergy upon antigen encounter . In addition to increased Treg cells , we also observed a higher proportion of previously activated , anergic CD4 T cells in BH3 and Bcl-2 Tg mice marked by upregulation of the markers FR4 and CD73 in CD44hi T cells ( Figure 5E–F; Martinez et al . , 2012 ) . Therefore , increased diversion to the Treg lineage/Treg suppression and CD4 T cell anergy might cooperate to suppress the emergence of autoimmunity in these mice . 10 . 7554/eLife . 03468 . 008Figure 5 . Alternative tolerance mechanisms in BH3 Tg mice . ( A ) Flow cytometric analysis of GFP expression by mature CD4 and CD8 lymph node T cells . Quantification indicates the fold change in GFP mean fluorescence intensity ( M . F . I . ) relative to WT/Nur77GFP CD4 and CD8 samples . GFP M . F . I . was normalized between experiments by subtracting the M . F . I . of a GFP– control . n ≥ 10 per genotype . ( B ) Comparison of GFP expression by total CD4 SP thymocytes vs Foxp3+ CD4 SP Treg cells . ( C ) Quantification of flow cytometric analysis of CD25+ vs CD25− Treg cells ( CD4+ Foxp3+ ) in the thymus and lymph nodes . n ≥ 7 per genotype . ( D ) Comparison of GFP expression levels in CD25+ vs CD25− Treg cells from BH3 Tg mice . ( E ) Flow cytometric analysis of non-Treg , anergic CD4 T cells in the lymph nodes . Treg cells were gated out by high expression of GITR and CD25 . Anergic cells , indicated by the bolded gates , express high levels of CD44 , FR4 and CD73 . ( F ) Quantification of anergic CD4 T cells gated as described in ( E ) . Mice were 6-weeks-old in ( A–D ) and 7-weeks-old in ( E–F ) . All data are representative of or compiled from at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 00810 . 7554/eLife . 03468 . 009Figure 5—figure supplement 1 . Nur77 GFP expression in mature T cells is maintained by TCR signaling . Flow cytometric analysis of Nur77 GFP expression over time in mature T cells removed from MHC by culturing . Histograms are representative of duplicate samples and two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 009 Despite similarly engaging alternative mechanisms to promote tolerance , BH3 Tg mice , unlike Bcl-2 Tg mice , have a considerably shortened life-span , with a median length of survival of about 50 weeks for both BH3 A and BH3 B lines ( Figure 6A ) . Most mice were sacrificed due to general deterioration of health , especially dramatic weight-loss , but some mice of the BH3 A line were sacrificed due to a severe inflammatory skin condition . No Bcl-2 Tg mice had to be sacrificed over a 60-week period . Upon examination of a large cohort of mice between 45 and 60 weeks-of-age , we found that all BH3 Tg mice displayed severe splenomegaly and lymphadenopathy and had significantly increased numbers of both T cells and B cells , especially in the mesenteric lymph nodes ( MLN ) and spleen ( Figure 6B , C ) . In comparison , Bcl-2 Tg lymphoid organs were only mildly enlarged , but a significant increase in T cell numbers was observed . We also observed a dramatic increase in CD3- B220- Myeloid/other cells in the MLN and spleen of BH3 Tg , but not Bcl-2 Tg mice ( Figure 6D ) . A large proportion of these cells were Ly6Clo CD11b+ neutrophils , indicative of increased inflammation in these mice ( Figure 6E ) . 10 . 7554/eLife . 03468 . 010Figure 6 . BH3 Tg mice have a shortened lifespan and exhibit lymphoid hyperplasia . ( A ) Kaplan-Meier curves for BH3 Tg lines A and B and Bcl-2 Tg mice compared to littermate WT controls . Statistical significance was calculated using the Log-rank test . ( B–D ) Quantification of CD3+ T cells ( B ) , B220+ B cells ( C ) and CD3− B220− Myeloid/other cells ( D ) in the peripheral lymph nodes ( PLN ) , mesenteric lymph nodes ( MLN ) and spleen of 45 to 60-week-old mice . WT controls were littermates to BH3 Tg and Bcl-2 Tg mice . ( E ) Flow cytometric analysis of CD11b and Ly6C expression on CD3− B220− splenocytes from 50-week-old mice . Ly6Chi CD11b+ inflammatory monocytes ( IM ) and Ly6Clo CD11b+ neutrophils ( Nϕ , also Ly6G+ ) are gated . Plots are representative of at least five mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 010 CD4 and CD8 T cells in aged BH3 Tg mice were highly activated , as indicated by increased CD44 expression and decreased CD62L expression , reminiscent of the phenotype of the disease-promoting T cells in AIRE-deficient mice [Figure 7A; ( Anderson et al . , 2002 ) ] . We also observed a significant increase in CD44hi/CD62Lhi central memory CD8 T cells . An increase in activated T cells was also sometimes observed in Bcl-2 Tg mice , but the phenotype was greatly attenuated . While we did not observe an increase in activated T cells in young mice ( data not shown ) , starting as early as 20 weeks-of-age , there was significant accumulation of activated T cells in the peripheral blood of BH3 Tg mice ( Figure 7B ) . This increase became very dramatic and nearly ubiquitous in 40 to 50-week-old mice . In comparison , only a couple Bcl-2 Tg mice at 40 to 50 weeks were found to have increased activated T cells . To directly test the possible autoreactivity of BH3 Tg T cells , we stimulated T cells with syngeneic or allogeneic irradiated splenocytes and measured T cell proliferation in a mixed lymphocyte reaction ( MLR ) . All T cells were able to proliferate in response to allogeneic stimulators , although CD4 T cells proliferated less than CD8 T cells and the presence of Bcl-2 further reduced proliferation ( Figure 7C ) . This datum is consistent with the known inhibitory effect of Bcl-2 on cell cycle progression ( Linette et al . , 1996; Cheng et al . , 2004 ) . When presented with syngeneic stimulators , little proliferation was observed for all the Tg and wild-type CD4 T cells , even with Treg depletion , most likely due to a combination of the lackluster ability of CD4 T cells to proliferate in this system and the inhibitory effect of Bcl-2 on cell cycle progression . In contrast , CD8+ BH3 Tg T cells exhibited significant proliferation in response to syngeneic stimulators , suggesting that the BH3 Tg mice contain autoreactive T cells ( Figure 7C ) . Despite our inability to detect autoreactive CD4 T cells in BH3 Tg mice by MLR , we found that the activated CD4 T cells in aged mice appear to have increased effector function , as splenic CD4 T cells from 50-week-old mice secrete increased amounts of IFNγ ( Figure 8A ) . Only IFNγ and not IL-17 or IL-4 was increased , suggesting that the activated T cells are Th1 polarized ( Figure 8A ) . Autoimmunity manifested as significant infiltration of lymphocytes into the liver , lung and kidney of BH3 , but not Bcl-2 Tg mice ( Figure 8B ) . In addition , autoantibodies against multiple non-lymphoid tissues ( pancreas , eye , lung , liver , kidney , stomach ) were detected by immunoblot of whole tissue extracts with sera from 40 to 50-week-old BH3 Tg mice ( Figure 8C ) . Interestingly , autoantibody binding was detected for all the BH3 Tg mice examined and most if not all the mice had autoantibodies against multiple tissues . Common banding patterns were observed amongst the BH3 Tg mice within a given tissue suggesting common autoantibody targets . Occasionally , autoantibodies were also found in some Bcl-2 Tg mice , but to a much lesser extent . An occasional band was sometimes observed in one of the wild-type controls , but these bands did not match the patterns observed in the BH3 Tg mice . A few BH3 Tg mice also exhibited a significant increase in anti-nuclear antibodies ( ANA ) ( Figure 8D ) . However , for the majority of the BH3 Tg mice , the levels were not appreciably higher than wild-type or Bcl-2 Tg mice , perhaps consistent with the autoimmune pathology in these mice being more predominantly T cell- rather than B cell-driven . In a few mice with high levels of ANA , significant disruption of kidney morphology was observed , reminiscent of glomerulonephritis ( Figure 8B ) . Aged BH3 Tg mice also had significantly increased numbers of both CD25+ and CD25− Treg cells in their lymphoid organs , making up approximately 50% of all CD4 T cells ( Figure 8E–F ) . This dramatic increase in Treg cells was still apparently insufficient to inhibit autoimmunity in these mice . 10 . 7554/eLife . 03468 . 011Figure 7 . BH3 Tg mice accumulate activated , autoreactive T cells . ( A ) Flow cytometric analysis of activated CD4 and CD8 peripheral lymph node T cells by CD44 and CD62L expression in 50-week-old mice . For CD4 T cells , gates indicate activated ( CD44hi CD62Llo ) and naïve ( CD44lo CD62Lhi ) populations . For CD8 T cells , activated effector memory ( CD44hi CD62Llo ) and activated central memory ( CD44hi CD62Lhi ) vs naïve ( CD44lo CD62Lhi ) populations are gated . ( B ) Compiled analyses of the percentage of activated T cells ( CD44hi CD62Llo ) in the peripheral blood of aged mice . Controls ( Ctrl ) are littermates to the Tg mice as denoted by data point shape . ( C ) Mixed lymphocyte reaction: purified T cells from 20-week-old mice were cultured with allogeneic ( Balb/c ) or syngeneic ( C57BL/6 ) irradiated splenocytes at a 1:1 ratio . Proliferation was measured by CFSE dilution over 4 days . Histograms are representative of triplicate samples and five independent experiments , two with Treg cell-depletion ( shown ) and three without . Gates indicate the percentage of total cells that have undergone at least one round of proliferation . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 01110 . 7554/eLife . 03468 . 012Figure 8 . T cell autoimmune pathology is apparent in BH3 Tg mice . ( A ) Quantification of flow cytometric analysis of T helper cell cytokine expression . Splenic CD4 T cells from 50-week-old mice were stimulated with PMA and Ionomycin for 4 hr in the presence of brefeldin A to allow cytokine accumulation . n ≥ 3 mice per genotype . ( B ) Hematoxylin and Eosin staining for lymphocyte infiltrates in liver , lung and kidney sections from 50-week-old mice . Images were captured at 20x magnification . Data are representative of at least three mice per genotype . ( C ) Sera immunoblots for detection of autoantibodies . Whole tissue extracts were probed with sera from 40 to 50-week-old mice . Each lane corresponds to sera from an individual mouse: WT n = 2 , BH3 Tg n = 7 , Bcl-2 Tg n = 6 . ( D ) Serum anti-nuclear antibody ( ANA ) quantification by ELISA . Dashed line represents two standard deviations above the WT mean . Samples above this line are considered positive for ANA . ( E ) Quantification of CD4+ Foxp3+ Treg cells in the lymphoid organs of aged mice . ( F ) Percentage of CD4 T cells that are Treg cells ( CD25+ vs CD25− ) in the mesenteric lymph nodes of 45 to 60-week-old mice . n ≥ 5 mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 03468 . 012 Bcl-2 family proteins and the Nur77 family of orphan steroid receptors are critical mediators of thymocyte apoptosis during negative selection . However , whether thymocyte deletion is necessary to maintain immunological self-tolerance remains under debate . Here , we find that T cell-specific over-expression of a Bcl2 BH3 mutant transgene results in T cell-driven multi-organ autoimmunity , strongly suggesting that a defect in thymic deletion is indeed sufficient to cause a breakdown in tolerance . Consistent with previous observations , we found that T cell-specific over-expression of wild-type Bcl-2 did not cause significant autoimmune pathology ( Linette et al . , 1995 ) . However , our results show that Bcl-2 Tg mice may develop autoimmunity at low penetrance , as some mice exhibited a modest accumulation of activated T cells and autoantibodies . In contrast , expression of a Bcl2 BH3 mutant transgene significantly increased the severity of the disease and resulted in greater than 70% lethality by one year-of-age . A significant accumulation of activated T cells was observed starting at 20 to 30 weeks-of-age and ultimately resulted in massive lymphoid hyperplasia , inflammation , lymphocyte infiltration into non-lymphoid organs and organ-targeted auto-antibody accumulation . This phenotype of low ANA antibody titers , activated T cells and organ-targeted autoantibodies is suggestive of a primarily T cell-mediated disease . Our data in the Nur77GFP and superantigen mouse models indicate that the Bcl-2 BH3 mutant is better able to rescue autoreactive thymocytes from negative selection than wild-type Bcl-2 . Abolishing the Nur77-dependent pro-apoptotic function of Bcl-2 by BH3 mutation might therefore enhance the anti-apoptotic effects of Bcl-2 over-expression . Alternatively , it is also possible that slightly higher BH3 transgene expression levels in DP and CD4 SP cells could be responsible for this phenotype , although the levels of transgene expression do not necessarily correlate with GFP expression as demonstrated by the higher GFP M . F . I . in CD4+ T cells of Bcl-2 over BH3 Tg mice ( Figure 5A ) . Cause set aside , a greater defect in negative selection could allow a threshold to be crossed in which other mechanisms of maintaining tolerance are no longer sufficient to prevent disease . This would be consistent with the recent study comparing the severity of disease in Bim−/− vs Bim−/−Bbc3−/− mice . Bim−/− mice on the C57BL/6 background were reported to have attenuated autoimmunity and 80% of them still survive at 70 weeks-of-age ( Bouillet et al . , 2001; Labi et al . , 2014 ) . Additional loss of Puma ( Bbc3 ) , however , enhanced the thymic deletion defect of Bim−/− mice and resulted in more severe , T cell-driven multi-organ autoimmune pathology ( Gray et al . , 2012 ) . It remains to be seen if loss of Bim and Puma in T cells alone would result in the same phenotype . Our observations in a T cell-specific model provide strong evidence that a severe thymic clonal deletion defect is sufficient to cause autoimmune pathology . Arguably , a defect in peripheral T cell deletion due to increased T cell survival may also contribute to the BH3 Tg phenotype . Although we cannot completely rule out this possibility , the fact that Bcl-2 Tg mice do not develop autoimmunity would suggest a minimal role for defective peripheral tolerance in this model . Despite the large disparity in autoimmune phenotype , we found that both BH3 and Bcl-2 Tg thymocytes were significantly protected from apoptosis in TCR transgenic models of negative selection . Previously , Bcl-2 over-expression was directly compared with Bim-deficiency in the HY model and Bcl-2 over-expression fell significantly short of Bim-deficiency in rescuing thymocytes from deletion ( Bouillet et al . , 2002 ) . However , in our thymic slice assay of HY TCR-mediated deletion , we find that Bcl-2 over-expression almost completely blocks thymocyte death . Similarly , over-expression of Bcl-2 in F5 TCR transgenic mice dramatically blocked thymocyte deletion by thymic slice assay and in vivo peptide injection . One possible explanation for the discrepancy between our results and previously published observations might be the negative selection models used . Apoptosis of HY TCR transgenic mice is notoriously difficult to block due to accelerated T cell development and early expression of the transgenic TCR ( Baldwin et al . , 2005 ) . Indeed , using thymic organ culture , another group has previously reported a significant inhibition of negative selection by Bcl-2 in the F5 model ( Williams et al . , 1998 ) . Regardless , however , we have demonstrated that Bcl-2 over-expression has a potent anti-apoptotic effect on negative selection in a normal , polyclonal system . Mirroring observations in Bim−/−/Nur77GFP mice ( Stritesky et al . , 2013 ) , we observed a striking increase in GFP-high post-selection thymocytes in Bcl-2/Nur77GFP mice , indicating a significant rescue of high affinity TCR clones from deletion . Here , over-expression of the BH3 mutant transgene resulted in substantially greater rescue of GFP-high DP and CD4 SP thymocytes compared to over-expression of wild-type Bcl2 . BH3 transgene expression also enhanced rescue of vβ5 and vβ6 DP and CD4 SP thymocytes from deletion in a polyclonal superantigen model of negative selection . Increased rescue of these cells in the thymus might lead to the development of a more self-reactive T cell repertoire , accelerating the expansion of activated T cells in BH3 Tg mice . Differential expression of the BH3 vs wild-type Bcl2 transgenes in the CD8 T cell compartment might also contribute to the development of autoimmunity in BH3 Tg , but not Bcl-2 Tg mice . Cd4-driven transgene expression in the BH3 transgenic mice preferentially enhances CD4 T cell seeding of the peripheral lymphoid organs . BH3 Tg mice have about a 3:1 ratio of CD4 to CD8 T cells in the lymph nodes compared to a 2:1 ratio in wild-type mice . In Bcl-2 Tg mice , the proportions are skewed in the opposite direction , resulting in a 1:1 ratio . This is likely due to a combination of transgene-biased rescue of cells from negative selection and increased survival of transgene-expressing T cells . BH3 Tg mice are therefore expected to have increased numbers of autoreactive CD4 T cells , which as key mediators of autoimmunity , could be responsible for the acceleration of disease in these mice ( Palmer and Weaver , 2010 ) . Alternatively , as Bcl-2 over-expression reduces the ability of T cells to proliferate ( Linette et al . , 1996; Cheng et al . , 2004 ) , downregulation of the BH3 transgene in CD8 T cells might enhance their functional capacity over that of Bcl-2 Tg CD8 T cells . As CD8 T cells play a well-documented role in T cell-mediated , tissue-specific autoimmune diseases ( i . e . multiple sclerosis and diabetes ) , more functionally competent autoreactive CD8 T cells in the BH3 Tg mice might contribute to the development of disease ( Gravano and Hoyer , 2013 ) . In response to defective clonal deletion in the thymus , we observed compensatory induction of alternative tolerizing mechanisms that might prevent earlier onset of autoimmunity in BH3 Tg mice . An increase in the proportion and number of Treg cells was noted in both the thymus and lymph nodes of young mice . Interestingly , CD25− Treg cells accounted for most of the increase in total Treg numbers and these cells have been reported to have reduced suppressive function ( Zhan et al . , 2011 ) . These CD25− Treg cells are thought to be derived from diverted , autoreactive cells that were not deleted in the thymus . Consistent with this , CD25− Treg cells in the thymus expressed GFP at a higher level than CD25+ Treg cells and at a similar level to the most GFP-high naïve CD4 SP cells in the BH3/Nur77GFP mice . Hence , diverting a portion of these GFP-high autoreactive cells to this lineage might be the more effective tolerizing strategy . Notably , our observation that CD25+ Treg proportions were not increased in the BH3 Tg mice suggests that T cells with self-reactive TCRs are not automatically committed to the Treg lineage and that there is likely a saturated niche for CD25+ Treg development determined by availability of other factors , such as IL-2 and IL-15 ( Stritesky et al . , 2012 ) . In addition to increased Treg cells , we observed a 2-fold increase in the proportion of anergic CD4 T cells in the BH3 and Bcl-2 Tg mice . This was similarly observed in Bim−/− mice ( Stritesky et al . , 2013 ) . Induction of anergy could cooperate with diversion to the Treg lineage/Treg suppression to inhibit T cell-mediated promotion of autoimmunity and prevent early onset of disease . However , differential induction of these tolerizing mechanisms was not observed in BH3 Tg vs Bcl-2 Tg mice and therefore cannot account for the eventual development of autoimmunity in BH3 Tg , but not Bcl-2 Tg mice . Interestingly , in aged BH3 Tg mice , Treg cells made up almost 50% of the CD4 T cell pool in mice displaying autoimmune pathology . However , this protective response was still apparently insufficient to suppress autoimmunity . Finally , the enhanced ability of the Bcl2 BH3 mutant transgene to rescue autoreactive thymocytes from apoptosis compared to wild-type Bcl2 could suggest that the Bcl-2 BH3 domain might indeed have some pro-apoptotic function . This finding may provide in vivo support for the notion that Nur77 promotes thymocyte apoptosis through conversion of Bcl-2 to a pro-apoptotic effector . However , this still needs to be further confirmed in a knock-in model . Additionally , analysis of a Bcl-2 protein that cannot interact with Nur77 would also be required to confirm that Bcl-2 pro-apoptotic function is Nur77-dependent . Intriguingly , it has previously been suggested that Nur77 and Bim might function in the same pathway , as Bim-deficiency was unable to synergize with Nur77-deficiency to inhibit thymocyte apoptosis in the OTII TCR transgenic model of negative selection ( Fassett et al . , 2012 ) . If Bcl-2 conversion is indeed critical for Nur77-mediated thymocyte apoptosis , this would support the idea that Nur77 and Bcl-2 family members come together at the mitochondria to cooperatively regulate cell death during negative selection . The mutant allele for the Bcl2 BH3 mutant transgenic mice ( BH3 Tg ) was generated by QuikChange site-directed mutagenesis ( Agilent Technologies , Santa Clara , CA ) from a pCI-human Bcl2 vector using the following primers: BH3 forward CTGACCCTCCGCCAGGCCGCGGCCGCCTTCTCCCGCCGCTACCGC , BH3 reverse GCGGTAGCGGCGGGAGAAGGCGGCCGCGGCCTGGCGGAGGGTCAG . The nucleotide sequence of amino acids 101-103 was changed from GGCGACGAC to GCGGCCGCC , creating a restriction enzyme site change from BglI to NotI . The mutant transgene was then cloned into the XhoI site of the pTG4 construct ( Adlam and Siu , 2003; Xue et al . , 2010 ) . The pTG4 construct was then modified to include the Cd4 locus control region ( LCR ) . PCR amplification of the 1 kb LCR sequence from C57BL/6 genomic DNA was performed with the following primers: LCR forward GACATCGATAGCTAGCACACGCCGGTAAGCCCATTCCCCACGC , LCR reverse GACATCGATGCGGTACCGATCCCAACCAAACTGCGGCCCTTTCA . The LCR was added to the 5′ end of the Cd4 silencer at a ClaI restriction site . The transgenic mice were generated on the C57BL/6 background using standard procedures . Founders were identified by PCR genotyping with the following primers: HGH forward GACACAAACTCACACAACGATGACGC , HGH reverse ATGCCTGGAATCCCAACAACTCGG . The presence of the BH3 GDD to AAA mutation in the incorporated transgene was confirmed by PCR amplification of the transgene from genomic DNA and successful digestion with NotI , but not BglI enzyme . The BH3 Tg mice were compared with the T cell-specific wild-type Bcl2 transgenic line LckPr-Bcl2 ( Sentman et al . , 1991 ) , which have been back-crossed to C57BL/6 for more than 10 generations , referred to here as Bcl-2 Tg . The BH3 and Bcl-2 Tg mice were crossed with the following transgenic lines: Nur77GFP ( Moran et al . , 2011 ) , F5 TCR ( Mamalaki et al . , 1993 ) , and HY TCR ( Teh et al . , 1990 ) . All mice were on the C57BL/6 background . All experimental protocols involving animals were approved by the UC Berkeley Animal Care and Use Committee . Lymphoid organs were dissociated through 40 μm cell strainers to obtain a single cell suspension and a red blood cell lysis was performed . Cells were surface stained with fluorochrome-conjugated antibodies ( eBioscience , San Diego , CA , BD Biosciences , San Jose , CA and Tonbo Biosciences , San Diego , CA ) in 1% fetal bovine serum in PBS . For Annexin V staining , after surface staining , cells were stained with FITC-conjugated Annexin V ( BD Biosciences ) in 1x Annexin V binding buffer ( 10x: 0 . 1M HEPES , pH 7 . 4; 1 . 4M NaCl; 25 mM CaCl2 ) . For intracellular staining , prior to surface staining , cells were stained with Tonbo Ghost Dye ( Tonbo Biosciences ) to label dead cells , then fixed and permeabilized with a Cytofix/Cytoperm Fixation/Permeabilization Kit ( BD Biosciences ) . For Bcl-2 intracellular staining , cells were then stained with FITC-conjugated anti-human Bcl-2 ( Clone 124 , Dako , Carpinteria , CA ) or a FITC-conjugated mouse IgG1 isotype control ( Dako ) . For Cleaved Caspase 3 staining , cells were incubated with 5% Normal Donkey Serum and anti-CD16/32 Fc block ( Cone 2 . 4G2 , UCSF Hybridoma Facility , San Francisco , CA ) for 15 min then stained with anti-Cleaved Caspase 3 ( Asp175 , Cell Signaling , Danvers , MA ) or a rabbit IgG isotype control ( Santa Cruz Biotechnology , Dallas , TX ) for 45 min . For detection of the cleaved caspase-3 , a PE donkey anti-rabbit secondary antibody ( Jackson ImmunoResearch , West Grove , PA ) was added for 20 min . For intracellular cytokine staining , cells were incubated with FITC-IFNγ ( Clone XMG1 . 2 ) , APC-IL-17 ( Clone eBio17B7 ) and PE-IL-4 ( Clone 11B11 ) or similarly conjugated isotype controls , all from eBioscience , for 30 min . Intracellular staining for Foxp3 was performed with an eBioscience PE anti-Foxp3 staining kit ( Clone FJK-16 s ) following dead cell and surface staining . All samples were analyzed on the BD Biosciences LSR Fortessa or LSR II . Lymphocyte single cell suspensions were pelleted , washed with PBS , and lysed in 1% NP-40 lysis buffer ( 150 nM NaCl , 1 mM EDTA , 50 mM Tris–HCl , pH 7 . 6 , 1 mM NaVO4 , 1 mM NaF , 1 mM DTT , 1 mM PMSF and Sigma protease inhibitor cocktail ) for 20 min on ice . Whole tissue extracts were prepared by tissue homogenization in 1% NP-40 lysis buffer and incubation on ice for 15 min . Lysates were cleared by centrifugation at 13 , 000 rpm for 10 min . Protein was quantified with the Bio-Rad DC Protein Assay ( Bio-Rad , Hercules , CA ) . 50 μg ( 10-well comb ) or 1 mg ( preparative comb ) of protein was run out on 10% SDS-PAGE gels . Proteins were transferred to a nitrocellulose membrane and blocked with 5% bovine serum albumin or 5% non-fat milk in 1% Tween TBS ( TBST ) . The following antibodies were used for immunoblotting: mouse Bcl-2 ( BD Pharmingen , San Jose , CA , Clone 3F11 ) , human Bcl-2 ( BD Pharmingen , Clone 6C8 ) , or β-actin ( Sigma , St . Louis , MO ) . For sera immunoblots , membranes were placed in a Surf-Blot apparatus ( #5055 , Idea Scientific , Minneapolis , MN ) and probed with sera diluted at 1:200 in 5% non-fat milk in TBST for 2 hr at room temperature . Membranes were washed with TBST then incubated with an HRP-conjugated sheep anti-mouse IgG antibody ( 1:5000; GE Healthcare , Pittsburgh , PA ) for 45 min and visualized with SuperSignal West Pico Chemiluminescent Substrate ( Thermo Fisher Scientific , Waltham , MA ) by autoradiography . Lymphoid organs were dissociated through 40 μm strainers into RPMI media ( L-glutamine , Sodium Pyruvate , Non-essential amino acids , Hepes , β-mercaptoenthanol and Penicillin-Streptomycin ) . For assessment of TCR-mediated apoptosis , thymocytes were plated at 2 × 106 cells/well in triplicate in 96-well flat-bottom plates . Cells were left untreated or stimulated with either plate-bound anti-CD3 and anti-CD28 antibodies ( clones 2C11 and PV-1 , UCSF Hybridoma Facility ) at the indicated concentrations for 18 hr . Death was assayed by Annexin V staining as described above . For assessment of cytokine production by splenic T cells , total splenocytes were stimulated for 4 hr with PMA and Ionomycin at the above concentrations in the presence of BD GolgiPlug per the manufacturer's protocol ( Brefeldin A , BD Biosciences ) . To assess GFP downregulation in Nur77GFP cells , single cell thymocyte suspensions and mature T cells , purified by negative selection columns ( T Cell Enrichment Columns , R&D Systems , Minneapolis , MN ) , were plated in 48-well plates at 2 . 5 × 106 cells/well and 1 × 106 cells/well , respectively . GFP expression by live cells was assessed every 24 hr over a 96 hr time course by flow cytometry . For F5 in vivo peptide injection studies , F5 TCR transgenic mice were injected intraperitoneally with 50 nmol of Influenza NP366-374 Strain A/NT/60/68 peptide ( AnaSpec , Fremont , CA ) in 200 μl of PBS or PBS only for controls . A second injection was performed 24 hr after the first . Thymocytes were collected at 48 hr after the initial injection . For F5 and HY thymic slice assays , tissue preparation was performed as previously described ( Dzhagalov et al . , 2012 , 2013 ) . In brief , thymic lobes were embedded in 4% GTG-NuSieve Low-melt Agarose ( Lonza , Walkersville , MD ) in HBSS . 500 μm slices were cut by Vibratome ( 1000 Plus , Leica , Buffalo Grove , IL ) and placed in 0 . 4 μm Cell Culture Inserts ( BD Biosciences ) in 6-well plates containing 1 ml of RPMI media . TCR-specific ( NP366-374 Strain A/NT/60/68 and Smcy HY738-746 , AnaSpec ) and control ( NP366-374 Strain A/PR/8/35 and OVA257-264 , AnaSpec ) peptides were diluted to 100 ng/ml in RPMI media and 1 ml volume was added to the slices . Slices were incubated at 37°C in a plastic bag filled with 80% O2 + 15% N2 + 5% CO2 ( Blood Gas , Praxair , Danbury , CT ) for 30 min . The peptide was then removed and slices were further incubated for the indicated times . Thymic slices were dissociated to create single cell suspensions and cell death was assessed by Cleaved Caspase 3 staining as described above . Balb/c ( allogeneic ) and C57BL/6 ( syngeneic ) splenocytes were harvested as described above and subjected to 2000 rads of γ-irradiation . Splenocytes were washed with RPMI media several times and plated at 1 × 105 cells/well in a round-bottom 96-well plate . T cells ( C57BL/6 responders ) were purified from lymph nodes and spleen by negative selection columns ( T Cell Enrichment Columns , R&D Systems ) or for Treg-depleted MLRs by staining with PE-conjugated antibodies to CD25 , CD19 , CD11b , CD11c and Ter-119 ( eBioscience and BD Biosciences ) and negative selection using an EasySep Immunomagnetic PE Positive Selection Kit ( StemCell Technologies , Vancouver , BC , Canada ) . The purified T cells were then labeled with 1 μM Cell Trace CFSE ( Invitrogen , Life Technologies , Grand Island , NY ) according to the manufacturer's protocol and seeded on top of stimulators at 1 × 105 cells/well in triplicate . Proliferation was assessed by CFSE dilution after 4 days . Serum anti-nuclear antibodies were quantified using a Mouse Anti-Nuclear Antibody Total Ig kit ( Alpha Diagnostic International , San Antonio , TX ) per the manufacturer's protocol . Briefly , samples were diluted at 1:100 in Low NSB Sample Diluent and added to a 96-well plate pre-coated with purified ENA . Quantification was performed relative to provided standards . The plate containing diluted sera and standards was incubated for 1 hr at room temperature . Anti-mouse Ig HRP was added for 30 min followed by TMB substrate for 15 min . The reaction was stopped and the plate was read at 450 nm . Absorbance at 630 nm was subtracted as background . Organs were fixed in formalin , embedded in paraffin , and sliced into 5 μm sections by microtome . Sections were stained with Hematoxylin and Eosin ( Thermo Fisher Scientific ) by standard procedures . Images were captured with LAS Core V4 . 0 software ( Leica ) on a Leica DFC500 microscope equipped with a Leica DM2500 camera at 20x magnification ( HCX PL-FLUOTAR 20X/0 . 50 objective , Leica ) .	Our immune system protects us from disease by recognizing and mounting a defence against harmful pathogens that enter our bodies . T cells , a type of white blood cell , play a key role in this process . Each T cell has a unique protein called a T cell receptor on its surface that is able to recognize particular pieces of pathogens . Together , the millions of T cells in our bodies , each with its own unique T cell receptor , can initiate an immune response to eliminate a vast array of potential pathogens . T cells are made in an organ called the thymus . During this production process , immature T cells are generated , including some with T cell receptors that recognize the harmless molecules that make up our bodies . If allowed to enter the bloodstream and left to their own devices , these T cells could trigger an immune response against the body , leading to the development of autoimmune disease . Normally , many of these ‘auto-reactive’ T cells are instructed to die in the thymus by a process called negative selection . Furthermore , auto-reactive cells that escape into the blood can also be shut down by additional failsafe mechanisms . Given the success of these failsafe mechanisms , notably the effectiveness of a class of T cells called T regulatory cells , some researchers have begun to ask if negative selection is necessary to prevent autoimmunity . During negative selection T cells die as a result of a process called apoptosis . Multiple proteins have been implicated in T cell apoptosis , including Bim , Puma and the Nur77 family of nuclear receptors . Blocking the function of some these proteins individually can rescue some autoreactive T cells from death: however , this rarely results in the development of autoimmune disease . Burger et al . have now created a mouse strain with T cells that produce large amounts of a mutant form of the anti-apoptotic protein , Bcl-2 , which can block the function of multiple pro-apoptotic proteins , including Bim and Puma . Additionally , it has been proposed that Bcl-2 can be converted to a pro-apoptotic protein by Nur77 proteins , but the Bcl-2 proteins in the mutant strain are able to resist this process . T cells in the thymus of the mutant mice were highly resistant to apoptosis accompanying negative selection . Moreover , as the mice aged , they accumulated autoreactive T cells in the blood , which led to symptoms of autoimmune disease and early death . While various failsafe mechanisms were engaged , they did not provide sufficient protection . The work of Burger et al . thus provides strong evidence that negative selection in the thymus is crucial for the prevention of autoimmune disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2014	T cell-specific inhibition of multiple apoptotic pathways blocks negative selection and causes autoimmunity
One manifestation of individualization is a progressively differential response of individuals to the non-shared components of the same environment . Individualization has practical implications in the clinical setting , where subtle differences between patients are often decisive for the success of an intervention , yet there has been no suitable animal model to study its underlying biological mechanisms . Here we show that enriched environment ( ENR ) can serve as a model of brain individualization . We kept 40 isogenic female C57BL/6JRj mice for 3 months in ENR and compared these mice to an equally sized group of standard-housed control animals , looking at the effects on a wide range of phenotypes in terms of both means and variances . Although ENR influenced multiple parameters and restructured correlation patterns between them , it only increased differences among individuals in traits related to brain and behavior ( adult hippocampal neurogenesis , motor cortex thickness , open field and object exploration ) , in agreement with the hypothesis of a specific activity-dependent development of brain individuality . Individualization is the process of developing unique traits and thus divergence from the inborn and genetically determined makeup . The behavioral and molecular bases of such divergence were traditionally investigated in human twin studies . However , the difficulty in conducting longitudinal studies in humans , as well as the limited range of phenotypes that could be assessed in each twin cohort , leave many fundamental questions open . In particular , underlying mechanisms at the levels of cells , tissues , systems or the entire brain and their interaction across these scales cannot be determined in human subjects because it is not possible to collect all relevant phenotypes with sufficient depth and precision or to manipulate the processes in question experimentally . Thus , addressing these problems calls for a suitable animal model in which both environment and genotype can be strictly controlled . Individualization involves an increasingly differing response of initially highly similar individuals to exposure to seemingly the same environment . We propose that activity-dependent structural plasticity is a central mechanism contributing to the individualization of the brain . The iterative nature of the feedback loops between plasticity and behavior result in increasingly different brains , behavioral trajectories and life courses . In this model , small initial differences are augmented through self-reinforcement . In support of this hypothesis , we previously showed that large groups of isogenic mice that were exposed to an enriched environment ( ENR ) developed stable and unique social and exploratory behavioral patterns that diverged between individuals over time ( Freund et al . , 2013; Freund et al . , 2015 ) . What differed between the mice of this cohort was their unique experience of that same environment and their resulting differential behavior . Because this ‘non-shared environment’ relates to the individual’s own experience and actions , the paradigm revealed a dimension that was previously largely hidden in group effects , but which is of greatest interest for studies addressing sources of variance in a system . We and others had previously described the stimulatory effect of ENR on mean levels of adult hippocampal neurogenesis ( Kempermann et al . , 1997; Nilsson et al . , 1999; Tashiro et al . , 2007 ) , the lifelong activity-dependent generation of granule cells in the mammalian dentate gyrus . Furthermore , we showed that longitudinal individual behavioral trajectories correlated with the within-group differences in numbers of new neurons among the enriched mice , underpinning the suitability of adult neurogenesis as a biologically relevant readout of activity-dependent brain plasticity ( Freund et al . , 2013 ) . This previous experiment suggested an increased variance in the numbers of new neurons integrated into the hippocampal circuit of ENR mice as compared to that of mice living in standard laboratory cages , but the effect could not be claimed unequivocally because the control group was small in size when compared to the experimental group . Moreover , because behavioral assessment was based on monitoring animals in the ENR enclosure , the same constructs were not accessible for control mice . Finally , we could not determine the degree to which the effect of ENR on variance ( and hence individuality ) was specific to adult neurogenesis and exploratory behavior . The experiment to address these questions is presented here . Because ENR has been shown to influence a broad range of body and brain-related parameters in rodents , including metabolic states ( Wei et al . , 2015 ) , volumes of certain brain areas ( Diamond et al . , 1985Diamond et al . , 1964; Diamond et al . , 1966; Diamond et al . , 1985 ) , and different behavioral aspects ( Clemenson et al . , 2015; Garthe et al . , 2016 ) , we were particularly interested in testing the ENR effect on the variance of these parameters . If increases in variance were general across all domains , this would suggest a common , non-specific causality . From a mechanistic perspective , the paradigm would thus be less feasible as a model that could be used to study the emergence of brain individuality . A more specific and selective induction of variance in response to enrichment would indicate that the observed individualization of the brain does not arise as a mere epiphenomenon of broader effects . To investigate whether long-term environmental enrichment triggers the specific development of inter-individual differences between mice , we performed a cross-sectional study and analyzed differences in variance between groups of mice housed in one large enriched environment or in control cages ( CTRL ) . Both ENR and CTRL groups consisted of 40 female C57BL/6JRj mice that were randomly assigned to their respective housing conditions , where they stayed for 105 days ( Figure 1A ) . In addition to the social complexity introduced by the number of animals in the enrichment cage , the complexity of the ENR was increased by the large size and the compartmentalization of the enclosure ( Figure 1B ) . A total of 28 morphological , behavioral and metabolic variables were assessed ( Supplementary file 1 and 2 ) . To determine the effects of the ENR on gross body morphology , we monitored the body weight of all animals over the 105 days of the study ( Figure 2A ) . At the beginning of the experiment , no differences in weight existed between the two groups , confirming initial similarity between the randomized experimental mice . However , five weeks after the start of the experiment , ENR mice were significantly lighter than mice housed in control cages ( CTRL ) . The difference in body weight remained constant throughout the experiment and indicated , together with the significantly shorter body length in ENR mice ( Figure 2B ) , that housing of mice in ENR reduces body size . By contrast , no differences in brain weights were detected between ENR and CTRL mice ( Figure 2C ) . The groups did not differ in the variances of body length , body weight and brain weight at any measured time point , suggesting that long-term ENR does not stimulate the development of inter-individual differences in gross body morphology . To analyze whether ENR increased inter-individual variability in behavior , all mice were subjected to the open field ( OF ) , novel object recognition ( NOR ) and rotarod tests ( Figure 3A–B ) . ENR mice traveled longer distances during the first OF trial , but showed less locomotion in the second OF trial ( Figure 3C ) and throughout the NOR test ( Figure 3D ) . No significant differences were found in the variance of locomotion in any of the OF or NOR trials . In our previous work , we introduced roaming entropy ( RE ) as a measure of the territorial coverage and exploratory activity of mice in order to introduce a qualitative aspect into activity measurements ( Freund et al . , 2013; Freund et al . , 2015 ) . To investigate the effects of ENR on spatial exploration , we computed RE for all mice in the OF arena ( Figure 3E–F ) . On both days , ENR mice had significantly lower RE than CTRL animals . Moreover , on day 2 , ENR mice showed a significantly greater variance in RE , suggesting a higher range of habituation to the OF among ENR mice . Indeed , both ENR and CTRL animals habituated to the OF arena , as indicated by a decrease in RE between the trials ( Figure 3G ) . However , habituation was more pronounced and exhibited higher variance in ENR mice . In the NOR test , ENR mice showed a significantly higher variance in the duration of their exploration of the objects when compared to CTRL mice ( Figure 3H ) , indicating that ENR increases the inter-individual variability in exploratory behavior . Although some individuals among ENR mice explored objects for much longer than any of the control animals , the median of the entire ENR group was not shifted compared to that of the CTRL group . Finally , to examine the effect of ENR on the recognition memory of individual mice , we analyzed the ability to discriminate a new object from an old one in the NOR test . A trend towards a preference for the new object was found only in the ENR mice and not in the CTRL group ( Figure 3I ) . The performance of ENR mice on the rotarod was superior to that of CTRL mice in all trials ( Figure 3J ) , indicating that ENR stimulates motor coordination and , presumably , fitness . Initially , ENR mice also showed a greater variance in their performance compared to that of CTRL , but while both groups improved in the task , this difference gradually disappeared . Together , we conclude that ENR promotes the development of inter-individual differences in specific interactions with the environment , but not in pure locomotor activity . To assess whether the observed behavioral variability is reflected in differences in brain plasticity , we quantified the rates of adult neurogenesis in the dentate gyrus of the hippocampus . To estimate the proliferation of precursor cells , we stained mouse brain sections for the proliferation marker Ki67 ( Figure 4A–B ) , whereas new-born cells that survived initial selection processes were identified by the presence of BrdU , which was injected 3 weeks before the end of the experiment ( Figure 4C–E ) . No differences in the means or variances of the numbers of proliferating cells in the subgranular zone of the dentate gyrus were observed between ENR and CTRL mice ( Figure 4F ) . By contrast , we found a significant increase in the means and variances of the numbers of BrdU-positive cells in animals housed in ENR ( Figure 4G ) , highlighting the specific effect of ENR on the survival of new-born cells . Co-localization of BrdU-positive cells with the neuronal marker NeuN and the astrocytic marker S100β ( Figure 4E ) showed that the variances in the survival of both neurons and astrocytes were higher in the ENR group than in the CTRL animals ( Figure 4H–I ) . An increase in the total number of cells was , however , only found in the neuronal cell population . These results indicate that ENR increases inter-individual variability in the survival of new-born neurons and astrocytes but not in the proliferation of precursor cells in the dentate gyrus . ENR has been long known to induce broad changes in brain structure in rodents , such as thickening of the cerebral cortex ( Diamond et al . , 1966 ) and increases in the volume of the dentate gyrus ( Kempermann et al . , 1997 ) . We also showed that the volume of the mossy fibers increases upon environmental stimulation concomitantly with adult neurogenesis in mice ( Römer et al . , 2011 ) . To further assess whether ENR increases inter-individual variability in brain plasticity beyond adult neurogenesis , we estimated the volumes of the hippocampus and its substructures: the dentate gyrus , infra- and suprapyramidal mossy fiber tracts ( IMF and SMF ) and the hilus ( Figure 5A–D ) . The volume of the entire hippocampus did not differ between ENR and CTRL mice ( Figure 5H ) , but the volume of the dentate gyrus was significantly increased in ENR mice ( Figure 5I ) . Furthermore , IMF ( Figure 5J ) and the hilus ( Figure 5L ) , but not SMF ( Figure 5K ) , were significantly larger in ENR animals than in CTRL mice . None of these parameters showed different variances depending on housing conditions . This suggests that ENR differentially influenced various aspects of hippocampal plasticity , while showing that the increased inter-individual variability that was triggered by ENR was specific to adult neurogenesis . Next , we measured the thickness of the entorhinal , cingulate and motor cortex ( Figure 5E–G ) as enrichment might specifically increase cortex thickness and structure in these areas ( Diamond , 2001Diamond et al . , 1964 , Diamond , 2001 ) . Although we did not detect differences in the thickness of any of these cortices between CTRL and ENR mice ( Figure 5M–O ) , the motor cortex thickness showed a significantly higher variance in the ENR group ( Figure 5O ) . Since ENR reportedly had beneficial effects on metabolism in outbred mice ( Wei et al . , 2015 ) , we compared the weights of the liver and adrenal glands as organs playing a role in metabolic and hormonal regulation and analyzed basic blood biochemistry . In agreement with lower body weights , ENR animals had smaller adrenal glands and livers ( Figure 6A–B ) . The levels of plasma corticosterone , which is synthesized in the adrenal gland and is used as an indicator of animal stress , did not differ between the groups ( Figure 6C ) . Moreover , significantly lower plasma cholesterol levels ( Figure 6D ) in ENR mice but no differences in triglyceride and glucose levels were seen when comparing ENR and CTRL mice ( Figure 6E–F ) . No variance differences between the two groups were detected in any of the measured metabolic parameters . To analyze the impact of ENR on relationships between phenotypes , we calculated correlations separately for CTRL and ENR mice ( Figure 7 ) . Strong correlation between two traits would point towards shared regulatory mechanisms in the control of such parameters . We observed few significant correlations in either housing condition , which suggests that the majority of the traits were independent of each other . In both animal groups , we found significant positive correlations of roaming entropy and object exploration between trials of the OF and NOR tests , respectively ( Figure 7A ) . These correlations indicate that the recorded behaviors were reliable and characteristic for each individual . ENR , however , weakened correlations between trials in NOR , as well as the negative association of object exploration to OF habituation , hinting towards more specific responses of animals to the environment ( e . g . exposure to novel objects or their placement ) . Housing in ENR led to remodeling of the associations between the brain structures and behavior ( Figure 7A ) . ENR uncoupled the negative correlations of object exploration in NOR test and of rotarod performance to the volume of the hippocampus . Similarly , habituation to the OF arena was negatively associated with the size of the IMF and positively correlated with the motor cortex thickness in CTRL mice , but not ENR mice . Hippocampal neurogenesis did not show significant correlation with any of the assayed phenotypes . Metabolic phenotypes , namely plasma glucose , cholesterol and triglycerides were positively correlated in both housing conditions , but these relationships were weakened by ENR ( Figure 7B ) . As expected , plasma triglycerides correlated positively to the liver size in both groups . Epidemiological studies in humans suggest that brain and cognition are linked to metabolism ( Kapogiannis and Mattson , 2011; Panza et al . , 2012 ) . We observed few associations between the measured phenotypes in our mice ( Figure 7B ) . ENR changed the sign of the correlations between object exploration in NOR test and plasma cholesterol from positive to negative . It also promoted negative correlation between plasma glucose and rotarod performance , suggesting that the fitness acquired by ENR mice has a metabolic component . As medicine acknowledges inter-individual differences as a key determinant in diagnosis and treatment , understanding the biological mechanisms underlying individuality becomes increasingly important . Here we established that ENR is a suitable model to dissect processes leading to brain individualization . The purpose of our multivariate cross-sectional study of ENR effects in mice was to provide an insight into the magnitude of the individualization of phenotypes spanning broad aspects of physiology . We have shown that different traits are not uniformly affected by the stimuli and , besides effects on the mean , effects on variance could also be observed for certain parameters . The effects in which ENR increases differences between individuals in the group cluster in variables related to behavior and adult neurogenesis . To evaluate the effects of ENR , we used statistical tests that were appropriate for each given data distribution ( as described in the 'Materials and Methods' ) and interpreted p-values smaller than 0 . 05 as sufficient evidence for the influence of ENR on a particular phenotype . p-values were introduced by Ronald A . Fisher as an informal index to indicate whether or not the null hypothesis of no effect fails to account for the whole of the observations ( summarized in Lehmann , 1993 and in Goodman , 1993 ) . Lower p-values imply lower likelihood of the null hypothesis given the obtained data , with strength of evidence being interpreted as weak to moderate in the range between 0 . 1 to 0 . 01 , and strong to very strong at or below 0 . 001 ( Goodman , 1999 ) . The negative findings , though , are less straightforward to evaluate . With 40 individuals in each experimental group , a 5% pre-set significance level in a two-tailed variance test enabled us to detect an effect size of 2 . 5 ( ratio of variances ) with power of 0 . 8 . This is a relatively large effect and moderate power , which implies that smaller changes in variance , as well as some of the stronger effects , could have been missed . At the same time , we would argue that the identified effects are robust enough to be biologically relevant . Furthermore , we adopted marginal , that is individual , interpretation of responses to ENR for each of the assayed phenotypes and did not control for an experiment-wide type I error rate . Correction of p-values upon multiplicity of tests stems from posing the universal null hypothesis that no association exists between any pair of variables under investigation ( Rothman , 1990 ) . The question we posed was , however , not whether ENR influences the variance or mean of any trait ( a universal null hypothesis ) , in which case the control of experimental error rate would be necessary , but rather each of the specific responses was of interest ( for the distinction of various scenarios see Cook and Farewell , 1996 ) . Most importantly , however , the correction for multiple tests leads to inflation of type II errors ( false negatives ) and hence introduces a ‘penalty for peeking’ , that is , the more parameters are investigated , the less likely each of the true associations is to be detected ( Perneger , 1998; Rothman , 1990 ) . Significant p-values in our experiment were not uniformly distributed across all phenotypes , supporting our view that we achieved a balance between false-negative and false-positive error rates sufficient to provide a coherent overview of the effects of ENR on a wide variety of traits . Our behavioral data highlight a significant effect of ENR on animals’ active interaction with their environment: improved fitness and coordination , as assessed by the rotarod task , and modified patterns of exploratory activity and habituation in OF and NOR tests . It has long been known that ENR elicits profound effects on brain plasticity and behavior ( Mohammed et al . , 2002; Sale et al . , 2014 ) . The effects of ENR housing on animal behavior in variations of OF and NOR tests have been reported in earlier studies . We have previously found , for example , that ENR mice habituate faster to an open field , and this observation has been interpreted as improved spatial processing ( Kempermann and Gage , 1999 ) . Since Ennaceur and Delacour introduced the spontaneous object recognition task ( Ennaceur and Delacour , 1988 ) , its modifications have been used to dissect components and neural bases of recognition memory ( reviewed in Ameen-Ali et al . , 2015 and Antunes and Biala , 2012 ) . Short 2 min trials in our NOR task precluded efficient familiarization with objects ( Melani et al . , 2017 ) , which explains the lack of preference towards the new object . The multi-trial paradigms were developed to reduce extra-experimental variance and thus to improve reproducibility and reduce the numbers of animals needed for the experiments ( Albasser et al . , 2010; Ameen-Ali et al . , 2012 ) . Although not a multi-trial paradigm in this classical sense , our design involved multiple trials of the NOR and OF tests , which allowed us to confirm that the exploratory behaviors were stable and idiosyncratic , as indicated by the high intra-group correlations between the trials of each task . The current study now highlights that ENR also induced substantial inter-individual variability in specific behavioral parameters . Particularly in the NOR test , ENR mice showed much greater variability in object exploration times compared to the relatively homogenous CTRL group . In the OF test , ENR mice not only exhibited greater habituation to the arena , but this response was also more variable than that in CTRL animals . Locomotion , a less specific aspect of exploration , decreased in ENR mice and its variance was not affected by housing . These observations corroborate our previous finding of ENR-induced individualization of spontaneous interactions of mice with their environment ( Freund et al . , 2013 ) . We have previously argued that the active participation with the outer world and habitat that manifests itself in the individual range of locomotion within that world ( roaming entropy ) is a major driving force of brain plasticity , presumably not limited to adult hippocampal neurogenesis ( Freund et al . , 2013 ) . The current data are in agreement with this hypothesis . The improvement of rotarod performance in the ENR mice observed in this study implies that , even if the running wheels are not supplied , large cage area and toys to interact with provide considerable motor stimulation ( Kempermann and Gage , 1999 ) , which , together with elevated variance of the motor cortex thickness , suggest that ENR strongly affects the plasticity of motor responses . It has been proposed that activation of motor cortex by ENR has widespread modulatory effects on other cortical areas ( Sale et al . , 2014; Di Garbo et al . , 2011; Niell and Stryker , 2010 ) . Our previous report suggested that long-term ENR induces variability in the survival of new-born neurons ( Freund et al . , 2013 ) . The large size of both the control and the ENR groups in the present study allowed us to corroborate that finding ( F-test: p=0 . 00004 ) . Although new astrocytes did not increase in numbers , we observed a small effect on the variance of these . By contrast , there were no differences in the variance of the number of proliferating cells between ENR and CTRL animals . Although the effect of ENR on the mean numbers of proliferating cells did not reach the significance threshold applied here ( p=0 . 09 ) , together with our previous report ( Kempermann et al . , 2002 ) , this study suggests that over prolonged periods of time , enrichment might have subtle pro-proliferative effects . Proliferating cells are a substrate on which selection mechanisms can act , but their behavioral significance as such has not yet been shown . The fact that ENR does not trigger individuality in precursor cell proliferation indicates that individualization mechanisms act selectively only on those aspects of neurogenesis that are relevant for interactions with the environment . In contrast to the study by Freund et al . , 2013 , in which 21% of the variance in adult neurogenesis could be explained by differences in cumulative roaming entropy ( RE ) , an aggregated measure of the longitudinal behavioral trajectory , the current analysis detected no correlation between the number of newborn neurons and any of the cross-sectionally measured behaviors , including RE measured in the OF test . RE is a convenient single parameter describing the uniformity of coverage of a given space ( Freund et al . , 2013; Freund et al . , 2015 ) , but its interpretation depends on the context in which it is used . Here , we calculated RE to describe the exploration in the OF , because it carries more information than the traditional measures . Specifically , it does not rely on hard boundaries , such as the periphery and center of the open field , and also takes into account exploration within each of these zones . ( For example , a mouse that remains in the corner of an OF arena and a mouse that visits the entire perimeter of an arena might have spent the same amount of time in the periphery but differ in the RE measure . ) In the present study , RE in the OF was lower among ENR mice than in CTRL mice . The behavior in the familiar environment of the ENR cage , however , is presumably driven by other factors and can be influenced by the group interactions ( Shemesh et al . , 2013 ) , whereas the OF test reflects an individual response of an animal to novel situations and might , therefore , constitute a different construct . Perception of novelty and incentive to explore an empty OF arena are likely to be different for CTRL animals , which spent their entire life in small cages , and the ENR group . The mossy fiber projection , and especially its infra-pyramidal blade ( IMF ) , is highly plastic in mice ( Crusio et al . , 1989; Schwegler et al . , 1981 ) and ENR can modulate its size ( Römer et al . , 2011 ) . We found an increased volume of the IMF but no effect on the variance of this volume after 3 months of ENR , suggesting that even for aspects of hippocampal plasticity there is no simple parallelism in the effects of ENR . Similarly , we observed an increase in the volume of the dentate gyrus , but did not detect changes in the mean volume of the hippocampus or in the variance of this phenotype . Adult-generated neurons contribute to the IMF ( Römer et al . , 2011 ) , yet we did not observe a correlation between numbers of new neurons and IMF volume within either housing group , impliying that , under physiological conditions , mechanisms other than adult neurogenesis determine the bulk of the IMF . This finding is in agreement with the results from the screen in the mouse genetic reference population ( Krebs et al . , 2011 ) . Cortex thickness changes upon enrichment of the environment have been reported in the older literature and were the cornerstone of the growing impact of the ENR paradigm in the 1970s ( reviewed in Diamond , 2001; Rosenzweig and Bennett , 1996 ) . Increases in cortical thickness do not strictly mirror volume changes ( Hammelrath et al . , 2016; Winkler et al . , 2010 ) , but they are an indication of massive cellular rearrangements in the cerebral cortex ( Diamond et al . , 1964Diamond et al . , 1966 ) . In our study , the only effect that we found in response to ENR was an increase in the variance of the motor cortex thickness . The key difference between our work and classical studies is that we worked with mice , whereas essentially all classical studies had been done in rats . The dynamics of three-dimensional brain development during first months of life differ between these two species ( Hammelrath et al . , 2016 ) . Furthermore , the majority of old experiments compared enriched animals to impoverished littermates , which were kept in social isolation . Such impoverishment negatively affects brain size ( Fabricius et al . , 2010 ) , thus amplifying the relative effects of enrichment ( Bennett et al . , 1964 ) . We believe that the impact of ENR on cortical plasticity deserves still more specific analyses with much greater resolution . It had been shown that ENR influences metabolism ( Wei et al . , 2015 ) : keeping outbred mice in ENR resulted in decreased body weight , mostly through reducing fat content; lowered blood cholesterol , triglycerides , and glucose; and improved insulin and leptin signaling . It has to be noted that the cages in the experiments performed by Wei and colleagues were equipped with running wheels to stimulate physical exercise . In the present study , we also observed decreases in body and liver weights , as well as lowering of plasma cholesterol , which indicates that ENR alone has a moderate beneficial effect on metabolism even in the absence of intense physical exercise . Although we routinely recorded reduced body weights in mice living in ENR conditions ( Kempermann and Gage , 1999 and unpublished observations ) , this response might be subjected to local conditions that are unique to specific animal facilities ( Crabbe et al . , 1999 ) as no such effect was observed in our previous experiment ( Freund et al . , 2013; Freund et al . , 2015 ) . ENR animals also had smaller adrenal glands and even though corticosterone levels were similar , this points towards reduced stress in ENR mice compared to CTRL . Finally , we did not detect differences in the variances of any of the metabolic parameters , further substantiating the conclusion that individualization of behavior and brain plasticity by ENR is not an epiphenomenon of more global physiological divergence . Although the issue of variance was brought up in very early studies ( Walsh and Cummins , 1979 ) , the ENR literature has not been much concerned with variance effects and inter-individual differences . The focus has always been on mean group effects . The question of ENR effects on variance came up , however , in the context of a movement in animal husbandry to provide larger space and enriching cage accessories in order to improve animal wellbeing and to provide more species-appropriate conditions . Variability induced by ENR , the concern went , would work against the desired standardization and stability of animal experiments in the life sciences . A widely cited study in mice by Wolfer et al . , however , confirmed that ENR ‘increases neither individual variability in behavioral tests nor the risk of obtaining conflicting data in replicate studies’ ( Wolfer et al . , 2004 ) . The results presented here ( Figure 3 ) stand in clear contrast to the first part of this statement and potentially also the second . As we did not test a full spectrum of behavioral tasks , we must not generalize our conclusion beyond open field and novel object recognition tests ( this study ) , or free roaming in the cage ( Freund et al . , 2013 ) . We would hypothesize that behavioral traits related to exploration and adjusting to novel situations , including hippocampal learning , are more strongly affected than other traits . The conclusion from Wolfer et al . requires a careful qualification . Nevertheless , we fully agree with the overall conclusion that the ‘housing conditions of laboratory mice can be markedly improved without affecting the standardization of results’ , especially if group sizes are sufficiently large . For most variables , even 3 months of ENR did not increase variability or alter correlations with other phenotypes . Furthermore , systemic variation might actually improve reproducibility ( Richter et al . , 2011; see also Richter et al . , 2010; with comments and re-analysis in Jonker et al . , 2013 and in Wolfinger and Reanalysis of Richter , 2013 ) . And finally , the attempt to ignore the within-group variation as an expression of a differential response to the same nominal stimulus might actually contribute to the ‘reproducibility crisis’ to a much larger extent than previously appreciated . The mechanisms by which ENR increases variance are currently unknown . We hypothesize that increases in variability are a result of the progressive amplification of initially small inter-individual differences that existed before the start of the experiment , or that were introduced by stochastic events in the initial period of ENR housing . Potential sources of pre-existing variation include prenatal influences on the pregnant mother , intrauterine positioning of the fetus and early postnatal experiences ( Lathe , 2004 ) . Early life-experiences especially are known to change epigenetic modifications in the brain , which contribute to the long-term control of gene expression . For instance , differences in maternal care in rats led to differences in the DNA methylation state of the glucocorticoid receptor in the hippocampus of the offspring ( Weaver et al . , 2004 ) . Moreover , human twin studies have suggested that monozygotic twins increasingly differ in epigenetic marks from early life to adulthood , presumably as a result of their different experiences ( Cheung et al . , 2018; Fraga et al . , 2005 ) . Environmental enrichment builds on the initial variation and amplifies the differences by providing opportunity for the development of individual behavior . According to our hypothesis of positive feedback through experience , the individualization that occurs in the first months of ENR housing should lead to a permanent discordance of behavior , responses to cognitive challenges , and possibly also brain morphology . In support of this , we have previously shown that mice establish stable behavioral trajectories in the first two months of ENR housing that are maintained for the time of monitoring ( Freund et al . , 2013 ) and are presumably kept up long-term . Whether life-long ENR housing would increase the inter-individual variability even more over time and whether the inter-individual differences are stable after withdrawal of the ENR stimulus are the subjects of current investigations . The positive influence of ENR on neurogenesis and behavior is independent of age ( Kempermann et al . , 1998 ) , but the housing of mice in ENR for longer than 3 months does not further increase experience-dependent neurogenesis ( Kempermann and Gage , 1999 ) . Concurrently , the exploratory activity of mice was shown to decrease with time in ENR ( Freund et al . , 2013 ) , suggesting , together with the age-related decline in neurogenesis , that the strength of the iterative feedback between neurogenesis and behavior decreases , which could result in a plateauing of the individuality effect with time . The question arises of whether ENR is unique among activity-dependent plasticity experiences in inducing behavioral and structural divergence . Published studies tend not to provide sufficient information about the individualizing effects of other manipulations because of modest group sizes . ENR is a complex paradigm , in which inanimate aspects of the environment and social interactions intertwine over prolonged periods of time . We hold the view that both this complexity and duration are essential elements in the consolidation of the induced changes . Accordingly , in our previous study ( Freund et al . , 2013; Freund et al . , 2015 ) , the patterns of general activity , RE and both social and non-social behaviors recorded towards the end of the ENR exposure could not be predicted by the initial differences between animals . In this study , ENR increased variation within a group of female mice . We have used females to avoid the inter-animal conflict behavior that unrelated males show when put together at the delivery age of 4 weeks and to build on our previous ENR experiments ( Kempermann and Gage , 1999; Freund et al . , 2013; Freund et al . , 2015 ) . Male mice are known to respond similarly to ENR with increased hippocampal neurogenesis ( Zhang et al . , 2018 ) . By contrast , several studies reported sex differences in behavioral responses towards ENR , with female mice being more susceptible to the positive effects of ENR on cognition ( Coutellier and Würbel , 2009; Hendershott et al . , 2016; Wood et al . , 2010 ) . As male mice build stronger hierarchies than females , we expect that the contribution of the social interaction on individuality development is stronger in a male group than in a female group of mice . Dominant males influence the behavior and stress levels of subordinate males ( Curley , 2016 ) , which could lead to an even stronger and faster individualization in ENR that is less instructed by activity-dependent brain plasticity . On the other hand , increased social distress in subordinate animals might blur individualization effects . However , whether ENR housing leads to the development of inter-individual differences in behavior and brain plasticity in male mice is currently unknown and should be addressed in future experiments . Despite the ample literature on ENR , few studies addressing larger numbers of dependent variables have been conducted , and to our knowledge , we are first to investigate the interactions between an extended panel of variables in a correlation matrix . Similarly , there has been little insight into the isometry or allometry of the induced changes . Because our experiment employed large groups of animals , we could survey the inter-individual correlation patterns between the variables separately within each environmental condition , thus avoiding spurious relationships that could arise from mean differences between groups . Correlation matrices revealed the extensive relative independence of outcome measures , suggesting that the choice of traits for the analysis was broad enough to reflect distinct underlying causalities . Furthermore , ENR restructured correlation patterns by strengthening or weakening some associations ( for details see Figure 7 ) , further demonstrating the uneven regulatory influence of ENR on various aspects of physiology and plasticity . Thus , even in the absence of global mean effects on these parameters , ENR seemed to induce broad adaptations in brain plasticity and metabolism . In conclusion , ENR does not generally increase variability across all domains . ENR-induced increases in variance were specific to exploratory behavior , adult neurogenesis and motor cortex thickness . The correlation pattern of these parameters with other traits was complex , with ENR remodeling many of the associations . We do not think , however , that increased structural variability is limited only to neurogenesis and motor cortex , but rather that the induced changes are very specific and can be revealed only when appropriate aspects of plasticity are examined . ENR arises from this study as a more holistic paradigm than often assumed , and proves to be a decent tool with which to investigate the bases of experience-dependent brain individualization . In the laboratory setting , animals are relieved from pressures present in nature and therefore they are free to choose the degree of interaction with their environment . In our previous longitudinal study , we made the case that in a situation in which both genes and ( nominal ) environment are kept constant , individuality emerges as a consequence of the so-called ‘non-shared environment’ , that is the individual response to that environment and activity ( Turkheimer , 2011 ) . This situation is comparable with monozygotic twins , which—even when raised in the same household—develop differences in behavior , appearance and disease susceptibility over time . The underlying mechanisms that drive this divergence are , however , unknown and difficult to address in human studies . Here , we present an animal model that can be used to study the influence of the non-shared environment on individualization and its relation to brain plasticity . Our data suggest that multivariate studies with a large number of individuals and , ideally , a longitudinal design are needed to elucidate the exact contribution of the non-shared environment to the overall outcome of increased individualization . In perspective , the model of long-term ENR can be extended to analyze the development of individuality in a genetically variable population to provide insights into the interaction of genes with the non-shared environment . 80 female C57BL/6JRj mice were purchased from Janvier at the age of 4 weeks and housed in standard polycarbonate cages ( Type III , Tecniplast ) in groups of five until the start of the experiment ( Figure 1A ) . At the age of 5 weeks , 40 mice were randomly selected and transferred into the enriched environment , where they stayed for three months ( no restricted randomization ) . The number of animals used in each group was decided on the basis of the sample size used in the initial study conducted by Freund et al . , 2013 , which the present study builds on . The enriched environment consisted of four quadratic polycarbonate cages ( 0 . 74 × 0 . 74 m ) that were assembled in a row and connected by two plastic tubes each . In total , the enriched environment covered an area of 2 . 19 m2 ( Figure 1B ) . Food and water were provided in every compartment of the cage . To provide sensory stimulation , each compartment of the cage was equipped with plastic toys , tunnels and hideouts , which were cleaned and rearranged once each week . The bedding material was replaced on a weekly basis . Once a month , the entire enclosure was cleaned . Control animals were housed for the same period of time in standard polycarbonate cages ( 36 . 5 × 20 . 7×14 cm ) connected to an individually ventilated cage system in groups of five . Control and enriched animals were receiving the same fortified chow ( #V1534; Sniff , Germany ) with 9% of energy from fat , 24% from protein and 67% from carbohydrates . All mice were maintained on a 12 hr light/12 hr dark cycle with humidity maintained at 55 ± 10% and food and water provided freely . The room was furnished with metal shelves containing laboratory equipment . Three weeks before sacrifice , the mice were injected intraperitoneally with bromodeoxyuridine ( BrdU; 50 mg/kg body weight; dissolved in 0 . 9% NaCl ) . Injections were performed once per day for three consecutive days . All experiments were conducted in accordance with the applicable European and national regulations ( Tierschutzgesetz ) and were approved by the responsible authority ( Landesdirektion Sachsen ) . Before starting the behavioral experiments , every mouse was visibly marked at the tail . To simplify handling , during the morning of every test session enriched animals were placed into standard cages in groups of five , which remained consistent throughout testing , and returned into the enriched environment cage in the evening . Mice were tested in the same order in all behavioral tasks . The sequence of the behavioral experiments is shown in Figure 3A . Mice were assessed for locomotor abilities using an Economex Rotarod from Columbus Instruments . The rotating cylinder started with a speed of 4 rpm and accelerated by 0 . 1 rpm . At a final speed of 34 rpm and a maximum time of five minutes , the test was stopped manually . The trial was completed when an animal fell off or reached the maximum duration . The mice were trained on three consecutive days with three trials per day . The rotarod was cleaned after every session . The open field ( OF ) enclosure consisted of a 120 × 120 cm square apparatus subdivided into four identical arenas of 60 × 60 cm , allowing for the simultaneous testing of four mice in the apparatus . The 40 cm high white plywood walls were marked with a green tape on the intersections to provide additional spatial clues . The only light source in the room , a 100 watts light bulb , was installed 1 . 5 m above the intersection of the middle walls , next to the camera ( Logitech ) . Paths were recorded using EthoVision software ( Noldus ) . Mice were placed in the middle of the empty arena and were allowed to explore the arena freely for 5 min in each trial . A total of two trials were performed on two consecutive days . Roaming entropy ( RE ) , a measure of territorial coverage , was calculated according to Freund et al . , 2013Freund et al . , 2013 . Each arena was divided into 10 × 10 subfields . The probability pi of a mouse being in a subfield i was estimated as a proportion of trial time spent in that subfield . Shannon entropy of the roaming distribution was then calculated as:RE=-∑i=1k ( pilog⁡pi ) /log⁡kwhere k is the number of subfields in the arena ( k = 100 ) . Dividing the entropy by the factor log ( k ) scales the RE to the range from zero to one . RE is minimal for the mice that stay in one place and maximal for the mice that spent equal amount of time in each subfield of the arena . Data from eight CTRL animals were lost in the second trial . The two OF trials were considered to serve as habituation for the NOR task ( Figure 3A ) . For this task , the same arenas were equipped with two of three different objects: object A was a 1 . 5 cm high blue cylinder with a diameter of 3 . 5 cm , object B was a black box of 8 . 5 × 9 . 5 × 2 cm , and object C was 4 . 5 cm long and transparent with a more complex geometric shape ( Figure 3—figure supplement 1 ) . All objects were made of plastic . For object placement in subsequent trials , see Figure 3B . On day 1 , following the OF trial , mice were presented with objects A and B . On day 2 , the animals were first exposed to the same objects and then in the following trial object A was replaced with object C . The same combination of objects was presented on day 3 , followed by a trial in which object B was moved into the adjacent quarter of the arena . Each trial lasted 2 min . Discrimination index was calculated for trial 3 on the basis of the exploration time for the new and old object as follows: DI = ( new object – old object ) / ( new object + old object ) , and ranged from −1 ( preference for the old object ) to 1 ( preference for the new object ) , while 0 indicated no preference ( Miyauchi et al . , 2016 ) . Eleven ENR and three CTRL mice that did not explore object A in any of the first two trials or that did not explore any object in trial 3 were excluded from the calculation of DI . Two days after the last behavioral experiment was performed , the mice were deeply anesthetized with a mixture of ketamine and xylazine and transcardially perfused with 0 . 9% NaCl . Directly after the perfusion , the liver , heart and adrenal glands were harvested and weighed . Brains were removed from the skull and postfixed in 4% paraformaldehyde overnight at 4°C and equilibrated with 30% sucrose in phosphate buffered saline ( PBS ) . For immunohistochemistry , brains were cut into 40 µm coronal sections using a sliding microtome ( Leica , SM2000R ) and stored at 4°C in cryoprotectant solution ( 25% ethyleneglycol , 25% glycerol in 0 . 1 M phosphate buffer , pH 7 . 4 ) . For detection of BrdU- , Ki67- and synaptoporin-positive cells , immunohistochemistry was performed using the peroxidase method as previously described ( Steiner et al . , 2008 ) . Briefly , free-floating sections were incubated in 0 . 6% H2O2 for 30 min to inhibit endogenous peroxidase activity . After washing , non-specific antibody-binding sites were blocked using 10% donkey serum and 0 . 2% Triton-X100 in Tris buffered saline ( TBS ) for 1 hr at room temperature . For BrdU detection , prior to blocking , sections were incubated in pre-warmed 2 . 5 M HCl for 30 min at 37°C , followed by extensive washes . Primary antibodies were applied overnight at 4°C as follows: monoclonal rat anti-BrdU ( 1:500 , Serotec ) , rabbit anti-Ki67 ( Novocastra , 1:500 ) , and rabbit anti-Synaptoporin ( Synaptic Systems , 1:500 ) . Sections were incubated with biotinylated secondary antibodies for 2 hr at room temperature ( 1:500 , Dianova ) . Primary and secondary antibodies were diluted in TBS supplemented with 3% donkey serum and 0 . 2% Triton-X100 . Detection was performed using the Vectastain ABC-Elite reagent ( 9 μg/ml of each component , Vector Laboratories , LINARIS ) with diaminobenzidine ( 0 . 075 mg/ml; Sigma ) and 0 . 04% nickel chloride as a chromogen . All washing steps were performed in TBS . BrdU- and Ki67-stained sections were mounted onto glass slides , cleared with Neo-Clear ( Millipore ) and cover-slipped using Neo-Mount ( Millipore ) . BrdU- and Ki67-positive cells were counted , by applying the simplified version of the optical fractionator principle as previously described ( Kempermann et al . , 1997 ) on every sixth section along the entire rostro-caudal axis of the dentate gyrus , using a brightfield microscope ( Leica DM 750 ) . Synaptoporin-stained sections underwent a Nissl-staining before mounting them with Entellan ( Merck ) . To prepare sections for Nissl staining , they were incubated for 20 min in each of the following solutions: staining buffer ( 4% sodium acetate , 0 . 96% acetic acid ) , followed by permeabilization solution ( 75% ethanol , 0 . 025% Triton-X100 ) and staining buffer . Staining solution ( 0 . 1% cresyl violet in staining buffer ) was applied for 20 min followed by differentiation of sections in 95% ethanol for 30 s and dehydration with isopropanol and xylene for 10 min each . Immunofluorescent staining was performed for co-labeling of BrdU-positive cells with NeuN and S100β as described ( Steiner et al . , 2008 ) . Briefly , sections were treated with 2 M HCl , washed extensively with PBS and blocked in PBS supplemented with 10% donkey serum and 0 . 2% Triton-X100 for 1 hr at room temperature , followed by incubation with primary antibodies overnight at 4°C ( rat anti-BrdU 1:500 , Serotec; mouse anti-NeuN 1:100 , Merck Millipore; and rabbit anti-S100β 1:200 , Abcam ) . Secondary antibodies were incubated for 4 hr at room temperature ( anti-rat Alexa 488 1:500; anti-mouse Cy5 1:500; and anti-rabbit Cy3 1:500; all from Jackson ImmunoResearch ) . Nuclei were counterstained using 4′ , 6-diamidino-2-phenylindole ( DAPI; 3 . 3 µg/ml ) for 10 min . All washing steps were performed in PBS . Sections were mounted onto glass slides and cover-slipped using Aqua-Poly/Mount ( Polysciences , Inc . ) . Imaging was performed with the ZEISS Apotome and the Software AxioVision software with optical sectioning mode . To determine total numbers of new-born neurons and astrocytes , 100 randomly selected BrdU immuno-positive cells along the rostro-caudal axis of the dentate gyrus were investigated for co-expression with NeuN or S100β . The final numbers of surviving new neurons and astrocytes were obtained by multiplying the total number of BrdU-positive cells ( as determined by peroxidase-based immunohistochemistry ) by the ratio of NeuN/BrdU-positive cells and S100β/BrdU-positive cells . The mossy fiber ( MF ) projections are characterized by a high content of the presynaptic vesicle protein synaptoporin ( Krebs et al . , 2011; Singec et al . , 2002 ) , therefore the volumes of the MF projections were estimated on sections immunolabeled against synaptoporin and counterstained with Nissl for a better distinction between neuronal cell layers . Volumetric analysis was performed on every sixth section with a semiautomated morphometric system consisting of a CCD camera ( Hitachi ) connected to a light microscope ( Leica DM-RXE ) using a 10x objective and the Stereoinvestigator 7 software ( MBF Bioscience ) . Structures were overlaid with the Cavalieri estimator probe grid of 25 µm and every grid point belonging to the particular structure of interest was selected . Volume estimates were calculated in the software taking into account the sampling interval ( 240 µm ) and the section thickness ( 40 µm ) . For the analysis of the cortex thickness , the areas of motor , entorhinal and cingulate cortices were defined as described by Diamond et al . ( Diamond et al . , 1964; Diamond et al . , 1985 ) . We used the following coordinates of bregma: motor cortex −1 . 06 to −1 . 46 , entorhinal cortex −2 . 30 to −2 . 80 , cingulate cortex 1 . 34 to 0 . 50 . Two to three constitutive sections from each animal were analyzed . Sections were scanned with a slide scanner ( Axio Scan . Z1 , Zeiss , Germany ) and measured using the ZEN blue software ( Zeiss , Germany ) . Sections from several animals had to be excluded because of insufficient staining quality or damage to the tissue in the respective areas: hippocampus volumetry , 1 CTRL , 1 ENR mouse; motor cortex , 2 CTRL , 2 ENR mice; entorhinal cortex , 1 CTRL , 2 ENR mice; cingulate cortex , 3 CTRL mice . Blood was collected into EDTA-coated tubes ( Sarstedt ) from the abdominal cavity during the perfusion immediately after the right ventricle was opened . Blood samples were incubated for 1 to 2 hr at room temperature , and centrifuged at 2000 × g for 15 min at 4°C . Plasma was centrifuged a second time and stored at −80°C . Plasma samples were assayed for glucose ( Amplex red glucose/glucose oxidase assay kit , Invitrogen ) , cholesterol ( Amplex red cholesterol assay kit , Invitrogen ) , triglycerides ( Triglycerides colorimetric quantification kit , Abcam ) and corticosterone ( Corticosterone ELISA kit , Enzo ) following the manufacturers’ instructions . Log-logistic concentration curves were calculated from standards in R using the drm function from the drc package ( Ritz et al . , 2015 ) . Corticosterone and triglyceride measures were log-transformed to normality . All experiments were carried out with the experimenter blind to the experimental group . The data from this study have been deposited at Dryad ( Körholz et al . , 2018 ) . Statistical analyses were carried out using the statistical software R ( R Core Team , 2014 ) . Data were tested for normality using the Shapiro-Wilk-test . For normally distributed measures , we used Welch’s t-test to compare means and F-test to test for equality of variance between groups . For repeated measures ( longitudinal data ) , a linear mixed regression was performed using the lmer function from the lme4 package ( Bates et al . , 2015 ) , and p-values were obtained by the likelihood ratio test of the full model against the model without the analyzed effects . For non-normal data , we performed the Wilcoxon rank sum test using the function wilcox . test as a non-parametric equivalent for the t-test , or the Brown-Forsythe test using the leveneTest function from the car package with the parameter center set to median as a more robust form of Levene’s test to compare the variances between groups . Longitudinal non-normal or heteroscedastic data were analyzed using a rank-based non-parametric test using the nparLD function from the nparLD package , which reports a Wald-type test statistic for each of the effects and their interactions ( Noguchi et al . , 2012 ) . All tests were two-tailed and differences were considered to be statistically significant at a p<0 . 05 . Data were visualized using the ggplot2 package ( Wickham , 2011 ) . In the box-whisker plots , center line and plus sign mark the median and mean , respectively . Upper and lower hinges indicate first and third quartiles . The upper whisker extends from the hinge to the largest value no more than 1 . 5 times the interquartile range ( IQR , a distance between the first and third quartiles ) ; the lower whisker extends from the hinge to the smallest value at most 1 . 5 times IQR . Full results of statistical tests are available in Supplementary file 2 .	Even identical twins who share genetics and the same environment develop individual traits . But how individuality emerges and the biological mechanisms behind it are not clear . It is hard to study people for a long time , so scientists turn to animal studies to answer such questions . One way to study the respective effects of genes and the environment is to study differences in genetically identical mice housed in either small cages with few animals and little to do , or in larger cages with toys and more animals . Comparing how these different environments affect individual animals and their biology may help scientists understand individuality . If individual traits emerge in groups of genetically identical animals housed in the same environment it is likely the result of the individual animal's behaviors or unique experiences . It might also be due to chance . Learning more about the biological processes that underlie individuality may help doctors better match therapies to individuals . It may also help scientists design better studies and help them avoid errors caused by individual variations between animals . Now , Körholz , Zocher , Grzyb et al . show that living in an enriched environment increases mouse individuality in certain brain and behavioral traits . Other biological traits , like metabolism , did not differ much between the animals in the enriched environment . In the experiments , genetically identical mice housed in either normal laboratory conditions or enriched environments underwent a series of behavioral and biological tests . The mice housed in more interesting environments showed greater variability in how they responded to behavioral tests that exposed them to a new object or an open space than their typically housed peers . There were also more differences in the number of newborn brain cells in the mice living in enriched housing . These findings may have very important implications for researchers , which could help scientists to better understand how individual behaviors or experiences may affect healthy aging and resilience to disease . Many researchers are also trying to improve the wellbeing of laboratory animals by housing them in more interesting environments . More studies using experiments like those conducted by Körholz et al . may help them understand how enriched animal housing may affect their experiments' results .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Selective increases in inter-individual variability in response to environmental enrichment in female mice
With experience we become accustomed to the types of environments that we normally encounter as we navigate in the world . But how does this fundamental knowledge develop in the first place and what brain regions are involved ? To examine de novo environmental learning , we created an ‘alien’ virtual reality world populated with landmarks of which participants had no prior experience . They learned about this environment by moving within it during functional MRI ( fMRI ) scanning while we tracked their evolving knowledge . Retrosplenial cortex ( RSC ) played a central and highly selective role by representing only the most stable , permanent features in this world . Subsequently , increased coupling was noted between RSC and hippocampus , with hippocampus then expressing knowledge of permanent landmark locations and overall environmental layout . Studying how environmental representations emerge from scratch provided a new window into the information processing underpinning the brain's navigation system , highlighting the key influence of the RSC . We continually encounter new environments and to operate effectively within them we must be able to form dependable representations of these surroundings . Numerous brain areas , including the hippocampus ( O'Keefe and Nadel , 1978; Rosenbaum et al . , 2004; Spiers and Maguire , 2006; Bird and Burgess , 2008; Manns and Eichenbaum , 2009; Howard and Eichenbaum , 2014 ) , entorhinal ( Hafting et al . , 2005; Sargolini et al . , 2006; Doeller et al . , 2010; Chadwick et al . , 2015 ) , parahippocampal ( PHC; Janzen and van Turennout , 2004; Kravitz et al . , 2011; Mullally and Maguire , 2011; Epstein and Vass , 2014 ) , retrosplenial ( RSC; Chen et al . , 1994; Cho and Sharp , 2001; Vann et al . , 2009; Auger et al . , 2012; Auger and Maguire , 2013 ) and posterior parietal ( DiMattia and Kesner , 1988; Husain and Nachev , 2007 ) cortices have been implicated in representing space and facilitating navigation therein , with generally concordant findings across species . We still lack , however , a precise account of how an environmental representation evolves de novo and the roles played by space-sensitive brain regions in this process . It is surprising that such a gap in our knowledge exists given that learning new environments is ubiquitous and relevant for independent living and even survival . Moreover , examining the genesis of environmental representations would seem an ideal means of leveraging our understanding of the specific roles played by relevant brain areas , when they come online , what they respond to , and how and when regions interact with each other to support the emerging representation . In recent years , there has been a move to study how neuronal responses in the medial temporal lobes develop in rodent pups when they interact with the world for the first time ( Langston et al . , 2010; Wills et al . , 2010 ) . This exciting research is only just starting to provide clues about when specific brain areas become functional and in response to what , leaving much still to learn about the neural development of spatial cognition and memory ( Mullally and Maguire , 2014; Wills and Cacucci , 2014 ) . In adult humans it is possible to study how multiple brain areas are engaged during navigation in large-scale environments by having subjects navigate in virtual reality ( VR ) while being scanned using techniques such as functional MRI ( fMRI ) . In the majority of these experiments subjects become familiarised with an environment before scanning and then typically perform tasks during scanning based on the environmental representation they formed pre-scan ( e . g . , Janzen and van Turennout , 2004; Spiers and Maguire , 2006 ) . A smaller number of VR scanning studies have focused on the acquisition phase of environmental knowledge ( e . g . , Wolbers et al . , 2004; Wolbers and Buchel , 2005; Iaria et al . , 2007; Baumann et al . , 2010 ) . However , as far as we are aware , in every case the environments contained landmarks and structures that were readily recognisable and nameable ( e . g . , shops and houses ) . Thus , the neural substrates of learning about an environment from scratch , with no prior experience of , or knowledge about , key elements within it , has never been examined in humans . How children and adult humans learn about new environments has been studied extensively in cognitive and environmental psychology . Prominent features in an environment , namely landmarks , have been posited to play a fundamental role ( Tolman , 1948; Lynch , 1960; Siegel and White , 1975; Downs and Stea , 1977; Golledge , 1991; Lew , 2011 ) . It has been further suggested that encoding landmarks facilitates the development of route knowledge , and learning how routes relate to each other then provides the navigator with a ‘survey’ representation of an environment ( Siegel and White , 1975; Epstein and Vass , 2014 ) . Despite landmarks being heavily implicated in many neuroscientific experiments in animals and humans over the decades , the effect of landmark properties on navigation has only recently been studied . Auger et al . ( 2012 ) ; ( see also Mullally and Maguire , 2011; Konkle and Oliva , 2012; Auger and Maguire , 2013 ) using fMRI found that PHC responded to visuospatial features of landmarks . By contrast RSC was engaged only when subjects viewed landmarks that had a permanent location and never moved ( a finding that has since been replicated—Marchette et al . , 2014; Troiani et al . , 2014 ) . This suggests that a function of the RSC may be to code for permanent landmarks on which a stable spatial representation of an environment can then be built . But , as with the work mentioned above , the landmarks examined in these studies were of a type already familiar to subjects . Key questions therefore remain . How does knowledge of landmark features , including permanence , but also properties such as size and visual salience , evolve de novo during environmental learning where no prior semantic knowledge exists about landmarks ? What brain areas support this learning , at what point do they come online , and to what aspects of the environment do they respond ? Furthermore , how might this information be used in building an overall environmental representation ? We addressed these questions in the current study . To do so we developed a new VR environment that subjects learnt during repeated exposures while undergoing fMRI scanning , and their accruing knowledge was tested during and after scanning . The environment was populated by entirely novel , ‘alien’ landmarks ( Figure 1A ) about which subjects had no pre-conceived ideas . These were located along different paths ( Figure 1B; ‘Materials and methods’ ) . On each path , half the landmarks were permanent , each remaining fixed in a single place , while the rest were transient and changed location on every exposure . The landmarks were developed and characterised in an initial experiment with separate subjects ( detailed in the ‘Materials and methods’ ) ensuring that the permanent and transient landmark groups were matched in terms of visual salience , how well they could be remembered , as well as other features . 10 . 7554/eLife . 09031 . 003Figure 1 . The virtual reality environment ‘Fog World’ . ( A ) Examples of the ‘alien’ landmarks . ( B ) Landmarks positioned within the virtual world . ( C ) An overhead perspective of the environment showing the five different coloured , intersecting paths—note this aerial view was never seen by participants during learning . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 003 Before scanning , subjects were instructed to learn the layout of the environment and told that they would be tested in a variety of ways after scanning without the specific nature of those tasks being revealed . They were informed that some of the landmarks would always remain in the same location whereas others would appear in a different place every time they saw them . The world contained five different coloured intersecting straight paths ( yellow , red , grey , blue and green; Figure 1C ) . Each path had 12 landmarks ( six permanent , six transient ) evenly distributed alongside it ( Figure 1B ) . While undergoing fMRI scanning , subjects learned the layout of the environment and its landmarks by viewing first person perspective videos travelling along each of the five paths , one at a time . Each trial consisted of a single journey along one of the paths and at the end of a video subjects were immediately shown the next learning trial on a different path . In these videos , the environment was covered in a shroud of fog to restrict the field of view thus ensuring we had complete control over the exposure subjects had to each landmark ( Figure 2A ) —hence we refer to the environment as ‘Fog World’ ( see Video 1 ) . 10 . 7554/eLife . 09031 . 004Figure 2 . The Experimental paradigm . While undergoing functional MRI ( fMRI ) scanning , subjects were presented with videos travelling along the various paths . ( A ) An example sequence of video frames with a landmark emerging through the fog , the camera turning towards it before returning back to the middle of the path—see also Video 1 . ( B ) After viewing videos of each of the five paths once , subjects answered a series of questions about individual landmarks to test their learning throughout the experiment . A learning ‘sweep’ consisted of one round of videos of the five paths and the questioning period which followed . There were 12 learning sweeps . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 00410 . 7554/eLife . 09031 . 005Video 1 . This is a short clip from one of the videos which subjects viewed inside the MRI scanner when learning the environment . It demonstrates the first-person perspective presented to subjects and shows how , when a landmark emerges through the fog , the camera turns to bring it into the centre of view whilst continuing along the path . It also provides an example of what happens at an intersection . See also Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 005 When all five paths had been travelled once , there came a questioning period to gauge how much information subjects had learned by that point in the experiment ( Figure 2B ) . In these questioning periods , participants were first shown an image of a single landmark displayed , in isolation , on a plain grey background for 2 s . They were then asked whether or not they remembered the landmark from the environment ( ‘Have you seen this item in the environment ? ’ , Yes/No ) . If they remembered seeing it , they were then asked about its permanence ( ‘How many locations in the environment have you seen it in ? ’ , Only 1/More than 1 ) , before being questioned about another landmark . Within each questioning period , subjects were asked about 13 landmarks: five permanent , five transient and three previously unseen . The combination of a 13 landmark questioning period and videos of the five different paths preceding it are referred to as a learning ‘sweep’ . There were three such sweeps in each scanning run ( or quarter ) and four runs , so in total 12 learning sweeps . Once out of the scanner after learning had concluded , subjects' knowledge of Fog World landmarks ( recognition memory or ‘memorableness’ , permanence , visual salience and size—see ‘Materials and methods’ ) , as well as their ability to volitionally navigate within Fog World were assessed . We elected to show participants videos of movement through the environment during scanning rather than have them navigate volitionally because this allowed us to control exactly what they saw and ensured that all participants had the same learning experience . Participants knew that they should pay close attention to learning the environment and its layout for testing post-scan . Similar to previous experiments ( e . g . , Janzen and van Turennout , 2004; Wolbers and Buchel , 2005; Doeller et al . , 2007; Schinazi and Epstein , 2010; Auger et al . , 2012; Konkle and Oliva , 2012; Auger and Maguire , 2013; Troiani et al . , 2014; Chadwick et al . , 2015 ) , we compared fMRI responses while subjects viewed images of individual , isolated landmarks displayed during the questioning periods at the end of each sweep , unless otherwise stated . Using this time period , rather than when landmarks were viewed during the navigation videos , removed potential problems associated with visual confounds ( e . g . , path colour ) and the more unconstrained neural responses that may have been associated with the minute long learning videos . During scanning , recognition memory for the landmarks was assessed in the questioning periods at the end of each learning sweep . We first wanted to establish whether or not subjects had learned to recognise the two types of landmark equally well . To do this , we performed separate linear regression analyses for permanent and transient landmarks to assess how the accuracy with which subjects recognised them changed throughout the learning phase in the scanner . We then directly compared the slopes and found that there was no difference in the rate at which subjects learned to recognise permanent and transient landmarks ( mean difference in rate = 0 . 0084 , SD = 0 . 063; t31 = 0 . 763 , p = 0 . 45 ) . We then examined recognition accuracy in each of the four learning quarters and observed a significant time ( quarters ) by landmark type ( permanent , transient ) interaction ( F3 , 29 = 8 . 045 , p = 0 . 0005 ) . Interrogating this result further , we found that that participants recognised permanent and transient landmarks equally in the first quarter ( mean accuracy permanent landmarks = 57 . 7% ( SD 9 . 7 ) ; transient landmarks = 58 . 1% ( SD 17 . 7 ) ; t31 = 0 . 12 , p = 0 . 9 ) and final quarter ( permanent landmarks = 79 . 6% ( SD 18 . 3 ) ; transient landmarks = 77 . 3% ( SD 14 ) ; t31 = 0 . 55 , p = 0 . 6 ) . In the second quarter ( permanent landmarks = 72 . 7% ( SD 14 . 4 ) ; transient landmarks = 60 . 4% ( SD 13 . 1 ) ; t31 = 3 . 517 , p = 0 . 001 ) and third quarter ( permanent landmarks = 80 . 0% ( SD 15 . 6 ) ; transient landmarks = 70 . 4% ( SD 16 . 3 ) ; t31 = 2 . 361 , p = 0 . 02 ) , however , participants were more accurate in their recognition of permanent than transient landmarks . Consistent with the result in the final quarter of learning , in the post-scan testing phase , there was no difference in how well subjects recognised permanent or transient landmarks ( ‘memorableness’: permanent mean accuracy = 82 . 9% ( SD 4 . 9 ) ; transient mean accuracy = 76 . 3% ( SD 4 . 4 ) ; t31 = 1 . 745 , p = 0 . 09 ) . Subjects were also accurate at identifying as novel landmarks which they had not seen before ( mean = 93 . 0% , SD 2 . 3 ) . Post-scan , subjects also rated other features of the landmarks . These included whether they thought an item was permanent or transient , how visually salient they found them , and finally the size that they were in Fog World ( see ‘Materials and methods’ ) . We compared these ratings with the corresponding actual values of permanence and size , and the salience scores from the separate initial landmark characterisation study ( see ‘Materials and methods’ ) in order to test the validity of the scan subjects' ratings and to confirm whether or not subjects had successfully learned about the landmarks ( full details in Table 1 ) . 10 . 7554/eLife . 09031 . 006Table 1 . Correlations between features of the 60 ‘alien’ landmarksDOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 006Permanence: actualPermanence: post-scanSalience: beh'al studySalience: post-scanSize: actualSize: post-scanPermanence: actual1 . 000–––––––––––Permanence: post-scan0 . 793†1 . 000––––<0 . 0001–––––Salience: beh'al study0 . 087−0 . 0011 . 000–––0 . 51 . 0––––Salience: post-scan0 . 3150 . 3250 . 3141 . 000––0 . 010 . 010 . 02–––Size: actual0 . 0000 . 0680 . 0880 . 428†1 . 000–1 . 0000 . 60 . 50 . 001––Size: post-scan0 . 1170 . 0930 . 1290 . 749†0 . 726†1 . 0000 . 40 . 50 . 3<0 . 0001<0 . 0001–Beh'al = ratings that came from the initial behavioural landmark characterisation study . Correlations are shown between: mean salience scores from the initial characterisation study , the actual size and permanence of landmarks in Fog World , and ratings of permanence , salience and size from the fMRI subjects post-fMRI scan . Each cell shows the Pearson correlation r value above the corresponding p value . Significant correlations are highlighted in bold text . Correlation is significant at the 0 . 05 level ( 2-tailed ) . †Correlation is significant at the 0 . 01 level ( 2-tailed ) . Permanence ratings made post-fMRI scan were strongly correlated with the actual values ( r = 0 . 793 , p < 0 . 0001 ) , indicating that subjects had successfully learned this information . Similarly , the size ratings in the post-scan session were significantly correlated with the actual landmarks sizes in Fog World ( r = 0 . 726 , p < 0 . 0001 ) . Comparing the visual salience ratings from the initial landmark characterisation study and the fMRI study was particularly interesting . While the correlation between the two was significant ( p = 0 . 02 ) , the slope of the correlation was not particularly marked ( r = 0 . 314 ) . The landmarks in the characterisation experiment were viewed one at a time and in isolation ( so not as part of Fog World ) . By contrast , there was a tendency for subjects post-fMRI scan to rate landmarks as more salient if they had been experienced in Fog World to be large ( r = 0 . 428 , p = 0 . 001 ) or permanent ( r = 0 . 315 , p = 0 . 01 ) . In other words , the visual salience of landmarks ( or how ‘attention grabbing’ they were ) was not just an inherent property; it was also influenced by how and where they had been experienced within the environment . Because we had a number of separate measures of landmark features , we then sought to establish if some of these variables loaded onto common underlying components . We therefore submitted the ratings and scores of permanence , size and salience of landmarks made by the scanning participants along with their memorableness scores and the actual permanence of landmarks to a principal components factor analysis using a varimax rotation and Kaiser normalization ( see ‘Materials and methods’ ) . The features clearly separated onto four orthogonal factors which accounted for 96 . 4% of the variance in the data . These four factors were strongly related to the permanence , memorableness , size and salience of the landmarks ( Figure 3A ) . Thus the factor analysis confirmed the presence of four independent components in the landmark features , which included permanence of landmarks as a distinct factor . 10 . 7554/eLife . 09031 . 007Figure 3 . Changes in the brain regions engaged by different landmark features over the course of learning . ( A ) The loading values of each landmark feature to the four principal component factors . Values above 0 . 5 are highlighted in bold . Factor 1 was strongly related to landmark permanence , factor 2 to their memorableness , factor 3 to their size and factor 4 to the visual salience of landmarks . ( B ) The bar graphs to the left show how strongly each of the four factors was related to the various features rated by subjects in the post-scan debrief . The associated brain regions responding to these four factors in the first and last quarters of learning are shown to the right . All activations are shown on a structural MRI brain scan of single representative subject . Each factor's activations are shown on the same sagittal slice and using a whole brain uncorrected threshold of p < 0 . 00001 for display purposes . The colour bars indicate the Z-score associated with each voxel . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 007 The post-scan testing session also involved subjects volitionally navigating to the locations of target landmarks within Fog World . This was a challenging test given that subjects had only been exposed to the environment ( with its five paths and 60 landmarks ) for the previous 40 minutes or so . Participants were first shown an image of a landmark and instructed that they would have to navigate to where they thought it was located in the environment by as direct a route as possible . On each trial , subjects were placed within a version of the environment in which there was no fog and the target landmark had been removed . They moved their way to where they thought that landmark belonged ( using the arrow keys on a keyboard ) and then indicated their chosen location by pressing the space bar . There were 12 trials ( nine involving permanent and three involving transient landmarks ) . If they thought the target landmark was transient ( and so could not be placed in a single location ) , subjects were instructed to press the space bar and indicate that they thought it was transient . For the navigation task , each trial was scored out of 3 , giving a maximum score of 36 . One point was awarded for locating a landmark on the correct path , 1 point for the correct part and side of the path , and a final point was awarded if they had taken a direct route to the landmark . If they correctly identified that the target landmark was transient ( and so could not be located in a single position ) , they were awarded 3 points . As we expected and hoped , there was a good deal of variance in navigation task performance ( mean score = 12 . 8 , SD = 8 . 1 ) . We reasoned that having such variation between individuals enabled us to examine more meaningful relationships between fMRI responses and the learning of landmark permanence . If a brain region's response was directly linked to learning of landmark permanence , one would expect activity within such a region to directly track the acquisition of this knowledge . It was therefore important to have variance in the amount of learning by subjects in order to capture such a relationship ( see ‘Results’ section , fMRI: accounting for subject-specific learning differences ) . Participants performed marginally better at identifying the transient landmarks , however the scores for the transient landmarks still only constituted an average of 38 . 8% ( SD 27 . 3 ) of the total marks ( from 25% of the trials ) . This shows that the transient landmark trials did not have a disproportionately large effect on the overall results . The distribution of scores ( out of the total marks ) on the post-scan navigation task was as follows: identifying the path 33 . 8% ( SD 16 . 6 ) , identifying the part and side of path 38 . 1% ( SD 16 . 5 ) , and using a direct route 28 . 0% ( SD 15 . 8 ) . This suggests that participants were not merely making simple path colour to landmark associations , and in fact identifying the part and side of path was where more marks were scored ( part vs path: t31 = −2 . 121 , p = 0 . 042 ) . Finally , in the post-scan debriefing session we also asked participants what they had been thinking throughout the navigation videos and how they approached learning the layout of the environment . This revealed a great deal of variety in what participants were thinking . Some tended to focus on visual characteristics of the landmarks , others paid greater attention to the overall layout of the environment's paths or to individual landmarks' locations along paths . Sometimes they were thinking about the landmark in view , but on other occasions were considering what came before , or what landmark might be next . Only two out of thirty two participants made reference to specifically focussing on the permanence of the landmarks . The variance in thought revealed by this feedback was not just large between participants but also within participants trial-to-trial making it impossible to model the fMRI time series during the videos in an accurate or meaningful way . This supported our decision to design the experiment such that the fMRI analyses focused on the inter-sweep questioning periods where landmarks were presented individually and devoid of environmental context , and may be why so many previous fMRI studies have adopted a similar approach ( e . g . , Janzen and van Turennout , 2004; Wolbers and Buchel , 2005; Doeller et al . , 2007; Schinazi and Epstein , 2010; Auger et al . , 2012; Konkle and Oliva , 2012; Auger and Maguire , 2013; Troiani et al . , 2014; Chadwick et al . , 2015 ) . In summary , the behavioural data showed that participants learned the basic identities of landmarks . There was a brief discrepancy in learning to recognise permanent and transient landmarks in the second and third quarters in favour of the permanent landmarks , but overall the rates of learning were similar . Participants possessed excellent knowledge of landmark permanence/transience by the end of the scanning , as well as landmark size and visual salience . They also demonstrated some knowledge , with variance across subjects , about the overall layout of Fog World , in the volitional navigation task . We next asked what underpinned this learning in the brain . We first examined the fMRI data by directly comparing permanent with transient landmarks across the whole scanning experiment . There were numerous large clusters of increased activity more for permanent than transient landmarks: one including left RSC and PHC ( −21 , −49 , −8 , z = 6 . 01; −12 , −43 , 1 , z = 5 . 72 ) , another including right RSC and PHC ( 9 , −52 , 4 , z = 5 . 10; 30 , −49 , −5 , z = 5 . 14 ) , and others in left occipital cortex ( −15 , −88 , 25 , z = 5 . 78; −9 , −85 , 4 , z = 4 . 95 ) , left and right superior posterior parieto-occipital sulcus ( POS; −3 , −76 , 40 , z = 5 . 24; 15 , −61 , 19 , z = 4 . 94 ) as well as left lateral temporal cortex ( −51 , −43 , 4 , z = 5 . 61 ) , right occipital cortex ( 18 , −88 , 22 , z = 5 . 50; 27 , −82 , 31 ) and posterior parietal cortex/precuneus ( 0 , −34 , 46 , z = 4 . 89 ) . No regions were more active for transient than permanent landmarks . Having established the brain areas that were more active overall for permanent landmarks compared to transient , we then investigated when these differences arose . By the final quarter of learning ( the last three learning sweeps ) there were significantly greater responses to the permanent landmarks compared to the transient in right ( 6 , −53 , 5; Z = 5 . 41 ) and left ( −6 , −55 , 10; Z = 5 . 90 ) RSC , as well as right POS ( 9 , −73 , 31; Z = 5 . 01 ) and posteriorly in the left occipital lobe ( −6 , −79 , −8; Z = 5 . 00 ) ( Figure 4 ) . There were also activations in the hippocampus ( −21 , −28 , −11; Z = 3 . 71 ) and PHC ( 21 , −37 , −14; Z = 4 . 12 ) but at a reduced threshold ( p < 0 . 0001 uncorrected; compared with the whole-brain FWE corrected p < 0 . 05 reported otherwise ) . The increased responses to permanent landmarks were even present as early as the third quarter of scanning ( sweeps 7–9 of 12 ) in similar regions ( right RSC: 12 , −51 , 3; Z = 5 . 31; left RSC: −12 , −55 , 6; Z = 5 . 72; left POS: −6 , −76 , 40; Z = 4 . 93; left occipital: −15 , −76 , −11; Z = 5 . 53 ) , but not in the hippocampus or PHC . There were no differences between responses to permanent and transient landmarks in either of the first two quarters of learning . No regions were more active for transient than permanent landmarks . 10 . 7554/eLife . 09031 . 008Figure 4 . Brain regions more engaged by permanent than transient landmarks by the end of learning . ( A ) Shows activations in retrosplenial cortex ( RSC ) and posterior parieto-occipital sulcus ( POS ) at the default threshold of p < 0 . 05 ( FWE ) . The colour bar indicates Z-score associated with each voxel . ( B ) Shows a plot of mean blood oxygenation level-dependent ( BOLD ) responses ( ±1 SEM ) within the RSC cluster ( circled in green ) . In the first two quarters of scanning , responses to permanent ( blue ) and transient ( red ) landmarks did not differ , but as subjects learned landmark permanence , BOLD responses increased for permanent landmarks with a corresponding decrease for transient landmarks . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 008 We next broadened the analyses to also include the other landmark features by considering how fMRI blood oxygenation level-dependent ( BOLD ) signals related to each of the four principal components from the factor analysis . We examined whether or not they had any identifiable neuronal correlates , and if so , how they might have evolved over the course of learning . We used the principal component scores from the factor analysis rather than the raw behavioural data because this allowed us to have fully orthogonalized regressors in the fMRI analyses . To do this , we created factor score estimates for every landmark corresponding to each of the four orthogonal principal components and then used these values to generate parametric regressors for a whole brain fMRI analysis ( see ‘Materials and methods’ ) . This enabled us to examine activity that was linearly modulated by each factor . Considering first the permanence-related factor ( factor 1 ) , namely fMRI responses associated with increasing values of this factor across the entire scanning session , the results were in line with the categorical contrast reported above: left lateral temporal cortex ( −60 , −40 , 1; z = 6 . 14 ) , left occipital cortex ( −15 , −91 , 25 , z = 5 . 86; −9 , −64 , −2 , z = 4 . 94; −15 , −73 , −8 , z = 4 . 86; −12 , −31 , −5 , z = 4 . 66; −9 , −85 , 4 , z = 4 . 64 ) , left PHC ( −21 , −49 , −5 , z = 5 . 29 ) , left RSC ( −9 , −43 , 1 , z = 5 . 03; −9 , −52 , 10 , z = 4 . 61 ) , right RSC ( 9 , −49 , 4 , z = 4 . 90 ) and posterior parietal cortex/precuneus ( 0 , −58 , 43 , z = 5 . 01 ) . No areas showed responses associated with decreasing values of this permanence factor . As before , having established the brain areas that were associated with increasing values of the permanence-related factor over the whole scan experiment , we then investigated when these differences arose . Increasing values of the permanence-related factor were associated with significant activations in bilateral RSC ( left: −12 , −55 , 7; Z = 5 . 60; right: 9 , −52 , 4; Z = 5 . 12 ) , as well as POS ( left: −9 , −91 , 25; Z = 5 . 59; right: 9 , −73 , 31; Z = 5 . 54 ) in the final quarter of learning , but only right RSC in the third quarter ( 12 , −48 , 1; Z = 5 . 03 ) and no regions in either of the first 2 quarters ( top of Figure 3B , in blue ) . Once again there were also increased responses in the left hippocampus ( −18 , −28 , −11; Z = 3 . 99 ) and right PHC ( 24 , −37 , −14; Z = 4 . 24 ) but at a reduced threshold ( p < 0 . 0001 uncorrected ) in the last quarter of scanning . No areas showed responses associated with decreasing values of this permanence factor . We then considered factor 2 , the ‘memorableness’ factor , namely those landmarks that were better remembered , and again began by examining fMRI responses associated with increasing values of this factor across the entire scanning session . The better remembered landmarks were associated with increased activations in left occipital ( −9 , −79 , −8 , Z = 6 . 54 ) , left postero-lateral parietal ( −48 , −52 , 34 , Z = 6 . 29 ) , and left temporal ( −63 , −37 , −2 , Z = 5 . 81 ) cortices , as well as bilateral POS ( left: −6 , −67 , 34 , Z = 5 . 87; right: 6 , −64 , 43 , Z = 5 . 62 ) and bilateral RSC ( left: −6 , −49 , 7 , Z = 5 . 10; right: 6 , −52 , 13 , Z = 4 . 69 ) . Investigating the source of these changes showed that the more easily remembered landmarks ( those with greater values for factor 2 ) did not produce any significant activation in the final quarter of learning . However , at the start of learning ( in the first quarter ) , there was a greater response in POS to landmarks which participants went on to later remember better in both the right ( 15 , −70 , 31; Z = 5 . 35 ) and left ( −9 , −76 , 28; Z = 5 . 30 ) hemispheres ( second row of Figure 3B , in red ) . There were also similar significant activations in the middle two quarters of learning ( second quarter , left: −3 , −76 , 40; Z = 4 . 94; right: 9 , −67 , 31; Z = 4 . 98; third quarter , left: −6 , −64 , 31; Z = 4 . 88; right: 3 , −64 , 37; Z = 4 . 76 ) . Intriguingly , these bilateral regions both overlapped with those which later went on to respond to permanent items in the final quarter . No areas showed responses associated with decreasing values of the memorableness factor , either across the whole scanning session or in any of the four quarters . Thus , the POS initially responded to memorable landmarks but then switched its response to permanent ones ( Figure 5A ) . Given this overlap , we plotted the response profiles of voxels in this overlapping region for the two factors . We extracted contrast estimates of the principal eigenvariate of responses within the overlapping voxels for factor one ( permanence ) and two ( memorableness ) in each of the four scanning runs using the MarsBaR toolbox and averaged across all subjects ( see ‘Materials and methods’ ) . Figure 5B shows the clear switch in responses within this region , with large activations initially present for the most memorable items ( factor 2 ) , but as subjects learned about the landmarks it instead ( in the middle of the third quarter ) became increasingly engaged by those which were permanent ( factor 1 ) . 10 . 7554/eLife . 09031 . 009Figure 5 . Response profile in POS . ( A ) POS responded to memorable landmarks ( those with higher factor 2 values ) in the first quarter of learning ( red ) and permanent ones ( with higher values for factor 1 ) in the final quarter ( blue ) . The overlap of these activations is shown in purple . ( B ) The response profile of the overlapping ( purple ) voxels for the two factors throughout whole scan . Responses were initially greater for memorable landmarks but then switched over the course of learning to eventually become responsive to permanence . Plots show mean BOLD responses ±1 SEM . Activations are shown on a structural MRI brain scan of single representative subject at the default threshold of p < 0 . 05 ( FWE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 009 Responses to the size and visual salience related factors ( factors 3 and 4 respectively ) remained constant throughout learning , with the greatest activations ( associated with increased values on these factors ) consistently occurring in posterior , visual areas ( Figure 3B in green and purple respectively ) . For example , average responses across learning were greatest for larger landmarks in a cluster located in superior posterior parts of the occipital lobes ( 18 , −88 , 22; Z = 7 . 71 ) , whereas a smaller cluster in just the right hemisphere was most active for salient landmarks ( 21 , −91 , 16; Z = 5 . 54 ) . No areas showed responses associated with decreasing values of these two factors . In summary , as subjects learned the permanence of landmarks , a representation emerged within the RSC ( factor 1 ) . POS also developed responses to permanent landmarks , but this region was also initially activated by landmarks which were subsequently better remembered ( factor 2 ) . The hippocampus and PHC were eventually more engaged by the most stable items , but later on and less strongly than RSC and POS . Perceptual features of the landmarks , their visual salience and size , were associated with tonic responses throughout learning in posterior visual areas ( factors 3 and 4 ) . In the above fMRI analyses , we used the amount of time that people had been exposed to the environment to probe the development of neural representations of the various landmark features . However , even though subjects would inevitably have learned more about the landmarks with more exposure to them , this measure does not account for individual differences in how much participants had learned at different points throughout the experiment . This variation between individuals was important for allowing us to examine permanence-related fMRI activity in greater detail . It enabled us to identify fMRI responses which might directly track the variation in learning of landmark permanence ( both between subjects and within individual people over time ) . This would provide more compelling evidence of whether activity in any brain region ( s ) might have a direct relationship with learning of landmark permanence , rather than looking at more simple , generalised associations . To characterise the dynamics of each subject's learning more precisely , we used their scores from each sweep's questioning period to construct a range of different models of their learning-state throughout the experiment . Of the models tested ( see ‘Materials and methods’ ) a Bayesian implementation of a ‘state-space’ model provided the best fit to the data ( Smith et al . , 2007 ) . We used this to create subject-specific parametric regressors of each subject's estimated learning state during each sweep of the scan , examples of which are shown in Figure 6 . We used these regressors to look for regions , anywhere in the brain , where responses matched how well a subject knew about the permanence of landmarks . The greatest activation was in the RSC ( 9 , −58 , 22; Z = 4 . 38 ) , a second peak was also present in the body of the caudate nucleus ( 18 , −10 , 25; Z = 4 . 03 ) . Therefore as subjects learned to distinguish permanent from transient landmarks , responses within their RSC directly reflected their knowledge of this difference . 10 . 7554/eLife . 09031 . 010Figure 6 . Examples of permanence learning curves and the associated fMRI response . Data from three examples subjects are shown . Learning curves were calculated and used to create subject-specific parametric regressors corresponding to the amount of permanence knowledge acquired throughout the scan . A whole brain comparison of fMRI responses to permanent vs transient landmarks according to how well subjects knew their permanence revealed responses only in RSC which were directly related to these curves . The learning curves show the estimated learning state ( coloured line ) and the 95% confidence interval ( coloured shaded area ) . The activation is shown on a structural MRI brain scan of single representative subject using a whole brain uncorrected threshold of p < 0 . 001 for display purposes . The colour bars indicate the Z-score associated with each voxel . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 010 We next examined changes in the functional connectivity between regions associated with learning landmark permanence using psychophysiological interactions ( PPI ) . A PPI analysis asks whether anywhere in the brain has a stronger relationship with a seed region during one condition compared to another . In this instance , we performed separate whole brain PPI analyses using the parts of RSC and POS which responded to landmark permanence ( Figure 4 ) as seed regions . In the second half of learning both regions showed increased functional coupling with the hippocampus when viewing permanent compared to transient landmarks . For both RSC ( right RSC: 21 , −16 , −23; Z = 4 . 23; left RSC: −30 , −10 , −26; Z = 3 . 42 ) and POS ( right POS: 30 , −7 , 20; Z = 3 . 70; left POS: −30 , −13 , −20; Z = 3 . 89 ) , this greater functional connectivity was with anterior parts of the hippocampus bilaterally . Neither RSC nor POS showed any differences in connectivity related to permanence during the first half of learning; these only emerged as subjects learned the stability of landmarks . Moreover , there were no other brain areas that showed functional connectivity with RSC and POS . To directly assess changes in connectivity associated with learning of permanence , we performed a further PPI analysis using the subject-specific permanence learning models . Using the same parts of RSC where greater responses emerged as subjects learned the permanence of landmarks as a seed region ( Figure 6 ) , the greatest increase in functional coupling developed with the left hippocampus ( −30 , −28 , −14; Z = 3 . 99 ) . Thus , as subjects learned the permanence of landmarks , their RSC not only developed greater responses to the permanent landmarks but also increased its functional connectivity with the hippocampus . In other words , the more that subjects learned about landmark permanence , the more their RSC-hippocampal functional coupling increased when viewing permanent landmarks . We next sought to go beyond features of the landmarks and consider whether information about landmark locations was present within different brain regions . This required us to assess the multi-voxel representations in relation to a continuous variable ( i . e . , how much individuals knew about permanent landmark locations ) . Widely-used approaches based on linear support vector machines , such as multi-voxel pattern analysis , can only be used to make categorical classifications , and so were not appropriate for our purpose . We therefore employed an alternative type of multivariate analysis method known as multivariate Bayes ( MVB ) . This is a model-based decoding method ( Friston et al . , 2008; FitzGerald et al . , 2012; Chadwick et al . , 2014 ) which compares competing hypotheses about the mapping between multi-voxel response patterns to a psychological target variable using a hierarchical approach known as parametric empirical Bayes ( see ‘Materials and method’ ) . Specifically , we used MVB to look for patterns of voxel activity within permanence responsive regions which mapped onto knowledge of permanent landmark locations as assessed in the post-scan navigation test . We reasoned that a representation relating to knowledge of permanent landmark locations would be strongest while subjects explicitly viewed them in that location . Therefore , unlike our other fMRI analyses , we examined fMRI responses during the learning videos , rather than during the questioning period images in which landmarks were isolated and devoid of more complex spatial information . We looked for patterns of multi-voxel activity , while people viewed permanent landmarks in situ , which related to how well they were able to subsequently locate them in the post-scan navigation task . RSC , POS , hippocampus and PHC had all been engaged by permanent items in some shape or form during the experiment so we focused these regions . RSC , hippocampus and PHC were defined independently using bilateral anatomical masks delineated by an experienced researcher , not involved in this project , guided by Duvernoy ( 1999 ) and Vann et al . ( 2009 ) on an averaged structural brain scan from a different set of n = 30 participants . For the one permanence-responsive region which did not relate to an easily defined anatomical locus—namely the POS—we used the cluster of voxels which were activated there in the main contrast of permanent vs transient landmarks outlined at the start of the fMRI results above . RSC , POS and PHC did not have any activity related to knowledge of permanent landmark locations at any point throughout the scanning experiment ( Figure 7 ) . Similarly , there were no significant results for the hippocampus during the first three quarters of learning . However , in the final quarter of the learning period , hippocampal responses emerged which were significantly related to the amount of information about the locations of permanent landmarks ( log model evidence = 12 . 8; posterior probability = 1 . 0 ) . Using the permutation function within MVB , with 100 samples , the hippocampal result in the final quarter of scanning gave a significant randomisation p value ( p = 0 . 0396 , ) , whereas all others were not significant . 10 . 7554/eLife . 09031 . 011Figure 7 . Multivariate Bayes analysis of responses which mapped onto knowledge of permanent landmark locations . The log model evidence values for response patterns within the RSC ( blue ) , posterior POS ( red ) , hippocampus ( HC; green ) and parahippocampal cortex ( PHC; purple ) relating to knowledge of permanent landmark locations are shown in each of the four quarters of scanning . By the final quarter of learning , the pattern of activity in the HC mapped onto the amount subjects knew about where permanent landmarks were located in the environment . The dashed black line indicates the threshold at which log model evidence values are considered to be strong ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09031 . 011 In summary , in contrast to the RSC which responded to accruing knowledge of landmark permanence per se , by the end of the learning phase , activity patterns within the hippocampus mapped onto how much subjects knew about where those permanent landmarks were located within Fog World . Given the short timescale over which subjects learned the permanence of completely novel landmarks within an alien VR world , it is notable how strong the RSC permanence representations were . It did not even take the whole scanning session for RSC to develop robust and selective sensitivity to the stable landmarks . This demonstrates the remarkable adaptability of the process and hints at it being fundamental for learning about and orientating within our surroundings . Reassuringly , there was no difference in recognition of the permanent and transient landmarks in the first or final quarters . There was a brief discrepancy in the recognition of permanent and transient landmarks in the second and third quarters of learning , in favour of the permanent landmarks . We are not sure why this is the case , but speculate that perhaps when faced with entirely novel items , we remember them better if they are in a consistent setting . In this experiment participants were aware from the initial task instructions that some landmarks would move and other would not , and they also realised that this knowledge was probed during the inter-sweep questioning . Consequently , it could be argued that this awareness of landmark permanence may have influenced responses in RSC . In the post-scan debriefing session , however , only 2 out of 32 participants made reference to specifically focusing on the permanence of landmarks during the navigation videos . This shows that participants were not overwhelmingly concerned with the permanence of landmarks , but rather seemed to instead be focused on their overall goal , namely the requirement to learn the environment and its layout . Additionally , in previous work ( Auger et al . , 2012; Auger and Maguire , 2013 ) , we have demonstrated that RSC responds to landmark permanence ( for everyday outdoor items ) when people are engaged in a completely incidental task . Whether this is also true when landmark permanence is learned incidentally should be probed in future studies . Previous experiments involving landmarks and VR have used every day , outdoor items as stimuli ( e . g . , Wolbers et al . , 2004; Wolbers and Buchel , 2005; Iaria et al . , 2007; Baumann et al . , 2010; Auger et al . , 2012; Auger and Maguire , 2013 ) . These familiar , real-world items inevitably came with semantic and associative ‘baggage’ making it difficult to achieve precise experimental control over numerous different features of the items which are often correlated with one another ( Troiani et al . , 2014 ) and could potentially confound any conclusions drawn ( Sugiura et al . , 2005 ) . Here , however , VR allowed us to investigate ‘pure’ representations of landmark permanence while this new information was freshly acquired . An additional advantage of using VR was that it provided a more ecologically relevant means with which to test the representation of landmark properties; it ensured subjects were actively engaged with the landmarks as they learned , in situ , the layout of the environment . However , one potential drawback of such a naturalistic , freely-behaving task is that it could introduce noise and variability to subjects' neural responses . This limitation was avoided by measuring fMRI activity when landmarks were viewed in complete isolation , as images presented between the learning videos . This not only ensured that subjects were focused on the specific relevant individual landmark at the necessary time , but also removed potential visual confounds which would have been present during the videos ( e . g . , path colour ) . Taken together with the prior studies , the current results highlight the flexible nature of RSC permanence representations . They can exist for items which are both real-world and alien; familiar or newly-encountered; viewed passively or when actively interacted with; and when attention is explicitly drawn , or not , to their permanence . By scanning subjects as they learned about Fog World , we also uncovered new information about how representations of landmark properties develop before they have become fully established . At the start of the learning process , RSC was not responsive to any feature of landmarks . Indeed , it was only during the third quarter of scanning that RSC became engaged at all , specifically reacting to permanent landmarks . The same was not true , however , for the POS region . It became strongly responsive to permanent landmarks at a similar time to RSC , but unlike RSC was engaged initially by another feature of the items—their memorableness—before their permanence became apparent . This is an important distinction as it indicates that the representation that develops in RSC is not only very sensitive to permanence but it is also highly specific . In the absence of reliable information regarding the permanence of landmarks , RSC did not just respond to some other feature . This finding is notable for another reason . In the literature , RSC and POS are sometimes grouped together and called ‘retrosplenial complex’ ( Vass and Epstein , 2013 ) , with no differentiation made between the two brain areas . However , RSC and POS are cytoarchitecturally distinct and have different connectivity ( Vann et al . , 2009 ) , and given that we have now pinpointed differences in their responsivity , we suggest caution in using the over-general term retrosplenial complex . Modelling the learning-state of each individual subject throughout the experiment provided further insight into how RSC processes landmark permanence . Activity in the RSC was directly related to how well people knew the permanence of items . This reveals that the permanence representation that emerges is not just a simple binary response which indicates whether or not an item is known to be stable . Instead it appears to be more informative , relating to the precision with which the permanence of landmarks is known . In addition , therefore , to identifying the most stable environmental cues , the RSC could also indicate how reliable subsequent representations based upon such cues might be . Overall , these results point to sophisticated , selective and specific processing of landmarks within RSC based upon their permanence . As noted above , superior posterior parts of POS displayed an intriguing profile of response; first activated by the items which were subsequently better remembered by subjects , and only later becoming engaged by the stable landmarks once their permanence was known . This region was not implicated in processing permanent landmarks in previous experiments ( e . g . , Auger et al . , 2012; Auger and Maguire , 2013 ) . So what is it about the memorable landmarks at the start and permanent landmarks by the end of learning in this study which engaged POS ? It is not entirely clear , but we can speculate about several possible reasons . For example , it could be that POS is tuned to navigationally ‘useful’ landmarks . Early in learning , in the absence of knowledge about landmark permanence , the memorable landmarks may have seemed more useful . Focus would then be expected to shift to the permanent landmarks which ultimately had greater utility for navigating and orienting . Alternatively , our results could indicate POS plays a role in relating landmarks with a specific location . This view is in keeping with a previous interpretation of processing within ‘retrosplenial complex’ ( Vass and Epstein , 2013 ) , which often extends into posterior parts of the POS . At the start of the learning process , encounters with the most memorable landmarks ( and the location they were experienced in ) would have been particularly evident and so elicited the largest early responses . The level of activity for memorable landmarks would then diminish as those which are not fixed are repeatedly encountered in conflicting locations . The region would , at the same time , become increasingly engaged by the permanent landmarks with repeated experience of them in the same place . The large response at the start of the scan to the most memorable landmarks also explains why , unlike RSC , activity in POS was not directly related to the permanence learning-state . This interpretation also accounts for the lack of POS engagement in previous experiments ( e . g . , Auger et al . , 2012; Auger and Maguire , 2013 ) , where the stimuli were never associated with specific locations . Therefore , while RSC appears specialised in processing the permanence of landmarks , the response profile of POS is more consistent with it playing some role in indexing navigational utility , or perhaps relating landmarks with the discrete locations where they have been encountered . Another region which responded to permanent landmarks in the present study was the hippocampus . In contrast to POS , the hippocampus was not activated by any landmarks early on , only later becoming engaged once the permanent landmarks became apparent . Furthermore , unlike RSC , activity in the hippocampus was not directly related to how well subjects had learned the permanence of landmarks . Whereas RSC tracked the acquisition of permanence knowledge , the hippocampus was only engaged by landmarks which were both known to be permanent and associated with a specific place . The MVB analysis indicated that once the hippocampal representation had emerged , it contained spatial detail related to how much subjects knew about where the permanent landmarks were located . This is in contrast to the other areas where no such decoding was possible , including in POS . This does not mean that POS or other regions do not contain detailed spatial information related to navigation performance; but any such representations , if they do exist , are significantly less evident than in the hippocampus . Interestingly , while the response profile in POS may suggest it initially represents landmark-location associations , as outlined above , this was not maintained over time . Moreover , the landmark-location representations in POS may have been discrete and local to a particular landmark , while the spatial detail decoded in the hippocampus was related to the wider knowledge of the environment's overall layout and the location of all the permanent landmarks within it . We observed increased functional coupling between RSC and hippocampus towards the end of the experiment , when the hippocampus started to come online . It seems then that the RSC may code for the stability of features within an environment , potentially providing this as an input to the hippocampus , which could in turn utilise the information to build detailed spatial representations based upon the most reliable cues . In terms of RSC-hippocampus connectivity , very few anatomical tracer studies have been conducted in humans . In the macaque monkey , RSC sends extensive cortical efferent projections to the medial temporal lobes ( presubiculum , parasubiculum , entorhinal and PHC cortex; Kobayashi and Amaral , 2007 ) . It receives inputs from similar regions , most notably the hippocampal formation ( entorhinal cortex , subiculum , presubiculum and parasubiculum ) , PHC and perirhinal cortices . In particular , Area 29 receives a large majority of projections from the hippocampus ( both subiculum and presubiculum ) and entorhinal cortex ( Aggleton et al . , 2012 ) . Thus , it is not surprising then that we observe functional connectivity between RSC and the hippocampus , and in particular anterior hippocampus . This is also in keeping with many studies that have reported RSC and anterior hippocampus co-activing in fMRI studies ( e . g . , Zeidman et al . , 2014 ) . In addition , inactivation of RSC in rats is associated with disruption of hippocampal place fields ( Cooper and Mizumori , 2001 ) , and previous studies have reported hippocampal involvement in retrieving spatial information about objects ( Save et al . , 1992; Manns and Eichenbaum , 2009; Baumann et al . , 2010; Ekstrom et al . , 2011 ) . We did not find responses relating to the landmarks in any other brain regions , but object-centred firing has also been observed in rodent lateral entorhinal ( Deshmukh and Knierim , 2012 ) and anterior cingulate cortex ( Weible et al . , 2012 ) , even for locations formerly ( but no longer ) occupied by objects ( Tsao et al . , 2013 ) . It will be interesting for future work to explore whether similar representations may also exist in humans and how they relate to the current findings . The present study provides an alternative interpretation of previous work which found that RSC is engaged when making judgements about locations relative to stable items ( Committeri et al . , 2004; Sulpizio et al . , 2013 ) . These previous studies concluded that activity within the RSC reflected coding of space relative to stable landmarks . However , our findings suggest it could in fact indicate a more fundamental representation of the landmark itself , specifically that of its inherent permanence . We suggest that the elementary discrimination between stable and moving landmarks demonstrated here within RSC is used to anchor representations of surrounding space . Sulpizio et al . ( 2013 ) also found RSC to be sensitive to viewpoint direction within a room but not relative to unstable objects . This could be linked to the presence of head direction cells within the rodent RSC ( Chen et al . , 1994; Cho and Sharp , 2001 ) , perhaps suggesting that head direction cell firing is centred upon permanent landmarks and this information is integrated within RSC . The discrimination between permanent and transient landmarks in RSC could play a crucial role in a number of fundamental computations involving space . It may go some way towards explaining its strong , ubiquitous engagement during fMRI studies of scene processing ( Epstein , 2008; Howard et al . , 2014; Zeidman et al . , 2014 ) , while navigating ( Maguire , 2001a; Hartley et al . , 2003; Spiers and Maguire , 2006; Spreng et al . , 2009; Vann et al . , 2009 ) , recalling autobiographical memories ( Maguire , 2001b; Gardini et al . , 2006; Steinvorth et al . , 2006; Svoboda et al . , 2006; Cabeza and St Jacques , 2007; Spreng et al . , 2009 ) and imagining future and fictitious events ( Addis et al . , 2007; Hassabis et al . , 2007; Szpunar et al . , 2007; Botzung et al . , 2008; Summerfield et al . , 2009 ) . All of these involve picturing scenes or events , with RSC possibly signalling the presence of permanent features ( Auger and Maguire , 2013 ) . The specificity of the RSC permanence response may also provide insights into failures of spatial navigation . Auger et al . ( 2012 ) and Auger and Maguire ( 2013 ) found that otherwise healthy individuals who were poor navigators had difficulty reliably identifying permanent landmarks , and had reduced responses in RSC compared to good navigators when viewing permanent landmarks . In neuropsychological studies , damage involving the RSC is never focal , but nevertheless leaves patients unable to derive orientation information from landmarks which they can otherwise recognise ( Maguire , 2001a; Vann et al . , 2009 ) . Metabolic and structural changes in Alzheimer's Disease/Mild Cognitive Impairment have been found to be first centred upon RSC ( Villain et al . , 2008; Pengas et al . , 2010 , 2012; Tu et al . , 2015 ) , and spatial disorientation is often one of the first signs of the disease . If representations of space are founded upon permanence information from RSC , as seems to be indicated by our results , then it needs to be dependable . If this is not the case , afflicted individuals may be falling at the first hurdle , so to speak , and one can understand why spatial disorientation then arises , because upstream regions that rely on good quality input cannot perform optimally to build reliable environmental representations . We believe that the current results provide compelling evidence that the RSC may be fundamentally concerned with coding for permanent landmarks . We further propose that this may account for its involvement in scene processing , navigation , autobiographical memory and future-thinking , providing input for upstream areas such as the hippocampus to then construct models of the world based upon reliable cues ( Maguire and Mullally , 2013; Zeidman et al . , 2014 ) . Future studies will be needed to establish the boundaries within which the RSC operates , for instance , does its permanence coding only function within the spatial domain ? The mechanisms underpinning the permanence response we report here also needs to be determined . While head direction cells have been found in rodent RSC ( Chen et al . , 1994; Cho and Sharp , 2001 ) , it has been estimated that these only account for approximately 8% of cells ( Chen et al . , 1994 ) . What are the other 92% of cells in RSC doing , and how does this relate to landmark permanence ? Intracranial recordings from human RSC have also been recently reported , showing increased responsivity to autobiographical memory retrieval ( Foster et al . , 2013 ) . We hope that the current results will encourage those conducting electrophysiological recording in humans and non-humans to explore the RSC and in particular its response to landmark permanence . By giving RSC its chance to be centre stage , we believe that new and important insights into how the brain performs the fundamental computations involved in representing and adapting to changes in the world will be forthcoming . Ten subjects ( five female , mean age 28 years , SD 4 . 8 ) took part in a landmark characterisation study . None of these subjects took part in the subsequent fMRI study . All were healthy , right-handed , highly proficient in English , and had normal vision . Each participant gave written informed consent for participation in the study , for data analysis and for publication of the study results . ‘Materials and methods’ were approved by the University College London Research Ethics Committee . We created 134 unique 3D ‘alien’ items to be used as potential landmarks in Fog World ( examples in Figure 1A ) . The landmarks were made with the animation software Blender 2 . 61 ( Blender Foundation , Amsterdam , Netherlands , http://www . blender . org/ ) . We then had the subjects characterise the following features of this novel set of landmarks:Salience ( ‘To what extent does this item grab your attention ? ’; 5 point scale: 1 = Not at all… 5 = Very much ) . Other associations ( ‘Does this remind you of anything ? ’; Yes/No ) . Likeableness ( ‘How do you feel about this item ? ’; Like/Dislike ) . Animateness ( ‘Does this item look like it could be alive or not ? ’; Alive/Not Alive ) . Memorableness ( ‘Have you already seen this item ? ’; Yes/No—answered having seen the items numerous times while rating the other features ) . Using these ratings , we selected two groups of 30 landmarks each ( from the original set of 134 ) to be used as the permanent and transient landmarks within the VR environment . These object groups were carefully selected to ensure that they did not differ in terms of any of the features rated ( t-tests of the two landmark groups: Salience: t58 = 0 . 669 , p = 0 . 51; Other associations: t58 = 0 . 000 , p = 1 . 0; Likeableness: t58 = 0 . 312 , p = 0 . 76; Animateness: t58 = −1 . 089 , p = 0 . 28; Memorableness: t58 = 0 . 247 , p = 0 . 81 ) . The landmarks groups did not differ on two other factors—the actual size of the items ( when in the VR environment: Small/Medium/Large; [t58 = 0 . 000 , p = 1 . 0] and mean spatial frequency [t58 = −0 . 562 , p = 0 . 58] ) . Having selected the two groups of landmarks , one was randomly allocated as the permanent set of landmarks and the other as the transient set . 32 different subjects ( 16 female , mean age 23 . 7 years , SD 2 . 4 ) took part in the fMRI study . All were healthy , right-handed , highly proficient in English , and had normal vision . Each participant gave written informed consent for participation in the study , for data analysis and for publication of the study results . ‘Materials and methods’ were approved by the University College London Research Ethics Committee . Fog World was created using the jMonkeyEngine 3 . 0 beta game engine ( http://jmonkeyengine . org ) , Java JDK 1 . 6 ( Sun Microsystems , Santa Clara , California ) and Blender ( Stichting Blender Foundation , Amsterdam ) . The world ( Figure 1 ) contained five different coloured intersecting straight paths ( yellow , red , grey , blue and green ) . Each path had 12 landmarks ( six permanent , six transient ) evenly distributed alongside it . A trial consisted of travelling along one of these paths . There were 60 trials in total , with the five paths being travelled 12 times each ( i . e . , once per learning sweep ) . Permanent landmarks remained in the same location on each trial , whereas transient landmarks appeared in a different location on every exposure . The locations in which all 60 landmarks appeared on each of the 60 trials were meticulously designed so that permanent and transient landmarks were equally distributed either side and along the whole length of each path . This ensured that the permanent and transient landmarks , as well as being matched for their perceptual features ( size , visual salience , and other features—see above ) , were placed in equivalent locations within the environment . Having determined the identities and precise locations of permanent and transient landmarks within Fog World on all 60 trials , we created a video for each trial to be presented to subjects while they underwent fMRI scanning . Each video took a first person perspective travelling along one of the paths . In these videos , the environment was covered in a shroud of fog to restrict the field of view and ensure close control over the exposure subjects had to all the landmarks . On each trial , the camera travelled along a path in a straight line . When a landmark emerged out of the fog , the camera turned to bring the landmark into the centre of the screen , where it was positioned for 2 s , the camera then panned back to the middle of the path as it continued travelling forwards ( Figure 2A from top to bottom; see also Video 1 ) . The paths were always travelled in the same direction , with the same start and end point each time . During scanning , the ordering of trials along the five different paths within each learning sweep was pseudorandomised so there were no biases in when the paths were travelled relative to each other . To encourage subjects to learn an integrated representation of the whole environment , the paths intersected one another . Each path intersected with two others ( Figure 1C ) . The first intersection was located three landmarks after the start of the path and the second was three landmarks before the end , with six landmarks between the two intersections . When the videos came to one of these intersections , the camera turned either left or right and the fog cleared enough to reveal three landmarks on the adjoining path . After 3 s , the landmarks were obscured by the fog again with the camera returning to the centre while continuing along the route . There were equal numbers of left and right turns at each intersection throughout the whole experiment and the ordering of the turns was pseudorandomised to ensure it was not predictable . The number of times each landmark was viewed during one of these intersection turns was also controlled so that overall exposure to all the landmarks remained identical . These 60 videos ( corresponding to the 60 trials ) were each approximately 1 min in length . Before scanning , subjects were shown an example trial ( containing landmarks and a path which did not appear during the main experiment ) to familiarise them with the general format of the main fMRI task . The main fMRI task has been described earlier . To summarise , once all five paths had been travelled once , there came a questioning period to gauge how much information subjects had learned by that point in the experiment . The combination of a 13 landmark questioning period and videos of the five different paths preceding it are referred to as a learning ‘sweep’ . In the questioning period between learning sweeps , the ordering of the three types of landmark ( permanent , transient or unseen ) was also pseudorandomised . There were a total of 12 learning sweeps throughout the experiment ( divided into four scanning runs , or ‘quarters’ , comprising three sweeps each ) . Each 3-sweep scanning run lasted 15–20 min and subjects could take a short break ( while remaining in the scanner ) between scanning runs when necessary . Once out of the scanner after the learning experiment had concluded , subjects were shown images of individual landmarks ( all 60 from the environment and 26 previously unseen landmarks ) and indicated whether or not they recognised them from the environment ( ‘Do you remember seeing this item in the environment ? ’ , Yes/No—the memorableness measure ) . After that , questions were only asked about the landmarks from the environment . Participants first rated the permanence of the environment's landmarks ( ‘How many positions in the environment do you think this item was in ? ’ , Only 1/Many ) ; next they rated the salience of each landmark ( ‘To what extent does this item grab your attention ? ’ , Not at all/A bit/A lot ) and finally the size that landmarks were in the environment ( ‘What size is this item ? ’ , Small/Medium/Large ) . A different randomised order of landmarks was used for each of these questions . A final active navigation task provided a thorough examination of how well participants had learned the layout of the whole environment . This task and scoring are described in the main text . During scanning , two ratings were collected from subjects to gauge how well they recognised landmarks and knew their permanence . We assessed the rates at which subjects came to recognise permanent and transient landmarks . To do this , we performed separate linear regression analyses for permanent and transient landmarks to assess how the accuracy with which subjects recognised them changed throughout the learning phase in the scanner . Using a paired t-test ( threshold p < 0 . 05 ) we then directly compared the slopes using these linear estimates across all subjects in order to establish whether or not subjects had learned to recognise the two types of landmark equally . The principal components factor analysis was conducted using the mean ratings for each feature of all 60 landmarks from every subject; it used a varimax rotation and Kaiser normalisation . We then generated orthogonal factor score estimates using the Anderson-Rubin method for use in a whole brain fMRI analysis . The factor analysis and all statistical tests were performed using SPSS version 20 ( http://www . spss . com ) . T2*-weighted single-shot echo-planar images with BOLD contrast were acquired on a 3T Magnetom Allegra head-only MRI scanner ( Siemens Healthcare , Erlangen , Germany ) operated with the standard transmit-receive head coil . fMRI data were acquired across four sessions with a sequence which was optimized to minimize signal dropout in the medial temporal lobe and used a descending slice acquisition order with a slice thickness of 2 mm , an interslice gap of 1 mm , and an in-plane resolution of 3 × 3 mm ( Weiskopf et al . , 2006 ) . 48 slices angled at −45° to the anterior–posterior axis were collected covering the entire brain , with a repetition time of 2 . 88 s , 30 ms echo time and 90° flip angle . A 3D MDEFT T1-weighted structural scan was also acquired for each participant with 1 mm isotropic resolution ( Deichmann et al . , 2004 ) . The first 6 ‘dummy’ volumes from each of the four sessions were discarded to allow for T1 equilibration effects . FMRI data were analysed using SPM8 ( www . fil . ion . ucl . ac . uk/spm ) . Images were realigned and unwarped using field maps which were acquired with a double-echo gradient field map sequence ( TE = 10 and 12 . 46 ms , TR = 1020 ms , matrix size 64 × 64 , with 64 slices , voxel size = 3 mm3 ) and then normalised to a standard EPI template in MNI space with a resampled voxel size of 3 × 3 × 3 mm and smoothed using an 8 mm FWHM Gaussian kernel . We compared fMRI responses while subjects viewed images of individual , isolated landmarks displayed during the questioning periods at the end of each sweep . Our primary interest was in seeing whether a neural representation of landmark stability emerged for previously unseen items over the course of the scanning experiment . We therefore directly contrasted fMRI BOLD responses to permanent and transient landmarks in the whole brain . We first considered responses across the whole scanning session , and in a second analysis divided the scanning experiment into quarters ( which corresponded to the four scanning runs each consisting of three learning sweeps ) in order to assess changes as subjects learned about the items . Potential issues associated with incidental changes in the BOLD signal over time were avoided by specifically comparing changes in the difference between permanent and transient landmarks . Permanent and transient landmark regressors were convolved with the haemodynamic response function . A separate regressor was created for the learning video time periods . This , along with participant-specific movement regressors were treated as covariates of no interest . We calculated subject-specific parameter estimates pertaining to each regressor of interest ( β ) for each voxel . Second level random effects analyses were then run using one-sample t-tests on these parameter estimates . We report all fMRI results at a whole brain threshold of p < 0 . 05 ( FWE ) , unless otherwise stated . We also examined fMRI responses in relation to the four factors from the principal components analysis . As above , we first analysed fMRI BOLD responses across the whole scanning session , and in a second analysis divided the scanning experiment into quarters . We created parametric regressors from the orthogonal factor score coefficients for every landmark for each of the four principal components using the Anderson-Rubin method . Parametric regressors from these scores were then entered into a whole brain GLM fMRI analysis . This enabled us to examine activity that was linearly modulated by each factor . These parametric regressors were each convolved with the haemodynamic response function . Analyses were then run in the same way as for the permanent vs transient landmark comparison described above . We used the MarsBaR toolbox for analysing responses within the overlapping voxels for the permanence and memorableness factors . MarsBaR is an SPM toolbox which extracts fMRI data from specified regions of interest . We defined the region of interest ( purple in Figure 5A ) as the contiguous voxels which showed significant activation for both the memorableness factor in the first quarter of learning and the permanence factor in the fourth quarter . For each subject , we then extracted the principal eigenvariate of responses within this region for the memorableness and permanence factor parametric regressors , in each of the four quarters of the scanning period . The mean ±1 SEM of these subjects' responses are plotted in Figure 5B . The ‘state-space’ learning models were created with the MATLAB- and WinBUGS-based software provided at http://www . neurostat . mit . edu . We compared the mean squared error ( MSE , in arbitrary units ) and percentage variance explained by three different types of learning model: a ‘state-space’ model estimated by maximum likelihood using an expectation maximisation algorithm ( Smith et al . , 2004 ) , a ‘state-space’ model estimated by a Bayesian approach ( Smith et al . , 2007 ) and a moving average of accuracy across each sweep and the sweeps immediately preceding and following it . A Bayesian implementation of a ‘state-space’ model provided the best fit ( Smith et al . , 2007 ) by both accuracy measures ( MSE = 204 , SD = 88; r2 = 64 ) than the ‘state-space’ model estimated by maximum likelihood ( MSE = 226 , SD = 85; r2 = 44 ) and the moving averages model ( MSE = 221 , SD = 86; r2 = 36 ) . As the state-space model estimated by a Bayesian approach ( Smith et al . , 2007 ) provided the best fit of the data , we used this to create subject-specific parametric regressors of each subject's estimated learning state during each sweep of the scan for use in a whole brain GLM analysis . Separate regressors were created for permanent and transient landmarks so that we could contrast responses to the two types of landmark in direct relation to how well people knew the permanence of landmarks . As SPM automatically mean centres parametric regressors within each scanning block , we concatenated the four sessions into one and added extra regressors to model the mean signal for each session . Parametric regressors are additionally mean centred by SPM at the first-level , so in order to accurately reflect between-subject differences in the overall extent of learning across the whole experiment , we added their overall performance on the post-scan navigation test as a second-level covariate of interest . Significant clusters are reported at a whole brain uncorrected threshold of p < 0 . 001 for the RSC and p < 0 . 05 FWE corrected for the rest of the brain . We chose this statistical threshold for the RSC given the more subtle nature of this specific contrast ( compared to the simpler categorical comparison of all permanent with all transient landmarks ) and our specific prior hypotheses about RSC processing landmark permanence . However , it should be noted that the RSC activation was also significant after applying small volume correction within a bilateral RSC mask ( threshold FWE corrected p = 0 . 01 ) . Functional connectivity was examined using the gPPI toolbox ( McLaren et al . , 2012 ) . We looked for changes in the functional connectivity of regions associated with learning landmark permanence . The seed regions and contrasts used for these analyses were all based upon the corresponding univariate whole-brain comparisons described above . First , for any regions responsive to landmark permanence , we looked for brain areas with which they showed increased functional coupling for permanent compared to transient landmarks . Early and late parts of the scanning session were compared separately ( learning sweeps 1–6 and 7–12 respectively ) . We analysed the two halves instead of four quarters in order to increase the number of trials and power with which to detect these potentially subtle effects . In a second connectivity analysis , implemented in the same way , we used a seed region and contrast which additionally accounted for inter-individual differences in learning . The MVB analysis uses the same design matrix as a standard univariate SPM analysis , with columns for experimental variables of interest as well as regressors of no interest . A contrast is then specified ( in this instance from subjects' scores on the post-scan navigation task ) and a ‘target’ variable is derived from this after accounting for potential confounds ( e . g . , head movement ) . The patterns of voxel activity within each region of interest are then fitted to this target variable , producing a model evidence value . These model evidence values are then compared to a null model to determine the log model evidence . For each model , voxel weights are calculated which map every voxel to knowledge of permanent landmark locations . The priors and variance of the patterns of these voxel weights are used to constrain the models at a second hierarchical layer . The pattern weights are iteratively bipartitioned into subsets based upon the size of the weights; each successive partition isolates the subset with the largest pattern weights . These subsets are optimised using a greedy search with a standard variational scheme under the Laplace assumption ( Friston et al . , 2007 ) . Once the optimal set size is reached , this final set of patterns represents the decoding model with the greatest evidence for the pattern weights . Log model evidence values therefore represent the mutual information shared by the psychological variable ( in this case knowledge of permanent landmark locations ) and the pattern of voxel responses within that brain region . The log evidence of different models can then be directly compared . MVB assumes that a small proportion of voxel activity patterns make a large contribution to decoding accuracy ( i . e . , information is sparsely coded ) . We used the SPM software's default settings for the MVB analyses , with nine greedy search steps and size of successive subdivisions set at 0 . 5 to test for evidence of sparse representations relating to knowledge of landmark location . We modelled the whole time period that permanent landmarks were in view during the learning videos . We defined the regions of interest anatomically ( with one exception—POS—see main text ) , as exploring responses within the whole bilateral anatomical regions rather than smaller functionally-defined clusters within them provided maximal multi-voxel information for the characterisation of representations . We report all analyses with a log model evidence value above three as significant , as is common practise ( Kass and Raftery , 1995; Penny et al . , 2004; Friston et al . , 2008 ) . We conducted further control analyses using the permutation function within MVB ( with 100 samples ) to check there was no bias towards the MVB procedure producing a positive result .	Throughout our lives , we encounter novel environments that we must learn to find our way around , from a new office to a new city . Studies of brain activity in humans and rodents have revealed that many brain regions are involved in navigation , most notably the hippocampus . However , these experiments have typically involved humans navigating around environments filled with familiar objects and landmarks , and therefore tell us relatively little about how the brain builds up a map of a completely new environment in the first place . To address this issue , Auger et al . scanned the brains of healthy human volunteers as they experienced an ‘alien’ virtual reality world called ‘Fog World’ , so-named because of the dense fog used to precisely control what the volunteers could see . In contrast to previous virtual reality environments , which have contained houses , shops and other recognisable objects , Fog World contains only abstract landmarks that bear little resemblance to anything in the real world . The volunteers watched videos that simulated journeys through Fog World with the goal of learning the layout of the environment so that they could navigate within it . Of note , half of the landmarks in Fog World remained in fixed positions on all learning trials , while the other half changed location from one trial to the next . After each block of trials , the volunteers were shown single landmarks—some from Fog World and others not—while their brains were scanned . A region called the retrosplenial cortex showed increasing activity that closely tracked the volunteers' growing knowledge of which landmarks had fixed , permanent locations in Fog World . In later trials towards the end of the learning period , the hippocampus also became active , and at this time communication between the retrosplenial cortex and hippocampus was also heightened . By the end of learning , the hippocampal activity was related to the volunteers' knowledge of the locations of the permanent landmarks across Fog World . As well as revealing that the retrosplenial cortex may be essential for processing permanent landmarks , the work of Auger et al . shows how the hippocampus and retrosplenial cortex could work together to map new environments . These findings might also help us to better understand why some healthy individuals are bad navigators , and why disorientation is a common early symptom in neurodegenerative disorders such as Alzheimer's disease , where the retrosplenial cortex is often one of the first brain regions to become damaged .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	A central role for the retrosplenial cortex in de novo environmental learning
Neurotrophin-3 ( Ntf3 ) and brain derived neurotrophic factor ( Bdnf ) are critical for sensory neuron survival and establishment of neuronal projections to sensory epithelia in the embryonic inner ear , but their postnatal functions remain poorly understood . Using cell-specific inducible gene recombination in mice we found that , in the postnatal inner ear , Bbnf and Ntf3 are required for the formation and maintenance of hair cell ribbon synapses in the vestibular and cochlear epithelia , respectively . We also show that supporting cells in these epithelia are the key endogenous source of the neurotrophins . Using a new hair cell CreERT line with mosaic expression , we also found that Ntf3's effect on cochlear synaptogenesis is highly localized . Moreover , supporting cell-derived Ntf3 , but not Bbnf , promoted recovery of cochlear function and ribbon synapse regeneration after acoustic trauma . These results indicate that glial-derived neurotrophins play critical roles in inner ear synapse density and synaptic regeneration after injury . The trophic factors , neurotrophin-3 ( Ntf3 ) and brain-derived neurotrophic factor ( Bbnf ) play critical roles in the embryonic inner ear , contributing to the survival of cochlear and vestibular sensory neurons and to the establishment of their projections into the respective sensory epithelia ( Fritzsch et al . , 2004; Ramekers et al . , 2012 ) . Both Bdnf and Ntf3 continue to be expressed in inner ear sensory epithelia after birth . In the postnatal vestibular epithelia , Bdnf is expressed only by supporting cells ( Schecterson and Bothwell , 1994; Montcouquiol et al . , 1998 ) , whereas Ntf3 is expressed by both supporting cells and hair cells ( Farinas et al . , 1994 ) . In the postnatal organ of Corti , Bdnf is expressed by both inner ( IHCs ) and outer hair cells ( OHCs ) and supporting cells at early postnatal ages ( P1–P6 ) , but is only detected in supporting cells from P10 onwards ( Wiechers et al . , 1999 ) . All cells in the early postnatal organ of Corti appear to express Ntf3 , but in the adult , expression is restricted to the IHCs and their surrounding supporting cells , with higher levels in the apical ( low-frequency ) region than at the base of the cochlear spiral ( Sugawara et al . , 2007 ) . Thus Bdnf and Ntf3 could have significant functions in the postnatal inner ear , and alterations in expression of these neurotrophins may modulate structure and function in the adult inner ear . In this study , we investigated the roles of postnatal Ntf3 and Bdnf in both normal and damaged inner ears . Using cell-specific and inducible knockout or overexpression technology , we eliminated or increased neurotrophin expression from supporting cells or hair cells in the postnatal inner ear . We show that these neurotrophins , when expressed by the glia-like supporting cells in these sensory epithelia , are required for the formation and/or maintenance of ribbon synapses , a role distinct from the one they play in embryogenesis . Ntf3 has major effects only in the cochlea , while postnatal Bdnf appears to act only in the vestibular organs . Furthermore , we show that Ntf3 overexpression can elicit regeneration of the synaptic contacts between cochlear nerve terminals and inner hair cells after noise-induced synaptopathy , a type of neural damage which appears to be widespread even in ears exposed at sound levels well below those which cause hair cell damage and permanent threshold shifts ( Kujawa and Liberman , 2009 ) . To alter neurotrophin expression by supporting cells , we used a mouse line that expresses the tamoxifen-inducible Cre recombinase ( CreERT ) under the control of the proteolipid protein 1 promoter ( Plp1/CreERT ) . We previously showed that this line can be used to induce effective gene recombination in inner ear supporting cells , when tamoxifen is delivered at early postnatal ages ( Gomez-Casati et al . , 2010a ) . We crossed Plp1/CreERT with mice carrying a conditional Ntf3 knockout allele ( Ntf3flox ) ( Bates et al . , 1999 ) or an overexpression transgene ( Ntf3stop ) to eliminate or increase the expression of Ntf3 , respectively . RT-qPCR showed that Ntf3 expression was significantly reduced in both the cochlea and utricle of Ntf3flox:Plp1/CreERT mice and increased in those of Ntf3stop:Plp1/CreERT mice after tamoxifen treatment ( Figure 1A ) . Mice with either Ntf3 knockout ( Ntf3flox:Plp1/CreERT ) or overexpression ( Ntf3stop:Plp1/CreERT ) in supporting cells had normal vestibular evoked potentials ( VsEPs , Figure 1B , C ) , indicating that postnatally , supporting cell-derived Ntf3 is not necessary for vestibular function . This is not the case for the embryonic inner ear , where the loss of Ntf3 results in the loss of a subpopulation of vestibular neurons ( Ernfors et al . , 1995 ) . In contrast , Ntf3 knockout from supporting cells resulted in elevated thresholds for auditory brainstem responses ( ABRs ) at high frequencies , while supporting cell Ntf3 overexpression resulted in reduced ABR thresholds at the same frequencies ( Figure 1E ) . Distortion product otoacoustic emissions ( DPOAEs ) were not affected by the abnormal Ntf3 expression ( Figure 1D ) , suggesting that the ABR changes were due to dysfunction of the inner hair cells ( IHCs ) , the cochlear nerve fibers , or the synapses that connect them . Correspondingly , the amplitude of ABR peak 1 ( P1 ) , the summed activity of the cochlear nerve , was reduced by supporting cell Ntf3 knockout and increased by overexpression , also at high frequencies ( Figure 1F ) . The latency of ABR P1 and the width of ABR wave 1 were not altered by the knockout or overexpression ( Figure 1G , H ) , suggesting that altering Ntf3 levels does not affect cochlear nerve myelination . Control mice carrying only the Plp1/CreERT allele showed normal cochlear responses , proving that the phenotype was not due to the expression of the CreERT transgene ( Figure 1—figure supplement 1 ) . These results demonstrate that supporting cell derived Ntf3 plays a critical role in normal physiology of the postnatal cochlea , specifically in the basal ( high-frequency ) regions . 10 . 7554/eLife . 03564 . 003Figure 1 . Ntf3 expression by postnatal supporting cells is required for cochlear , but not vestibular function . ( A ) RT-qPCR shows that postnatal tamoxifen injection reduces or increases Ntf3 mRNA levels in adult Ntf3flox:Plp1/CreERT or Ntf3stop:Plp1/CreERT inner ears , respectively; n = 5–6 . p < 0 . 05 , p < 0 . 01 by two-tailed unpaired t test . ( B and C ) Postnatal Ntf3 knockout ( blue ) or overexpression ( red ) from supporting cells does not alter VsEP thresholds ( B ) or their peak 1 ( P1 ) amplitudes at 0 dB ( C ) ; n = 4–8 . ( D–F ) Postnatal knockout or overexpression of Ntf3 from supporting cells reduces or enhances cochlear function , respectively . Ntf3 knockout ( blue ) elevates ABR thresholds ( E ) and decreases ABR P1 amplitudes ( F ) , without changing DPOAE thresholds ( D ) ; n = 16–17 . Ntf3 overexpression ( red ) reduces ABR thresholds ( E ) and increases ABR P1 amplitudes ( F ) , without changing DPOAE thresholds ( D ) ; n = 21 . ABR P1 amplitudes were assessed at 70 dB SPL . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 by two-way ANOVA . ( G ) ABR P1 latencies are not affected by either Ntf3 knockout ( blue ) or Ntf3 overexpression ( red ) at all frequencies examined . Key in C applies to A–G . ( H ) Mean ABR waveforms from responses to 32 kHz tone pips from Ntf3 knockouts and their controls ( upper ) and Ntf3 overexpressors and their controls ( lower ) . Gray shading indicates ABR wave 1 . Both ABR P1 latencies ( G ) and waveforms ( H ) results were assessed at 70 dB SPL; n = 13–17 . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 00310 . 7554/eLife . 03564 . 004Figure 1—figure supplement 1 . Expression of the Plp1/CreERT allele does not affect the cochlear function . Plp1/CreERT ( +/− ) and Plp1/CreERT ( −/− ) littermates exhibit similar DPOAE thresholds ( A ) , ABR thresholds ( B ) , ABR P1 amplitudes ( C ) , and ABR P1 latencies ( D ) at all frequencies tested . ABR P1 amplitudes and latencies were assessed at 70 dB SPL . Tamoxifen was injected at P0-1 and physiological tests were performed at P42; n = 4 . Key in D applies to all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 004 We then examined the role of supporting cell derived Bdnf in the postnatal inner ear using analogous conditional alleles ( Bdnfflox [Rios et al . , 2001] and Bdnfstop [Chang et al . , 2006] ) in combination with the Plp1/CreERT . RT-qPCR showed that inner ear Bdnf expression was significantly altered ( Figure 2A ) . Neonatal knockout of Bdnf in supporting cells resulted in elevated VsEP thresholds and decreased VsEP peak 1 ( P1 ) amplitudes ( Figure 2B , C; blue ) . This phenotype is identical to that seen when supporting cell Bdnf knockout was induced during late embryogenesis ( E14 . 5-17 . 5 ) ( Gomez-Casati et al . , 2010b ) , demonstrating that the postnatal period is when supporting cell-derived Bdnf plays a critical role in the vestibular epithelium . Surprisingly , vestibular function was unaffected in mice overexpressing Bdnf ( Figure 2B , C; red ) , indicating that Bdnf is expressed at sufficient amounts in the postnatal vestibular epithelium . Cochlear function was unaffected by either increased or reduced Bdnf expression as seen by DPOAE thresholds ( Figure 2D ) , ABR thresholds ( Figure 2E ) , ABR amplitudes ( Figure 2F ) , and ABR latencies ( Figure 2G , H ) , demonstrating that supporting cell-derived Bdnf is not important in the postnatal cochlea . Together , our results show that supporting cell-derived Ntf3 and Bdnf play complementary roles in the postnatal cochlea and vestibular organs . 10 . 7554/eLife . 03564 . 005Figure 2 . Bdnf expression by postnatal supporting cells is required for vestibular , but not cochlear function . ( A ) RT-qPCR shows that postnatal tamoxifen injection results is reduced or increased Bdnf mRNA levels in adult Bdnfflox:Plp1/CreERT or Bdnfstop:Plp1/CreERT inner ears , respectively; n = 6 . ( B and C ) Postnatal Bdnf knockout ( blue ) from supporting cells leads to higher VsEP thresholds ( B ) and lower VsEP P1 amplitudes ( C ) at the high stimulus level ( 0 dB re 1 g/ms ) . Postnatal Bdnf overexpression ( red ) does not affect the vestibular function ( B and C ) ; n = 6–11 . p < 0 . 05; *p < 0 . 01 by two-tailed unpaired t tests . ( D–H ) Neither Bdnf knockout ( blue ) nor overexpression ( red ) from postnatal supporting cells alters cochlear function , as shown by normal DPOAE thresholds ( D ) , ABR thresholds ( E ) , ABR P1 amplitudes ( F ) , ABR P1 latencies ( G ) , and ABR P1 waveforms ( H ) at 70 dB SPL compared to control littermates; n = 6 . Key in C applies to A–G . Gray shading ( H ) indicates ABR wave 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 005 We then analyzed cochlear histopathology to determine the anatomical basis for the pathophysiology in Ntf3 mutants . Examination of plastic sections showed no obvious changes in the organ of Corti , the spiral ganglion , or accessory structures of the cochlear duct such as the stria vascularis , spiral ligament , etc . Quantitative analysis showed normal density of supporting cells and hair cells ( Figure 3A , C ) as well as peripheral axons of cochlear nerve fibers ( Figure 3B , D ) . In contrast , analysis of immunostained whole-mounts of the organ of Corti revealed alterations in the numbers of synaptic contacts between inner hair cells ( IHCs ) and cochlear nerve terminals . Immunostaining of pre-synaptic ribbon and post-synaptic receptor patches , using antibodies against CtBP2 and GluA2 , respectively ( Figure 4A ) , showed that supporting cell Ntf3 knockout reduced the number of synaptic puncta ( Figure 4A; upper panels and Figure 4D; blue ) , while overexpression increased synapse number ( Figure 4A; lower panels and Figure 4D; red ) . Effects on synaptic counts were restricted to basal ( high-frequency ) cochlear regions ( Figure 4D ) , as seen for ABR thresholds and amplitudes ( Figure 1E , F ) . Similar results were obtained when separately counting either pre-synaptic ribbons or post-synaptic receptor patches ( Figure 4B , C ) . The strong correlation between the changes in ABR amplitudes and synaptic counts ( Figure 4E ) suggests that levels of supporting cell-derived Ntf3 influence cochlear function by regulating the number of IHC ribbon synapses . Previous studies have suggested that alterations in ribbon synapses affect ABR responses by changing sound-evoked peak discharge rates ( Davies , 1996; Glowatzki et al . , 2006; Gomez-Palacio-Schjetnan and Escobar , 2013 ) . Thus , increases ( or decreases ) in the number of synapses per fiber in Ntf3 overexpressors or knockouts could increase ( or decrease ) the probability of synchronous EPSPs , thereby increasing ( or decreasing ) onset spike rate in response to tone pips and ultimately altering the amplitude of ABR P1 , the summed activity of the cochlear nerve . The observation that Ntf3 overexpression can increase synapse density indicates that Ntf3 is normally expressed in limited amounts , as expected from the neurotrophic hypothesis ( Davies , 1996 ) . 10 . 7554/eLife . 03564 . 006Figure 3 . Ntf3 expression by postnatal supporting cells does not affect the organ of Corti morphology , hair cell or axon numbers . ( A and B ) Photomicrographs of the basal turn of an Ntf3flox:Plp1/CreERT cochlea ( ∼32 kHz region ) showing a cross-section of the organ of Corti ( A; scale bar = 100 µm ) and a tangential section through the osseous spiral lamina showing peripheral axons of cochlear nerve fibers ( B; scale bar = 10 µm ) . ( C ) The number of IHCs per 100 µm of organ of Corti is not altered by either Ntf3 knockout ( blue ) or overexpression ( red ) at any cochlear regions; n = 5–6 . ( D ) The number of cochlear-nerve peripheral axons per 100 µm of osseous spiral lamina ( near the habenula perforata ) is not affected by either Ntf3 knockout ( blue ) or overexpression ( red ) ; n = 6 . Key in D applies to C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 00610 . 7554/eLife . 03564 . 007Figure 4 . Ntf3 expression by postnatal supporting cells regulates hair cell ribbon synapse density at high frequencies . ( A ) Representative confocal images ( maximal projection from a focal series ) of IHC synapses from 32 kHz region of Ntf3flox , Ntf3flox:Plp1/CreERT , Ntf3stop , and Ntf3stop:Plp1/CreERT cochleae immunolabeled for pre-synaptic ribbons ( CtBP2-red ) and post-synaptic receptor patches ( GluA2-green ) ( scale bar = 10 µm ) . The dashed lines show the approximate outline of one IHC . CtBP2 antibody also weakly stains IHC nuclei . ( B–D ) Quantitative data shows that Ntf3 knockout reduces , and overexpression increases , the number of pre-synaptic ribbons ( B ) , post-synaptic GluA2 receptor patches ( C ) , and putative ribbon synapses , defined as juxtaposed CtBP2- and GluA2-positive puncta ( D ) at high frequency cochlear regions; n = 5–6 . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 by two-way ANOVA . ( E ) Relative synaptic counts vs relative ABR P1 amplitudes of Ntf3 knockouts ( blue ) or overexpressors ( red ) shows a linear correlation . Data points were obtained by normalizing synaptic counts and ABR P1 amplitudes of Ntf3 mutants to the values of their respective controls at each of the frequency regions analyzed . Key in D applies B–E . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 007 Since both supporting cells and IHCs express Ntf3 in the postnatal organ of Corti ( Sugawara et al . , 2007 ) , it was important to assess the contribution of hair cell-derived Ntf3 . We generated a mouse line carrying tamoxifen-inducible Cre under the control of the Pou4f3 promoter ( Pou4f3/CreERT ) , which is active only in hair cells in the postnatal inner ear ( Fujioka et al . , 2011 ) . Using a Rosa26tdTomato reporter line ( Madisen et al . , 2010 ) , we found that the Pou4f3/CreERT line drives inducible gene recombination in about 50% of inner and outer hair cells , with similar recombination efficiency at all cochlear regions ( Figure 5A ) , and with a pattern similar to another hair cell promoter , Atoh1/CreERT ( Weber et al . , 2008 ) . We generated mice with knockout or overexpression of Ntf3 in hair cells by crossing Pou4f3/CreERT with Ntf3flox or Ntf3stop mice and treating them with tamoxifen at early postnatal ages . RT-qPCR analysis showed that Ntf3 expression was significantly reduced and increased in cochleae of Ntf3 knockouts and overexpressors , respectively ( Figure 5B ) . Mice with hair cell-specific Ntf3 knockout had no cochlear phenotype: ABR thresholds and P1 amplitudes were indistinguishable from controls ( Figure 5C , D; blue ) . Thus , hair cell-derived Ntf3 appears to be dispensable in the postnatal cochlea . In contrast , mice overexpressing Ntf3 in postnatal hair cells showed significant reduction in ABR threshold at 32 kHz ( Figure 5C; red ) as well as increased ABR P1 amplitudes at frequencies ≥16 kHz ( Figure 5D; red ) , similar to the patterns seen for supporting cell specific Ntf3 overexpression ( Figure 1 ) . 10 . 7554/eLife . 03564 . 008Figure 5 . Ntf3 overexpression by postnatal hair cells increases cochlear sensitivity and synaptic densities at high frequencies . ( A ) Pou4f3/CreERT allows for hair cells specific inducible gene recombination . Rosa26tdTomato:Pou4f3/CreERT mice and their littermate Rosa26tdTomato controls were injected with tamoxifen at P1-P3 , and the cochleas were collected at P60 . Inducible recombination is seen as tdTomato fluorescence co-localized with MyoVIIa immunostaining . Scale bar = 50 µm . ( B ) RT-qPCR shows that postnatal tamoxifen injection reduced Ntf3 mRNA in Ntf3flox:Pou4f3/CreERT cochlea and increased Ntf3 expression in Ntf3stop:Pou4f3/CreERT cochlea , compared to their respective controls; n = 4–5 . p < 0 . 05 , *p < 0 . 01 by two-tailed unpaired t tests . ( C and D ) Postnatal overexpression of Ntf3 in hair cells ( red ) reduced ABR thresholds ( C ) and increased ABR P1 amplitudes ( D ) at high frequencies; n = 9–10 . Postnatal knockout of Ntf3 from hair cells ( blue ) had no effect on these measures; n = 8–11 . ABR P1 amplitudes were assessed at 70 dB SPL . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 by two-way ANOVA . ( E ) Confocal maximal projection of 7 adjacent IHCs from the 32 kHz region of a hair cell-specific Ntf3 overexpressor , after immunostaining for synapses as in Figure 4A; tdTomato indicates recombined hair cells . Scale bar = 10 µm . ( F ) Synaptic counts are increased in recombined ( tdTomato+ , Ntf3 overexpressing ) IHCs compared to neighboring unrecombined cells; n = 6 cochleae , with at least 100 hair cells in each group . p < 0 . 05 , *p < 0 . 01 by two-tailed paired t tests . Key in B applies to B–D . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 008 To probe the mechanisms underlying the effects of hair cell-derived Ntf3 overexpression , we analyzed IHC synaptic counts . For this , we generated Ntf3stop:Rosa26tdTomato:Pou4f3/CreERT mice , which allowed identification of hair cells in which the Ntf3 overexpression transgene had been activated , based on tdTomato fluorescence ( Figure 5E ) . Pre-synaptic ribbons and post-synaptic receptor patches were increased specifically in the recombined ( tdTomato ( + ) and Ntf3 overexpressing ) IHCs ( Figure 5F ) . This observation suggests that the Ntf3 released by IHCs remains close to its source , possibly because the supporting cells create a diffusion barrier , preventing cross talk between adjacent IHCs . Together , these results show that supporting cells are the key endogenous source of Ntf3 for synaptic survival , but that Ntf3 overexpression in hair cells can increase synaptic density , even in the postnatal cochlea . A significant component of the acute response to acoustic trauma ( AT ) is the swelling of cochlear nerve terminals at afferent synapses , suggestive of noise-induced glutamate excitotoxicity ( Robertson , 1983 ) . Such damage can lead to a permanent loss of IHC synapses and reduction in ABR amplitudes , followed by a slow death of spiral ganglion neurons , despite complete recovery of cochlear thresholds within 1–2 weeks ( Kujawa and Liberman , 2009 ) . To determine if Ntf3 overexpression could either prevent , or promote recovery from , this noise-induced synaptic loss and attenuation of cochlear responses , we exposed control and Ntf3 overexpressing mice to noise levels known to cause this type of neuropathy ( 8–16 kHz at 100 dB SPL for 2 hr ) ( Kujawa and Liberman , 2009 ) . In Ntf3stop control mice , this exposure caused immediate threshold shifts as large as 50 dB , which recovered partially by 3 days and almost completely ( except 32 kHz ) by 10 days after exposure ( Figure 6A; left panel ) . However , even after 10 days , ABR P1 amplitudes ( at ≥16 kHz ) remained significantly lower than before exposure ( Figure 6B; left panel ) . In contrast , in Ntf3 overexpressing mice , ABR thresholds recovered completely by 3 days post-exposure ( Figure 6A; right panel and Table 1 ) , and P1 amplitude recovery was significantly enhanced ( Figure 6B; right panel and Table 1 ) . In contrast to Ntf3 , a parallel set of experiments on mice overexpressing Bdnf in supporting cells showed no effects on the extent or time course of the noise-induced pathophysiology ( Figure 6C , D and Table 1 ) . 10 . 7554/eLife . 03564 . 009Figure 6 . Overexpression of Ntf3 , but not Bdnf , promotes recovery from noise-induced attenuation of cochlear responses . ( A and B ) Ntf3 overexpression accelerates the recovery of ABR thresholds ( A ) and promotes recovery of ABR P1 amplitudes ( B ) after acoustic trauma ( AT ) ; n = 5–7 . ( C and D ) Bdnf overexpression does not affect the recovery of ABR thresholds ( A ) and ABR P1 amplitudes ( B ) after acoustic trauma; n = 5–7 . p < 0 . 05 , **p < 0 . 001 by two-way ANOVA . Gray shading indicates the noise exposure spectrum . Key in B applies to all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 00910 . 7554/eLife . 03564 . 010Table 1 . Statistical analysis ( two-way ANOVA ) of ABR threshold and P1 amplitude changes between control and neurotrophin overexpressing ( Plp1/CreERT ) mice after acoustic traumaDOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 010Ntf3 overexpressor vs controlBdnf overexpressor vs controlp ValueF Statisticp ValueF StatisticABR thresholdAT + 1 day0 . 00518 . 4690 . 68010 . 1716AT + 3 days<0 . 000119 . 000 . 67520 . 1774AT + 10 days<0 . 000117 . 340 . 98620 . 0003ABR P1 amplitudeAT + 1 day0 . 03424 . 6950 . 18991 . 7580AT + 3 days<0 . 000118 . 080 . 56920 . 3276AT + 10 days<0 . 000118 . 130 . 04214 . 3160 Analysis of immunostained cochlear epithelia from the noise-exposed ears showed that Ntf3 overexpression does not prevent noise-induced synaptic loss , but rather supports synapse regeneration . Immediately after exposure ( AT + 2 hr ) , control and Ntf3 overexpressing mice had the same degree of synaptic loss ( colocalized GluA2–CtBP2 puncta , Figure 7A; upper panels , Figure 7B and Table 2 ) . Similarly , Ntf3 overexpression did not alter the acute loss of pre-synaptic ribbons ( Figure 7C and Table 2 ) . However , at 14 days of post-exposure ( AT + 14 days ) , a clear effect of Ntf3 on IHC synapses was observed . Whereas the synaptic loss in control mice remained equal to that in AT + 2 hr , Ntf3 overexpressors showed significant synaptic recovery ( Figure 7A; lower panels , Figure 7B and Table 2 ) . In control mice , noise-induced loss of pre-synaptic ribbons was progressive , with further reductions seen between 2 hr and 14 days post-exposure . Remarkably , Ntf3 overexpression prevented the progressive loss of synaptic ribbons ( Figure 7C and Table 2 ) . 10 . 7554/eLife . 03564 . 011Figure 7 . Ntf3 overexpression from postnatal supporting cells promotes recovery from noise-induced synaptic degeneration . ( A ) Representative confocal images of IHC synapses from 32 kHz region of Ntf3stop and Ntf3stop:Plp1/CreERT cochleae immunolabeled for CtBP2 and GluA2 at 2 hr ( upper panels ) or 14 days after acoustic trauma ( AT ) ; scale bar = 10 µm . ( B and C ) Ntf3 overexpression promotes regeneration of IHC synapses ( B ) and prevents the progressive loss of IHC ribbons ( C ) between 2 hr ( AT + 2 hr ) and 14 days ( AT + 14 days ) after acoustic trauma . n = 3–8 . p < 0 . 05 , *p < 0 . 001 by two-way ANOVA . Gray shading indicates the noise exposure spectrum . Key in C applies to B–C . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 01110 . 7554/eLife . 03564 . 012Table 2 . Statistical analysis ( two-way ANOVA ) of synaptic density changes between control and Ntf3 overexpressing ( Plp1/CreERT ) mice after acoustic traumaDOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 012p ValueF StatisticSynapseAT + 2 hr0 . 97320 . 001AT + 14 days0 . 00658 . 358RibbonAT + 2 hr0 . 81340 . 056AT + 14 days0 . 000215 . 54 For Ntf3 to be a viable candidate to treat noise-induced hearing loss , it must be effective even if applied after noise exposure . To test if Ntf3 overexpression after noise exposure enhances synaptic regeneration and functional recovery , we could not use the Plp1/CreERT transgene , as this line does not produce effective recombination in cochlear supporting cells after P15 ( data not shown ) . Therefore , we used a line expressing CreERT under the control of the promoter for the glutamate aspartate transporter Slc1a3 ( Slc1a3/CreERT ) ( Wang et al . , 2012 ) . In the adult cochlea , Slc1a3 ( GLAST ) is expressed by inner phalangeal and inner border cells ( Furness and Lawton , 2003; Glowatzki et al . , 2006 ) , the same supporting cells that express Plp1 at early postnatal ages ( Gomez-Casati et al . , 2010a ) . RT-qPCR showed that tamoxifen treatment in adult Ntf3stop:Slc1a3/CreERT mice significantly increased cochlear Ntf3 expression as compared to controls lacking the Cre ( Figure 8A ) . Ntf3stop:Slc1a3/CreERT mice and control mice were then subjected to acoustic trauma , followed immediately ( <1 hr ) by tamoxifen treatment ( AT/Tmx; Figure 8B ) . As for mice injected with tamoxifen in the early postnatal period ( Figure 6A ) , controls and Ntf3 overexpressors showed comparable loss and gradual recovery of ABR thresholds when tamoxifen was administered in adult mice immediately after noise exposure ( Figure 8C and Table 3 ) . In contrast , the recovery of ABR P1 amplitudes was significantly enhanced by Ntf3 overexpression ( Figure 8D ) . Interestingly , Ntf3 had no significant effects on ABR P1 amplitudes 3 days after noise , but by 14 days after exposure , Ntf3 overexpressors showed significantly higher ABR P1 amplitudes compared to control littermates ( Figure 8D and Table 3 ) . The delay in Ntf3's impact on cochlear responses might suggest a slow increase in Ntf3 availability or a delay in the effects of Ntf3 on synapse regeneration . 10 . 7554/eLife . 03564 . 013Figure 8 . Ntf3 overexpression from adult supporting cells after acoustic trauma promotes auditory function recovery and synaptic regeneration . ( A ) RT-qPCR shows that tamoxifen treatments of adult mice increased Ntf3 expression in Ntf3stop:Slc1a3/CreERT cochlea; n = 6 . p < 0 . 001 by two-tailed unpaired t tests . ( B ) Time line of the experiment showing the ages of mice for ABR measurements , acoustic trauma ( AT ) , tamoxifen inductions ( Tmx ) , and sample collections for synaptic counts . ( C–D ) The effects of Ntf3 overexpression from adult supporting cells on ABR thresholds ( C ) and P1 amplitudes ( D ) after noise exposure; n = 5–9 . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 by two-way ANOVA . Key in D applies to C–D . ( E ) Representative confocal images of IHC synapses from 32 kHz region of Ntf3stop and Ntf3stop:Slc1a3/CreERT cochleae immunolabeled for CtBP2 and GluA2 . The samples were collected from mice without AT/Tmx ( Pre-AT/Tmx ) or 14 days after AT/Tmx ( AT/Tmx + 14 days ) . Scale bar = 10 µm . ( F ) Ntf3 overexpression after acoustic trauma promotes regeneration of IHC ribbon synapses; n = 5–6 . p < 0 . 01 , **p < 0 . 001 by two-way ANOVA . Gray shading indicates the noise exposure spectrum . DOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 01310 . 7554/eLife . 03564 . 014Table 3 . Statistical analysis ( two-way ANOVA ) of ABR threshold and P1 amplitude changes between control and Ntf3 overexpressing ( Slc1a3/CreERT ) mice before and after acoustic trauma and tamoxifen treatmentsDOI: http://dx . doi . org/10 . 7554/eLife . 03564 . 014p ValueF StatisticABR thresholdPre-AT/Tmx0 . 77740 . 081AT/Tmx + 3 days0 . 16741 . 965AT/Tmx + 14 days0 . 53010 . 400ABR P1 amplitudePre-AT/Tmx0 . 28261 . 166AT/Tmx + 3 days0 . 36950 . 821AT/Tmx + 14 days0 . 000812 . 71 To define the cellular mechanisms of the Ntf3 effects , we analyzed the density of ribbon synapses before and after noise exposure in these animals . As expected , the basal level of synaptic counts was similar between Cre− and Cre+ mice ( Figure 8E , top panels and Figure 8F ) . Importantly , 14 days after noise exposure and tamoxifen treatment , cochleae from Ntf3 overexpressors had significantly higher density of pre-synaptic ribbons , post-synaptic GluA2 receptor patches , and putative synapses ( Figure 8E , bottom panels and Figure 8F ) . Together with the results from Ntf3stop:Plp1/CreERT mice ( Figure 6 ) , our study demonstrates that Ntf3 overexpressed either before or immediately after noise exposure can effectively promote recovery from noise-induced cochlear synaptopathy . Neurotrophins are key molecular mediators for synaptic development and function in the central nervous system ( Gomez-Palacio-Schjetnan and Escobar , 2013 ) . In this study , we show that Ntf3 and Bdnf , which are necessary for sensory neuron survival in the developing inner ear ( Fritzsch et al . , 2004; Ramekers et al . , 2012 ) , continue to play important and complementary roles in the postnatal inner ear , specifically by modulating the number of synapses between hair cells and sensory neurons . Our results indicate that supporting cells of the sensory epithelia are the key source of these neurotrophins in the postnatal inner ear . Importantly , we show that increasing the availability of Ntf3 , but not Bdnf , promotes the recovery of both cochlear responses and IHC synapses after acoustic trauma , even when Ntf3 expression is induced after noise exposure . The specificity of the effects of each neurotrophin on each organ cannot be explained simply by the spatio-temporal expression pattern of the components for these signaling pathways , as both Bdnf and Ntf3 , as well as their respective receptors Ntrk2 ( TrkB ) and Ntrk3 ( TrkC ) , are expressed in the postnatal and adult cochlea and vestibular organs ( Pirvola et al . , 1992; Fritzsch et al . , 1999; Gestwa et al . , 1999; Wiechers et al . , 1999; Farinas et al . , 2001; Stankovic and Corfas , 2003; Sugawara et al . , 2007 ) . A similar specificity in the biological roles of these neurotrophins has been shown in embryonic development , that is , constitutive knockout of Bdnf or Ntf3 reveals predominant pro-survival roles on vestibular or cochlear neurons , respectively ( Farinas et al . , 1994; Jones et al . , 1994; Ernfors et al . , 1995; Schimmang et al . , 1995; Bianchi et al . , 1996 ) . However , while genetic replacement of one neurotrophin with the other can almost completely rescue neuronal survival deficits caused by constitutive deletion of either Bdnf or Ntf3 ( Coppola et al . , 2001; Agerman et al . , 2003 ) , we observed that Ntf3 and Bdnf have distinct and non-overlapping roles in postnatal synapse formation . Thus , it appears that vestibular and auditory primary sensory neurons can respond equally to endogenous levels of either Bdnf or Ntf3 during development , but this ability is lost after birth . In the vestibular system , supporting cell-derived Bdnf is the sole neurotrophin necessary for postnatal formation and maintenance of hair cell synapses ( Gomez-Casati et al . , 2010b ) , and the levels of Bdnf expressed by these non-neuronal cells is not limiting , as Bdnf overexpression does not alter vestibular function . While previous studies of constitutive knockouts showed that Ntf3 is necessary for the survival of a subpopulation of vestibular neurons during embryogenesis ( Ernfors et al . , 1995 ) , our data indicate that Ntf3 , which like Bdnf is expressed primarily by supporting cells in the postnatal vestibular sensory epithelia ( Sugawara et al . , 2007 ) , is dispensable in the vestibular system after birth . In contrast to the vestibular organs , supporting cell-derived Ntf3 , but not Bdnf , is necessary for the establishment of normal synapses and auditory function in the cochlear base . The correlation of Ntf3 expression levels with synapse numbers and cochlear sensitivity in both knockout and overexpression models indicates that supporting cell-derived Ntf3 is not only a critical but also a limiting factor in the postnatal cochlea . Interestingly , postnatal knockout of supporting cell-derived Ntf3 and Bdnf did not affect the survival of the sensory neurons themselves , in either the vestibular ( Gomez-Casati et al . , 2010b ) or the cochlear ( spiral ) ganglion , indicating that endogenous Ntf3 and Bdnf become dispensable for neuronal survival after birth . These sensory neurons may become independent of trophic support in the adult , or alternatively , other trophic factors , such as insulin-like growth factor-1 ( Igf1 ) and macrophage migration inhibitory factor ( Mif ) , may promote survival after birth . Both Igf1 and Mif are expressed in the postnatal cochlea , and , for both , loss results in ganglion cell death or altered innervation after birth ( Camarero et al . , 2001; Bank et al . , 2012 ) . In addition , glial cell line-derived neurotrophic factor ( Gdnf ) is also expressed in postnatal inner ear ( Stankovic and Corfas , 2003 ) , and it has been suggested that vestibular neurons switch trophic sensitivity from Bdnf to Gdnf after target innervation ( Hashino et al . , 1999 ) . Although adult inner hair cells also express Ntf3 ( Wheeler et al . , 1994; Sugawara et al . , 2007 ) , we found no cochlear dysfunction in mice lacking Ntf3 expression in these cells . The lack of phenotype in the hair cell-specific Ntf3 knockout is unlikely due to the partial recombination in hair cells because ( a ) recombination in supporting cells and IHCs had similar efficiency ( ∼60% ) ; ( b ) cochlear Ntf3 expression was reduced to a similar extent by deletion from either supporting cells or hair cells; and ( c ) Ntf3 overexpression by hair cells had similar effects on cochlear function and synapse density as that seen in Ntf3 overexpression by supporting cells , indicating that modulation of Ntf3 expression in a subset of hair cells is sufficient to produce phenotypic outcomes . These observations support the conclusion that endogenous Ntf3 expressed by postnatal supporting cells , but not inner hair cells , is necessary for normal cochlear function . The mosaic recombination pattern in the Pou4f3/CreERT Ntf3 overexpression revealed that the effects of Ntf3 on hair cell synaptogenesis are precisely localized , that is , synapse density was increased only on hair cells in which the transgene was activated . Thus , it appears that supporting cells create a physical barrier , allowing for local signaling events without cross talk between adjacent hair cells . The lack of effect of postnatal Ntf3 knockout or overexpression on the total number of myelinated sensory axons in the osseous spiral lamina indicates that the changes in synapses is not due to alterations in the number of sensory neurons , and thus it is consistent with the notion that the effects of Ntf3 are local . Present results suggest that Ntf3 levels may regulate the number of synapses by influencing branching of the unmyelinated terminals of cochlear sensory neurons . Regardless of the cellular origin , the effects of Ntf3 knockout or overexpression on synaptic density and cochlear function are restricted to the high-frequency ( basal ) half of the cochlea . The pro-survival effects of embryonic Ntf3 have the same tonotopy , that is , the loss of spiral ganglion cells in mice with Ntf3 or Ntrk3 knockout are primarily seen at the cochlear base ( Fritzsch et al . , 1997; Tessarollo et al . , 1997; Coppola et al . , 2001; Farinas et al . , 2001 ) . Thus , it appears that Ntf3 is the key endogenous trophic factor for both embryonic development and postnatal function of high-frequency cochlear neurons . A previous study showed that Ntf3 promotes axonal growth and synaptogenesis in organ of Corti explants after excitotoxicity ( Wang and Green , 2011 ) . Our results indicate that in vivo , Ntf3 regulates ribbon synapse numbers without altering cochlear nerve axonal numbers , suggesting a direct effect of Ntf3 on IHC synaptogenesis . Most importantly , overexpression of Ntf3 , but not Bdnf , promotes recovery from noise-induced synaptic degeneration and the associated decrements in auditory evoked potentials . Since the Ntf3 receptor Ntrk3 is expressed by cochlear neurons , not IHCs ( Gestwa et al . , 1999 ) , our results suggest that Ntf3 overexpression acts first by promoting the recovery of post-synaptic terminals , which prevents the progressive loss of pre-synaptic ribbons , and enhances the regeneration of IHC synapses after noise exposure . There has been extensive exploration of the use of Bdnf and Ntf3 as therapeutics for sensorineural hearing loss , based on their pro-survival effects on spiral ganglion neurons after hair cell degeneration due to ototoxic drugs ( Ramekers et al . , 2012 ) . It has been suggested that these neurotrophins might promote long-term neuronal survival in cochlear implant users , who typically have few , if any , remaining hair cells ( Budenz et al . , 2012 ) . Recent studies in mice have shown that significant synaptic loss precedes hair cell death and spiral ganglion cell degeneration in both noise-induced and age-related hearing loss ( Kujawa and Liberman , 2009; Sergeyenko et al . , 2013 ) , and it has been suggested that this primary neuropathy is a major cause of problems hearing in a noisy environment , the most common complaint of those with sensorineural hearing loss . Therefore , treatment of cochlear synaptopathy presents a novel therapeutic approach for age-related and noise-induced hearing loss . While a number of trophic factors , including Bdnf and Ntf3 , can mediate the survival of injured sensory neurons ( Roehm and Hansen , 2005; Ramekers et al . , 2012 ) , our findings indicate that specific neurotrophins are required to promote regeneration and to regain function of the synaptic connections in damaged inner ear epithelia . Plp1/CreERT ( Doerflinger et al . , 2003 ) , Slc1a3/CreERT ( Wang et al . , 2012 ) , Bdnfflox ( Rios et al . , 2001 ) , Ntf3flox ( Bates et al . , 1999 ) , and Rosa26tdTomato reporter mice ( Madisen et al . , 2010 ) were obtained from Jackson Laboratory . The Bdnfstop mouse line was provided by Rudolf Jaenisch ( Chang et al . , 2006 ) . To generate the Ntf3stop mouse , we engineered the conditional Ntf3 overexpression transgene using the mouse Ntf3 cDNA ( kindly provided by Barbara Hempstead , Cornell University ) under regulation by the synthetic CAGGS promoter/enhancer/intron followed by a loxP-STOP-loxP cassette ( kindly provided by Laurie Jackson-Grusby , Children's Hospital Boston ) . Briefly , a loxP-STOP-loxP cassette was cloned immediately upstream of the Ntf3 cDNA present in a Bluescript plasmid . A SacI/KpnI fragment containing the mouse Ntf3 cDNA and the loxP-STOP-loxP cassette was cut out from the plasmid , blunt-ended , and ligated with an EcoRI cut/blunt-ended pCAGGSTurbo-cre vector to generate pCAGGS-loxP-STOP-loxP Ntf3 . A PstI fragment containing an ATG-FRT site was cut out from the pPGK-ATG-FRT ( no EcoRI ) vector and subcloned into the PstI site , 3′ of the CAGGS-loxP-STOP-loxP Ntf3 construct . The whole pCAGGS-loxP-STOP-loxP Ntf3-ATG-FRT plasmid was targeted downstream of the collagen 1a1 locus by frt/Flpase-mediated site-specific integration . To generate Pou4f3/CreERT mice , an 8 . 6-kb Pou4f3 ( Brn3 . 1 ) regulatory sequence ( Sage et al . , 2006 ) was cut out from pSP73-Pou4f3-alpha9 plasmid using SalI ( kindly provided by Doug Vetter , Tuffs University ) . The SP73-alpha9 fragment was re-ligated and digested with SmaI to remove alpha9 cDNA from SP73 backbone . The CreERT coding sequence ( Feil et al . , 1996 ) was obtained from EcoRI digestion of pCreERt plasmid ( kindly provided by Laurie Jackson-Grusby ) , blunted and ligated to SP73 backbone . SP73-CreERT vector was then cut with SalI and re-ligated with Pou4f3 regulatory region . The final plasmid ( 14 . 4-kb ) was digested with FspI and SbfI to release the 13-kb fragment containing Pou4f3/CreERT transgene , which is then purified for microinjection . Both Ntf3stop and Pou4f3/CreERT mice were generated at Mouse Gene Manipulation Facility at Children's Hospital Boston . To knockout Bdnf or Ntf3 from postnatal supporting cells , we crossed Bdnfflox:Plp1/CreERT or Ntf3flox:Plp1/CreERT mice with Bdnfflox or Ntf3flox mice , respectively . Tamoxifen was injected intra-peritoneally at 50 mg/kg/day from P0–P1 daily . To overexpress Bdnf or Ntf3 from postnatal supporting cells , we crossed Plp1/CreERT mice with Bdnfstop or Ntf3stop mice , respectively . Tamoxifen was injected at 33 mg/kg/day from P1–P7 daily . The recombination efficiency and specificity of Pou4f3/CreERT mice were examined by crossing these mice with homozygous Rosa26tdTomato reporter mice . To knockout Ntf3 from hair cells , Ntf3flox:Pou4f3/CreERT mice were mated with Ntf3flox mice . To overexpress Ntf3 from hair cells , we crossed Pou4f3/CreERT:Rosa26tdTomato mice with Ntf3stop mice . The hair cell-specific recombination was induced by tamoxifen injection at 50 mg/kg/day from P1–P3 daily . To overexpress Ntf3 from adult supporting cells , we crossed Slc1a3/CreERT mice with Ntf3stop mice . Ntf3flox:Plp1/CreERT mice and their controls were on C57BL/6 background . Ntf3stop:Slc1a3/CreERT and their controls were on a mixed background of C57BL/6 and FVB/N . All the other mice were on FVB/N background . For gene expression study , 10-week old mice were gavaged with tamoxifen ( 200 mg/kg/day ) for 3 days . For acoustic trauma studies , 12-week old mice were gavaged with tamoxifen for 3 days with the first dosage given immediately ( <1 hr ) after noise exposure . In all experiments , Cre negative littermates were used as controls . All ‘Materials and methods’ were performed in compliance with animal protocols approved by the Institutional Animal Care and Use Committee at Children's Hospital Boston . Mice ( 8- to 10-week old ) were euthanized in a CO2 chamber and the inner ears were extracted . Cochleas ( membranous labyrinths ) and utricles ( sensory epithelia ) were dissected from temporal bones , and total RNA was purified using RNeasy spin-columns ( Qiagen , Valencia , CA ) . Total RNA from cochlea ( 200 ng ) or utricle ( 50 ng ) was reverse transcribed ( RT ) using an iScript cDNA Synthesis Kit ( Bio-Rad , Hercules , CA ) in 20 µl reaction with a mixture of oligo ( dT ) and random hexamer primers . The reverse transcription was performed at 42°C for 1 hr followed by 85°C for 5 min . Quantitative PCR was carried out on a CFX96 machine using an iQ SYBR Green Supermix ( Bio-Rad ) . For each well of the 96-well plate ( Bio-Rad ) , the 20 μl reaction contained 10 μl of 2× iQ SYBR Green Supermix , 6 pmol of each forward and reverse primer , and 1 μl of cDNA sample . The cycling conditions were as follows: 95°C for 3 min followed by 40 cycles of 95°C for 30 s , 60°C for 30 s , and 72°C for 30 s . Each sample was loaded in duplicates . The following forward ( F ) and reverse ( R ) primers were used: Bdnf , F: GTGTGTGACAGTATTAGCGAGTGG , R: GATACCGGGACTTTCTCTAGGAC , which generates a 101 bp amplicon; Ntf3 , F: GCCCCCTCCCTTATACCTAATG , R: CATAGCGTTTCCTCCGTGGT , which generate an 83 bp amplicon; and Rpl19 , F: ACCTGGATGAGAAGGATGAG , R: ACCTTCAGGTACAGGCTGTG , which generates a 101 bp amplicon . Expression levels of Bdnf and Ntf3 were normalized to the reference housekeeping gene Rpl19 in the same samples . The relative expression level was calculated by the 2−ΔΔCt method as shown previously ( Stankovic and Corfas , 2003 ) . Inner ear physiology , including vestibular evoked potentials ( VsEPs , the summed activity of the vestibular afferent pathways to sudden head accelerations ) , auditory brainstem responses ( ABRs , the summed activity of auditory afferent pathways to short tone bursts ) , and distortion product otoacoustic emissions ( DPOAEs ) , was performed on mice anesthetized with xylazine ( 20 mg/kg , i . p . ) and ketamine ( 100 mg/kg , i . p . ) . Analysis of animals without noise exposure was performed on 5- to 6-week old mice . For experiments involving noise exposure , the first recording was performed at 14–15 weeks of age , followed by noise exposure at 16 weeks of age and the additional measurements 1 , 3 , and 10 days post-exposure . For VsEPs and ABRs , needle electrodes were placed into the skin ( a ) at the dorsal midline close to the neural crest , ( b ) behind the left pinna , and ( c ) at the base of the tail ( for a ground electrode ) . For VsEPs , mice were positioned on their backs , with the head coupled securely to a shaker platform . Stimuli were linear acceleration ramps , 2 ms in duration , applied in the earth–vertical axis at 17/s with alternating stimulus polarity . An accelerometer , mounted near the head , was used to calibrate the resultant jerk , which is expressed in dB re 1 . 0 g/ms . Electrophysiological activity was amplified ( 10 , 000× ) , filtered ( 0 . 3–3 kHz ) , and digitized ( 125 kHz ) , and 1024 responses were averaged at each stimulus level . We collected an intensity series in 5 dB steps encompassing stimulus levels above and below threshold . ABR potentials were evoked with 5 ms tone pips ( 0 . 5 ms rise-fall , with a cos2 envelope , at 40/s ) delivered to the eardrum at log-spaced frequencies from 5 . 6 to 32 kHz . The response was amplified ( 10 , 000× ) and filtered ( 0 . 3–3 kHz ) with an analog-to-digital board in a PC-based data acquisition system . Sound level was raised in 5 dB steps from 10 to 80 dB sound pressure level ( SPL ) . At each level , 1024 responses were averaged ( with stimulus polarity alternated ) after ‘artifact rejection’ . The DPOAEs in response to two primary tones of frequencies f1 and f2 were recorded at ( 2 × f1 ) −f2 , with f2/f1 = 1 . 2 , and the f2 level 10 dB is lower than the f1 level . The ear-canal sound pressure was amplified and digitally sampled at 4 µs intervals . DPOAE thresholds were defined as the f1 level required to produce a response at 0 dB SPL . These acoustic signals , generated by outer hair cells and measureable in the ear canal , are useful for differential diagnosis: attenuation of ABRs without a change in DPOAEs provides strong evidence for cochlear synaptic or neural dysfunction ( Kujawa and Liberman , 2009 ) . Mice ( at 12–16 weeks ) were placed within small cells in a subdivided cage , suspended in a reverberant noise exposure chamber , and exposed to an octave band of noise ( 8–16 kHz ) at 100 dB for 2 hr . Noise calibration to target SPL was performed immediately before each noise overexposure . Sound pressure levels varied by <1 dB across the cages . Mice were perfused intracardially with 4% paraformaldehyde in 0 . 1 M phosphate buffer . Cochleas were extracted , perfused intralabyrinthly , and post-fixed with 1 . 5% paraformaldehyde and 2 . 5% glutaraldehyde . The cochleas were then osmicated in 1% osmium tetroxide , decalcified in 5% EDTA , dehydrated and embedded in araldite . Serial sections ( 20 µm ) parallel to the modiolus were cut using a Leica RM2165 microtome and mounted on microscope slides in Permount . Cochlear regions of interest ( 6 , 16 , and 32 kHz ) were identified based on 3D reconstruction and cochlear mapping ( Hirose and Liberman , 2003 ) . Axonal counts were made by imaging tangential sections through the osseous spiral lamina near the habenula perforata . Several fascicles of the cochlear nerve fibers were present in these sections . All myelinated fibers from each section were imaged using 63× DIC optics and counted . The number of axonal fibers was then divided by the cochlear length at specific frequency regions . Analysis of animals without noise exposure was performed at 8–10 weeks . Analysis of noise-exposed animals was performed either immediately ( AT + 2 hr ) or 2 weeks after noise exposure ( AT + 14 days ) . Cochleae were fixed as described above , post-fixed in 4% paraformaldehyde in 0 . 1 M phosphate buffer for 2 hr , and decalcified in 5% EDTA . Cochlear tissues were then microdissected and permeabilized by freeze-thawing in 30% sucrose . The microdissected pieces were blocked in 5% normal horse serum with 1% Triton X-100 in phosphate-buffered saline ( PBS ) for 1 hr , followed by incubation in primary antibodies ( diluted in blocking buffer ) at 37°C for 16 hr . The primary antibodies used in this study were: anti-myosin VIIa ( rabbit anti-MyoVIIa; Proteus Biosciences , Ramona , CA; 1:500 ) , anti-C-terminal binding protein 2 ( mouse anti-CtBP2 IgG1; BD Biosciences , San Jose , CA; 1:200 ) , and anti-glutamate receptor 2 ( mouse anti-GluA2 IgG2a; Millipore , Billerica , MA; 1:2000 ) . Tissues were then incubated with appropriate Alexa Fluor-conjugated fluorescent secondary antibodies ( Invitrogen , Carlsbad , CA; 1:500 in blocking buffer ) and 1 μg/ml DAPI ( Invitrogen ) for 1 hr at room temperature . The tissues were mounted on microscope slides in Vectashield mounting media ( Vector laboratories , Burlingame , CA ) . All pieces of each cochlea were imaged at low power to convert cochlear locations into frequency using a custom ImageJ plugin ( http://www . masseyeandear . org/research/otolaryngology/investigators/laboratories/eaton-peabody-laboratories/epl-histology-resources/imagej-plugin-for-cochlear-frequency-mapping-in-whole-mounts ) . Confocal z-stacks of the 5 . 6 , 8 , 11 . 3 , 16 , 22 . 6 , and 32 kHz regions from each cochlea were taken using a Zeiss LSM510 microscope equipped with either 40× ( 1× digital zoom , Pou4f3/CreERT reporter study ) or 63× ( 2× digital zoom , synaptic counts ) oil immersion lens . The number of inner hair cells ( IHCs ) at specific cochlear regions was determined based on the DAPI nuclear counts at the IHC focal plane . For synaptic counts , the z-stacks ( 0 . 25 µm step size ) were set to span the entire length of IHCs so that all the synaptic specializations were imaged . Image stacks were imported to Amira software ( Visage Imaging , San Diego , CA ) , which produced three-dimensional ( 3D ) renderings of each confocal z-stack using the ‘connected components’ feature . CtBP2 and GluA2 puncta in each image stacks were then captured and counted automatically . To assess the appositions of CtBP2 with GluA2 puncta ( putative synapses ) or CtBP2/GluA2 puncta with tdTomato ( synaptic puncta on recombined IHCs ) , the z-stacks were re-imaged using custom software ( Source Code 1 ) that computed and displayed the x-y projection of the voxel space within 0 . 5 µm of the center of each puncta , as identified by Amira analysis ( Liberman et al . , 2011; Lin et al . , 2011 ) . The number of juxtaposed CtBP2 and GluA2 puncta or CtBP2/GluA2 puncta and tdTomato was visualized and counted from these miniature image arrays . Synaptic counts of each z-stack were divided by the number of IHC nuclei , which could be visualized by weak staining of CtBP2 antibody . Each image usually contained 8–10 IHCs .	Noise-induced hearing loss is common , and can result from prolonged exposure to moderate levels of noise that are not perceived as painful or even unpleasant . Some hearing loss can be attributed to the death of hair cells in a part of the inner ear called the cochlea . When sound waves hit the cochlea , they cause the fluid inside it to vibrate: the hair cells detect these vibrations and convert them into electrical signals that are sent along neurons to the brain . However , vibrations that are too strong can destroy hair cells . Increasing evidence suggests that hearing loss also results from damage to the synapses that connect the hair cells and the neurons in the cochlea . During development of the inner ear , molecules called growth factors are needed to ensure the survival of these neurons . Wan et al . predicted that these growth factors might also have a role in adult animals , and that producing more of them might help to safeguard hearing from the damaging effects of noise . Consistent with this , mice that were genetically modified to lack a growth factor called neurotrophin-3 had cochleae that did not work properly and had fewer synapses between hair cells and neurons compared to control mice . Conversely , mice that produced too much neurotrophin-3 had more synapses than controls and also recovered more quickly from the effects of 2 hr exposure to 100 dB noise ( roughly the volume of a pneumatic drill ) . Studies of the cochlea revealed that the extra neurotrophin-3 had boosted the regeneration of synapses damaged by the noise . The beneficial effects of neurotrophin-3 were still seen when overproduction was started shortly after noise exposure , suggesting that it could have therapeutic potential . This is particularly significant in the light of recent evidence that the loss of synapses often comes before the death of hair cells in both age-related hearing loss and noise-induced hearing loss .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Neurotrophin-3 regulates ribbon synapse density in the cochlea and induces synapse regeneration after acoustic trauma
The Hedgehog ( Hh ) signal is transduced across the membrane by the heptahelical protein Smoothened ( Smo ) , a developmental regulator , oncoprotein and drug target in oncology . We present the 2 . 3 Å crystal structure of the extracellular cysteine rich domain ( CRD ) of vertebrate Smo and show that it binds to oxysterols , endogenous lipids that activate Hh signaling . The oxysterol-binding groove in the Smo CRD is analogous to that used by Frizzled 8 to bind to the palmitoleyl group of Wnt ligands and to similar pockets used by other Frizzled-like CRDs to bind hydrophobic ligands . The CRD is required for signaling in response to native Hh ligands , showing that it is an important regulatory module for Smo activation . Indeed , targeting of the Smo CRD by oxysterol-inspired small molecules can block signaling by all known classes of Hh activators and by clinically relevant Smo mutants . The Hedgehog ( Hh ) signaling pathway controls the development of many tissues during embryogenesis ( McMahon et al . , 2003 ) . Even quantitative abnormalities in Hh signaling can lead to human birth defects ( Bale , 2002 ) . After development , Hh signaling regulates tissue stem cells and regenerative responses to injury ( Machold et al . , 2003; Shin et al . , 2011 ) . Aberrant Hh signaling can be oncogenic , and genes encoding Hh pathway proteins can function as oncogenes or tumor suppressor genes ( Scales and de Sauvage , 2009 ) . The most commonly damaged step in Hh-driven cancers involves the poorly understood interaction between two transmembrane ( TM ) proteins , Patched 1 ( Ptch1 ) and Smoothened ( Smo ) ( reviewed in Briscoe and Therond [2013] ) . Ptch1 , encoded by a tumor suppressor gene , is a 12-pass TM protein that serves as the receptor for Hh ligands , including Sonic Hedgehog ( Shh ) ( Marigo et al . , 1996; Stone et al . , 1996 ) . In the absence of Hh ligands , Ptch1 inhibits the function of Smo , a 7-pass TM protein that is encoded by a human oncogene . Shh binds and inactivates Ptch1 , unleashing Smo’s activity and allowing the Gli transcription factors to initiate target gene transcription . Despite the fact that Smo has become a drug target in oncology , with an FDA-approved Smo inhibitor in clinical use ( Von Hoff et al . , 2009 ) and others in ongoing trials , the mechanism by which Smo is regulated by Ptch1 remains a mystery . Current models suggest that Ptch1 , a protein with some homology to bacterial small molecule transporters , regulates Smo through an endogenous ligand whose identity is unknown ( Davies et al . , 2000; Taipale et al . , 2002 ) . Smo consists of an extracellular N-terminal region containing a cysteine rich domain ( CRD ) , a heptahelical transmembrane segment ( 7TM ) and an intracellular C-terminal tail ( C-term ) ( Figure 1A ) . Smo belongs to the G-protein coupled receptor ( GPCR ) superfamily of proteins , most closely related to the Frizzled ( Fz ) group of Wnt receptors ( Dann et al . , 2001; Fredriksson et al . , 2003 ) . Previous work on Smo has largely focused on the 7TM domain , which contains a binding site for cyclopamine , a sterol-like plant alkaloid that was the foundational Hh inhibitor ( Chen et al . , 2002a ) . A battery of subsequent small-molecule screens uncovered a set of exogenous ligands that regulate Smo activity through this site , either as agonists such as SAG or antagonists such as SANT-1 and the FDA-approved Hh-inhibitor Vismodegib ( Frank-Kamenetsky et al . , 2002; Chen et al . , 2002b; Robarge et al . , 2009 ) . The 2 . 5 Å crystal structure of the 7TM segment of Smo bound to a synthetic antagonist has provided a high-resolution view of this binding pocket , which is formed by the extracellular end of the 7TM helix bundle and connecting loops ( Wang et al . , 2013 ) . Smo drugs that occupy this ‘cyclopamine binding site’ are classified as such by their ability to compete with cyclopamine for Smo binding . No endogenous molecules are known that engage this site in the 7TM of Smo . 10 . 7554/eLife . 01340 . 003Figure 1 . The mouse Smo CRD is required to bind oxysterols . ( A ) Schematic of full-length ( FL ) , YFP-tagged mSmo and the ΔCRD and ΔC truncation mutants used in this study . ( B ) Structure of the 20 ( S ) -OHC beads used in Smo pull-down assays . ( C ) EndoH and PGNaseF sensitivity of YFP-mSmo , ΔCRD-YFP-mSmo and ΔC-YFP-mSmo stably expressed in Smo−/− cells and loaded on an 8% Tris-Glycine SDS-PAGE gel . The fraction of each protein with slower mobility on the gel was resistant to EndoH but sensitive to PGNaseF , suggesting post-Golgi localization . ( D ) 20 ( S ) -OHC beads captured YFP-mSmo and ΔC-YFP-mSmo , but not ΔCRD-YFP-mSmo from lysates of cells stably expressing each protein . Binding to beads was not seen when 50 μM free 20 ( S ) -OHC was added as a competitor . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 003 A second binding site on Smo has been defined by side-chain oxysterols , oxidized derivatives of cholesterol carrying an additional hydroxyl group on the iso-octyl chain . Specific oxysterols can fully activate Hh signaling in the absence of Hh ligands in multiple cell types and also induce the accumulation of Smo in the primary cilium , a trafficking step essential for Smo to activate downstream signaling ( Kha et al . , 2004; Corcoran and Scott , 2006; Dwyer et al . , 2007; Kim et al . , 2007; Rohatgi et al . , 2007; Johnson et al . , 2011 ) . We previously demonstrated that a specific side-chain oxysterol , 20 ( S ) -hydroxycholesterol ( 20 ( S ) -OHC ) , directly binds Smo in a manner that is highly stereospecific: the enantiomer , ent-20 ( S ) -OHC , or the epimer , 20 ( R ) -OHC , failed to bind Smo or to activate Hh signaling ( Nachtergaele et al . , 2012 ) . While this ‘oxysterol binding site’ showed allosteric interactions with the canonical cyclopamine binding site , it was clearly distinct since oxysterols did not show a competitive interaction with cyclopamine ( Dwyer et al . , 2007; Nachtergaele et al . , 2012 ) . Indeed , previous structural comparison studies have speculated that oxysterols bind to the extracellular CRD of Smo based on its relationship to the Fz CRD , which binds to the palmitoleyl group of secreted Wnt ligands ( Bazan and de Sauvage , 2009; Bazan et al . , 2012; Janda et al . , 2012; Sharpe and de Sauvage , 2012 ) . Wnt binding to the Fz CRD triggers signaling across the membrane , but the function of the Smo CRD has remained enigmatic . We find that the extracellular CRD of Smo in vertebrates is both necessary and sufficient to bind to 20 ( S ) -OHC , thus demonstrating that the cyclopamine and oxysterol binding sites occupy different domains in Smo . We determined the crystal structure of the zebrafish Smo CRD at 2 . 3 Å to provide a view of the oxysterol-binding pocket and to establish its relationship to the Fz CRD and other Fz-like CRDs that bind small hydrophobic ligands . Either deletion of the CRD or its inhibition by a new class of oxysterol-inspired small molecules can impair the signaling initiated by the native ligand Shh . Our results elucidate the molecular mechanism by which oxysterols activate Smo and show that the Smo CRD is a physiologically and therapeutically important target in the vertebrate Hh pathway . We previously developed a ligand affinity chromatography assay to measure the interaction between 20 ( S ) -OHC and detergent-solubilized , full-length Smo ( Nachtergaele et al . , 2012 ) . For the studies presented here , we used a similar strategy to assay the interaction between truncated versions of Smo ( Figure 1A ) and 20 ( S ) -OHC , using sepharose beads on which 20 ( S ) -OHC was immobilized through an amino group installed on the iso-octyl chain ( hereafter called 20 ( S ) -OHC beads; Figure 1B ) . We produced deletion mutants ( Figure 1A ) of yellow fluorescent protein ( YFP ) -tagged mouse Smo ( mSmo ) lacking the CRD ( ΔCRD-YFP-mSmo ) or the C-terminal intracellular domain ( ΔC-YFP-mSmo ) and confirmed that these proteins were folded when stably expressed in Smo−/− mouse embryonic fibroblasts ( MEFs ) ( Rohatgi et al . , 2009 ) . Both truncated proteins demonstrated a slower migrating species that was resistant to Endoglycosidase H ( EndoH ) , suggesting the presence of glycan modifications usually attached in the Golgi ( Figure 1C ) ( Chen et al . , 2002a ) . For both YFP-mSmo and ΔC-YFP-mSmo , this post-Golgi band was selectively captured on 20 ( S ) -OHC beads , showing that the C-terminal intracellular domain of Smo was dispensable for this interaction ( Figure 1D ) . In this and subsequent experiments , specificity of binding was established by competition with free 20 ( S ) -OHC . In contrast , ΔCRD-YFP-mSmo failed to show an interaction , suggesting that the CRD was required for oxysterol binding . Previous studies have shown that the truncated versions of Smo lacking either CRD or the C-terminal domain remain competent to bind cyclopamine and other cyclopamine-competitive ligands , consistent with these molecules interacting with the 7TM segment ( Chen et al . , 2002a; Wang et al . , 2013 ) . ΔCRD-YFP-mSmo also remained responsive to 7TM ligands ( described below ) , confirming proper folding . 10 . 7554/eLife . 01340 . 004Figure 2 . The isolated mSmo CRD can bind oxysterols . ( A ) Fractionation of the mSmo CRD-Fc protein on a Superose 6 gel-filtration column . The UV280 absorbance of each fraction ( blue curve ) is shown above the protein content of each fraction on a silver stained gel . Monodisperse protein ( fractions 13–15 ) elutes in a sharp peak and binds to 20 ( S ) -OHC beads ( panels below ) , while aggregated protein runs as a broad peak ( fractions 5–12 ) and fails to bind oxysterols . The indicated fractions ( red boxes ) were incubated with 20 ( S ) -OHC beads in the presence or absence of free 20 ( S ) -OHC competitor , and the amount of mSmo CRD-Fc protein captured on the beads or left in the flow through was assayed on an anti-Fc immunoblot . ( B ) A binding curve ( Kd ∼180 nM ) for the mSmo CRD-Fc-20 ( S ) -OHC interaction was measured by incubating a fixed amount of protein with increasing amounts of bead-immobilized sterol . ( C ) Binding of mSmo CRD-Fc to 20 ( S ) -OHC beads is inhibited in a dose-responsive fashion by free 20 ( S ) -OHC but not by the enantiomer ent-20 ( S ) -OHC . A competition assay was used to test the ability of various oxysterols ( D ) or Smo ligands ( E ) to inhibit the binding of mSmo CRD-Fc to 20 ( S ) -OHC beads . Anti-Fc immunoblots show the amount of protein in the input , captured on the beads , and left in the flow-through . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 004 To determine if the mSmo CRD was sufficient to bind 20 ( S ) -OHC , we purified isolated mSmo CRD fused to the constant region of the human IgG heavy chain ( mSmo CRD-Fc; Figure 2A ) . The mSmo CRD-Fc protein secreted into the media of 293F cells ran as a smear on an SDS-PAGE gel . Further purification by Protein A affinity chromatography followed by gel filtration allowed us to isolate monodisperse mSmo CRD-Fc ( Figure 2A , fractions 13–15 ) . This well-behaved protein bound to 20 ( S ) -OHC beads . A significant population of the protein was clearly misfolded , as it fractionated as a broad peak on a gel filtration column and failed to bind to 20 ( S ) -OHC ( Figure 2A , fractions 5–12 ) . Binding of mSmo CRD-Fc to 20 ( S ) -OHC beads was saturable ( Figure 2B ) , specific ( Figure 2C ) and followed the same requirements for oxysterol stereochemistry and regiochemistry as previously described ( Figure 2D ) ( Nachtergaele et al . , 2012 ) . Binding could be inhibited by free 20 ( S ) -OHC and free 20 ( S ) -yne , the ∼10-fold more potent alkyne analog of 20 ( S ) -OHC ( Nachtergaele et al . , 2012 ) . However , the enantiomer ent-20 ( S ) -OHC , the epimer 20 ( R ) -OHC , and 22 ( S ) -OHC ( all sterols that cannot activate Hh signaling ) were unable to inhibit binding ( Nachtergaele et al . , 2012 ) . Ligands known to engage Smo at the cyclopamine binding site , SAG and SANT-1 , failed to inhibit the binding of mSmo CRD-Fc to 20 ( S ) -OHC beads , as did Itraconazole , a purported Smo ligand that binds to an unknown site ( Figure 2E ) ( Chen et al . , 2002a , 2002b; Kim et al . , 2010 ) . While our manuscript was in preparation , an independent study also reported the interaction between oxysterols and the Smo CRD ( Nedelcu et al . , 2013 ) . Overall , our results show that the cyclopamine and oxysterol binding sites on Smo are distinct . For clarity , we hereafter refer to these sites as the 7TM and CRD sites , respectively . To investigate the function of the Smo CRD for signaling induced by the native ligand Shh , YFP-tagged mSmo variants were expressed by stable retroviral transduction in Smo−/− MEFs to avoid the confounding effects of endogenous Smo . These clonal Smo−/−:YFP-mSmo cells could activate a Hh target gene , Gli1 , when exposed to Shh ( which binds and inactivates Ptch1 ) or to the Smo agonists SAG and 20 ( S ) -OHC , which bind to the 7TM and CRD sites , respectively ( Figure 3A ) ( Rohatgi et al . , 2009 ) . In an independent , non-transcriptional measure of signaling , loss of the repressor form of Gli3 ( Gli3R ) was observed in response to all three agonists ( Figure 3A ) . In contrast , ΔCRD-YFP-mSmo failed to activate Hh target genes or to extinguish Gli3R levels in response to both Shh and 20 ( S ) -OHC , but retained its ability to respond to SAG ( Figure 3A ) . Identical results were obtained using a luciferase-based Hh reporter transiently expressed along with YFP-mSmo or ΔCRD-YFP-mSmo in Smo−/− cells ( Figure 3B ) ( Sasaki et al . , 1997; Varjosalo et al . , 2006 ) . SAG activated ΔCRD-YFP-mSmo remained susceptible to inhibition by cyclopamine , consistent with an intact 7TM site ( Figure 3C ) . As noted previously , ΔCRD-YFP-mSmo was not constitutively active , but it did demonstrate a higher level of basal activity in Hh reporter assays ( Taipale et al . , 2002; Aanstad et al . , 2009 ) . The SAG responsiveness shows that ΔCRD-YFP-mSmo is not a misfolded or inactive protein; instead , it supports the notion that the CRD of Smo mediates the response to oxysterols while the 7TM segment mediates the response to SAG . Most significantly , this result suggests that the CRD plays an important role in mediating the response to Shh and thus in mediating the interaction between Ptch1 and Smo . 10 . 7554/eLife . 01340 . 005Figure 3 . The mSmo CRD is required for Shh- and oxysterol-mediated activation of Hh signaling . ( A ) Smo−/− cells stably expressing full-length ( FL ) YFP-mSmo or ΔCRD-YFP-mSmo were treated with Shh , SAG ( 100 nM ) or 20 ( S ) -OHC ( 10 μM ) . Levels of Gli1 and Gli3R protein , determined by immunoblotting after fractionation on an 8% Tris-glycine SDS-PAGE gel , were taken as a metric of pathway activation . An anti-YFP blot shows the levels of YFP-mSmo in each sample , and p38 levels are used as a loading control . ( B and C ) A luciferase-based Hh reporter gene was used to measure signaling in Smo−/− cells transiently transfected with constructs encoding YFP-mSmo or ΔCRD-YFP-mSmo and then treated with the indicated Smo ligands . In ( C ) , ΔCRD-YFP-mSmo activated with SAG ( 25 nM ) can be inhibited by the co-administration of cyclopamine ( 5 μM ) . Error bars denote S . D . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 005 We tested the binding of Smo from various species to 20 ( S ) -OHC beads ( Figure 4A , B ) . Both full-length Drosophila melanogaster Smo ( dSmo ) and the isolated dSmo CRD failed to bind 20 ( S ) -OHC beads . However , a truncated version of zebrafish Smo ( zSmo ) lacking the intracellular C-terminal region ( YFP-zSmoΔC ) , expressed in mammalian cells and solubilized with detergent , bound to 20 ( S ) -OHC beads , showing that this interaction is likely conserved in the vertebrate ( but not in the invertebrate ) Hh pathway . 10 . 7554/eLife . 01340 . 006Figure 4 . The Smo-oxysterol interaction is conserved in vertebrates . ( A ) The interaction of 20 ( S ) -OHC beads with full-length mSmo , full-length Drosophila Smo ( dSmo ) or zebrafish Smo ( zSmo ) carrying a truncation of the intracellular C-terminal tail ( zSmoΔC ) was tested in the presence of free 20 ( S ) -OHC or its enantiomer . ( B ) The zSmo ectodomain ( which includes the CRD ) can bind to 20 ( S ) -OHC beads , but the dSmo CRD cannot . ( C ) Zebrafish embryos ( 30hpf ) carrying a GFP transgene driven by the engrailed2a promoter were treated with 20 ( S ) -OHC ( 50 µM ) or cyclopamine ( 40 µM ) and assessed for GFP expression by fluorescence and ptch2 expression by in situ hybridization . See Figure 4—figure supplement 1 for quantitation . ( D ) A binding curve ( Kd ∼170 nM ) for the zSmo ectodomain-20 ( S ) -OHC interaction was measured by incubating a fixed amount of protein with increasing amounts of bead-immobilized sterol . The amount of zSmo ectodomain captured on the beads ( shown in the graph ) was quantitated from a coomassie-stained SDS-PAGE gel shown above . ( E ) Binding of the zSmo ectodomain to 20 ( S ) -OHC beads was inhibited in a dose-responsive fashion by free 20 ( S ) -OHC but not by its enantiomer . ( F and G ) Coomassie-stained SDS-PAGE gels show the amount of zSmo ectodomain captured on 20 ( S ) -OHC beads in the presence of various oxysterols ( F ) or Smo ligands ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 00610 . 7554/eLife . 01340 . 007Figure 4—Figure supplement 1 . 20 ( S ) -OHC activates Hedgehog signaling in zebrafish embryos . Zebrafish embryos ( 30hpf ) carrying a GFP transgene driven by the engrailed2a ( eng2a:GFP ) promoter were treated with 20 ( S ) -OHC ( 50 µM ) or cyclopamine ( 40 µM ) and assessed for GFP expression by fluorescence . To quantify the effect of each treatment on eng2a:GFP expression , the width of the GFP-positive domain was measured at three points along the length of each embryo ( right panel , white lines ) and averaged . The average width is plotted as a scatter plot , with each point representing one embryo . 12–16 embryos per condition are depicted . Red bars represent mean ± SD . All conditions were significantly different from each other ( ***p<0 . 0001 , one-way ANOVA with Bonferroni correction for multiple comparisons . ) DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 007 We tested whether the zebrafish Hh pathway was responsive to oxysterols , because our structural studies described below focused on Smo protein from this species . Full-length zebrafish Smo was poorly expressed in mammalian cells , precluding tests of its responsiveness to oxysterols in cultured cells . Hh pathway activity underlies the specification of distinct muscle cell types in the zebrafish embryo , in part through the activation of the engrailed2 ( eng2 ) gene in subsets of slow-twitch and fast-twitch myoblasts ( Wolff et al . , 2003 ) . To investigate the in vivo significance of the interaction between 20 ( S ) -OHC and Smo , we treated embryos carrying an eng2a:GFP reporter construct ( Maurya et al . , 2011 ) with either 20 ( S ) -OHC or cyclopamine . As expected , cyclopamine treatment suppressed expression of the reporter gene; by contrast , 20 ( S ) -OHC treated embryos showed a significant increase in the number of GFP positive fast twitch muscles compared to vehicle treated embryos ( Figure 4C and Figure 4—figure supplement 1 ) . Consistent with this , 20 ( S ) -OHC treated embryos also showed a modest expansion of the expression domain of the endogenous Hh target gene , ptch2 . These data show that 20 ( S ) -OHC can induce Hh signaling in the context of a living vertebrate embryo and suggest that the in vitro interaction between zebrafish Smo and 20 ( S ) -OHC induces its activation in vivo . We succeeded in purifying large quantities of the zSmo ectodomain , encompassing both the CRD and the segment between the CRD and the first transmembrane helix . The zSmo ectodomain demonstrated saturable , specific binding to 20 ( S ) -OHC beads ( Figure 4D , E ) . Similar to the mouse protein , binding could be inhibited by oxysterols that activate Hh signaling but not by those that cannot ( Figure 4F ) ; 7TM site ligands also failed to compete for binding ( Figure 4G ) . To obtain molecular insights into the architecture of the Smo extracellular region , we crystallized the zSmo ectodomain and determined its structure using selenomethionine-labeled protein for phasing ( Table 1 , Figure 5—figure supplement 1A ) . Refinement resulted in an R-factor of 21 . 6% ( R-free: 26 . 0% ) with two zSmo molecules in the crystallographic asymmetric unit , each composed of a well-defined model that included residues 41–158 ( root mean square deviation [RMSD]: 0 . 60 Å for 118 Cα positions , Figure 5—figure supplement 1B ) . Although we set-up crystallization trials with the entire zSmo ectodomain ( residues 29–212 ) , the N- and C-terminal regions could not be traced due to missing electron density and thus were not included in the final model ( Figure 5—figure supplement 2 ) . The portion of the zSmo ectodomain spanning residues 41–158 ( visible in our structure ) shows sequence similarity to the previously identified CRD in the Fz protein family ( Dann et al . , 2001 ) and thus will hereafter be called the zSmo CRD . 10 . 7554/eLife . 01340 . 008Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 008SeMet-substituted zSmo CRDNative zSmo CRDData collection BeamlineESRF-ID23-EH1DIAMOND I03 Wavelength0 . 9791 . 000 Space groupP43212P43212 Cell Dimension ( Å ) a , b = 68 . 2; c = 92 . 3a , b = 68 . 6; c = 95 . 3 Resolution48 . 0–2 . 3 ( 2 . 36-2 . 30 ) 31 . 0–2 . 6 ( 2 . 67-2 . 60 ) Completeness ( % ) 99 . 2 ( 92 . 8 ) 98 . 9 ( 94 . 7 ) Unique reflections10 , 0857307 ( 491 ) Rmerge ( % ) 10 . 0 ( 92 . 0 ) 13 . 4 ( 64 . 8 ) I/σ ( I ) 25 . 5 ( 3 . 9 ) 14 . 4 ( 2 . 4 ) Multiplicity26 . 5 ( 23 . 7 ) 8 . 8 ( 6 . 0 ) Refinement Resolution range ( Å ) 30 . 50–2 . 3031 . 00–2 . 60 No . reflections95767275 Rwork ( % ) 23 . 621 . 6 Rfree ( % ) 28 . 426 . 0 No . atoms ( protein/Zn/Na/water ) 1880/1/3/281880/1/2/48 B-factors ( Å2 ) ( protein/Zn/Na/water ) 57/60/43/4840/32/27/30 r . m . s . deviations Bond lengths ( Å ) 0 . 0120 . 004 Bond angles ( ° ) 1 . 6040 . 714 Ramachandran statistics Favored ( % ) 96 . 597 . 9 Disallowed ( % ) 0 . 40Each structure was determined from one crystal . Numbers in parentheses refer to the highest resolution shell . Rfree equals the R-factor against 5% of the data . The small interface between the two zSmo CRD molecules observed in the asymmetric unit of the crystal ( buried surface area of 490 Å2 ) and a crystal contact formed by a zinc ion bonded to three different protein chains ( one chain A and two chain B molecules; Figure 5—figure supplement 1C ) suggested that the dimeric arrangement observed in the crystal is not likely to be of functional significance . In agreement with this crystal packing analysis , purified zSmo ectodomain behaved as a monomer in solution at low concentration ( 5 µM ) when assessed using multi angle light scattering ( Figure 5—figure supplement 1D ) . The zSmo CRD monomer adopts a globular fold composed of four α helices ( α1: residues Q77-N92; α2: P94-Y108; α3: Q122-N130; α3′: S133-E138 ) and a short two-stranded β sheet ( β1: K43-S45 and β2: K116-E118; Figure 5A and Figure 5—figure Supplement 2 ) . This arrangement is stabilized by five disulfide bridges ( labeled , I , II , III , IV in Figure 5A ) . Disulfide bridges I , II , III , and IV lock the four helices together into a tight bundle , whereas disulfide bridge * , formed by the N- and C-terminal cysteines , orients the termini in close proximity and away from the helical bundle ( Figure 5A ) . Structure-based evolutionary analysis of zSmo CRD revealed that the closest structural relatives are the CRDs of Frizzled 8 ( Fz8; Dann et al . , 2001; Janda et al . , 2012 ) , secreted Frizzled-related protein 3 ( sFRP3 , Dann et al . , 2001 ) and muscle-specific kinase ( MuSK , Stiegler et al . , 2009 ) , shown clustered in the blue branch in Figure 5B ( Figure 5—figure supplement 3 ) . These three structures show a similar helical bundle arrangement compared to the zSmo CRD , with the exception of a rearrangement of helix α3 and α3′ , which forms a continuous helix in Fz8 and sFRP3 . Strikingly , 4 out of 5 disulfide bridges are highly conserved ( I , II , III , and IV ) , retaining the overall fold of the helix bundle . Only one disulfide bridge ( labeled with an asterisk * in Figure 5A ) is not conserved , resulting in a rearrangement of the relative orientations of the two termini compared to the zSmo CRD ( Figure 5C–E ) . 10 . 7554/eLife . 01340 . 009Figure 5 . Structural analysis of the zebrafish Smo CRD . ( A ) Ribbon diagram of zSmo CRD in rainbow coloring from blue ( N-terminus ) to red ( C-terminus ) with the secondary structure elements numbered . The four disulfide bridges ( black sticks ) conserved in all Fz-like CRDs are depicted with Roman numerals , and the non-conserved disulfide bridge is marked with an asterisk ( * ) . N- and C-termini are labeled . ( B ) Structural phylogenetic analysis of the CRDs . Structural superposition of CRDs from zSmo , Frizzled 8 ( Fz8 , PDB ID 4F0A , Janda et al . , 2012 ) , secreted Frizzled-related protein 3 ( sFRP3 , PDB ID 1IJX , Dann et al . , 2001 ) , muscle-specific kinase ( MuSK , PDB ID 3HKL , Stiegler et al . , 2009 ) , Niemann-Pick C1 protein ( NPC1 , PDB ID 3GKI , Kwon et al . , 2009 ) , riboflavin-binding protein ( RFBP , Monaco , 1997 ) , and folate receptor α ( FRα , PDB ID 4LRH , Chen et al . , 2013 ) were superimposed using SHP ( Stuart et al . , 1979; Riffel et al . , 2002 ) . CRDs that form ligand-binding pockets ( red background ) or grooves ( blue background ) form two distinct evolutionary branches . In addition , CRDs show distant structural similarity to the extracellular domains of glypicans ( Pei and Grishin , 2012 ) . However , analysis of the crystal structures of glypicans Dally-like protein and glypican 1 revealed no apparent grooves or pockets that could accommodate small molecules ( Kim et al . , 2011; Svensson et al . , 2012 ) and thus were not included in our structural analyses . ( C–H ) Ribbon diagrams of superimposed Fz-like CRD domains from the structural phylogenetic analysis in ( B ) . ( C ) Fz8-palmitoleyl complex , ( D ) sFRP3 , ( E ) MuSK , ( F ) NPC1-cholesterol complex , ( G ) RFBP-riboflavin complex , ( H ) FRα-folate complex . Color coding and labeling follows ( A ) . Ligands are shown as spheres in atomic coloring ( carbon: slate; oxygen: red; nitrogen: blue ) . In ( F–H ) the conserved disulfide bridges are highlighted with a circle . NPC1 ( F ) does not contain disufide bridge IV . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 00910 . 7554/eLife . 01340 . 010Figure 5—figure supplement 1 . Electron density of the zSmo CRD structure and oligomeric state of the zSmo ectodomain . ( A ) SeMet SAD-phased and density modified map from RESOLVE ( Terwilliger , 2003 ) calculated to 2 . 3 Å resolution and contoured at 1 . 0 σ showing the two core zSmo CRD helices α1 and α2 . ( B ) SigmaA-weighted 2FO-FC map of the final model of SeMet-labeled zSmo ectodomain from REFMAC ( Murshudov et al . , 1997 ) at 2 . 3 Å resolution and contoured at 1 . 0 σ . View is the same as in ( A ) . ( C ) Close-up view of the zinc-binding site in the zSmo CRD crystal structure . The anomalous difference Fourier map ( yellow , contoured at 5 σ ) and SigmaA-weighted 2FO-FC map ( blue , contoured at 1 . 0 σ ) of the final model of native zSmo CRD were calculated to 2 . 6 Å . Note that zinc is present in a crystal contact formed by three different zSmo chains . ( D ) Multi angle light scattering of the glycosylated zSmo ectodomain ( expressed in mammalian cells ) indicates a molecular mass ( red scattered dots ) of 24 . 43 ± 0 . 9 kDa and is in agreement with the theoretical molecular mass for a non-glycosylated monomer ( 20 . 4 kDa ) . The zSmo ectodomain has two predicted N-linked glycosylation sites ( each accounting for 2 kDa ) , which explains the difference between the theoretical and MALS-derived molecular mass . Protein concentration at the elution peak was 8 . 123×10−5 g/ml . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 01010 . 7554/eLife . 01340 . 011Figure 5—figure supplement 2 . Sequence alignment of the ectodomains of Smo family members and the CRD of mFz8 . Sequences were aligned using ClustalW ( Larkin et al . , 2007 ) and adjusted manually for mFz8 . Secondary structure assignments of zSmo CRD and mFz8 ( PDB ID 4F0A , Janda et al . , 2012 ) are displayed above the alignment and color-coded as in Figure 5 . Disulfide bonds are highlighted and numbered as in Figure 5A . Smo disulfide bond * , which is not conserved in the CRD protein family , is marked in yellow . The two cysteine residues of mFz8 forming the rearranged disulfide bond ( marked with * in Figure 5C ) are highlighted in violet . The box indicates the zSmo residues visible in our crystal structure . Residues lining the oxysterol binding groove in Smo are highlighted in red for zSmo and residues lining the palmitoleyl-binding groove in mFz8 are in blue . Mutated mSmo residues that substantially reduced binding to 20 ( S ) -OHC beads are depicted with a plus ( + ) below the alignment . Mutated residues that did not reduce binding are marked with a number sign ( # ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 01110 . 7554/eLife . 01340 . 012Figure 5—figure supplement 3 . Structural comparison of CRDs . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 012 Using structure fold recognition methods , Bazan and de Sauvage identified an additional group of Fz-like CRD containing proteins ( Bazan and de Sauvage , 2009 ) . These include the Niemann-Pick C1 protein ( NPC1 ) and the riboflavin-binding protein ( RFBP ) . Our evolutionary structural analysis confirmed their findings and allowed us to add the folate receptor α ( FRα ) to this group ( Figure 5B , red branch and Figure 5—figure supplement 3 ) ( Chen et al . , 2013 ) . Structural comparison of these proteins to the zSmo CRD revealed the common features identified for Fz-like CRDs , namely the helical bundle ( formed by helices α1 , α2 and α3 ) and the four conserved disulfide bonds that stabilize the fold and the relative orientations of the helices ( Figure 5F–H ) . A common feature of the Fz-like CRD family members is their ability to bind small , hydrophobic molecules in a pocket formed by the core helices α1 , α2 and α3 . While NPC1 , RFBP and FRα bury their respective ligands in the protein core ( cholesterol in NPC1 , riboflavin in RFBP and folate in FRα ) with the help of extensive protrusions from the core CRD fold ( shown in gray in Figure 5F–H ) , Fz8 , the closest structural homolog of the zSmo CRD structure , binds the palmitoleyl moiety covalently linked to Wnt proteins in a shallow groove ( Figure 5C; Janda et al . , 2012 ) . To investigate the putative oxysterol binding site in the Smo CRD , we calculated the volumes of potential binding pockets in our zSmo CRD structure . The most prominent groove is indeed located at an equivalent position to the Fz8 palmitoleyl-binding groove ( Figure 6A , B ) . The residues forming this groove are highly conserved in all vertebrate Smo CRDs ( Figure 6C and Figure 5—figure supplement 2 ) , and the volume ( 551 Å3 ) and shape of the groove is sufficient for 20 ( S ) -OHC binding . Computational docking using AutoDock ( Morris et al . , 2009 ) showed that the hydrophobic groove on the zSmo CRD surface ( Figure 6A ) can accommodate 20 ( S ) -OHC with a favorable free energy of binding ( Figure 6—figure supplement 1A–C ) . The four rings of the oxysterol are predicted to lie on the base of the groove lined with zSmo residues W87 and L90 and make additional potential hydrophobic interactions with zSmo residues M86 , G89 , Y108 , G140 , P142 and F144 . 10 . 7554/eLife . 01340 . 013Figure 6 . Mapping and analysis of the zSmo oxysterol binding site . ( A and B ) Ribbon representations of the zSmo CRD ( A ) and the Fz8 CRD-palmitoleyl ( B ) structures . View and presentation follows Figure 5A . The palmitoleyl-binding pocket of Fz8 is depicted as a cyan wire mesh and the corresponding pocket in the zSmo CRD structure is in red wire mesh . Volumes were calculated using the program Volumes ( RE Esnouf , unpublished ) , with a 1 . 4 Å probe radius . The palmitoleyl moiety is shown as slate spheres . ( C ) The solvent accessible surface of the zSmo CRD is color-coded according to residue conservation ( from non-conserved , white , to conserved , black ) based on alignments containing amino acid sequences from >80 vertebrate Smo proteins . The right panel is rotated 90° around the y-axis relative to the left panel . Residues on the opposite face of the oxysterol-binding pocket that were subjected to mutagenesis are labeled . ( D ) Close-up view of the potential 20 ( S ) -OHC binding site in the zSmo CRD structure . Residues predicted to make contacts with 20 ( S ) -OHC are shown in stick representation and highlighted in red . Boundaries of the hydrophobic groove are marked with dotted lines . zSmo residues are numbered , with the corresponding mSmo residues in parentheses . ( E ) The indicated full-length mSmo point mutants were tested for their interaction with 20 ( S ) -OHC beads in the absence or presence of free 20 ( S ) -OHC competitor . Well-folded Smo mutants ran as a double band on a 4–12% Bis-Tris gradient gel ( arrowheads ) , with only the slower-migrating species being captured on 20 ( S ) -OHC beads . ( F ) A Hh reporter assay was used to measure signaling in Smo−/− cells transiently transfected with constructs encoding various mSmo point mutants and then treated ( 48 hr ) with Shh , SAG ( 100 nM ) or 20 ( S ) -OHC ( 10 μM ) . The maximum reporter response for each mutant was set to 100% . Error bars denote S . D . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 01310 . 7554/eLife . 01340 . 014Figure 6—figure supplement 1 . Molecular modeling analysis of the zSmo CRD . ( A ) Close-up view of the Smo oxysterol-binding groove . Presentation is as in Figure 6D . Boundaries of the potential binding site are marked with dashed lines . ( B ) The computationally docked structure of 20 ( S ) -OHC ( stick representation; carbon: orange , oxygen: red ) in complex with the zSmo CRD suggests energetically favorable interactions between the two molecules with an estimated free binding energy of −9 . 0 kcal/mol and estimated inhibition constant , Ki , equal to 260 nM . View is as in ( A ) . ( C ) Close-up view of the palmitoyl binding site in the Fz8-Wnt complex ( PDB ID 4F0A , Janda et al . , 2012 ) in the same orientation as in ( A ) . ( D ) Homology model of the Drosophila Smo-CRD ( dSmo ) based on the zSmo structure ( sequence identity: 42% ) reveals a substantially wider groove compared to grooves of the CRDs from zSmo and Fz8 . Three key residues ( Met86 , Tyr108 , Gly140 ) are absent in dSmo that have been shown to be essential for 20 ( S ) -OHC binding to vertebrate Smo . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 01410 . 7554/eLife . 01340 . 015Figure 6—figure supplement 2 . Mutagenesis of the putative oxysterol binding site in the mSmo CRD . ( A ) The indicated full-length mSmo point mutants were tested for their interaction with 20 ( S ) -OHC beads in the absence or presence of free 20 ( S ) -OHC competitor . ( B ) Heat map summarizing the signaling properties of all Smo mutants tested in this study in response to Shh , SAG and 20 ( S ) -OHC , using assays of the type shown in Figure 6E , F . Mutants that were not responsive to SAG and/or ran as a single band on an SDS-PAGE gel ( indicating lack of glycan chains attached in the Golgi apparatus ) were deemed not folded ( NF ) . Mutants were only assessed for Shh and 20 ( S ) -OHC responsiveness if they had SAG responsiveness that was at least 75% of that seen with wild-type mSmo . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 015 To test this model for the oxysterol-binding pocket , we mutated Smo residues that map to this pocket and , as controls , other residues that point away from the pocket or that are on the opposite face of the molecule ( Figure 6C , D ) . All mutations were made in full-length mouse Smo , and mutant proteins were tested for binding to 20 ( S ) -OHC beads after detergent-solubilization from membranes ( Figure 6E ) . Figure 6D shows corresponding mouse and zebrafish residue numbers , and hereafter the residues are numbered according to the mouse sequence . Only those mutants that fractionated as a doublet on an SDS-PAGE gel were evaluated because this property demonstrates post-Golgi trafficking and hence correct folding ( Figure 1C ) . Mutations in residues on the opposite face of the putative sterol binding pocket ( E162A , P120A/E/G , P128S/E/R , P88N , L150A/D/S ) or at the periphery of the pocket ( R165A/E and N118A ) did not disrupt binding to 20 ( S ) -OHC beads ( Figure 6E and Figure 6—figure supplement 2 ) . In contrast , mutations in residues that frame the putative oxysterol pocket ( L112A , L112D , G115F , L116A , Y134F , G166F , P168A , F170A ) substantially reduced binding to 20 ( S ) -OHC beads ( Figure 6E and Figure 6—figure supplement 2 ) . Taken together , our mutagenesis data support the structure-based model for the interaction between oxysterols and the Smo CRD . To understand why Drosophila Smo does not bind oxysterols , we constructed a homology model of the dSmo CRD based on the zSmo structure ( Figure 6—figure supplement 1D ) . Despite the notable sequence identity between zebrafish and Drosophila Smo CRDs ( ∼42% ) and the conserved disulfide bond pattern , the homology model revealed a substantially different oxysterol-binding groove on the dSmo CRD surface . 5 out of 8 residues that are essential for vertebrate Smo interactions with oxysterols ( zSmo residues M86 , W87 , G89 , Y108 and G140 ) are different in dSmo ( corresponding dSmo residues D129 , Y130 , A132 , F151 and F187; Figure 6—figure supplement 1D ) , potentially providing an explanation for why dSmo does not bind to oxysterols . Finally , we tested a subset of these mSmo mutants for their ability to rescue Hh signaling in Smo−/− cells treated with Shh , SAG or 20 ( S ) -OHC . The mutations that preserved 20 ( S ) -OHC binding also preserved mSmo responsiveness to all three agonists ( Figure 6F and Figure 6—figure supplement 2B ) . The most informative mutations were G115F , P168A and Y134F , the last a conservative change that substitutes a Drosophila residue ( F ) for the corresponding mouse residue ( Y ) . All three mutants were responsive to SAG , showing that they were not disabled , but demonstrated substantially reduced 20 ( S ) -OHC binding and responsiveness , with the mSmo Y134F being completely unresponsive ( Figure 6F ) . Interestingly , Shh-responsiveness was unaffected in mSmo G115F but significantly reduced in both mSmo Y134F and P168A . Finally , there were a few mutants ( e . g . , L116A ) that did not show strong binding to 20 ( S ) -OHC beads in our in vitro assay but still modestly responded to 20 ( S ) -OHC when introduced into Smo−/− cells . This discrepancy may be due to the fact that our signaling assay in intact cells is more sensitive than the binding assay with solubilized proteins , which is conducted in the presence of high detergent to maintain Smo solubility after extraction from membranes . The current generation Smo inhibitors that have entered the clinic , including the FDA-approved drug Vismodegib , all engage the 7TM site on Smo ( Frank-Kamenetsky et al . , 2002 ) . However , mutations that prevent drug binding or drug activity can lead to clinically relevant resistance to these agents ( Yauch et al . , 2009; Dijkgraaf et al . , 2011 ) . Antagonists that engage the oxysterol binding site in the CRD would represent an orthogonal strategy for Smo inhibition . To design such inhibitors , we considered two observations from our prior structure–activity relationship ( SAR ) studies on 20 ( S ) -OHC ( Nachtergaele et al . , 2012 ) . First , the stereochemistry at position 20 that determines the spatial relationship between the ring system and the iso-octyl chain is critical for the ability of 20 ( S ) -OHC to activate Smo , since 20 ( R ) -OHC is inactive . Second , the replacement of the iso-butyl group at the end of the iso-octyl chain with an alkyne group increased Hh-activation potency by ∼10-fold ( Figure 7A ) . Starting from this high-potency Smo activator 20 ( S ) -yne , we inverted the stereochemistry at position 20 to make 20 ( R ) -yne or oxidized the hydroxyl group to a ketone , changing carbon 20 to a planar sp2 hybridized center , to make 20-keto-yne ( Figure 7A ) . Both molecules blocked the binding of mSmo CRD-Fc to 20 ( S ) -OHC beads but did not affect the binding of a fluorescent cyclopamine derivative ( bodipy-cyclopamine ) to Smo-expressing cells , showing that they engaged the CRD site but not the 7TM site ( Figure 7B , C ) . The alkyne group was an important structural feature required for competition , as both 20 ( R ) -OHC and 20-keto-cholesterol ( Figure 7A ) failed to inhibit the CRD–20 ( S ) -OHC interaction ( Figure 7B ) . 10 . 7554/eLife . 01340 . 016Figure 7 . Partial agonists that target the Smo CRD . ( A ) Structure and synthetic logic for 20 ( R ) -yne and 20-keto-yne . 20 ( R ) -OHC and 20-keto-cholesterol are related analogs that lack the alkyne moiety . ( B ) Immunoblots show the amount of mSmo CRD-Fc captured on 20 ( S ) -OHC beads in the presence of the indicated oxysterols added as competitors . ( C ) Binding of bodipy-cyclopamine to cells expressing full-length mSmo was determined by FACS in the presence of various Smo ligands . The bodipy-cyclopamine fluorescence in a cell population is expressed as a cumulative distribution function ( CDF ) , which denotes the percentage of cells that show a given level of fluorescence or lower . Bodipy-cyclopamine binding can be competed by SANT-1 and Vismodegib , two 7TM site ligands , but not by any of the CRD-binding oxysterols . ( D ) Hh reporter activity in cells treated with increasing concentrations of the indicated oxysterols . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 016 Despite binding to the mSmo CRD , 20 ( R ) -yne and 20-keto-yne were weak activators of signaling in the absence of Shh , reinforcing the importance of stereochemistry at position 20 for Smo activation ( Figure 7D ) . However , both molecules inhibited signaling induced by the native ligand Shh , the CRD agonist 20 ( S ) -OHC or the 7TM agonist SAG ( Figure 8A–C and Figure 8—figure supplement 1 ) . Both the molecules also reduced signaling by mSmoM2 , a constitutively active , oncogenic Smo mutant ( Taipale et al . , 2000 ) , and mSmo D477H , the mouse version of a human Smo mutant that is resistant to the FDA-approved drug Vismodegib ( Figure 8D , E ) ( Yauch et al . , 2009 ) . We hereafter call these molecules oxysterol-based inhibitors or OBIs . This activity profile shows that the OBIs are CRD-targeted partial agonists of Smo that can reduce signaling by Smo activators and by clinically relevant Smo mutants . Our OBIs seem to inhibit Smo by a different mechanism compared to another recently reported CRD antagonist , 22-azacholesterol , which does not block signaling induced by SAG or by mSmoM2 ( Nedelcu et al . , 2013 ) . The broader Hh inhibitory activity of OBIs is instead reminiscent of the glucocorticoids Budesonide and Ciclesonide , which also fail to compete with cyclopamine for binding to Smo ( Wang et al . , 2012 ) . 10 . 7554/eLife . 01340 . 017Figure 8 . 20 ( R ) -yne and 20-keto-yne can inhibit Hh signaling . ( A–C ) Hh reporter activation by 100 nM SAG ( A ) , 5 μM 20 ( S ) -OHC ( B ) and Shh ( C ) can be inhibited by both 20 ( R ) -yne and 20-keto-yne . ( D ) Hh reporter activity in Smo−/− cells transfected with wild-type mSmo or mSmo D477H and then treated ( 48 hr ) with Shh in the presence of 20 ( R ) -yne , 20-keto-yne ( both at 25 µM ) or vismodegib ( 100 nM ) . ( E ) Hh reporter activity in NIH 3T3 cells transfected with constitutively active , oncogenic mSmoM2 and then treated ( 12 hr ) with 20 ( R ) -yne and 20-keto-yne ( 10 µM each ) or cyclopamine ( 1 µM ) . ( F ) Accumulation of endogenous mSmo in cilia of NIH 3T3 cells treated ( 4 hr ) with the indicated Smo ligands in the presence or absence of Shh . Each point represents the Smo fluorescence in a single cilium and the red lines denote the median and the interquartile range of mSmo fluorescence ( n = 60 for each condition ) . ( G ) The binding of mSmo CRD-Fc to 20 ( S ) -OHC beads can be inhibited by cyclopamine but not by the structurally-related alkaloid tomatidine . Error bars denote S . D . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 01710 . 7554/eLife . 01340 . 018Figure 8—figure supplement 1 . Table of IC50 values for the OBIs . In Figure 8A , B , C , we demonstrate the inhibitory activity of 20 ( R ) -yne and 20-keto-yne in the presence of three different Hh activators ( SAG , 20 ( S ) -OHC and Shh ) . Using curve fitting ( described in detail in methods ) , we derived IC50 values for each inhibitor in the presence of the three different activators . Values represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01340 . 018 An early step in signaling that precedes transcription is the Shh-induced accumulation of Smo in the primary cilium ( Corbit et al . , 2005 ) . Antagonists that bind to the 7TM site display striking differences in their impact on this key trafficking step . Cyclopamine and cyclopamine derivatives ( that contain a sterol-like tetracyclic ring structure ) do not block Smo ciliary accumulation ( Figure 8F ) and in fact can drive Smo accumulation in cilia even in the absence of Shh ( Rohatgi et al . , 2009 ) . On the other hand , non-sterol 7TM antagonists like SANT-1 and Vismodegib prevent Shh-induced Smo accumulation in cilia ( Rohatgi et al . , 2009 ) . The CRD-targeted OBIs both behaved like cyclopamine in this assay—they induced Smo accumulation in cilia when added alone and also did not block Shh-induced ciliary accumulation of Smo ( Figure 8F ) . This similarity between the OBIs and cyclopamine led us to consider the possibility that cyclopamine might not be a pure 7TM inhibitor like SANT-1 and Vismodegib but instead may also engage the CRD . Indeed , unlike the non-sterol 7TM inhibitors ( Figure 2E ) , cyclopamine blocked the interaction between the mSmo CRD-Fc and 20 ( S ) -OHC beads ( Figure 8G ) , suggesting that it is capable of binding the CRD in this in vitro assay . Our work provides both structural and mechanistic insights into the enigmatic CRD of Smo in Hh signaling . The CRD of Fz proteins binds to Wnt ligands . While the Fz CRD is related to the Smo CRD , no protein ligand has been identified to date that directly binds to the Smo CRD , and its role in Smo function has not been defined . In Drosophila , deletion of the Smo ectodomain or the mutation of specific cysteine residues in the CRD completely inactivates the protein ( Nakano et al . , 2004 ) . In contrast , in cultured mouse cells , ΔCRD-Smo has basal activity in overexpression experiments ( Murone et al . , 1999; Taipale et al . , 2002 ) . In zebrafish embryos , ΔCRD-Smo can rescue phenotypes dependent on low-level signaling but not on high-level signaling , and it shows higher levels of basal accumulation in cilia ( Aanstad et al . , 2009 ) . Our work now shows that the Smo CRD in vertebrates binds to oxysterols and mediates the ability of these lipids to activate Hh signaling . Structure-guided mutagenesis studies revealed that the Smo CRD binds to 20 ( S ) -OHC in the region that was previously identified as the binding site for small hydrophobic molecules in other CRDs , formed by the evolutionary conserved helical bundle of the CRD core . This supports the hypothesis that CRDs evolved from an ancestral domain that sensed hydrophobic molecules ( Bazan and de Sauvage , 2009 ) . Our structural analysis showed that the Smo CRD oxysterol binding site is most similar to the palmitoleyl-binding site in Fz CRDs ( Janda et al . , 2012 ) ; however , the binding grooves are built of divergent residues ( Figure 5—figure supplement 2 and Figure 6—figure supplement 1 ) , suggesting that they accommodate different classes of hydrophobic ligands . Smo activity can be regulated by two distinct binding sites in the CRD and the 7TM segments . Oxysterols and their derivatives regulate Smo through the CRD site , while SANT-1 , SAG , and Vismodegib bind to the 7TM segment . Binding of agonists like 20 ( S ) -OHC and 20 ( S ) -yne to the CRD must be communicated to the 7TM helix bundle for transduction across the membrane . Indeed , while the 7TM and CRD sites are separable , the dramatic synergy between 20 ( S ) -OHC and SAG we have previously reported suggests a positive allosteric link between these domains ( Nachtergaele et al . , 2012 ) . This synergy also implies that 7TM and CRD ligands can bind to Smo simultaneously . A speculative but intriguing insight into the interaction between the CRD and 7TM domains comes from our unexpected finding that the Hh inhibitor cyclopamine , an established 7TM ligand , also inhibits the binding of the isolated CRD to 20 ( S ) -OHC beads . This result was unexpected because oxysterols do not decrease the binding of bodipy-cyclopamine to cells expressing full-length Smo ( Figure 7C and Dwyer et al . , 2007 ) . We believe that this discrepancy is due to the fact that the cell-based bodipy-cyclopamine binding assay is mostly measuring the interaction between cyclopamine and its high affinity ( Kd∼10 nM , Rominger et al . , 2009 ) binding site in the 7TM domain . Cell ( or membrane ) binding assays can often miss lower affinity ( ∼1–10 μM ) interactions , which can be detected by ligand affinity chromatography assays ( Phizicky and Fields , 1995 ) . It is also possible that cyclopamine binds the CRD more weakly when the CRD is embedded in the context of the whole protein . As noted above , cyclopamine is a sterol that induces the accumulation of Smo in primary cilia , both properties that distinguish it from the pure 7TM site inhibitors SANT-1 and Vismodegib . One possibility is that cyclopamine can bridge the two ligand-binding sites on Smo and engage both a high-affinity interaction with the 7TM segment and a lower affinity interaction with the CRD . Alternatively , two molecules of cyclopamine could engage the CRD and 7TM sites separately or cyclopamine could be involved in a ‘hand-off’ interaction between the CRD and the 7TM segments analogous to the manner in which cholesterol is transferred between NPC2 and NPC1 ( Kwon et al . , 2009; Wang et al . , 2010 ) . While the relevance of this interaction for the inhibition of Smo by cyclopamine in cells remains to be established , the puzzling ability of cyclopamine to induce Smo accumulation in cilia ( while inhibiting Smo activity ) may be related to its ability to engage the CRD . This represents a third mechanism by which ligands can engage Smo , one that is distinct from pure 7TM and CRD ligands . Interestingly , glucocorticoids have been shown to fall into two distinct classes of Smo modulators—cyclopamine-competitive ligands that presumably bind to the 7TM potentiate signaling and a second class of inhibitors that do not compete with cyclopamine but appear to engage a distinct site ( Wang et al . , 2012 ) . The CRD of Smo is also important for signaling by Shh , since ΔCRD-Smo cannot be efficiently activated by either Shh or 20 ( S ) -OHC but remains responsive to SAG . While we observed very little activation of ΔCRD-YFP-Smo by Shh ( Figure 3B ) , another study ( Nedelcu et al . , 2013 ) reported that a ΔCRD-Smo-mCherry protein retained a low level of Shh responsiveness , suggesting that the CRD is not absolutely required for signaling initiated by Shh . This difference in the degree of Shh-responsiveness may be due to the position of the fluorescent protein tag , differences in the tendency of the YFP and mCherry tags to oligomerize or differences in the expression systems used in the two studies . The striking decrease in Shh-responsiveness when the CRD is deleted raises two questions—does Ptch1 regulate Smo through the oxysterol binding site in the CRD and is 20 ( S ) -OHC an endogenous ligand for Smo ? Our mutagenesis of the putative oxysterol binding site in the CRD sheds light on the first question . We find mutations in the mSmo CRD ( Y134F and G115F , Figure 6F ) that can dissociate the Shh and oxysterol responses . These mutants fail to bind or respond to 20 ( S ) -OHC but can still respond to Shh . The simplest interpretation of these data is that the endogenous Smo ligand regulated by Ptch1 does not bind Smo in precisely the same site as 20 ( S ) -OHC . In fact , we have previously reported ( Nachtergaele et al . , 2012 ) that cyclopamine is much less potent against Shh-activated Smo compared to 20 ( S ) -OHC-activated Smo , likely because the conformation adopted by Smo is different in response to these two agonists . Both of these findings suggest that 20 ( S ) -OHC is not the Ptch1-regulated ligand that modulates Smo activity in response to Shh reception . It remains possible that a Shh-regulated ligand binds to the CRD in a manner that is distinct from that of 20 ( S ) -OHC . The CRD is required for Smo to adopt a fully active conformation in response to Shh ( but it is dispensable when Smo is activated by the synthetic 7TM ligand SAG ) . In this view , the CRD would serve as a domain that allosterically activates the 7TM helix bundle in response to Shh . Some mutations ( Smo Y134F , P168A ) that abolish 20 ( S ) -OHC responses do indeed substantially dampen the ability of Shh to activate Smo . The observation that CRD point mutations in Smo that block oxysterol binding also impair signaling by Hh ligands has been used to infer that oxysterol binding is required for physiological Smo signaling ( Nedelcu et al . , 2013 ) . While this hypothesis has substantial implications for Hh regulation in development and cancer , it remains to be determined if the CRD site in cells is occupied by oxysterols or by a different ligand , or if perturbations in endogenous oxysterol levels can modulate Hh signaling . Finally , testing the activity of oxysterol binding site mutants in the context of embryonic development or Hh-driven tumors is essential for elucidating the physiological function of this site and whether it plays a role in graded , low-level or high-level signaling . We have developed partial agonists of Smo that bind to the CRD . Understanding the structural and mechanistic basis for this partial agonism is an important future goal . Remarkably , the simple inversion of the stereochemistry at C-20 converts a potent agonist into a weak partial agonist and an effective inhibitor of signaling . This stereochemical inversion presumably allows the molecule to trap Smo in a poorly active confirmation , likely one similar to that stabilized by cyclopamine , in which Smo is localized in cilia but is inactive . The structures of the OBIs suggest that Smo activation potential depends critically on the spatial orientation between the ring system and the iso-octyl chain of 20 ( S ) -OHC . Regardless of the mechanism , inhibitors targeting the Smo CRD would provide an orthogonal approach to modulate Hh signaling in regeneration and cancer . Partial agonists offer the possibility of blocking unrestrained signaling ( such as that seen in cancer ) while preserving lower-level , physiological signaling ( Riese , 2011 ) . This ability to attenuate Smo activity may be useful since currently used Smo antagonists cause significant side-effects , leading nearly half of the patients in some trials to discontinue treatment ( Tang et al . , 2012 ) . Perhaps the most important question moving forward is to identify the Shh-regulated ligand that mediates the communication between Ptch1 and Smo and to understand how it regulates Smo through the 7TM and CRD sites . Structural studies of a Smo construct carrying both the 7TM segment and the CRD in complex with various ligands that engage either site or both sites will be essential to understand how Smo transmits the Hh signal across the membrane . NIH 3T3 and 293T cells were obtained from ATCC ( Manassas , VA ) , and 293F cells were obtained from Life Technologies ( Grand Island , NY ) . The production of Smo−/−:YFP-mSmo , Smo−/−:ΔCRD-YFP-mSmo and Smo−/−:ΔC-YFP-mSmo stable lines is described below . SAG ( >95% ) was from Enzo Life Sciences ( Farmingdale , NY ) , cyclopamine ( >98% ) from Toronto Research Chemicals ( Toronto , Ontario , Canada ) , Itraconazole ( >98% ) from Sigma ( St . Louis , MO ) and SANT-1 ( >95% ) from EMD Millipore ( Billerica , MA ) . All sterols except ent-20 ( S ) -OHC , 20 ( S ) -yne , 20 ( R ) -OHC , 20 ( R ) -yne , 20-keto-cholesterol , 20-keto-yne and 20 ( S ) -amine were purchased from Steraloids ( purity >98% ) ( Newport , RI ) . All mSmo mutants were made using Quickchange or PCR methods in the context of a previously described construct encoding full-length mouse Smo ( UniProt P56726 ) with the coding sequence for yellow fluorescent protein ( YFP ) inserted immediately after the signal sequence ( pCS2:YFP-mSmo; Rohatgi et al . , 2009 ) . The ΔCRD-YFP-mSmo construct lacked amino acids ( a . a . ) 68–184 , and the ΔC-YFP-mSmo construct was truncated after a . a . 574 . The construct encoding mSmo CRD-Fc was made by cloning the mSmo sequence encoding the CRD ( a . a . 1–183 ) into a pCX vector carrying a C-terminal human Fc tag . Constructs for mammalian expression of the extracellular region of zebrafish Smo ( UniProt Q90X26; zSmo ectodomain: a . a . 29–195 ) , of a C-terminal , intracellular truncation of zSmo ( UniProt Q90X26; zSmo ΔC: a . a . 29-624 ) or the CRD of Drosophila Smo ( UniProt P91682; dSmo CRD: a . a . 32–204 ) , fused C-terminally with either a hexa-histidine , mono Venus or 1D4 epitope-tag that can bind selectively the Rho 1D4 antibody ( Molday and MacKenzie , 1983 ) , were cloned into the pHLsec vector ( Aricescu et al . , 2006 ) . A construct for bacterial expression of the extracellular region of zebrafish Smo ( UniProt Q90X26; zSmo-ectodomain: a . a . 29–212 ) , fused C-terminally with a with a hexa-histidine ( His6 ) tag , was cloned into the pET22b vector . Stable cell lines expressing YFP-mSmo , ΔCRD-YFP-mSmo and ΔC-YFP-mSmo were made by infecting Smo−/− cells with a retrovirus carrying these constructs cloned into pMSCVpuro . Retrovirus was generated by transfecting the MSCV:YFP-mSmo constructs into Bosc23 cells . The virus-containing media were used to infect Smo−/− MEFs , and stable integrants were selected with puromycin and cloned by FACS . We have previously reported the chemical synthesis of ent-20 ( S ) -OHC , 20 ( S ) -yne , 20 ( R ) -OHC and 20-keto-cholesterol ( Nachtergaele et al . , 2012 ) . Full synthetic procedures are provided below for 20-keto-yne , 20 ( R ) -yne , and 20 ( S ) -amine . Melting points were determined on a Kofler micro hot stage and were uncorrected . NMR spectra were recorded in CDCl3 , at 300 MHz ( 1H ) or 75 MHz ( 13C ) . Chemical shifts ( δ ) were reported downfield from internal Me4Si ( δ: 0 . 00 ) . HR FAB-MS determinations were made with the use of JEOL MStation ( JMS-700 ) Mass Spectrometer , matrix m-nitrobenzyl alcohol , with NaI as necessary , using mass spectrometry facilities located at the University of Missouri–St . Louis . HIRES-MS determinations were made with the use of Thermo Orbitrap Velos Mass Spectrometer , using the facilities located at Washington University in St . Louis . IR spectra were recorded as films on a NaCl plate or in KBr . Elemental analyses were carried out by M–H–W laboratories . Optical rotations were measured on a Perkin-Elmer polarimeter , Model 341 . Chromatography was performed using flash chromatography grade silica gel ( 32–63 μm; Scientific Adsorbents , Atlanta , GA ) . Dichloromethane was distilled over CaH prior to application . Tetrahydrofuran was distilled over Na/benzophenone just prior to application . All other chemicals were used as purchased without further purification . Organic extracts were dried over anhydrous Na2SO4 . 20 ( S ) -amine was prepared as a 10 mM stock in 1:1 chloroform/methanol . For each coupling reaction , 250 μl ( packed volume ) of FastFlow 4 NHS-activated sepharose ( GE Healthcare , San Francisco , CA ) was washed extensively into DMSO . 300 μl of DMSO , 2 . 5 μl of the 10 mM 20 ( S ) -amine stock and 1 . 5 μl of triethylamine were added to the washed resin , and the reaction was rotated for 4 hr at room temperature , protected from light . After coupling , the beads were spun down , the supernatant removed and 1 ml of 5% ethanolamine in DMSO was added to block the remaining free reactive sites ( 4 hr , room temperature , protected from light ) . For reporter assays in NIH 3T3 cells , a 10-cm plate of cells was transfected with 8 μg of a 4:1 wt/wt ratio of firefly luciferase reporter driven by an 8xGli-responsive promoter ( Sasaki et al . , 1997 ) and a Renilla luciferase reporter driven by a constitutive TK promoter ( Promega , Madison , WI ) . The next day , transfected cells were seeded into a 96-well plate , grown to confluence , and treated overnight with drugs diluted in media containing 0 . 5% fetal bovine serum ( FBS ) . For reporter assays in Smo−/− cells , 25 , 000 cells per well were seeded in a 24-well plate 24 hr prior to transfection . The next day , after a media replacement step , each well was transfected with 1 ng Smo construct and 500 ng of the reporter mix described above , using Xtreme Gene HP ( Roche , Mannheim , Germany ) . After overnight transfection , the media were once again changed to fresh media . Cells were grown to confluence and treated with drugs diluted in media with 0 . 5% FBS for 48 hr . Activity of both reporters was measured using the Dual-Luciferase Reporter kit ( Promega ) and read on a Synergy H1 Hybrid Multi-Mode Microplate Reader ( BioTek , Winooski , VT ) . The Gli luciferase to Renilla luciferase ratio is reported as ‘Hedgehog reporter activity’ . Each experiment , which included three technical replicates , was repeated at least three times . pCX-mSmo CRD-Fc was produced by secretion ( 96 hr ) into the media of 293F suspension cells ( Life Technologies , Grand Island , NY ) transfected with an expression construct . The collected media were cleared by centrifugation ( 10 min , 1000×g , 4°C ) , adjusted to pH 8 . 5 , filtered through a 0 . 22 μm PVDF membrane and applied to a 1 ml Protein A Hitrap column ( GE Healthcare ) . mSmo CRD-Fc was eluted from the Protein A column with 100 mM citrate pH 3 . 5 , immediately adjusted to pH 8 . 5 and then loaded on a Superose 6 ( 10/300 , GE Healthcare ) gel filtration column equilibrated in 20 mM Tris pH 8 . 5 , 150 mM NaCl . Monodisperse protein that eluted as a sharp peak ( Figure 2A ) was collected and used for binding assays . The purified mSmo CRD could not be cleaved away from the Fc tag efficiently and thus was used in assays as the fusion . In addition , it could not be heated above 37°C prior to SDS-PAGE electrophoresis because it underwent irreversible aggregation . The zSmo ectodomain and dSmo CRD were expressed by transient transfection in HEK-293T cells ( using an automated procedure , Zhao et al . , 2011 ) . 5 days post-transfection , the conditioned medium was dialyzed ( for 48 hr at 4°C ) , and the ectodomain constructs of zSmo or dSmo were purified by either immobilized Rho 1D4 antibody affinity chromatography using CNBr-Activated Sepharose ( GE Healthcare ) as described previously ( Molday and MacKenzie , 1983 ) or IMAC using Talon beads ( Clontech , Mountain View , CA ) . Proteins were concentrated and further purified by size-exclusion chromatography ( Superdex 200 16/60 column; GE Healthcare ) in buffer containing 10 mM HEPES , pH 7 . 5 , 150 mM NaCl . The zSmo ectodomain used for crystallization and oxysterol binding assays was expressed in E . coli Rosetta ( DE3 ) pLysS cells ( Novagen/EMD Millipore ) as inclusion bodies and purified as follows ( protocol adapted from Brown et al . ( 2002 ) ) . After cell lysis , the inclusion body pellets were washed four times and then solubilized in 8 M urea , 50 mM Tris-HCl , pH 8 , and 100 mM NaCl . The solubilized protein was then purified via IMAC ( Ni-Sepharose FastFlow; GE Healthcare ) under denaturing conditions . After IMAC purification the eluted protein was reduced with 10 mM DTT and added drop-wise to 1 l of rapidly-stirring refold buffer ( 3 M urea , 150 mM Tris pH 8 . 5 , 200 mM L-arginine , 1 . 5 mM reduced glutathione [GSH] , 0 . 15 mM oxidized glutathione [GSSG] ) , which was then further stirred gently overnight at room temperature . The solution was then dialysed into 25 mM Tris pH 8 . 5 , 10 mM NaCl at 4°C , filtered , loaded onto a 5 ml HiTrap QFF column ( GE Healthcare ) and eluted with an NaCl gradient ( from 10 mM to 1 M NaCl ) . The eluted protein was concentrated and further purified via size exclusion chromatography ( Superdex 75 16/60 [GE Healthcare] in 10 mM HEPES pH 7 . 5 , 150 mM NaCl ) . SeMet-labeled zSmo ectodomain was produced in E . coli strain B834 ( DE3 ) ( Novagen/EMD Millipore ) . Cells were grown in 2 l cultures at 310 K for 4 hr and after induction with 300 μM isopropyl β-D-1-thiogalactopyranoside , the temperature was then lowered to 298 K . Following incubation for further 20 hr , the cells were harvested and the protein was purified as described for the unlabeled zSmo ectodomain . Cultured cells stably expressing YFP-mSmo , ΔCRD-YFP-mSmo , or ΔC-YFP-mSmo were scraped into ice-cold PBS containing SigmaFast Protease inhibitor cocktail ( Sigma ) and collected as a pellet by centrifugation ( 1000×g , 10 min , 4°C ) . Cells were lysed ( 1 hr , 4°C ) by agitation in modified RIPA buffer ( 50 mM sodium-Tris pH 7 . 4 , 150 mM sodium chloride , 2% NP-40 , 0 . 5% deoxycholate , 0 . 1% sodium-dodecyl sulfate [SDS] , 1 mM dithiothreitol and the SigmaFast Protease inhibitor cocktail ) . After clarification ( 20 , 000×g , 45 min , 4°C ) , the protein concentration of each lysate was measured using the bicinchoninic acid assay ( BCA , Pierce/Thermo Scientific , Rockford , IL ) . Lysate aliquots containing equal amount of total protein were fractionated on an 8% SDS-PAGE gel and transferred to a nitrocellulose membrane for immunoblotting with anti-Gli1 antibody ( #L42B10 , 1:500; Cell Signaling , Denvers , MA ) , anti-GFP antibody ( 1:5000; Novus , Littleton , CO ) , anti-Gli3 antibody ( AF3690 1:200; R&D , Minneapolis , MN ) and anti-p38 antibody ( ab31828 , 1:2000; Abcam , Cambridge , MA ) . In Figures 1–4 , vertical dashed black lines represent non-contiguous lanes from the same immunoblot juxtaposed for clarity . 293T cells transfected with constructs encoding mSmo variants were lysed in hypotonic SEAT buffer ( 250 mM sucrose , 1 mM EDTA , 10 mM acetic acid , 10 mM triethanolamine and the SigmaFast EDTA-Free protease inhibitor cocktail ) . After the removal of nuclei by centrifugation ( 900×g , 5 min , 4°C ) , membranes were pelleted by ultracentrifugation ( 95 , 000×g , 30 min ) and solubilized in a n-dodecyl-β-D-maltopyranoside ( DDM ) extraction buffer ( 50 mM Tris pH 7 . 4 , 150 mM NaCl , 10% vol/vol glycerol , 0 . 5% wt/vol DDM and the SigmaFast EDTA-Free protease inhibitor cocktail ) for 2 hr at 4°C , followed by removal of insoluble material by ultracentrifugation ( 100 , 000×g , 30 min ) . This DDM membrane extract was incubated with 20 ( S ) -OHC beads for 12 hr at 4°C to allow binding to equilibrium . After extensive washing , proteins captured on the beads were eluted with reducing SDS sample buffer . The presence of YFP-mSmo in these eluates was determined by quantitative immunoblotting with an anti-YFP antibody ( NB600-308 , 1:5000; Novus ) and infrared imaging ( Li-Cor Odyssey ) . For ligand affinity chromatography with purified mSmo CRD-Fc or zSmo ectodomain , protein was diluted in 20 mM Tris pH 8 . 5 , 150 mM NaCl , 0 . 3% octyl-glucoside prior to addition of competitors and 20 ( S ) -OHC beads . After binding was allowed to proceed for 12 hr at 4°C , the resin was washed and captured protein was eluted as described above . The presence of mSmo CRD-Fc was measured by an anti-human HRP-coupled antibody ( 1:20 , 000 ) or anti-human IR800-coupled antibody ( 1:10 , 000; for all quantitation , detected by LiCor Odyssey ) . The presence of zSmo ectodomain protein was measured by colloidal Coomassie staining ( GelCode Blue , Pierce/Thermo Scientific ) . Cells were fixed with cold 4% paraformaldehyde ( 10 min , room temperature [RT] ) , washed with phosphate buffered saline ( PBS; 3 times , 5 min each ) , placed in blocking solution ( PBS , 1% ( vol/vol ) normal donkey serum , 0 . 1% ( vol/vol ) Triton X-100 , 10 mg/ml bovine serum albumin ) for 30 min at RT and then stained with primary antibodies ( overnight , 4°C ) : anti-acetylated tubulin ( #T6793; Sigma ) at 1:3000 ( vol/vol ) dilution and anti-Smo ( Rohatgi et al . , 2007 ) at 1:500 ( vol/vol ) dilution in blocking solution . After washing ( three times , 5 min in PBS + 0 . 1% Triton X-100 ) , Alexa-coupled secondary antibodies ( Jackson ImmunoResearch ) were applied ( 1:500 [vol/vol] dilution , 1 hr , RT ) . Finally , stained cells were washed in PBS ( three times , 5 min ) and mounted onto glass slides with Prolong Gold mounting media with DAPI ( Life Technologies ) . The fixed cells were imaged with a Leica SP8 laser scanning confocal microscope , using a 63× oil objective ( NA 1 . 40 ) and 1 . 3× zoom . For the quantitative analysis of Smo levels in cilia , all images used for comparisons were taken with identical gain , offset , and laser power settings on the microscope . Non-manipulated maximum projections of z-stacks were used for quantitation ( Fiji ) . A mask , constructed by automatically applying a threshold to the acetylated tubulin image , was then applied to the corresponding anti-Smo image to measure Smo fluorescence at cilia . Local background correction was performed by moving the mask to measure fluorescence at a nearby region , and this value was subtracted from the ciliary Smo fluorescence . All statistical analysis and curve fitting were done in GraphPad Prism . For microscopy data , the Smo fluorescence for each cilium was individually plotted , generating a scatter plot that represents variability in the data . To compare Smo levels between different conditions , the median and interquartile range are provided ( n = 60 for each condition ) . For Hh reporter assays , each point is reported as the mean ± standard deviation ( SD ) derived from triplicates . Each result in the paper was repeated at least three times with similar outcomes . Relative luciferase activity was calculated by dividing Gli luciferase by Renilla luciferase luminescence . Fold-change in reporter activity was calculated by dividing each replicate by the mean reporter activity of the vehicle-treated control . Normalized ( % of max ) Hh reporter activity was calculated by setting the maximum value of a set to 100% and zero to 0% using the ‘normalize’ function of GraphPad Prism . In all graphs , dotted lines are straight connectors between points , and solid lines represent non-linear curve fits of the data ( all done in GraphPad Prism ) . In Figures 2C and 4E , the curves were fit using the ‘log ( inhibitor ) vs response—variable slope’ function of GraphPad Prism . The model used for this function was Y = Bottom + ( Top–Bottom ) / ( 1+10^ ( [LogIC50-X]HillSlope ) , where ‘Y’ represents bound zSmo as a percentage of the maximum bound ( with zero competitor ) , ‘Top’ and ‘Bottom’ represent the plateaus at the beginning and end of the curve , respectively , and ‘X’ represents the concentration of free competitor added to the binding reaction . In Figures 2B and 4D , the curve was fit using the ‘one site—total and nonspecific binding’ function . The equation used for this fit incorporates both specific binding ( specific = BmaxX/[X+Kd] ) and non-specific binding ( nonspecific = NSX + Background ) . ‘X’ in this case represents the sterol immobilized on the resin . In Figure 8A , B , C , the same ‘log ( inhibitor ) vs response—variable slope’ function as above was used to asses the IC50s of the OBIs in a Hh reporter assay . Prior to crystallization , the zSmo ectodomain from bacterial expression was concentrated to 7 mg/ml . Crystallization trials , using 100 nl protein solution plus 100 nl reservoir solution in sitting drop vapor diffusion format , were set up in 96-well Greiner plates using a Cartesian Technologies robot ( Walter et al . , 2005 ) . Crystallization plates were maintained at 20 . 5°C in a TAP Homebase storage vault and imaged via a Veeco visualization system ( Mayo et al . , 2005 ) . zSmo ectodomain native and selenomethionine-substituted crystals were obtained out of mother liquor containing 100 mM HEPES pH7 . 0 , PEG 6000 20% , 10 mM ZnCl2 . X-ray diffraction data were collected at 100 K and crystals were treated with 25% ( vol/vol ) glycerol in mother liquor for cryo protection . Data were collected at beamline I03 at the Diamond Light Source , UK ( native zSmo ectodomain ) , and at beamline ID23-EH1 ( selenomethionine-substituted zSmo ectodomain ) at the European Synchrotron Radiation Facility ( ESRF ) , France . X-ray data were processed and scaled with the HKL suite ( Otwinowski and Minor , 1997 ) and XIA2 ( Evans , 2006; Kabsch , 2010; Winter , 2010 ) . Data collection statistics are shown in Table 1 . The zSmo ectodomain crystal structure was determined by single anomalous dispersion ( SAD ) analysis . The positions of three selenium atoms were determined using SHELXD ( Schneider and Sheldrick , 2002 ) . This solution was used as an input into the AutoSol module of the PHENIX suite ( Adams et al . , 2002 ) for phase calculation and improvement . The resulting map was of high quality and allowed tracing of the whole polypeptide chain ( Figure 5—figure supplement 1A ) . An initial model was built automatically using RESOLVE ( Terwilliger , 2003 ) and completed manually using COOT ( Emsley and Cowtan , 2004 ) . Iterative rounds of refinement in autoBUSTER ( Blanc et al . , 2004 ) , PHENIX ( Adams et al . , 2002 ) and REFMAC ( Murshudov et al . , 1997 ) applying non-crystallographic symmetry restraints as well as manual building in COOT ( Emsley and Cowtan , 2004 ) resulted in a well-defined model for zSmo ectodomain that included two molecules in the asymmetric unit both composed of residues 41–158 ( Figure 5—figure supplement 1B ) . The zSmo ectodomain N- and C-terminal regions could not be traced due to missing electron density and were not included in the final model . The native structure was solved by molecular replacement using PHASER ( McCoy et al . , 2007 ) with the SeMet-labeled structure as a search model and refined as described above for the SeMet-labeled protein . Crystallographic and Ramachandran statistics are given in Table 1 . Stereochemical properties were assessed by MolProbity ( Davis et al . , 2007 ) . Superposition of CRD structures and root mean square deviation ( RMSD ) values were calculated for equivalent Cα atoms using program SHP ( Stuart et al . , 1979; Riffel et al . , 2002 ) . The phylogenetic tree for CRDs ( Figure 5B ) was prepared with program PHYLIP ( Felsenstein , 1989 ) with the summed structural correlation data presented in Figure 5—figure supplement 3 to construct a distance matrix . The program VOLUMES ( RE Esnouf , unpublished ) was used with a 1 . 4 Å radius probe to analyze the CRD binding grooves of zSmo and Fz8 . The analysis of evolutionary conserved residues among the CRDs of the Smoothened family members was based on 80 amino acid sequences of vertebrate Smo CRDs and was mapped onto the zSmo CRD crystal structure using ProtSkin ( Deprez et al . , 2005 ) . The refined atomic coordinates of the zSmo CRD crystal structure were kept rigid during the molecular docking . The guanidinium group of Arg139 forms a hydrogen bond with a carbonyl oxygen of Arg139 in the neighboring molecule and occludes the oxysterol-binding pocket . Thus , the mmt180 rotamer ( Lovell et al . , 2000 ) of Arg139 was used during docking and pocket analysis . Atomic coordinates of 20 ( S ) -OHC were downloaded from PubChem ( compound ID 121935 , Wang et al . , 2009 ) and kept flexible during docking in AutoDock 4 . 2 . 5 . 1 using the Lamarckian genetic algorithm and default parameters ( Morris et al . , 2009 ) . Estimated inhibition constant , Ki ( dissociation constant of the zSmo CRD-20 ( S ) -OHC-complex ) , was calculated using formula Ki = exp ( ΔG/[RT] ) , where ΔG is a free energy of binding in kcal/mol , R is the gas constant 1 . 987 cal K−1 mol−1 , and T = 298 . 15 K . The homology model of dSmo CRD ( Ile82-Thr204 , UniProtKB ID P91682 ) was built using program MODELLER 9 . 9 ( Eswar et al . , 2008 ) with the zSmo CRD structure as a template . The amino acid sequence identity between the corresponding CRD regions is 42% . MALS analysis of purified and glycosylated zebrafish Smo ectodomain ( expressed from mammalian cells ) was performed using an analytical Superdex S200 10/30 size exclusion chromatography column ( GE Heathcare ) eluted in 150 mM NaCl , 10 mM HEPES pH 7 . 5 ( flow rate 0 . 5 ml/min ) with static light scattering ( DAWN HELEOS II , Wyatt Technology , Santa Barbara , CA ) , differential refractive index ( Optilab rEX , Wyatt Technology ) and Agilent 1200 UV ( Agilent Technologies , Santa Clara , CA ) detectors . Data were analyzed using the program ASTRA ( Wyatt Technology ) . Adult fishes were maintained on a 14-hr light/10-hr dark cycle at 28°C in the AVA ( Singapore ) certified IMCB Zebrafish Facility . Zebrafish strain used was Tg ( eng2a:eGFP ) i233 ( Maurya et al . , 2011 ) . The embryos of Tg ( eng2a:eGFP ) i233 were dechorinated using pronase ( Roche ) at one cell stage . The well-developing ones at the 50% epiboly stage were selected and grown in fish water containing 50 μM 20 ( S ) -OHC or 40 μM cyclopamine . Control embryos were kept in water containing the same amount of ethanol , used as the vehicle for the drugs . Standard in situ hybridization ( ISH ) was performed with anti-Dig alkaline phosphatase and chromogenic substrate NBT/BCIP as previously described ( Oxtoby and Jowett , 1993 ) . ptch2 ( formerly ptc1 ) RNA probe was prepared from template as previously described ( Concordet et al . , 1996 ) .	Just over 30 years ago , researchers identified a new signaling molecule with an important role in the development of fruit flies . Embryos lacking this molecule were thought to resemble a hedgehog , eventually leading to this cell–cell communication system being designated the “Hedgehog” pathway . This pathway has subsequently been shown to be involved in the development of many other animals , as well as in the repair of damaged tissues in adult organisms . Abnormal Hedgehog signaling has also been implicated in both human birth defects and in cancers of the skin and the brain . Many such tumors are driven by the unrestrained activation of a membrane-bound protein called Smoothened , which has led to the development and clinical use of small molecules that prevent Hedgehog from activating Smoothened . The existing anti-tumor drugs all bind to the same region of the Smoothened receptor , namely the part that sits within the cell membrane . A second group of molecules , known as oxysterols , can activate Smoothened , but exactly how they do this has been unclear . Now , Nachtergaele et al . have shown that oxysterols bind to a region of the Smoothened receptor that lies outside the cell , and that is rich in the amino acid cysteine . By solving the crystal structure of this part of the receptor from zebrafish , Nachtergaele et al . were able to map the oxysterol binding site at high resolution . This revealed strong similarities between this binding site and those in related receptors belonging to the Wnt signaling pathway . Deleting the cysteine-rich domain significantly impaired Hedgehog signaling , as did a new class of small molecule inhibitors designed specifically to target the oxysterol binding site . In addition to providing new insights into the structure and function of the Smoothened receptor , the work of Nachtergaele et al . opens up possibilities for novel therapeutic agents that could be used in the treatment of cancers caused by abnormal Hedgehog signaling .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Structure and function of the Smoothened extracellular domain in vertebrate Hedgehog signaling
Primaquine is the only drug available to prevent relapse in vivax malaria . The main adverse effect of primaquine is erythrocyte age and dose-dependent acute haemolytic anaemia in individuals with glucose-6-phosphate dehydrogenase deficiency ( G6PDd ) . As testing for G6PDd is often unavailable , this limits the use of primaquine for radical cure . A compartmental model of the dynamics of red blood cell production and destruction was designed to characterise primaquine-induced haemolysis using a holistic Bayesian analysis of all published data and was used to predict a safer alternative to the currently recommended once weekly 0 . 75 mg/kg regimen for G6PDd . The model suggests that a step-wise increase in daily administered primaquine dose would be relatively safe in G6PDd . If this is confirmed , then were this regimen to be recommended for radical cure patients would not require testing for G6PDd in areas where G6PDd Viangchan or milder variants are prevalent . Plasmodium vivax accounts for over half the world’s malaria burden outside sub-Saharan Africa ( Gething et al . , 2012 ) . The control and elimination of vivax malaria require both cure of the blood stage infection ( the stage that causes acute illness ) and the prevention of later relapses which derive from dormant hypnozoites in the liver ( radical cure ) . Hypnozoites are formed from sporozoites , which do not develop immediately following mosquito inoculation but instead remain dormant in hepatocytes for weeks or months before developing and causing recurrent blood stage infections called relapses . In general , P . vivax infections in tropical regions are associated with frequent relapses ( with intervals as short as three weeks ) whilst relapses in P . vivax infections from Central America , Northern India and temperate regions are associated with longer intervals from acute infection to first relapse ( White , 2011 ) . Primaquine , an 8-aminoquinoline , is currently the only widely available antimalarial drug for the radical cure of P . vivax infections . Primaquine causes predictable oxidant haemolysis in G6PD deficiency ( G6PDd ) one of the most common genetic abnormalities of man ( Cappellini and Fiorelli , 2008 ) . Throughout Asia , the Mediterranean littoral and Africa , allele frequencies for this enzyme deficiency vary between 3% and 35% in the populations at risk from vivax malaria ( Howes et al . , 2013 ) . As G6PDd has sex-linked inheritance , males are either deficient ( hemizygotes ) or normal , whereas women can be deficient ( homozygotes ) , normal or partially deficient ( heterozygotes ) in proportions determined by the Hardy-Weinberg equilibrium . Because of Lyonisation , there is substantial variability in the proportion of red cells which are deficient in individual heterozygote females ( Beutler et al . , 1962 ) . The degree of haemolysis following primaquine depends on the dose administered and the severity of the enzyme deficiency ( and in heterozygote females the proportion of erythrocytes which are deficient ) . The more severe G6PDd variants found in South East ( SE ) Asia ( for example , Viangchan , Mahidol , Coimbra , Union ) and the Middle East/West Asia ( for example , Mediterranean ) are generally associated with more severe haemolysis compared to the common African A- variant . For G6PD normal patients , the primaquine regimen for radical cure that is recommended in SE Asia and Oceania ( where relapse rates are high ) is 0 . 5 mg base/kg/day for 14 days . Elsewhere it is 0 . 25 mg/kg/day for 14 days . For patients with G6PDd , a weekly dose is recommended; 0 . 75 mg/kg/week given for a total of 8 doses . Unfortunately G6PDd testing is not widely available despite the recent introduction of point-of-care rapid diagnostic tests ( RDTs ) for G6PDd . These RDTs are currently too expensive to deploy on a wide scale and can be difficult to interpret , and thus are not generally available ( Brito et al . , 2016-08; Satyagraha et al . , 2016; Oo et al . , 2016 ) . Thus , primaquine is commonly not given to patients to avoid the risk of haemolysis so the burden of vivax malaria remains high , causing considerable morbidity and economic loss ( Price et al . , 2007 ) . The mechanisms regulating red blood cell production and turnover have been well characterised . Red blood cells ( RBCs ) transport oxygen which is reversibly bound to the main red cell protein , haemoglobin . RBC production in the bone marrow is regulated to maintain oxygen carrying capacity . When the haemoglobin concentration in the blood falls , this reduces oxygen carriage and RBC production is up-regulated , a process mediated largely by the renal hormone , erythropoietin . At times of increased bone marrow production , reticulocytes appear in increased numbers in the circulation ( the upper limit of normal is ≈1 . 5% ) . Normal RBCs in healthy people have a very stable life expectancy of around 120 days . This is well modelled by a Gumbel distribution with low variance . In nucleated cells G6PD can be newly synthesised , but the red cells lose their nucleus before leaving the bone marrow so very young red cells ( reticulocytes ) have the highest G6PD activity , and this declines as the RBCs age . In most G6PDd variants , the mutant enzyme degrades more rapidly compared to the normal enzyme . Older erythrocytes may have up to five times less G6PD activity than reticulocytes . G6PDd results in lowered NADPH and a reduced ability to regenerate reduced glutathione . Reduced glutathione protects normal RBCs against oxidant stresses such as the haemolytic effects of primaquine metabolites and certain foods , classically fava beans . G6PD is also important for the function of catalase , another oxidant defence mechanism . As these non-reusable oxidant defence reserves are ‘used up’ , the aging erythrocyte becomes increasingly vulnerable to oxidant haemolysis ( Beutler et al . , 1954a; Dern et al . , 1954; Beutler , 2008; Recht et al . , 2014 ) . As young red cells have more functional enzyme than older cells , the degree of oxidant haemolysis depends on the genetic variant of G6PDd and the age distribution of the red cell population . Once the older cells have haemolysed , the remaining younger erythrocytes are essentially resistant to further damage by the same dosing regimen ( that is , drug exposure ) ( Beutler et al . , 1954a ) . However , higher primaquine doses do cause further haemolysis . This explains the fall then rise in haemoglobin with continued daily primaquine administration in mild and moderate severity variants of G6PDd . This temporary primaquine insensitivity in G6PDd individuals with the continued primaquine administration was characterised by Beutler and colleagues in a series of studies conducted over sixty years ago ( Beutler et al . , 1954a , 1954b , 1955; Dern et al . , 1954; Beutler , 1959 ) and later exploited by Alving et al . to develop the once weekly regimen in G6PDd ( Alving et al . , 1960 . ) By experimenting with high-dose weekly regimens and low-dose daily regimens , Beutler and colleagues showed haemoglobin would first fall as a result of oxidant haemolysis and then rise despite continued exposure to the same doses of primaquine which had caused the initial haemolysis . This resulted from reactive erythropoiesis ( reticulocytosis ) that introduced a younger red cell population to the circulation which was essentially ‘resistant’ to the haemolytic effects of that primaquine dose . Intermittent primaquine administration resulted in progressively smaller cycles of haemolysis followed by reticulocytosis as the red cell population became younger . These results led to a recommendation for a high-dose , once weekly primaquine regimen for radical cure in vivax malaria patients with G6PDd ( 8 once weekly adult doses of 45 mg ) ( Alving et al . , 1960 ) . This regimen was devised based on studies in subjects with the African A- variant of G6PDd , which is one of the mildest deficiencies . Safety was not formally assessed in more severe deficiencies . A recent trial of this regimen in vivax malaria patients with the more severe Viangchan G6PDd variant from Cambodia showed a greater fall in haemoglobin and a delayed recovery from anaemia in G6PDd compared to G6PD normal patients with one patient requiring a blood transfusion ( Kheng et al . , 2015 ) . These data suggest that weekly primaquine may not be the optimal regimen for the more severe G6PDd variants prevalent outside Africa . Reconsideration of the detailed haematological studies that laid the foundation for the weekly regimen suggests that an ascending-dose regimen of primaquine , with a schedule that matches the dynamics of red blood cell production , could induce a safe ‘slow burn’ haemolysis , even in individuals with severe G6PDd variants , and would still deliver a total therapeutic dose for radical cure . Accordingly , our study had two objectives; first , to construct a compartmental model for red blood cell dynamics which could be used to analyse all available data from past studies of haemolysis in G6PDd individuals , and second to predict an optimal ascending dose regimen which would be safe and efficacious yet practical and could , therefore , be recommended without G6PD testing . Figure 1 shows hypothetical data simulated from the compartmental model with a primaquine regimen of 45 mg weekly for eight weeks fitted to data from adult G6PD deficient Cambodian patients . Parameters were randomly drawn from the Bayesian posterior distribution . 10 . 7554/eLife . 23061 . 003Figure 1 . Comparison between the data from Kheng et al . ( 2015 ) ( shown in green , population median in thick black line ) and posterior predictive 80% credible intervals ( shown in red , median: thick line; 10&90% boundaries: dashed lines ) in which adult Cambodian patients who were G6PD deficient were given weekly primaquine ( 45 mg ) for eight weeks . Left: reticulocyte response; Right: haemoglobin response . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 003 The signal-to-noise ratio in the reticulocyte data is low and this is apparent from the median reticulocyte count which varies considerably during the 56 days . In comparison , simulations from the mechanistic model show that a substantial rise in the reticulocyte count should occur approximately one week after the first dose , with a peak after the third dose , and then return to normal slowly over the subsequent six weeks . The serial haemoglobin data on the other hand show a clear trend with a large fall after the first dose , a smaller fall after the second and then a gentle recovery with no major effect from subsequent primaquine doses . This trend is reproduced by the model and the posterior distribution also characterises satisfactorily the variance observed in steady state haemoglobin concentrations . Combining the data from Figures 2–4 , it is possible to estimate a primaquine dose-haemoglobin response curve for G6PDd individuals whose severity is similar to the ‘moderate severity’ variants G6PDd Mahidol/Viangchan . The data at different dosing levels are sparse and the studies have been done in very different contexts; however , the strong mechanistic assumptions used to construct the compartmental model regularize the problem enough to compare the studies in a principled way . The data from G6PDd Mediterranean are excluded from this dose-response curve estimation because the haemolysis observed with this variant is considerably greater than for G6PDd Mahidol/Viangchan . However , the observed falls in haemoglobin after 5 daily doses of 30 mg in G6PDd Med Sardinians are shown by the red triangles in Figure 5 , right plot , for comparison . 10 . 7554/eLife . 23061 . 004Figure 2 . Comparison between approximate model fits ( red ) and data ( green ) extracted from four primaquine studies with a single dose or daily regimens all at 30/45 mg adult doses . Dosing periods are shaded in blue . The top two plots are for Mahidol and Viangchan variants , respectively . The bottom two plots are for the Mediterranean variant . From top left to bottom right: single 45 mg dose given to 7 G6PDd Mahidol Thais ( Charoenlarp et al . , 1972 ) ; 14 daily doses of 30 mg given to 15 G6PDd presumed Viangchan variant Khmer soldiers ( only mean and extreme values reported ) ( Everett et al . , 1977 ) ; 1 G6PDd Med Sardinian given two courses of daily 30 mg doses ( Pannacciulli et al . , 1965 ) ; 2 G6PDd Med Sardinians given 5 daily doses of 30 mg ( Salvidio et al . , 1967 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 00410 . 7554/eLife . 23061 . 005Figure 3 . Comparison between approximate model fits ( red ) and data ( green ) extracted from four primaquine studies on the same individual with G6PDd African A- ( Alving et al . , 1960 ) . Dosing periods are shaded in blue . The top two plots are for weekly dosing regimens ( 8 doses ) : left is 60 mg per week; right is 45 mg per week; the bottom two plots are daily dosing regimens ( 14 doses ) : left is 15 mg per day; right is 30 mg per day . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 00510 . 7554/eLife . 23061 . 006Figure 4 . Time series data of reticulocyte count ( top row ) and haemoglobin concentrations ( bottom row ) from the Cambodian study on G6PDd individuals ( n=18 , left column ) and G6PD normals ( n=57 , right column ) ( Kheng et al . , 2015 ) . The faint green lines show individual patient data; the thick black lines represent the population median values at each time-point; the dashed black lines show the interquartile range . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 00610 . 7554/eLife . 23061 . 007Figure 4—source data 1 . This provides the source data for the reticulocyte counts and haemoglobin concentrations over time from the Kheng et al . ( 2015 ) study on weekly high-dose primaquine . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 00710 . 7554/eLife . 23061 . 008Figure 5 . Estimating the dose-response curve for moderate/severe G6PDd . Left: estimates of the log⁡d parameter as a function of the administered dose plotted with a linear regression curve ( red cross: Viangchan; red circles: posterior estimates from model fitted to data from G6PDd Viangchan; blue cross: Mahidol; green crosses: African A- ) . Right: dose-response curve ( thick black line ) with 90% credible intervals ( dotted black lines ) as measured by fall in haemoglobin ( y-axis ) after five days at a given dose ( x-axis ) based on draws from the posterior distribution . The red and green crosses are the estimated falls after five days from Viangchan and African A- studies , respectively ( see Figures 2 , 3 ) . The red triangles show the falls observed in G6PDd Med studies from Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 008 The posterior MCMC samples inferred from the Kheng data can be used to approximate model uncertainty around the median dose-response curve . The right plot of Figure 5 shows the posterior predictive dose-response curve with 90% credible intervals , where the ‘response’ is defined as the drop in haemoglobin after five days at a given dosing level . Overlaid are estimates of the falls in haemoglobin induced by 5 daily doses from studies in Figures 2 , 3 , and an extrapolated estimate from the posterior distribution of the model fitted to data from weekly dosing in Viangchan variant . It is of interest to compare the fitted dose-response relationship in Figure 5 ( right: thick black line ) —corresponding to the more severe variants of G6PDd—with the green crosses corresponding to observed and fitted haemolysis in G6PDd African A- ( mild variant ) . As would be expected , for the mild variant the dose-response relationship has the same shape but is shifted to the right . The currently recommended dose for the radical cure of vivax malaria in an adult in SE Asia and Oceania delivers 420 mg ( that is , 30 mg/d x 14 d ) of primaquine and is very effective ( John et al . , 2012 ) . The maximum primaquine dose administered in the weekly regimen is 360 mg ( 8 x 45 mg ) but the efficacy of this regimen has only been reported in Afghan refugees in Pakistan , a country with a relatively low relapse rate ( Leslie et al . , 2008 ) . The primary objective of our research is to design a novel primaquine regimen that could be given safely to individuals with G6PDd or of unknown status without G6PD testing and deliver a total dose that would be efficacious . The scientific hypothesis is that the same total dose could be given safely with tolerated declines in Hb over a longer duration by starting with a lower initial dose which is increased gradually over time . The ascending dose regimen would allow for a steady adjustment of the age distribution of RBCs by both slow primaquine-induced haemolysis and the resulting increased erythropoiesis . These results only concern ascending dose regimens given over 20 days . There are two reasons for this; first , adherence to long course regimens is likely to be poor , and second , the first relapses emerge from the liver about 14 days after starting treatment so the primaquine regimen has to provide sufficient drug to prevent the emergence or eliminate these parasites . Primaquine is widely recommended for the radical cure of vivax malaria but it is often not given because testing for G6PD deficiency is not widely available outside large centres . This has deleterious consequences for vivax malaria affected communities because it is the multiple relapses of vivax malaria from liver hypnozoites that cause substantial morbidity . Seminal research conducted over 50 years ago characterized the biology of oxidant haemolysis caused by primaquine and provided an alternative once weekly regimen for patients who were G6PDd based on controlled haemolysis . This was shown to be safer in adult subjects with the 'mild' African A- variant of G6PDd , but was recommended for all G6PDd variants with variable adoption by countries since . In some countries ( for example , Iran ) it is the standard radical treatment for all patients . The safety and effectiveness of the high dose weekly regimen have been studied little over the past five decades . Uncomplicated malaria treatment recommendations are usually a trade-off between dosing precision and operational feasibility . A regimen which is long or complicated may be adhered to poorly . In this particular case it must also be able to prevent or suppress relapsing P . vivax or P . ovale parasites which begin to emerge from the liver as early as two weeks ( becoming patent about one week later ) in SE Asia and Oceania . This modelling exercise , based on all available data , sought to devise a primaquine regimen which would be safer in G6PDd patients , and , therefore , might be deployed without G6PD testing . It was calibrated against recent data in Cambodian patients most of whom had the Viangchan G6PDd variant . Thus , the model predictions of the degree of haemolysis and the tolerability and safety profile would be expected to hold for variants with similar or less severe enzyme abnormalities , but it would not necessarily hold for more severe variants such as G6PDd Mediterranean where more clinical research is required . Under all circumstances , the ascending regimen proposed here would be expected to be safer than the current 14 day regimens in G6PDd hemizygous males and homozygous females , especially the 0 . 5 mg/kg regimen needed for frequent relapsing P . vivax . This is clinically relevant also for female heterozygotes . Even with current rapid testing methods ( for example , fluorescent spot test and RDTs ) which generally detect patients with ≤30% normal G6PD activity , the haemolytic risk in heterozygote females , who may be classified erroneously as 'G6PD normal' , could still be substantial . Up to ≈70% of their erythrocytes may be G6PD deficient , and clinically significant haemolysis may result from daily higher dose primaquine regimens given to female heterozygotes ( Chu et al . , 2017 ) . Although this compartmental model of RBC dynamics is highly simplified , it reproduces the essential dynamics of the body’s response to primaquine-induced haemolysis in both healthy individuals and malaria patients . It can therefore help to guide the design of a Phase I study to evaluate its predictions , and thereby develop an optimal ascending dose regimen of PQ . An adaptive design protocol has been developed to test the simplified regimen ( A ) in G6PDd Mahidol healthy volunteers . A study in healthy G6PDd volunteers is essential to characterise the haemolytic response . Data from such a study can then be used to determine an optimal regimen which would then be tested for safety , and efficacy ( that is , radical cure ) in vivax malaria patients in a Phase II ( that is , to define the PK-PD relationship in patients ) . Whether patients would adhere sufficiently to a longer regimen is an important operational concern so the optimised regimen would then need to be assessed for safety and effectiveness in larger field trials . This use of mathematical modelling such as this could also be readily applied to the slowly eliminated 8-aminoquinoline tafenoquine , currently being tested for safety and efficacy in humans ( Beck et al . , 2016 ) . Tafenoquine has the great advantage of being administered as a single dose for radical cure due to its long terminal elimination half-life . However , this means it could be dangerous in G6PD deficiency . Whereas the rapidly eliminated primaquine can be stopped if there is significant haemolysis , limiting the haemolytic effect , the haemolytic effect of the slowly eliminated tafenoquine cannot be readily reversed and so haemolysis will continue until all susceptible red cells are destroyed . Combined regimens for G6PDd patients in which primaquine is given initially to induce controlled haemolysis followed by tafenoquine might be possible , and would allow shorter total treatment durations . Data on the Hb response to different doses of tafenoquine would be necessary to calibrate the model . The results of this study show how care will need to be taken when designing an ascending primaquine dose regimen in order to minimize falls in haemoglobin . This is shown by the toxic regimen D ( Table 1 and Figure 7 ) . This gives an insight into the ‘memory’ property of the ascending dose regimens . The effect of a 30 mg dose will entirely depend on which doses were given during the previous days . In conclusion , these results suggest that an ascending PQ dosing regimen for vivax radical cure might be well tolerated and effective in mild or moderately severe G6PDd variants . These predictions should now be tested in an adaptive phase I study . 10 . 7554/eLife . 23061 . 012Figure 7 . Dynamics of ascending regimens . Left: Comparing the haemolytic effect of four regimens . Thick black line: proposed optimal regimen; thick black dashed line: more conservative regimen with lower total dose; thin black dashed line: longer duration regimen for more severe variants; thick red line: bad choice regimen . Right: Posterior predictions for the proposed ascending dose for a given starting haemoglobin ( steady state ) . Prediction using the median posterior values is shown by a thick black line . Predictions for 100 random draws from the posterior are shown by dashed blue lines . The horizontal line at a haemoglobin concentration of 9 is a proposed conservative ‘safety threshold’ . Horizontal line at a haemoglobin concentration of 8 is a proposed regimen limiting toxicity threshold . DOI: http://dx . doi . org/10 . 7554/eLife . 23061 . 012 The structure of the model of red cell dynamics is similar to the compartmental model developed by ( Savill et al . , 2009 ) . RBC dynamics are simulated by tracking the age distribution of the red blood cells in hourly blocks . The homeostatic dynamics , which maintain the number of red blood cells or haematocrit at a steady state are straightforward . At steady state , approximately 0 . 83% of RBCs are replaced each day and 1% of RBCs in the circulation are reticulocytes . Severe acute anaemia has two consequences in the bone marrow . Reticulocytes are released into the circulating blood at an earlier age and with increased erythropoiesis normoblasts may be released into the circulation ( reported as nucleated RBCs ) ( Hillman , 1969 ) . Previous iron turnover studies in humans following phlebotomy suggest sigmoid relationships for both of these processes ( Hillman , 1969 ) . The historical ( Figures 2 , 3 ) and Kheng ( Figure 4 ) data were used to fit the compartmental model; the former were used to select suitable prior distributions for parameters . Bayesian model fitting via MCMC was then applied to the data from the weekly high-dose primaquine in Cambodia ( Kheng et al . , 2015 ) . The likelihood of the parameters is defined by a deterministic simulation from the compartmental model for a given dosing regimen and assumes both the haemoglobin levels and reticulocyte counts are observed with Laplace distributed errors ( assuming Laplace errors is equivalent to minimizing absolute deviation ) . A Bayesian hierarchical structure was used for the steady state haemoglobin and the maximum increase in the production of red blood cells . This makes the assumption that each patient in the study is characterized by an individual steady state haemoglobin concentration Hb∗ and a maximum production capacity ρmax drawn from a population distribution ( normal distributions in both cases ) . Weekly informative priors were used for all parameters and the posterior distribution was estimated using MCMC with a Metropolis-Hastings proposal . Details of prior distributions and histograms of posterior distributions , together with convergence diagnostics and summary statistics are in the Appendix 1 , ‘Structure of hierarchical model and MCMC diagnostics' .	Malaria is the most important parasitic disease that affects humans . Over half of the malaria cases in Asia and South America are caused by a species of malaria parasite called Plasmodium vivax ( known as vivax malaria ) . This form of malaria results in repeated illness because dormant parasites in the liver wake at intervals to infect the blood . The only available drug that can stop these relapses is a drug called primaquine , which was developed seventy years ago . Unfortunately , primaquine causes dangerous side effects in certain individuals who are deficient in an enzyme called G6PD , which helps defend red blood cells against stresses . Primaquine damages these cells so that they burst , leading to anaemia . This is a major problem because G6PD deficiency is common in regions where malaria is present: in some areas up to 30% of the population may be G6PD deficient . Since G6PD testing is not widely available , doctors often avoid prescribing primaquine to treat malaria , which results in more cases of disease relapse . Failing to prevent vivax relapses causes extensive illness and hinders efforts to eliminate malaria . Is there a way to give this drug to patients that would be safer for people with G6PD deficiency ? Primaquine destroys older rather than younger red blood cells . Watson et al . used mathematical modelling to see whether it is possible to develop a primaquine treatment strategy that would allow a gradual destruction of older red blood cells in individuals with G6PD deficiency , which would be safer . The mathematical model incorporates data from previous studies in malaria patients and healthy volunteers with G6PD deficiency and combines this with knowledge of how red blood cells are produced and destroyed . Watson et al . predicted that giving primaquine over 20 days in a steadily increasing dose was safer than current recommendations . Mathematical models are simplifications of real world processes . The only way to test these findings properly will be to run a clinical trial that gives healthy volunteers who are G6PD deficient a course of primaquine treatment with a steadily increasing dose .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "computational", "and", "systems", "biology" ]	2017	Modelling primaquine-induced haemolysis in G6PD deficiency
Transcription is a highly stochastic process . To infer transcription kinetics for a gene-of-interest , researchers commonly compare the distribution of mRNA copy-number to the prediction of a theoretical model . However , the reliability of this procedure is limited because the measured mRNA numbers represent integration over the mRNA lifetime , contribution from multiple gene copies , and mixing of cells from different cell-cycle phases . We address these limitations by simultaneously quantifying nascent and mature mRNA in individual cells , and incorporating cell-cycle effects in the analysis of mRNA statistics . We demonstrate our approach on Oct4 and Nanog in mouse embryonic stem cells . Both genes follow similar two-state kinetics . However , Nanog exhibits slower ON/OFF switching , resulting in increased cell-to-cell variability in mRNA levels . Early in the cell cycle , the two copies of each gene exhibit independent activity . After gene replication , the probability of each gene copy to be active diminishes , resulting in dosage compensation . Gene expression is a stochastic process , consisting of a cascade of single-molecule events ( Coulon et al . , 2014; Sanchez and Golding , 2013 ) , which get amplified to the cellular level . A dramatic consequence of stochastic gene expression is that individual cells within a seemingly homogenous population often exhibit significant differences in the expression level of a given gene ( Raj and van Oudenaarden , 2008 ) . In fact , cell-to-cell variability in expression levels is the most commonly used proxy for the presence and magnitude of stochastic effects ( Elowitz et al . , 2002; Raj et al . , 2006; Raser and O'Shea , 2005 ) . The mapping between stochastic kinetics and population heterogeneity can be made rigorous by making specific assumptions about the kinetics of gene activity and using stochastic theoretical modeling to predict the copy-number statistics of mRNA or protein that would result from these kinetics ( Friedman et al . , 2006; Raj et al . , 2006; Shahrezaei and Swain , 2008; Thattai and van Oudenaarden , 2001 ) . The theoretical prediction is then compared to measured single-cell data , to validate the assumptions and estimate kinetic parameters . Using this approach , cell-cell variability in mRNA numbers has been successfully used to demonstrate the bursty , non-Poissonian nature of mRNA production in organisms from bacteria to mammals ( Bahar Halpern et al . , 2015b; Raj et al . , 2006; Senecal et al . , 2014; So et al . , 2011; Zenklusen et al . , 2008 ) , and to decipher how genetic and cellular parameters modulate these kinetics ( Jones et al . , 2014; Sanchez and Golding , 2013 ) . However , the ability to map back mRNA copy-number statistics to transcription kinetics is limited by a number of factors . First , the measured number of mRNA molecules in the cell represents temporal integration over the lifetime of mRNA molecules ( Raj et al . , 2006 ) . And while in bacteria this lifetime is very short ( ~mins [Chen et al . , 2015] ) , in higher organisms it can be as long as hours ( Schwanhäusser et al . , 2011 ) . Consequently , the measured mRNA level is a poor proxy for the instantaneous activity of the gene . Second , the cellular mRNA combines contributions from all copies of the gene of interest—for example , four copies in a diploid cell at G2 . Each of these gene copies acts individually and stochastically ( Hansen and van Oudenaarden , 2013; Levesque et al . , 2013 ) ; their combined contribution depends on whether they are correlated and how . Finally , the sampled population typically contains a mixture of cells at different phases of the cell cycle . As a result , deterministic changes in gene copy number and activity along the cell cycle add to the measured population heterogeneity , and may be erroneously interpreted as resulting from stochastic effects ( Zopf et al . , 2013 ) . Here we demonstrate how these limitations can be overcome , such that mRNA statistics is reliably used to infer the kinetic parameters of stochastic gene activity . Specifically , we investigate the transcriptional activity of Oct4 and Nanog , two key genes in the pluripotency network of mouse embryonic stem cells ( Young , 2011 ) . Elucidating the stochastic kinetics of these genes , and how it changes along the cell cycle , is crucial for understanding pluripotency and the path to differentiation . For one , Nanog expression has been reported to exhibit large cell-to-cell variability ( Filipczyk et al . , 2013; Kalmar et al . , 2009; Singer et al . , 2014 ) , and this variability was argued to play an important role in differentiation ( Abranches et al . , 2014; Chambers et al . , 2007; Silva et al . , 2009 ) , but both the sources and consequences of Nanog variability are still unclear ( Cahan and Daley , 2013; Torres-Padilla and Chambers , 2014 ) . It has also been shown that human stem cells’ propensity to differentiate varies significantly between different phases of the cell cycle ( Gonzales et al . , 2015; Pauklin and Vallier , 2013; Singh et al . , 2013 ) , but again , we are lacking a detailed picture of the underlying transcriptional activity of key pluripotency factors along the cell cycle . To elucidate Oct4 and Nanog kinetics along the cell cycle , we simultaneously measured the numbers of nascent ( actively transcribed ) and mature mRNA for each gene in individual cells , and used the DNA contents of the cell to determine its cell-cycle phase . We next used the single-cell data to test how gene activity depends on the presence of other copies of the same gene and how it changes as the gene replicates during the cell cycle . This information allowed us to construct a stochastic model for gene activity , which explicitly accounts for the presence of multiple gene copies and the progression of the cell cycle . We then used the cell-cycle-sorted single-cell data to calibrate the theoretical model and estimate the kinetic parameters that characterize Oct4 and Nanog activity . Our first goal was to measure simultaneously nascent and mature mRNA from the genes of interest . While both mRNA species reflect the same underlying kinetics of gene activity , the two are subject to very different kinetics of elimination: Nascent mRNA is eliminated ( by being converted to mature mRNA ) once elongation and splicing are complete , typically in a few minutes ( Coulon et al . , 2014; Martin et al . , 2013 ) . In contrast , mature mRNA is subject to active degradation , with a typical half-life of a few hours ( Sharova et al . , 2009 ) . A consequence of these very different time scales is that simultaneously measuring both species for the same gene would allow us to better constrain the theoretical model of gene activity and estimate the underlying parameters ( see below and Figure 1—figure supplement 1 ) . To detect nascent and mature mRNA in individual cells , we used single-molecule fluorescence in situ hybridization ( smFISH ) ( Femino et al . , 1998; Raj et al . , 2008; Skinner et al . , 2013 ) to label the gene of interest , with spectrally-distinct probes sets for the intron and exon sequences ( Hansen and van Oudenaarden , 2013; Senecal et al . , 2014; Vargas et al . , 2011 ) . Under this labeling scheme , nascent mRNA are expected to be bound by both probe sets , while mature mRNA will only exhibit exon-probe binding ( Figure 1A ) . Consistent with these expectations , Oct4 and Nanog labeling in mouse embryonic stem cells revealed numerous diffraction-limited spots containing exon-only signal ( Figure 1B , Figure 1—figure supplement 2 ) . In the same cells , only a small number of nuclear spots contained both intron and exon signals ( Figure 1B , Figure 1—figure supplement 2 ) . Neither type of spot was observed in Fibroblasts , where Oct4 and Nanog are not expressed ( Chambers et al . , 2003; Pesce et al . , 1998 ) ( Figure 1B , Figure 1—figure supplement 2 ) . We used automated image analysis to identify individual mRNA spots , allocate them to cells and discard false positive spots ( Skinner et al . , 2013 ) ( Figure 1C , Figure 1—figure supplement 3 , Materials and methods 5 ) . We identified the fluorescence intensity corresponding to a single mature mRNA ( Skinner et al . , 2013; Zenklusen et al . , 2008 ) and used this intensity value to convert the total fluorescence of exon spots in each cell to the numbers of nascent and mature mRNA ( Figure 1G ) . Our measured values for both the mean and coefficient of variation for Nanog mRNA per cell ( 126 ± 24 and 0 . 80 ± 0 . 05 , respectively; designates mean ± SEM throughout; 3 experiments with >600 cells per experiment; Figure 1D ) are in excellent agreement with the literature ( Abranches et al . , 2014; Faddah et al . , 2013; Grün et al . , 2014; Hansen and van Oudenaarden , 2013; Muñoz Descalzo et al . , 2013; Ochiai et al . , 2014; Singer et al . , 2014 ) ( Supplementary file 1A ) . For Oct4 , our measured mean ( 477 ± 67; 3 experiments with >700 cells per experiment; Figure 1D ) is higher than in previous reports ( Faddah et al . , 2013; Grün et al . , 2014; Singer et al . , 2014 ) while our coefficient of variation ( 0 . 34 ± 0 . 01 ) is in agreement with previous estimates ( Faddah et al . , 2013; Grün et al . , 2014; Singer et al . , 2014 ) ( Supplementary file 1A ) . The difference in mean values may reflect differences in cell lines or experimental conditions . 10 . 7554/eLife . 12175 . 003Figure 1 . Quantifying mature mRNA , nascent mRNA and cell-cycle phase in individual mouse embryonic stem ( ES ) cells . ( A ) Identifying nascent and mature mRNA . Introns ( red ) and exons ( green ) were labeled using different colors of smFISH probes . In the cell , pre-spliced nascent mRNA at the site of active transcription are bound by both probe sets , whereas mature mRNA are only bound by the exon probe set . ( B ) Mouse embryonic stem ( ES ) cells ( top row ) labeled for Oct4 exons ( left column , green ) and introns ( center column , red ) . Automated image analysis ( right column ) was used to identify the cell boundaries ( black line ) , intron ( red ) and exon ( green ) spots , as well as false-positive spots ( black circles , see Panel C ) . Co-localized exon and intron spots ( yellow ) were identified as nascent mRNA ( square ) , whereas spots found only in the exon channel were identified as mature mRNA . Fibroblasts ( bottom row ) were used as negative control . Scale bar , 5 µm . ( C ) The distribution of Oct4 mRNA spot intensities for mature mRNA ( green , >100000 spots ) , nascent mRNA ( red , >1000 spots ) , and spots found in Fibroblasts ( black , >1000 spots ) . The histograms were used to discard false positive spots ( gray region ) and to identify the signal intensity corresponding to a single mRNA . ( D ) The distributions of mature and nascent mRNA numbers per cell for Oct4 ( >700 cells ) and Nanog ( >1000 cells ) . ( E ) The same cells as in panel B , labeled for DNA using DAPI ( left column , blue ) . Automated image analysis ( right column ) was used to identify the nuclear boundary ( black line ) . The DNA content of each nucleus was used to estimate the phase of the cell cycle ( cyan , grey , and blue shading; see Panel F ) . ( F ) The distribution of DNA content per cell ( >700 cells ) , estimated from the nuclear DAPI signal ( panel E ) . The histogram of DNA content per cell was fitted to a theoretical model of the cell cycle ( black line ) , and used to identify which cells are in G1 phase ( cyan ) and which in G2 ( blue ) . ( G ) Overlay of the smFISH and DAPI channels for mouse embryonic stem cells ( top ) and fibroblasts ( bottom ) . The estimated number of mature ( green ) and nascent ( red ) mRNA , as well as the phase of the cell cycle ( blue ) , are indicated for the two stem cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00310 . 7554/eLife . 12175 . 004Figure 1—figure supplement 1 . Fitting both nascent and mature mRNA constrains model parameters . ( A ) The distributions of mature ( left ) and nascent ( right ) mRNA numbers were calculated ( dashed line ) and simulated using the Gillespie algorithm ( Gillespie , 1977 ) ( gray bars , 10000 simulations ) for the stochastic model of transcription kinetics described in the main text ( Figure 3A ) . The parameters used were: kON = 1 min-1 , kOFF = 1 min-1 , kINI = 5 min-1 , τRES = 1 min , kD = log ( 2 ) /60 min-1 . For simplicity , no cell-cycle effects were included . Each simulation was run for a total of 20000 min and an observation time tob was randomly selected from the last 10000 min . At tob , the number of mature mRNA was recorded , and the equivalent number of full-length transcripts of nascent mRNA was calculated from the timing of initiation events occurring between the times t = tob-τRES and t = tob . The simulated mature and nascent mRNA data were then each fitted back to the same model using maximum likelihood estimation ( Neuert et al . , 2013 ) , with kON , kOFF and kINI as fitting parameters . The best fits ( log-likelihood values differing from the maximum log-likelihood by <1% ) are indicated on the plots in green and red shading . ( B ) Convex hull of the estimated parameters kON , kOFF , kINI that obey the quality criterion above for mature ( green ) and nascent ( red ) mRNA . It can be seen that parameters estimated from mature or nascent mRNA data independently span ~2 orders of magnitude , while using the fits from both species significantly constrains the parameter space . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00410 . 7554/eLife . 12175 . 005Figure 1—figure supplement 2 . smFISH images of Nanog mRNA in ES cells and Fibroblasts . Mouse ES cells ( top row ) labeled for Nanog exons ( first column , green ) , Nanog introns ( second column , red ) and DNA ( DAPI , third column , blue ) . Fibroblasts ( bottom row ) were used as negative control . Spots in the exon and intron channels were seen in ES cells but not in Fibroblasts . Co-localized exon and intron spots were identified as nascent mRNA ( square ) , whereas spots found only in the exon channel were identified as mature mRNA . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00510 . 7554/eLife . 12175 . 006Figure 1—figure supplement 3 . Distribution of Nanog mRNA spot intensities . The distribution of exon-channel spot intensities for Nanog mature mRNA ( green , >10000 spots ) , nascent mRNA ( red , >1000 spots ) , and spots found in Fibroblasts ( black , >1000 spots ) . The histograms were used to discard false positive spots ( gray region ) and to identify the signal intensity corresponding to a single mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00610 . 7554/eLife . 12175 . 007Figure 1—figure supplement 4 . 3D reconstruction of nuclei from the DAPI channel . The boundary of each nucleus was detected in each focal plane . The nuclei boundaries were used to reconstruct the 3D shape of each nucleus . For more information see Materials and methods 4 . Pixel size is 130 nm × 130 nm . Focal planes have 500 nm spacing . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00710 . 7554/eLife . 12175 . 008Figure 1—figure supplement 5 . Fitting the DNA-content histogram to a cell-cycle model . ( A ) The DNA-content histogram from mouse ES cells ( gray , >700 cells ) was fitted to the Fried/Baisch cell cycle model ( Johnston et al . , 1978 ) ( black ) . ( B ) In the model ( black ) , the distribution of DNA contents per cell is the sum of multiple Gaussians ( colored lines ) with equal coefficients of variation ( CV = σ/μ , the ratio of the standard deviation to the mean ) : The DNA content of cells in G1 phase is represented by a single Gaussian distribution ( green ) with mean μ and standard deviation σ . The DNA of cells in G2/M phase is represented by a Gaussian distribution ( blue ) with mean 2μ and standard deviation 2σ . The DNA content of cells in S phase is approximated by a sum of 3 Gaussians ( brown ) . For more information see Materials and methods 6 . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 008 Next , to identify the cell-cycle phase of individual cells , we used the total DNA contents of each cell , estimated from the DAPI signal integrated over the three-dimensional nucleus ( Figure 1E , Figure 1—figure supplement 4 ) . The distribution of DNA contents from the cell population was well described by the Fried/Baisch model for the cell cycle ( Johnston et al . , 1978 ) ( Figure 1—figure supplement 5 ) . We therefore used the model to classify the cells into G1 , S and G2/M phases ( Figure 1F , G ) . Below we refine this analysis further by calculating , for each cell , its temporal position within the cell cycle and the gene copy number of Oct4 and Nanog ( see Figure 3 ) . At this stage , however , we could already identify sub-populations of cells at the G1 and G2 phases of the cell cycle ( Figure 1F ) , and use these cells to address the questions of gene-copy independence and dosage compensation . First , we tested whether individual copies of the same gene act independently of each other , rather than in a correlated manner . To do so , we examined cells in G1 , where each gene exists in two copies per cell . We measured the number of nascent mRNA at each copy of the gene . For both Oct4 and Nanog , we did not detect significant correlation between the nascent mRNA levels of the two gene copies in the cell ( r , Pearson correlation coefficient; Oct4: r = 0 . 05 ± 0 . 04 , p>0 . 05; Nanog: r = 0 . 07 ± 0 . 01 , p>0 . 05; 3 experiments with >200 cells per experiment ) ( Figure 2—figure supplement 1 ) . Furthermore , we found that , for both genes , the numbers of active transcription sites per cell followed a binomial distribution , consistent with the assumption that the two copies of the gene act independently of each other ( Figure 2A; χ2 goodness of fit test ( Singer et al . , 2014 ) gives p>0 . 05 for both Oct4 and Nanog; 3 experiments with >200 cells per experiment ) . Thus , our data indicate independent stochastic activity of each copy of the gene . 10 . 7554/eLife . 12175 . 009Figure 2 . Oct4 and Nanog exhibit independent allele activity and dosage compensation . ( A ) The distribution of number of active transcription sites for Oct4 ( left; >700 cells ) and Nanog ( right; >1 , 000 cells ) , in cells having two copies of each gene . In both cases , the measured distribution ( gray ) is described well by a theoretical model assuming independent activity of the two alleles ( binomial distribution , red ) . Error bars represent the estimated SEM due to finite sampling . ( B ) The fold change in transcriptional activity following gene replication for Oct4 , Nanog , and a control reporter gene ( CAG-lacZ ) . For Oct4 and Nanog , the average number of nascent mRNA ( left ) increases less than two-fold following gene replication , while a two-fold increase is observed in the control reporter gene . The change in number of nascent mRNA reflects an increase in the number of active transcription sites ( middle ) , with no change in the number of nascent mRNA at each transcription site ( right ) . Error bars represent SEM from 3 experiments with >200 cells per cell-cycle phase in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 00910 . 7554/eLife . 12175 . 010Figure 2—figure supplement 1 . Nascent mRNA correlation between two gene copies . ( A ) Heat maps of the number of nascent mRNA at the two gene copies within the same cell . Left , Oct4 ( 1 experiment with >200 cells ) . Right , Nanog ( 1 experiment with >200 cells ) . The Pearson’s correlation coefficient ( r; mean ± SEM from 3 experiments with >200 cells per experiment ) between gene copies is indicated on each plot , as well as the p-value ( mean ± SEM from 3 experiments with >200 cells per experiment ) obtained using a Student’s t-distribution ( calculated using the MATLAB function corr ) . ( B ) The data in Panel A were reshuffled by pairing the nascent mRNA at one gene copy from a given cell with the nascent mRNA from a gene copy at another , randomly selected cell . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 010 We next wanted to test how the activity of Oct4 and Nanog changes when each of the genes replicates during the cell cycle . Under the simplest assumption , each gene copy will maintain its transcriptional activity irrespective of the total number of gene copies in the cell . In that case , the prediction would be that the total amount of nascent mRNA doubles between G1 and G2 phases ( Note that the mature mRNA , due to its much longer lifetime ( Supplementary file 1 ) , is not expected to immediately follow the gene dosage in such a simple manner; Figure 3—figure supplement 1 ) . However , when we compared the nascent mRNA level between G1 and G2 phases , we found that , for both Oct4 and Nanog , the fold change was significantly lower than two ( Oct4: 1 . 28 ± 0 . 09 , Nanog: 1 . 51 ± 0 . 15; 3 experiments with >200 cells per phase in each experiment; Figure 2B ) . Thus , Oct4 and Nanog exhibit dosage compensation in their activity , analogous to the effect seen for X-chromosome genes between male and female ( Heard et al . , 1997 ) , as well as for some autosomal genes when their copy number is altered ( FitzPatrick et al . , 2002; Gupta et al . , 2006 ) . The change in gene activity between G1 and G2 was manifested in a <2 fold increase in the number of active transcription sites per cell , while the number of nascent mRNA per active site remained unchanged ( Figure 2B ) . In contrast to Oct4 and Nanog , a reporter gene expressed from a strong synthetic promoter ( Niwa et al . , 1991; Vintersten et al . , 2004 ) did not show dosage compensation , instead exhibiting a two-fold increase in nascent mRNA following gene replication ( 1 . 97 ± 0 . 07; 2 experiments with >200 cells per phase in each experiment; Figure 2B ) . 10 . 7554/eLife . 12175 . 011Figure 3 . Extracting the stochastic kinetics of Oct4 and Nanog . ( A ) A stochastic 2-state model for gene activity , which incorporates cell cycle and gene copy-number effects . Each gene copy stochastically switches between ‘ON’ and ‘OFF’ states . Transcription is stochastically initiated only in the ‘ON’ state . After initiation , the nascent transcript ( red ) elongates with constant speed , and is then converted into a mature mRNA molecule ( green ) . Mature mRNA are degraded stochastically . Gene copies are independent , and their number changes from 2 to 4 following gene replication ( left , cyan box ) . At the end of the cell cycle , mRNA molecules are binomially partitioned between the two daughter cells . Dosage compensation is included though a decrease in the rate of activation following gene replication ( left , grey box ) . ( B ) Estimating the gene replication time and the fold-change in transcriptional activity for Oct4 ( left; >700 cells ) and Nanog ( right; >1000 cells ) . The number of nascent mRNA was plotted against the time within the cell cycle for each cell ( grey points ) , and the data were binned into populations of equal cell number ( black markers ) . The binned data were fit to a step function ( red ) , used to estimate the gene replication time and the fold-change in number of nascent mRNA before/after gene replication . Error bars represent SEM . ( C ) The distribution of mature and nascent mRNA copy number over time , for Oct4 ( left; >700 cells ) and Nanog ( right; >1000 cells ) . The cell population was partitioned into 12 time windows , equally-spaced within the cell cycle ( rows; we discarded the first and last windows , where the low cell numbers lead to a large error in the ERA calculation [Kafri et al . , 2013] ) . The measured distributions ( gray ) are overlaid with the model predictions for mature ( green ) and nascent ( red ) mRNA . ( D ) The probabilistic rates of the transcription process and the gene elongation rate , for Oct4 ( blue ) and Nanog ( red ) . The rates were estimated from the best theoretical fit of the mature and nascent mRNA distributions ( panel C ) . The rate that varies most between Oct4 and Nanog is the probability of switching to an active transcription state , kON , which is ~5-fold higher for Oct4 ( inset ) . Error bars represent SEM from 3 experiments with >600 cells per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 01110 . 7554/eLife . 12175 . 012Figure 3—figure supplement 1 . Expected behavior of mature and nascent mRNA numbers over time . ( A ) A deterministic theoretical model of transcription , which includes the effects of gene replication and cell division , was used to predict the numbers of nascent ( red , top ) and mature ( green , bottom ) mRNA at different times in the cell cycle . For demonstration , the model parameters were given values measured for Oct4 . For more details see Materials and methods 9 . It can be seen that , even though the mRNA levels are cyclostationary ( i . e . the number of mRNA at the end of the cell cycle is twice that at the beginning ) , the level of mature mRNA does not reach steady state during the cell cycle . This is because the lifetime of mature mRNA ( 7 . 1 hr; Supplementary file 1 ) is comparable to the duration of individual cell cycle phases . In contrast , the number of nascent mRNA reaches steady state soon after gene replication because of its short lifetime ( residence time 3 . 5 min; Figure 3D ) . ( B ) The ratio of mean mRNA level in G2 phase to that in G1 is predicted to be 2 for nascent mRNA , but <2 for mature mRNA . ( C ) The predicted ratio of mean mRNA level in G2 phase to that in G1 as a function of the ratio of cell cycle duration to mRNA lifetime ( kD*τDIV; black line ) . The values for nascent ( red ) and mature ( green ) mRNA are indicated on the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 01210 . 7554/eLife . 12175 . 013Figure 3—figure supplement 2 . Mapping DNA content to time in the cell cycle using ergodic rate analysis . Ergodic rate analysis ( see Materials and methods 7 ) was used to transform the DNA content distribution ( left , model fit of the experimental data , see Figure 1—figure supplement 5 ) to a mapping between DNA content and time within the cell-cycle ( right ) . For example , the DNA contents values μ+σ and 2μ ( extracted from the cell cycle model ) are mapped to distinct times within the cell cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 01310 . 7554/eLife . 12175 . 014Figure 3—figure supplement 3 . Agreement between methods of measuring dosage compensation . The extracted fold change in nascent Oct4 ( left ) and Nanog ( right ) mRNA following gene replication , as measured using two methods: Method #1 , comparing the mean number of nascent mRNA of cells in G1 phase to that of cells in G2 phase ( see Figure 1F ) . Error bars represent SE . M . from 3 experiments with >200 cells per phase in each experiment . Method #2 , extracting the fold change from a step-function fit to the nascent mRNA over time ( see Figure 3B ) . Error bars represent SEM from 3 experiments with >600 cells per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 01410 . 7554/eLife . 12175 . 015Figure 3—figure supplement 4 . Estimated gene replication times fall within S phase . The boundaries of S phase were estimated from the fit of the cell cycle model ( see Figure 1F and Figure 1—figure supplement 5 ) . The gene replication times estimated from a step-function fit to nascent mRNA over time ( Figure 3B ) fall within S phase for both Oct4 and Nanog . Error bars represent SEM from 3 experiments with >600 cells per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 01510 . 7554/eLife . 12175 . 016Figure 3—figure supplement 5 . The effect of model representation of dosage compensation on the estimated rates of transcription . The probabilistic rates of the transcription process and the gene elongation rate for Oct4 ( blue ) and Nanog ( red ) . The rates were estimated from the best theoretical fit of the mature and nascent mRNA distributions ( Figure 3C ) , using two versions of the stochastic 2-state model for gene activity . The models differ in their representation of dosage compensation: The kON model ( circles ) includes a decrease in the rate of gene activation following gene replication ( Figure 3A ) , whereas the kOFF model ( squares ) includes instead an increase in the rate of gene inactivation . For each model , the amount of dosage compensation was calculated to reflect the measured increase in the number of nascent mRNA over time ( Figure 3B ) . Error bars represent SEM from 3 experiments with >600 cells per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12175 . 016 To extract the kinetics of Oct4 and Nanog from our single-cell data , we constructed a theoretical model describing the stochastic activity of each gene ( Figure 3A ) . In the model , each copy of the gene switches stochastically between active ( 'ON' ) and inactive ( 'OFF' ) states , with rates kON and kOFF . In the active state , transcription is initiated , again stochastically , with rate kINI . Following initiation , nascent mRNA remains at the transcription site for a finite residence time τRES , representing the combined duration of transcript elongation , splicing and release ( Coulon et al . , 2014; Hoyle and Ish-Horowicz , 2013 ) . The nascent mRNA is then deterministically converted into mature mRNA . Mature mRNA is degraded stochastically with rate kD . The copy number of each gene doubles from two to four at a time τREP during the cell cycle . Following gene replication , the rate of gene activation kON changes by a factor α , to allow for dosage compensation . Finally , at the end of the cell cycle , mature mRNA are partitioned binomially between the two daughter cells ( Golding et al . , 2005; Rosenfeld et al . , 2005 ) . To compare our single-cell data with model predictions , we first mapped the DNA content of each cell to the cell’s temporal position within the cell cycle ( Figure 3—figure supplement 2 ) . This was done using ergodic rate analysis ( Kafri et al . , 2013 ) , which uses static single-cell measurements from steady-state populations to obtain temporal information . We then plotted , for both Oct4 and Nanog , the amount of nascent mRNA as a function of time along the cell cycle ( Figure 3B ) . Fitting the data to a step function allowed us to estimate the gene replication time , τREP , and the fold change in gene activity , α . For both genes , the two parameters were consistent with the estimates using G1 and G2 phases , obtained earlier ( Figure 3—figure supplement 3 and Figure 3—figure supplement 4 ) . Next , we proceeded to estimate the kinetic parameters of gene activity for Oct4 and Nanog . For a given set of parameters , we solved the model above using a modified version of the finite state projection algorithm ( Munsky and Khammash , 2006 ) , extended to include the deterministic process of mRNA elongation , the contribution of multiple gene copies , and the progression of the cell cycle ( Materials and methods 8 ) . Solving the model yielded the copy-number distribution for both nascent and mature mRNA at different times along the cell cycle ( Figure 3C ) . We then used maximum-likelihood estimation ( Neuert et al . , 2013 ) to obtain the values of kON , kOFF , kINI and τRES ( Supplementary file 2A ) . For both Oct4 and Nanog , the estimated parameters provided a good fit between model predictions and the experimental histograms ( Figure 3C ) . The parameter values were also consistent with previous estimates , in cases where such estimates existed ( Supplementary file 1B ) . What are the kinetics revealed by the model ? The Oct4 and Nanog genes spend a comparable fraction of time in the active transcriptional state ( Oct4: kON/ ( kON+kOFF ) ≈ 34% for each gene copy prior to gene replication; Nanog: 22% Supplementary file 2B ) . During each of these 'ON' periods , Oct4 and Nanog produce , on average , similar numbers of mRNA ( Oct4: kINI/kOFF ≈ 110 , Nanog: 120 ) . However , where the two genes vary significantly is in the probabilistic rates of switching between the 'ON' and 'OFF' states , with Nanog switching more slowly in both directions ( kON ≈ 9×10-3 min-1 for Oct4 , 2×10-3 min-1 for Nanog; kOFF ≈ 2×10-2 min-1 for Oct4 , 7×10-3 min-1 for Nanog ) . In particular , the ~5-fold difference in kON results in a correspondingly longer average “OFF” duration for Nanog ( in G1 , τOFF= 1/kON ≈ 8 . 9 hr , compared to 1 . 8 hr for Oct4; Supplementary file 2B ) . The differences in transcription kinetics between Oct4 and Nanog also lead , unavoidably , to different degrees of cell-to-cell variability in mRNA numbers . In particular , the higher measured coefficient of variation for Nanog ( 0 . 80 , compared to 0 . 34 for Oct4 ) is a direct reflection of the lower value of kON ( Raj et al . , 2006 ) . In other words , the large heterogeneity in Nanog levels , highlighted in previous studies ( Abranches et al . , 2013; Chambers et al . , 2007; Filipczyk et al . , 2013; Kalmar et al . , 2009 ) does not require invoking more complex kinetics than those of other genes ( e . g . additional kinetic steps [Neuert et al . , 2013; Senecal et al . , 2014] ) , but merely a difference in the value of a single parameter . Following gene replication , both Oct4 and Nanog exhibit a decrease in the transcriptional activity of each gene copy . The effect of this dosage compensation is to equalize gene expression along the cell cycle and decrease the degree of cell-to-cell variability . The lower variability may be physiologically significant , as it has been reported that changes in Oct4 levels as small as two-fold may lead to different cell fates ( Niwa et al . , 2000 ) . The compensatory effect is achieved through a decrease in the probability of each gene copy to be active ( 0 . 72 fold for Oct4 and 0 . 76 fold for Nanog; Supplementary file 2A ) . Similar behavior was recently reported for a number of genes in cultured mammalian cells ( Padovan-Merhar et al . , 2015 ) . These authors also found that the cell volume ( independently of the cell cycle phase ) strongly affects the number of nascent mRNA at each transcription site . In our study , the cell-to-cell variability in volume within each cell-cycle phase was significantly smaller than that seen by ( Padovan-Merhar et al . , 2015 ) ( CV≈0 . 2 versus ≈0 . 5 ) , preventing us from exploring the effect of cell volume on gene activity . Interestingly , the synthetic reporter gene CAG-lacZ did not exhibit dosage compensation . Perhaps the viral enhancer elements included in the promoter ( Niwa et al . , 1991; Vintersten et al . , 2004 ) are more resistant to the regulatory mechanisms that create the compensatory effect in endogenous genes . We note that despite the complex stochastic kinetics of transcription , and the multiple ways that these kinetics can be modulated ( Sanchez et al . , 2013; So et al . , 2011 ) , some simple unifying features emerge . When comparing the activity of Oct4 and Nanog , we found that the kinetic parameter that varies the most between the two is the probabilistic rate of switching to the active state , kON , while the rates of gene inactivation and of transcription initiation are much closer ( Figure 3D ) . The dosage compensation effect following gene replication , observed in both Oct4 and Nanog ( Figure 2B ) , is also consistent with a change in kON . These two observations extend a number of recent studies in a range of systems ( including one of Nanog in mouse embryonic stem cells [Ochiai et al . , 2014] ) , all suggesting that varying expression level—along the cell cycle ( Padovan-Merhar et al . , 2015 ) , between different growth conditions ( Ochiai et al . , 2014 ) , or under regulation by a transcription factor ( Senecal et al . , 2014; Xu et al . , 2015 ) —is achieved by changing kON . The mechanistic basis for this prevalent phenomenology is yet to be elucidated ( Padovan-Merhar et al . , 2015; Sanchez and Golding , 2013 ) . We have shown how changes in gene copy number and in promoter activity along the cell cycle can be incorporated into the analysis of mRNA copy-number statistics . However , multiple additional factors may contribute to mRNA heterogeneity . First , as noted above , the cell volume has recently been shown to dramatically affect transcription kinetics ( Padovan-Merhar et al . , 2015 ) . Consequently , cell-cell variability in volume will translate into different mRNA levels . Second , the stochastic kinetics of mRNA processing downstream of transcription—splicing ( Coulon et al . , 2014 ) , export from the nucleus ( Bahar Halpern et al . , 2015a; Battich et al . , 2015 ) , degradation and partition at cell division ( Huh and Paulsson , 2011 ) —will too add to mRNA number heterogeneity . Finally , cell-to-cell differences in relevant kinetic parameters—of transcription and the subsequent mRNA processes , of the cell cycle , etc . ( so called 'extrinsic noise' ) —will also contribute to the observed mRNA heterogeneity . Additional work , both experimental and theoretical , is required to delineate the relative contribution of all these factors to the eventual mRNA statistics that we measure . Our protocol is based on Raj et al . ( Raj et al . , 2008 ) . Modifications were made to adapt the protocol to a suspension of mouse embryonic stem cells . Sterile , nuclease-free , aerosol-barrier pipette tips were used . Diethylpyrocarbonate ( DEPC ) -treated water ( Ambion , Carlsbad , CA , AM9922 ) was used whenever the protocol calls for water . We developed custom software in MATLAB to perform nucleus and cell segmentation . For each cell in the fluorescence image stacks , we reconstructed the nucleus in the DAPI channel and recognized the cell boundary in the Cy5 ( intron ) channel ( Figure 1 , Figure 1—figure supplement 4 ) . To begin reconstructing individual nuclei , a series of morphological operations was performed on each focal plane in the DAPI channel image stack . First , a Sobel filter was applied to obtain the edges of the nuclear slices ( the portions of the nuclei visible within the focal plane ) . Second , morphological filling was applied to fill the interiors of the nuclear slices . Third , the focal plane was smoothed using morphological opening . Finally , the optimal threshold value was determined for each nuclear slice following ( Xu et al . , 2015 ) . Briefly , a series of increasing threshold values was applied . At each threshold value , the area ( A ) and circularity ( 4πA/P2; where P is the perimeter length ) of the thresholded nuclear slice was calculated . Once the area and circularity satisfied the criteria: A>500 pixels and 4πA/P2>0 . 7 , the threshold value was used . The processed individual focal planes were stacked to form a 3-dimensional mask . Individual nuclei were identified in the mask as overlapping nuclear slices from neighboring planes . To recognize the cell boundary , we thresholded the Cy5 ( intron ) channel because it primarily had two levels of pixel values corresponding to 1 ) non-specific labeling and/or autofluorescence within cells and 2 ) the non-cell background . The threshold value was determined using Otsu’s method . The reconstructed nuclei were used to segment joined cells using a watershed algorithm with the nuclei as basins , and to remove above-threshold objects that did not contain nuclei . For each image stack , the output of the nucleus and cell segmentation program was visually inspected and refined using a graphical user interface . To examine how the observed ratios of both nascent and mature mRNA numbers before/after gene replication are affected by the relative timescales of mRNA lifetime and cell cycle duration , we created a simple deterministic model for the kinetics of the two species . The model includes only mRNA production and degradation , along with the cell-cycle effects of gene replication and cell division , but disregarding gene-state switching and dosage compensation . The level of each mRNA species is described by: ddtR ( t ) =g ( t ) kINI−kDR ( t ) , g ( t ) =10≤t<τREP2τREP≤t<τDIV , where R ( t ) and g ( t ) are the mRNA and gene copy-numbers , kINI and kD are the rates of mRNA transcription and degradation , τREP and τDIV are the times of gene replication and cell division . When solving for nascent mRNA using this formalism , an effective degradation rate is used , which corresponds to the residence time at the gene , kD =1/τRES . At the end of the cell cycle , mRNA are partitioned to the daughter cells . To obtain the cyclostationary solution , we imposed the boundary condition R ( τDIV ) =2R ( 0 ) . The solution is the following piecewise function: R ( t ) =kINIkD1−e−kD ( τDIV−τREP ) 2−e−kDτDIVe−kDt0≤t<τREPkINIkD2−e−kD ( t−τREP ) −e−kD ( τDIV−τREP ) 2−e−kDτDIVe−kDtτREP≤t<τDIV . R ( t ) is plotted in Figure 3—figure supplement 1A for both mature and nascent Oct4 mRNA using the measured gene replication time ( τREP; Figure 3—figure supplement 4 ) , the effective transcription initiation rate from averaging over ‘ON’/’OFF’ gene states ( kINI=0 . 6 min-1; Figure 3D ) , the literature average of mature mRNA degradation rate ( kD; Supplementary file 1 ) , the measured residence time ( τRES; Figure 3D ) , and the literature estimate of the cell division time ( τDIV=13 hr; [Cartwright et al . , 2005] ) . Next , we defined observation time windows for the early and late parts of the cell cycle , within which the numbers of mRNA are averaged: ⟨R ( 0≤t<τ1 ) ⟩=1τ1∫0τ1dtR ( t ) , and ⟨R ( τ2≤t<τDIV ) ⟩=1τDIV−τ2∫τ2τDIVdtR ( t ) , where τ1 is some time in the beginning of the cell cycle before the gene replication time , and τ2 is some time near the end of the cell cycle after the gene replication time . The ratio , RM , is defined as: RM≡⟨R ( τ2≤t<τDIV ) ⟩⟨R ( 0≤t<τ1 ) ⟩ . We calculated RM for nascent and mature Oct4 mRNA ( Figure 3—figure supplement 1B ) using the periods of G1 and G2 phases extracted from the cell cycle model ( Figure 1F ) as the first ( 0≤t<τ1 ) and second ( τ2≤t<τDIV ) observation time windows in addition to the parameters used above . To demonstrate the effect of varying mRNA lifetimes , we plotted RM against kDτDIV ( Figure 3—figure supplement 1C ) .	Scientific investigation requires researchers to use experimental observations to understand the biological process that resulted in these observations . One example is a cellular process called transcription , where the DNA of a gene is copied many times to make molecules of messenger RNA ( mRNA ) , which are later used as instructions to make proteins . Scientists indirectly measure the dynamics of transcription , that is , how often the gene produces mRNA , by counting how many mRNA molecules there are in many individual cells . These numbers are then compared to the predictions made by a mathematical model of transcription , and if the model and experiment agree well , this is interpreted to mean that the model properly describes how often this gene is transcribed . Unfortunately , this procedure is not straightforward because many factors complicate the relationship between the dynamics of transcription and the number of mRNAs that will be detected in each cell at any one point in time . For example , it is not possible to tell whether a detected mRNA has just been transcribed , or whether it was made hours ago . The age of the cell and how many copies of the template DNA are present also affect the dynamics of transcription . As a result , mRNA measurements may be misinterpreted , leading to wrong conclusions about how highly particular genes are transcribed . To address this problem , Skinner et al . first improved the experimental measurements by discriminating between mature mRNA and the new mRNA that is still being transcribed . The experiments also measured how much DNA each cell contains , which indicates how old the cell is . These measurements were incorporated into a new mathematical model that is able to predict the dynamics of transcription of particular genes . Skinner et al . applied their method to two mouse genes called Oct4 and Nanog , which regulate the transformation of embryonic stem cells into other types of cells . The experiments show that both genes can switch between an “on” state where they are being actively transcribed and an “off” state where little or no mRNA is being produced . However , Nanog switches between these two states less often than Oct4 , which results in larger variations between the numbers of mRNAs between different cells . The experiments also show that over the life of the cell , the level of transcription from each copy of the DNA decreases . Skinner et al . ’s approach can be used to refine our knowledge of the transcription of other genes . However , to further improve our understanding of transcription , measurements of other factors will need to be incorporated into the mathematical models .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "short", "report", "computational", "and", "systems", "biology" ]	2016	Single-cell analysis of transcription kinetics across the cell cycle
Lethal mutagenesis has emerged as a novel potential therapeutic approach to treat viral infections . Several studies have demonstrated that increases in the high mutation rates inherent to RNA viruses lead to viral extinction in cell culture , but evidence during infections in vivo is limited . In this study , we show that the broad-range antiviral nucleoside favipiravir reduces viral load in vivo by exerting antiviral mutagenesis in a mouse model for norovirus infection . Increased mutation frequencies were observed in samples from treated mice and were accompanied with lower or in some cases undetectable levels of infectious virus in faeces and tissues . Viral RNA isolated from treated animals showed reduced infectivity , a feature of populations approaching extinction during antiviral mutagenesis . These results suggest that favipiravir can induce norovirus mutagenesis in vivo , which in some cases leads to virus extinction , providing a proof-of-principle for the use of favipiravir derivatives or mutagenic nucleosides in the clinical treatment of noroviruses . Due to elevated error frequencies during the replication of their genetic material , RNA virus populations exist as complex distributions of mutant genomes also known as quasispecies ( Domingo et al . , 2012 ) . Genetic variability confers viral populations the flexibility to rapidly adapt to the environment , typically the host , and respond to different selective constraints such as the immune system or antiviral compounds ( Domingo et al . , 2008 ) . As a consequence , changes in the replication fidelity of a virus can affect its virulence and transmission during natural infections ( Pfeiffer and Kirkegaard , 2005; Vignuzzi et al . , 2006; Bull et al . , 2010 ) . Recent evidence suggests that RNA virus replication error rates are finely balanced to generate ample diversity while maintaining sufficient accuracy in the transmission of genetic information ( Pfeiffer and Kirkegaard , 2005; Vignuzzi et al . , 2006; Levi et al . , 2010; Gnädig et al . , 2012; Sanz-Ramos et al . , 2012; Rozen-Gagnon et al . , 2014 ) . Pioneering theoretical studies on self-replicating genomes suggested that any live organism has a maximum error rate tolerated to copy its genome . Given their elevated mutation rates , it was anticipated that RNA viruses exist close to their corresponding tolerated threshold ( Swetina and Schuster , 1982; Domingo , 2000; Eigen , 2002 ) . Hence , slight increases in virus mutation frequencies might result in the extinction of the replicating population ( Eigen , 2002; Domingo et al . , 2010 ) . These predictions led to the proposal of lethal mutagenesis of viruses as a new therapeutic approach based on reducing the fidelity of genome replication . Several nucleoside analogues ( i . e . , ribavirin , 5-fluorouracil , 5-azacytidine ) and non-nucleoside compounds ( amiloride ) display antiviral activities in cell culture against a wide range of RNA viruses and appear to act via increased mutagenesis ( Loeb et al . , 1999; Crotty et al . , 2000; Sierra et al . , 2000; Grande-Pérez et al . , 2002; Dapp et al . , 2009 ) . Increased mutation frequencies are accompanied with decreased virus progeny , infectivity and fitness , leading in some cases to the complete extinction of the virus population ( Domingo et al . , 2012 ) . Despite all the evidence in cell culture , data confirming lethal mutagenesis as a plausible therapeutic approach in vivo remains limited . Although some of the antiviral compounds with mutagenic activity in cell culture are also reported to have antiviral activity in vivo , controversy exists regarding the therapeutic mechanism . To date , the most successful antiviral compound with mutagenic activity in cell culture is purine analogue ribavirin ( Crotty et al . , 2000; Maag et al . , 2001; Brochot et al . , 2007; Perales et al . , 2009; Moreno et al . , 2011 ) . Ribavirin is commonly used in combination with pegylated interferon in the treatment of hepatitis C virus ( HCV ) infections . There are contradictory results on whether the mode of action of ribavirin on HCV in vivo is due to mutagenesis , immunomodulatory activities , or other antiviral mechanisms ( Graci and Cameron , 2006; Chevaliez et al . , 2007; Lutchman et al . , 2007; Perelson and Layden , 2007; Chung et al . , 2013; Dietz et al . , 2013 ) . Although increased HCV mutation frequencies have been reported in several cases ( Asahina et al . , 2005; Dietz et al . , 2013 ) , some studies have not found increased mutagenesis as a factor contributing to the associated antiviral activity ( Chevaliez et al . , 2007; Lutchman et al . , 2007 ) . Therefore there remains an open question as to whether or not ribavirin-mediated antiviral activity in vivo is as a result of mutagenesis or one of the other reported mechanisms ( Graci and Cameron , 2006; Chevaliez et al . , 2007 ) . Hence , further investigations are needed to demonstrate that lethal mutagenesis is a conceivable approach for the general treatment of RNA virus infections in vivo . Several novel compounds eliciting antiviral mutagenesis in cell culture were recently identified ( Levi et al . , 2010; Mullins et al . , 2011; Baranovich et al . , 2013; Dapp et al . , 2014 ) . Among them , favipiravir is a novel broad spectrum nucleoside analogue which is effective in the control of a vast number of RNA viruses in vivo ( Gowen et al . , 2007; Furuta et al . , 2009; Mendenhall et al . , 2011; Baranovich et al . , 2013; Caroline et al . , 2014; Oestereich et al . , 2014; Smither et al . , 2014 ) , although its therapeutic mechanism of action is still under study . Favipiravir was initially identified as an antiviral compound for the treatment of influenza virus infection ( Furuta et al . , 2002; Sidwell et al . , 2007 ) whose activity correlates with increased mutagenesis in cell culture ( Baranovich et al . , 2013 ) . A recent study demonstrated that favipiravir-triphosphate can be used as a substrate by the virus polymerase and incorporated ambiguously into RNA opposite C and U in the template molecule ( Jin et al . , 2013 ) . With the aim of demonstrating lethal mutagenesis as a conceivable approach to treat viral infections in vivo , we investigated whether ribavirin and favipiravir elicit antiviral activities in a mouse model of persistent norovirus infection . Favipiravir caused significant decreases in virus titres and viral RNA levels and led to the clearance of infectious virus in the faeces of seven out of nine animals ( only two out of ten animals did not shed detectable infectious virus in control group ) . Blind passaging of faecal and tissue samples in cell culture confirmed that favipiravir treatment led to viral extinction in some animals . In contrast , ribavirin showed limited efficacy in vivo which correlated with lower antiviral activity in cell culture when compared to favipiravir . Favipiravir antiviral activity in vivo was associated with a significant increase in viral mutation frequencies . Viral RNA isolated from faeces of treated animals showed decreased specific infectivity , and infectious virus from faeces showed decreased fitness in tissue culture . These data suggest that favipiravir causes increased mutagenesis leading to decreased infectivity and fitness of viral genomes , both features of viral populations approaching extinction during mutagenesis . These results constitute a proof of concept for lethal mutagenesis in vivo and support antiviral therapies based on mutagenic compounds at the clinical level . The data also support the use of favipiravir in the treatment of norovirus infections for which there are not yet licenced antivirals or vaccines available and highlight a need for further studies on favipiravir and other improved derivatives as possible broad-range antiviral strategies . To investigate whether ribavirin and favipiravir elicit any antiviral activity on norovirus replication , we employed two different murine norovirus ( MNV ) strains , MNV-1 and MNV-3 . MNV-1 was the first MNV strain isolated as the causative agent of a fatal infection in Stat1−/− mice ( Karst et al . , 2003 ) . Later studies identified many other MNV strains , including MNV-3 , wide spread in different mice laboratory colonies ( Hsu et al . , 2006; Thackray et al . , 2007; Barron et al . , 2011 ) . MNV is used as a model to study norovirus replication and pathogenesis and has facilitated a better understanding of the norovirus life cycle ( Wobus et al . , 2006 ) . MNV-1 efficiently replicates in cell culture although it shows limited virulence in wild-type mice . In contrast , MNV-3 typically produces lower yields in tissue culture , yet establishes long-term persistent infections in wild-type mice , being detected for at least 8 months after inoculation ( Arias et al . , 2012a; McFadden et al . , 2013 ) . Previous studies demonstrated that ribavirin and favipiravir elicit antiviral activity against MNV-1 and human norovirus ( HuNoV ) replicon in cell culture ( Chang and George , 2007; Rocha-Pereira et al . , 2012 ) . However , the mechanism of action and their utility in vivo have not been described . The treatment of RAW264 . 7 cells infected with MNV ( MOI of 0 . 01 TCID50/cell ) with ribavirin or favipiravir resulted in significant reductions in virus yields , reaching up to a 3-log10 decrease for ribavirin and a 4-log10 for favipiravir ( Figure 1 ) . A decrease in cell viability was observed when cells were treated with high concentration of ribavirin , although this reduction never exceeded 50% . In contrast , no significant decrease in cell viability was observed in cells treated with favipiravir ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03679 . 003Figure 1 . Ribavirin and favipiravir are efficient inhibitors of norovirus replication . ( A ) MNV-1 and MNV-3 viral yields obtained after infection of confluent monolayers of RAW264 . 7 cells in the absence ( white bars ) or presence of 200 ( light grey ) , 400 ( dark grey ) , or 800 μM ribavirin ( black ) . MNV was inoculated at an MOI of 0 . 01 TCID50/cell and infections were allowed to proceed for 24 hr when cultures were freeze-thawed for virus release . ( B ) MNV-1 and MNV-3 viral yields obtained after infection of confluent monolayers of RAW264 . 7 cells in the absence ( white bars ) or presence of 200 ( light grey ) , 400 ( dark grey ) , or 800 μM favipiravir ( black ) . MNV was inoculated at an MOI of 0 . 01 TCID50/cell and infections were allowed to proceed for 24 hr . Statistically significant differences are represented ( p < 0 . 05 , ; p < 0 . 01 , ; p < 0 . 001 , *; 2-way ANOVA test ) . ( C and D ) Kinetics of MNV-1 and MNV-3 infection in the presence of ribavirin or favipiravir . Confluent monolayers of RAW264 . 7 cells were infected with MNV-1 ( C ) or MNV-3 ( D ) at an MOI of 5 TCID50/cell . Infected cell cultures were treated with 200 μM ribavirin ( RBV ) or favipiravir ( FPV ) as explained in ‘Materials and methods’ . Replication kinetics of MNV-1 and MNV-3 in untreated infected cells are shown in parallel ( DMEM ) . Every time point is the average of three biological replicates ( ±SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 00310 . 7554/eLife . 03679 . 004Figure 1—figure supplement 1 . Favipiravir and ribavirin toxicity in RAW264 . 7 cells . ( A ) Ribavirin toxicity upon RAW264 . 7 cells was scored using both trypan blue , which measures the proportion of dead cells , and CellTiter-Blue Cell Viability Assay ( Promega ) , which accounts for living active cells . ( B ) Favipiravir toxicity was determined by CellTiter-Blue Cell Viability Assay ( Promega ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 004 To further investigate the mechanism of inhibition , we determined the effect of ribavirin and favipiravir upon MNV replication in single-cycle replication kinetics carried out by infecting RAW264 . 7 cells at high MOI . Favipiravir inhibition occurred in a gradual manner , with the greatest reduction in virus titre being observed at later time points ( Figure 1C , D ) . This result is suggestive of a cumulative antiviral activity often observed with increasing number of mutations in viral genomes during successive rounds of replication . In contrast , ribavirin inhibited norovirus replication from an early time point ( 8 hr ) , suggesting a different mechanism of action , an additional antiviral activity relative to favipiravir , or potentially due to a different mutational spectrum activity elicited by these two compounds . However , it is alternatively possible that the delayed inhibition shown by favipiravir compared to ribavirin reflects the fact that it takes several more steps to convert favipiravir to the active form ( Furuta et al . , 2005 ) . To investigate this possibility , we repeated the kinetics by incubating the cells overnight with favipiravir before infection to facilitate favipiravir conversion into favipiravir-triphosphate . No major differences were observed with cells pre-incubated during 1 hr or overnight with favipiravir ( data not shown ) , which suggests that ribavirin might be a more potent inhibitor of viral RNA replication . To confirm this possibility , we determined the viral RNA synthesis kinetics for MNV-3 in the presence of both compounds and confirmed that ribavirin inhibits viral RNA synthesis from an early time post-infection , while the effects of favipiravir are only observed at later time points ( data not shown ) . Although ribavirin and favipiravir cause increased mutagenesis in several RNA viruses ( Graci and Cameron , 2006; Moreno et al . , 2011; Baranovich et al . , 2013 ) , the mechanism of antiviral activity against noroviruses is not known . To determine whether ribavirin and favipiravir treatment resulted in greater mutation frequencies , we carried out sequence analysis of individual molecular clones isolated from populations subjected to 4 passages in the presence of either 200 μM ribavirin or favipiravir . We found that both compounds caused significant increases in the mutation frequencies of replicating virus . Ribavirin treatment resulted in a ∼threefold increase while favipiravir led to a five to sixfold increase in the number of mutations per nucleotide ( Figure 2 ) . Importantly , alterations in the transition frequency patterns were also observed . Ribavirin treatment resulted in greater proportion of G to A and C to U transitions than in untreated virus populations , while favipiravir led to a slight increase in A to G and U to C transition rates ( Table 1 ) . These alterations are in agreement with those observed for other viruses treated with the same compounds ( Agudo et al . , 2010; Levi et al . , 2010; Baranovich et al . , 2013; Ortega-Prieto et al . , 2013 ) . 10 . 7554/eLife . 03679 . 005Figure 2 . Increased mutation frequencies in virus quasispecies treated with ribavirin or favipiravir . Mutation frequencies are represented as the average number of mutations found every 10 , 000 nucleotides sequenced in MNV-1 and MNV-3 populations after 4 passages in RAW264 . 7 cells in the absence or presence of 200 μM ribavirin ( RBV ) or favipiravir ( FPV ) ( p < 0 . 05 , ; p < 0 . 001 , *** , Mann–Whitney U test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 00510 . 7554/eLife . 03679 . 006Table 1 . Mutation type distribution in MNV populations treated with ribavirin and favipiravirDOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 006UntreatedRibavirinFavipiravirMNV-1MNV-3MNV-1MNV-3MNV-1MNV-3A → G441395U → C131237G → A025220C → U414404Transversions131210Deletions101000Total nucleotides sequenced63 , 39565 , 82223 , 35820 , 24312 , 48816 , 003Total number of nucleotides sequenced in each different untreated or treated population analysed . Proportion of different types of mutations observed in untreated , or ribavirin- or favipiravir-treated MNV populations . A signature feature for error catastrophe in virus populations subjected to mutagenesis is a decrease in virus specific infectivity ( González-López et al . , 2005; Grande-Pérez et al . , 2005; Perales et al . , 2009 ) . To investigate whether these compounds reduced the infectivity of treated norovirus populations , we serially passaged MNV in the presence of either ribavirin or favipiravir ( Figure 3 , Figure 4 , respectively ) . In both cases , lower virus titres and encapsidated viral RNA levels were observed during passage in cell culture . After 5 passages , virus titres for MNV-1 treated with 400 μM favipiravir , and MNV-3 treated with either 400 μM ribavirin or favipiravir were close to or below the detection limit , indicating that ribavirin and favipiravir can cause lethal mutagenesis of MNV . We confirmed that MNV-3 was extinguished after 5 passages in the presence of 400 μM favipiravir , by three consecutive blind passages in RAW264 . 7 cells in the absence of the drug . The specific infectivity , a measure of the ratio of infectious virus per encapsidated genome , was consistently reduced during passages in the presence of favipiravir for both MNV strains . However , treatment with ribavirin resulted in decreased specific infectivity only for MNV-3 treated with the highest concentration ( 400 μM ) . This suggests that the lower mutagenic activity exhibited by ribavirin ( Figure 2 ) is responsible for this limited effect on specific infectivity . MNV-3 was more sensitive to favipiravir and ribavirin than MNV-1 ( Figures 3 and 4 ) , although this difference could not be associated with significant variations in the mutation frequency values between these strains during infections in treated or untreated cells ( Figure 2 ) . This different behaviour may be related to the observation that MNV-3 has lower fitness than MNV-1 ( compare virus titre yields for untreated infections in Figures 1 , 3 and 4 ) which can result in a greater sensitivity to mutagenesis , as previously reported for other viruses ( Sierra et al . , 2000; Pariente et al . , 2001 ) . Confirming this possibility , a tissue culture-adapted MNV-3 population ( 18 serial passages in cell culture ) responded similarly to MNV-1 when treated with 400 μM ribavirin ( data not shown ) . 10 . 7554/eLife . 03679 . 007Figure 3 . Murine norovirus titres decrease during serial passage in the presence of ribavirin . MNV-1 ( A , B , C ) and MNV-3 ( D , E , F ) were serially passaged in the absence ( white circles ) or presence of 200 ( grey triangles ) or 400 μM ribavirin ( black squares ) . Virus was inoculated at an MOI of 0 . 1 TCID50/cell in passage 1 . Subsequent passages were carried out with 200 μl ( 1/10 vol ) of neat virus recovered from the previous passage . The different graphs show the resulting virus titres ( A and D ) , genome copy equivalents ( B and E ) , and the resulting specific infectivity determined for encapsidated genomes ( C and F ) . Specific infectivity values were calculated as the number of infectious viruses ( TCID50 units ) found in 109 genome copies from data obtained in A , B , D , and E . To calculate the number of genome copy equivalents , non-encapsidated genomes were removed before RNA extraction by micrococcal nuclease treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 00710 . 7554/eLife . 03679 . 008Figure 3—figure supplement 1 . Reduced virus yields in ribavirin-mutagenised populations . RAW264 . 7 cells were infected with MNV-1 or MNV-3 recovered after 5 passages in the absence ( untreated ) or presence of ribavirin ( 200 or 400 μM ) with the exception of 400 μM ribavirin-treated MNV-3 where passage 2 was used due to insufficient virus titres in subsequent passages . Infections were performed at an MOI of 0 . 01 . After adsorption ( 1 hr at 37°C ) , virus inoculum was removed and complete media added to infected cultures . Infected cells were then incubated as mentioned in ‘Materials and methods’ for a total of 17 hr before freezing . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 00810 . 7554/eLife . 03679 . 009Figure 4 . Murine norovirus infectivity decreases during serial passages in the presence of favipiravir . MNV-1 ( A , B , C ) and MNV-3 ( D , E , F ) were serially passaged in the absence ( white circles ) , or presence of 200 ( grey triangles ) or 400 μM favipiravir ( black squares ) . Passage 1 infections were carried out at an MOI of 0 . 1 TCID50/cell . Subsequent passages were carried out with 200 μl ( 1/10 vol ) of neat virus recovered from the previous passage . The different graphs show the resulting virus titres ( A and D ) , genome copy equivalents ( B and E ) , and the resulting specific infectivity determined for encapsidated genomes ( C and F ) . Specific infectivity values were calculated as the number of infectious viruses ( TCID50 units ) found in 109 genome copies from data obtained in A , B , D , and E . To calculate the number of genome copy equivalents , non-encapsidated genomes were removed before RNA extraction by micrococcal nuclease treatment . The arrow in D indicates virus extinction confirmed by the absence of infectious virus and viral RNA ( qPCR ) after three serial passages in the absence of favipiravir . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 00910 . 7554/eLife . 03679 . 010Figure 4—figure supplement 1 . Reduced virus yields in favipiravir-mutagenised populations . RAW264 . 7 cells were infected with MNV-1 or MNV-3 recovered after 5 passages in the absence ( untreated ) or presence of favipiravir ( 200 μM ) Infections were performed at an MOI of 0 . 01 . After adsorption ( 1 hr at 37°C ) , virus inoculum was removed and complete media added to infected cultures . Infected cells were then incubated as mentioned in ‘Materials and methods’ for a total of 17 hr before freezing . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 010 To further investigate the effect of mutagenesis on virus fitness , RAW264 . 7 cells were infected with virus obtained after 5 passages in the presence of ribavirin or favipiravir ( Figures 3 and 4 ) at the same MOI ( 0 . 01 TCID50/cell ) . Viral populations previously subjected to either favipiravir or ribavirin mutagenic treatment resulted in decreased virus titres , in agreement with a loss of infectivity as a consequence of mutagenesis ( Figure 3—figure supplement 1 , Figure 4—figure supplement 1 ) . The efficacy of ribavirin and favipiravir as antiviral compounds in vivo was next investigated in C57BL/6 mice persistently infected with MNV-3 . Animals were infected 4 weeks before the beginning of the treatment to allow the establishment of a persistent infection . In a preliminary experiment ( Figure 5—figure supplement 1 ) , we found that all the animals treated with favipiravir had lower virus titres ( 5/5 ) in their faeces while only 2 out of 5 animals treated with ribavirin had consistently lower titres ( Figure 5—figure supplement 1 ) . This suggests that favipiravir is more efficient than ribavirin in the control of norovirus replication in vivo , in agreement with data obtained in tissue culture . Hence , we conducted further experiments to determine whether favipiravir can drive lethal mutagenesis of MNV-3 during persistent infection in mice . With this aim , we treated mice persistently infected with MNV-3 with either placebo or 600 mg/kg/day of favipiravir for 8 weeks ( Figure 5 ) . Favipiravir was effective in the control of persistent norovirus replication in vivo with decreased virus titres and RNA levels in faeces observed ( Figures 5 and 6 ) at very beginning of the treatment ( day 1 ) and throughout the entire treatment period ( day 53 ) . A predicted half-life of 31 days is calculated for virus clearance in favipiravir-treated mice in contrast with 122 days in the untreated group ( p = 0 . 0064 , log-rank test; Figure 5B ) . 10 . 7554/eLife . 03679 . 011Figure 5 . Favipiravir reduces infectious norovirus titres in mice faeces and tissues . Two groups of ten 4–5-week old C57BL/6 male mice were oral gavage-infected with 104 TCID50 units of MNV-3 . 4 weeks after virus inoculation , persistently infected animals were subjected to treatment with either 300 mg/kg animal of favipiravir twice a day ( FPV ) or with buffer ( Control ) for 8 weeks . From day 35 onwards , there are nine animals instead of 10 in the favipiravir-treated group ( due to the accidental death of one mouse during dosing ) . ( A ) Virus titres in faeces of animals untreated or treated with favipiravir . Virus titres were determined by TCID50 assays of faecal samples supernatant previously resuspended at 100 mg/ml in PBS ( p < 0 . 05 , ; p < 0 . 01 , ; p < 0 . 001 , *; 2-way ANOVA test ) . A dashed line indicates the limit of detection ( 3 . 02 Log10 TCID50/g stool ) . ( B ) Virus titres in caecum and colon of animals after 53 days of treatment with favipiravir . Virus titres were determined by TCID50 assays of homogenates of caecum and colon resuspended in DMEM at a concentration of 30 mg/ml ( p < 0 . 05 , ; 2-way ANOVA test ) . ( C ) Reduced positive shedding in animals treated with favipiravir . The percentage of animals shedding detectable virus titre along time , based on A , decreases faster in animals treated with favipiravir than in untreated animals ( p = 0 . 0064 , log-rank test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01110 . 7554/eLife . 03679 . 012Figure 5—figure supplement 1 . Antiviral activity in vivo of favipiravir and ribavirin . 4–5-week old male C57BL/6 mice were oral gavage-infected with 104 TCID50 units of MNV-3 . 4 weeks after infection , animals underwent daily treatment with mutagenic compounds for a total of 18 days ( day 0 to day 17 ) , with the exception of days 6 and 13 when the animals were not treated . Animals were dosed by oral gavage with 8 mg/day ( 3 animals ) or 16 mg/day ( 2 animals ) of either ribavirin or favipiravir for 12 days . Afterwards , the animals were treated with 16 mg/day of the same drug for six additional days . Virus titres obtained in faecal samples isolated 4 days before treatment began ( −4 ) and after 2 and 18 days of treatment ( **; p < 0 . 001 , 2-way ANOVA test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01210 . 7554/eLife . 03679 . 013Figure 5—figure supplement 2 . Infectious virus rebound after blind passage of faecal and tissue sample homogenates in RAW264 . 7 cells . Virus titres obtained after blind passage infection in RAW264 . 7 cells of faeces , caecum , and colon homogenates obtained from animals at treatment day 53 ( Figure 5A , B ) . An arrow indicates samples from the same three animals that remained negative after blind passage , indicating extinction . These samples remained negative by TCID50 assay and qPCR after three consecutive blind passages . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01310 . 7554/eLife . 03679 . 014Figure 5—figure supplement 3 . Sensitivity to favipiravir of virus isolated in animal faeces . Faecal virus samples obtained from untreated or treated mice during 53 days were amplified by blind passage in RAW264 . 7 cells and treated with favipiravir to examine the possible presence of resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01410 . 7554/eLife . 03679 . 015Figure 5—figure supplement 4 . Reduced weigh increase in favipiravir-treated mice . They are represented animal weights before ( −5 ) and during the treatment with placebo or favipiravir ( ; p < 0 . 05; 2-way ANOVA test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01510 . 7554/eLife . 03679 . 016Figure 6 . Long exposure to favipiravir results in significantly decreased norovirus RNA levels in animal tissues and faeces . Two groups of ten male C57BL/6 mice of 4–5-weeks were oral gavage-infected with 104 TCID50 units of MNV-3 . 4 weeks after virus inoculation , persistently infected animals were subjected to treatment with either 300 mg/kg animal of favipiravir twice a day ( FPV ) or with buffer ( Control ) for 8 weeks . At day 53 , there are nine animals instead of ten in favipiravir-treated group due to the accidental death of one mouse during dosing . A dashed line indicates the limit of detection ( 102 genome copy equivalent per mg of stool ) . ( A ) Viral genome copy equivalents isolated in faecal samples ( p < 0 . 05 , ; p < 0 . 01 , ; p < 0 . 001 , *; 2-way ANOVA test ) . Viral RNA extracted was then reverse transcribed and quantitated as described in ‘Materials and methods’ . ( B ) Viral genome copy equivalents isolated in caecum and colon ( p < 0 . 05 , ; 2-way ANOVA test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01610 . 7554/eLife . 03679 . 017Figure 6—figure supplement 1 . Semi-quantitative analysis of viral RNA in favipiravir-treated and control mice . Viral RNA was extracted and reverse transcribed as described in ‘Materials and methods’ ( Arias et al . , 2012a ) . Standard curve was obtained with known amounts of in vitro transcribed MNV-3 RNA ( 102 to 105 genome equivalents ) that were PCR amplified after reverse transcription . ( A ) RT-PCR amplification of faecal samples isolated from animals treated during 53 days with favipiravir or placebo . RNA isolated in 50 μg of faeces was RT-PCR amplified and agarose-gel resolved as explained above . ( B ) RT-PCR amplification of RNA extracted from colon samples isolated from animals treated for 8 weeks with favipiravir or placebo . 50 ng of RNA isolated from different animal colon tissues were RT-PCR amplified as explained above . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 017 After 53 days of treatment with favipiravir , infectious virus titres were undetectable in the faeces of 7 out of 9 animals , while only 2 out of 10 animals had undetectable titres in the control group , suggesting that the infections were cleared as a result of favipiravir treatment ( Figure 5A , C ) . Similar results were obtained when homogenates of caecum and colon collected at the end-point ( day 53 ) were analysed for the presence of MNV ( Figure 5B ) . To confirm whether these samples were negative , we carried out a blind passage in RAW264 . 7 cells . We confirmed that three treated animals were negative for all the samples analysed ( faeces , caecum , and colon ) while the remaining treated and untreated animals ( 6 and 10 respectively ) were positive for MNV ( Figure 5—figure supplement 2 ) . All these three animal samples remained negative by titration and qPCR after 3 passages in the absence of favipiravir . These results suggest that favipiravir has assisted in clearing the infection in 33% of treated mice . To investigate whether favipiravir treatment in vivo resulted in the selection of adapted MNV-3 variants , we carried out infections in RAW264 . 7 cells with virus samples recovered from mice . We did not identify differences in the sensitivity to favipiravir between samples isolated from treated and untreated animals , suggesting no adaptation to the treatment ( Figure 5—figure supplement 3 ) . Quantification of MNV RNA in faecal samples also showed lower viral levels in treated mice than in control animals for the duration of the study ( Figure 6 ) . The same three animals that were negative for MNV-3 above showed viral RNA levels in faeces below the detection limit after 53 days of treatment , determined both by RT-qPCR ( Figure 6 ) , and RT-PCR followed by agarose gel analysis ( Figure 6—figure supplement 1 ) , while all placebo-treated animals showed high levels of viral RNA . Quantification of viral RNA extracted from caecum and colon , the major tissues for virus replication during persistent infections ( Arias et al . , 2012a ) , also confirmed that viral RNA could not be detected in these three treated mice that contain no amplifiable infectious virus in faeces and tissues , further supporting that virus infection was cleared in these animals ( Figure 6 ) . To clarify if favipiravir antiviral activity observed in vivo was associated with mutagenesis , we examined the mutation frequency of viral populations shed in the faeces of five different animals in each group . The mutation frequencies found in virus samples from favipiravir-treated animals were greater than in placebo-treated mice with an average of a 2 . 9-fold increase relative to control animals being observed ( Figure 7A ) . We also determined the mutation frequencies in three samples isolated from ribavirin-treated animals in the preliminary experiment , and they were similar to those observed for placebo-treated animals ( 4 . 0 ± 2 . 2 vs 3 . 6 ± 1 . 4 substitutions per 10 , 000 nucleotides sequenced , respectively ) . These results suggest that the antiviral activity observed for favipiravir in vivo is linked to mutagenesis , and the clearance of infection in some of these animals is the consequence of lethal mutagenesis of persistently replicating virus . 10 . 7554/eLife . 03679 . 018Figure 7 . Increased mutation frequencies and decreased infectivity in virus populations isolated from favipiravir-treated animals . ( A ) Mutation frequency in virus isolated in faecal samples . Every value in the graph represents the virus mutation frequency in a different animal faecal sample after 18 days of treatment . Mutation frequencies are represented as the average number of nucleotide substitutions found in every 10 , 000 nucleotides sequenced . ( 0 . 001 < p < 0 . 05 , ; t test ) . ( B ) Decreased infectivity in viral RNA isolated from favipiravir-treated animals . Viral RNA isolated from placebo and favipiravir-treated animal faecal samples was quantified and 2 × 105 genome copy equivalents were lipofected in semiconfluent BHK-21 cell monolayers . At 24 hr post-transfection , cells were freeze-thawed and the resulting virus yields determined by TCID50 assays in RAW264 . 7 cells . They are represented as the virus titres obtained per 106 genome copies isolated from the faeces of infected animals before treatment ( day 0 ) and at treatment days 7 and 18 . ( C ) MNV recovered from favipiravir-treated animals shows reduced replication yields . Virus isolated from animal faeces treated with favipiravir were first amplified in RAW264 . 7 cells allowing virus replication for 24 hr . Recovered viruses were titrated and 0 . 01 TCID50 units/cell applied to new RAW264 . 7 cell monolayers . Virus infections were collected at 8 hr post-infection and the cultures freeze-thawed to release infectious virus ( ; p < 0 . 01; t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 01810 . 7554/eLife . 03679 . 019Figure 7—figure supplement 1 . Specific infectivity of viral RNA isolated from treated animals . 4–5-week old male C57BL/6 mice were oral gavage-infected with 104 TCID50 units of MNV-3 . 4 weeks after infection , animals underwent daily treatment with mutagenic compounds during a total of 18 days ( day 0 to day 17 ) with the exception of days 6 and 13 when the animals were not treated . Animals were dosed by oral gavage with 8 mg/day ( 3 animals ) or 16 mg/day ( 2 animals ) of either ribavirin or favipiravir for 12 days . Afterwards , the animals were treated with 16 mg/day of the same drug for six additional days . Viral RNA isolated from placebo- , ribavirin- , and favipiravir-treated animal faecal samples were quantified and 5 × 104 genome copy equivalents lipofected in semiconfluent BHK-21 cell monolayers . At 24 hr post-transfection cells were freeze-thawed and resulting virus yields determined by TCID50 assays in RAW264 . 7 cells . They are represented the infectivity values for RNA samples isolated from faeces collected at day 18 . DOI: http://dx . doi . org/10 . 7554/eLife . 03679 . 019 To investigate whether favipiravir-induced mutagenesis in vivo resulted in decreased infectivity of viral populations , we firstly isolated and quantified viral RNA from faeces at various times during the treatment of animals with favipiravir and used an equivalent genome copy number to transfect BHK-21 cells ( Figure 7B ) . BHK-21 cells support MNV replication but not virus ( re ) infection , resulting in a single round of replication for the transfected genomes and a better correlation between infectious genome units and virus titre . Thus , this approach provides an indirect measure of viral genome specific infectivity , as only viable genomes will result in the recovery of infectious virus . Viral titres recovered were consistently lower in samples obtained from favipiravir-treated animals than in placebo-treated animals ( Figure 7B ) , suggesting decreased specific infectivity as a consequence of increased mutagenesis . Virus titres for genomes isolated from ribavirin-treated animals showed no significant difference with untreated animals ( Figure 7—figure supplement 1 ) . To compare whether mutagenesis in treated animals resulted in lower fitness of virus samples recovered , we determined the relative replication rates of virus populations recovered from mice after 18 days of treatment . To this aim , we carried out MOI controlled infections ( 0 . 01 TCID50/cell ) in RAW264 . 7 cells using virus previously isolated from faecal samples and propagated once in cell culture . Virus yields obtained using virus from favipiravir-treated animal samples were lower on average than those isolated from placebo-treated animal samples , suggesting that mutagenesis induced by favipiravir results in a viral fitness cost in vivo ( Figure 7C ) . Thus , favipiravir causes increased mutagenesis and decreases the specific infectivity and fitness of norovirus in vivo which supports an antiviral activity mediated by enhanced mutagenesis . Lethal mutagenesis has been the subject of numerous studies in cell culture in the last few years as an alternative approach to classical antiviral therapies ( Perales et al . , 2011 ) . Due to the elevated mutation frequencies in RNA viruses , it was predicted that an increase in the replication error rate might result in the extinction of the viral population ( Eigen , 2002; Domingo et al . , 2010 ) . However , several in vivo studies carried out up to date have resulted in insufficient evidence to support lethal mutagenesis as a practical approach at the clinical level . In this study , we show that favipiravir , a purine nucleoside analogue , can cause the extinction of an RNA virus during replication in its natural host . The antiviral activity observed in vivo is associated with increased mutation frequencies and , importantly , reduced infectivity in virus samples isolated from the treated animals . These are features typically observed in viral quasispecies approaching extinction during lethal mutagenesis which constitutes a proof-of-principle for this antiviral strategy . Favipiravir was initially identified as an antiviral compound for the treatment of influenza virus infection ( Furuta et al . , 2002; Sidwell et al . , 2007 ) and currently is being tested in a phase 3 clinical trial . Previous studies on mice and data from clinical trials in humans show that favipiravir is well tolerated in vivo with much less toxicity exhibited than ribavirin ( Gowen et al . , 2007 ) . In this study , we observed no major side effects in the mice treated during the 8 weeks . The only apparent side-effect is that treated mice show a modest reduced rate of weight gain compared to the control group ( Figure 5—figure supplement 4 ) . In addition to its activity in vivo against influenza virus and data shown here on MNV , favipiravir is efficient in the control and clearance of a broad number of RNA viruses including picornavirus , parmyxovirus , bunyavirus , arenavirus , and togavirus ( Furuta et al . , 2002; Gowen et al . , 2007; Mendenhall et al . , 2011; Caroline et al . , 2014; Oestereich et al . , 2014; Smither et al . , 2014 ) . Importantly , favipiravir has shown efficient antiviral activity in mouse models for Ebola virus infections , which has led some African countries to consider using this drug for the control of the current outbreak ( Ikuomola , 2014 ) . Further studies are needed to elucidate if the mechanism of action of favipiravir against these other viruses is lethal mutagenesis , and if it constitutes a possible universal antiviral mutagen for the clinical treatment of viral diseases . In particular , an attractive possibility would be to study the effect of favipiravir in the control of HCV , particularly given the lower toxicity displayed by favipiravir . These results are also relevant for the development of antiviral strategies to control human norovirus ( HuNoV ) infections for which there are currently no licenced vaccines or antiviral therapies . Due to the absence of cell culture systems to recover and propagate infectious HuNoVs , MNV has been suggested as a potential surrogate system ( Wobus et al . , 2006 ) . HuNoVs are a significant cause of non-bacterial gastroenteritis with large economic losses typically associated with frequent outbreaks in contained environments ( >$160 million in UK hospitals alone ) . In addition , they have been linked to other important disorders such as ulcerative colitis and persistent diarrhoea ( Murata et al . , 2007; Ludwig et al . , 2008; Khan et al . , 2009; Capizzi et al . , 2011; Bok and Green , 2012 ) . Chronic norovirus infections constitute a major health problem in immunocompromised patients with no treatment yet available ( Bok and Green , 2012 ) . Favipiravir or other derivatives with improved pharmacokinetic properties may constitute an attractive candidate for the treatment of these patients . The rapid evolution and large mutation frequencies of norovirus replicating in immunocompromised patients ( Bull et al . , 2012 ) suggest that antiviral mutagenesis could be an effective approach . Favipiravir could also be considered as prophylactic treatment to reduce virus dissemination during the course of an outbreak , especially in contained environments such as hospitals or nurseries . The data obtained for persistently infected mice support this possibility with lower virus yields shed by treated animals since an early time during treatment ( Figure 5 ) . Recent evidence suggests that the combination of a mutagenic compound with a classical antiviral molecule can be more efficient in the extinction of a virus than the use of the compound alone or the combination of classical inhibitors only ( Pariente et al . , 2001; Perales et al . , 2009 ) . Inhibitors with antiviral activity in vivo have been identified for multiple RNA viruses , including norovirus ( Perry et al . , 2012; Rocha-Pereira et al . , 2013 ) , encouraging further studies in this direction . Given its significant efficacy in the control of different RNA viruses , favipiravir or other derivatives with improved pharmacokinetics constitute attractive candidates to become universal antiviral compounds against viral diseases via lethal mutagenesis . Studies with mice were performed in the Department of Pathology BSU Unit ( PCD 80/2802 ) after ethical review by the University of Cambridge Review Panel and subsequent approval by the UK Home Office ( PPL70/7689 ) . All animal procedures and care conformed strictly to the UK Home Office Guidelines under The Animals ( Scientific Procedures ) Act 1986 . Procedures for the cultivation of cells and MNV infections have been previously described ( Arias et al . , 2012a ) . Murine leukaemia macrophage cells RAW264 . 7 were used for the propagation and titration ( TCID50 assay ) of murine norovirus 1 and 3 ( MNV-1 and MNV-3 ) used in this study . Baby hamster kidney cells ( BHK-21 ) were used for the determination of infectivity in viral genomes isolated from animal samples . All the different cell lines were cultured in Dulbecco's modified Eagle medium ( DMEM ) with 10% FCS , 100 U/ml penicillin , and 100 mg/ml streptomycin ( complete DMEM ) and maintained at 37°C with 10% CO2 . MNV-1 and MNV-3 strains used in this study were obtained after reverse genetics recovery of infectious virus as previously described ( Arias et al . , 2012b ) . Recovered viruses were then subjected to two consecutive passages in RAW264 . 7 cells . The resulting population was titrated and used as a passage 0 stock . RAW264 . 7 cells were grown until they formed confluent monolayers ( ∼2 × 106 cells in 35 mm diameter dish ) . Supernatant was then removed and replaced by 1 ml of complete DMEM containing either 200 , 400 , or 800 μM ribavirin or favipiravir and the cells were incubated for two additional hours at 37°C and 10% CO2 . Cells were inoculated with 200 μl of virus at the multiplicity of infection ( MOI ) indicated and incubated for 1 hr at 37°C and 10% CO2 . Supernatants were removed , cells washed to eliminate unattached virus , and 2 ml of fresh media containing ribavirin or favipiravir were added to each well . Cell cultures were collected at 24 hr post-infection and virus released through two consecutive freeze–thaw cycles . For experiments involving serial passage of virus populations in the presence of favipiravir or ribavirin , passage 1 cells were infected at an MOI of 0 . 01 TCID50/cell with MNV-1 or MNV-3 . In subsequent passages , 200 μl of neat virus from the previous passage ( 1/10 of total virus ) were applied to a new monolayer of cells . Viral RNA was extracted from 100 μl of viral samples , either supernatant from lysed infected cultures or PBS-resuspended faeces from animals , using EconoSpin columns ( Epoch , Missouri City , TX ) , following protocols provided by the manufacturer . Viral RNA was quantified using a two-step qPCR approach following protocols described previously ( Arias et al . , 2012a ) . For the calculation of mutation frequency in any virus population , we have followed protocols previously described ( Beaucourt et al . , 2011 ) . Briefly , 4 μl of purified RNA were reverse-transcribed in 20 μl final volume using SuperScript III ( Roche , Switzerland ) as indicated by the manufacturer . 3 μl of cDNA were then PCR amplified using high fidelity KOD polymerase ( Toyobo ) using primers spanning genomic positions 3734 to 3770 and 6074 to 6034 in MNV-1 and 3734 to 3770 and 5738 to 5711 in MNV-3 . PCR products were purified with EconoSpin columns ( Epoch ) and directly ligated in plasmid PCR Blunt using Zero Blunt PCR cloning kit ( Life Technologies , Carlsbad , CA ) . Positive Escherichia coli colonies were identified by PCR screening with primers flanking the vector-cloning site and GoTaq polymerase ( Promega , Madison , WI ) . The resultant PCR products corresponding to individual MNV cDNA clones were sequenced and the mutation frequency in each population calculated . 4–5-week old male C57BL/6 mice were orally infected with 104 TCID50 units of MNV-3 as previously described ( Arias et al . , 2012a ) . After 4 weeks , persistently infected animals were subjected once or twice daily to oral gavage treatment with ribavirin , favipiravir , or placebo . Ribavirin was dissolved in PBS before inoculation into animals , while favipiravir was resuspended in 0 . 5% carboxyl methyl cellulose ( CMC ) in PBS . For the preliminary experiment ( Figure 5—figure supplement 1 ) , animals were treated once or twice a day with 8 mg of ribavirin or favipiravir ( ∼300 or 600 mg/kg animal/day ) for 18 days . For the larger experiment ( 10 mice per group ) ( Figures 5 and 6 ) , animals were treated with 300 mg/kg animal favipiravir twice a day ( 600 mg/kg animal/day ) for 8 weeks . Control animals were treated with 0 . 5% CMC in PBS . Faecal samples were collected at different time points along the treatment period and the presence of infectious particles and viral RNA determined . Animals were sacrificed after the 8-week treatment period and caecum and colon tissues collected to analyse the presence of viral RNA . To confirm the absence of infectious virus in those faecal and tissue samples that showed negative infectivity by TCID50 assays , 100 μl of samples homogenates were used to infect 105 cells . Infections were collected at 24 hr and subjected to two freeze-thawing cycles to release virus . The resulting virus was analysed by TCID50 assays . Those samples that remained negative after blind passage amplification were subjected to two additional blind passages in RAW264 . 7 cells as explained above , and the absence of infectious MNV was confirmed by TCID50 assays and qPCR . To determine whether virus replicating in vivo has acquired resistance to favipiravir , virus samples obtained from animal faeces at treatment day 53 were blind passaged in RAW264 . 7 cells as mentioned above . The amplified virus samples were applied to 105 RAW264 . 7 cell monolayers at an MOI of 0 . 01 , and infections were allowed to proceed for 48 hr in the absence ( DMEM ) or presence of 200 μM favipiravir .	Viruses can infect , take control of and replicate themselves inside the living cells of other organisms . Some viral diseases can be treated with antiviral drugs , which stop viral infections either by making it more difficult for viruses to enter cells or by preventing the virus replicating once inside . As antiviral drugs are currently only available to treat a handful of viral infections , efforts are underway to develop and test experimental antiviral drugs . One such experimental drug is called favipiravir , which is proving to be effective against several viruses that store their genetic information in the form of RNA molecules . These viruses include those that cause diseases such as influenza , gastroenteritis , and Ebola . Along with ongoing work determining how safe and effective favipiravir is for treating viral infections , researchers are also attempting to better understand how favipiravir works . Whenever a strand of RNA is copied to allow a new virus to form , there is a risk that mistakes—or mutations—that could harm the virus are introduced into the genetic code . Previous experiments performed on cells grown in the laboratory suggested that favipiravir works against RNA viruses by increasing how often these mutations occur . RNA viruses naturally experience a large number of mutations and the ability to make mutations is in fact a benefit for viruses as it allows them to evolve rapidly and to escape immune responses . However , there is a limit to how many mutations can be tolerated in the viral genome before it can no longer replicate . Therefore , a slight increase in how often mutations occur—as thought to be caused by favipiravir—is able to stop the RNA virus replicating and halt the infection . However , favipiravir's mode of action had yet to be confirmed in living animals . Using mice , Arias et al . tested favipiravir's ability to treat a persistent infection by norovirus—the most common cause of viral gastroenteritis in humans and also responsible for life-threatening chronic diarrhoea in immunodeficient patients . Treatment increased the number of mutations that occurred when the viral RNA replicated and could reduce the amount of virus in the mice to undetectable levels . In addition , favipiravir did not show toxicity in mice after 8 weeks of treatment . This suggests that favipiravir has the potential to be used safely and effectively to treat norovirus and other RNA viruses , although further studies are required before it can be developed into a clinical treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2014	Favipiravir elicits antiviral mutagenesis during virus replication in vivo
Nonsense-mediated mRNA decay ( NMD ) is a translation-dependent RNA quality-control pathway targeting transcripts such as messenger RNAs harboring premature stop-codons or short upstream open reading frame ( uORFs ) . Our transcription start sites ( TSSs ) analysis of Saccharomyces cerevisiae cells deficient for RNA degradation pathways revealed that about half of the pervasive transcripts are degraded by NMD , which provides a fail-safe mechanism to remove spurious transcripts that escaped degradation in the nucleus . Moreover , we found that the low specificity of RNA polymerase II TSSs selection generates , for 47% of the expressed genes , NMD-sensitive transcript isoforms carrying uORFs or starting downstream of the ATG START codon . Despite the low abundance of this last category of isoforms , their presence seems to constrain genomic sequences , as suggested by the significant bias against in-frame ATGs specifically found at the beginning of the corresponding genes and reflected by a depletion of methionines in the N-terminus of the encoded proteins . Recent advances in sequencing technologies led to the detection of a wealth of new RNA transcripts and revealed that eukaryotic genomes are pervasively transcribed . In human cells , roughly 75% of the genome gives rise to RNA transcripts of various length but only an estimated 2% corresponds to protein-coding messenger RNAs ( mRNAs ) ( Djebali et al . , 2012 ) . Even in a compact genome such as the budding yeast Saccharomyces cerevisiae ( S . cerevisiae ) , hundreds of such pervasive non-coding RNAs ( ncRNAs ) were identified in addition to the stable ncRNAs such as transfer RNAs ( tRNAs ) , small nucleolar RNAs ( sn ( o ) RNAs ) , and ribosomal RNA ( rRNAs ) ( Neil et al . , 2009; Xu et al . , 2009; van Dijk et al . , 2011; Geisler et al . , 2012 ) . Yeast ncRNAs are transcribed from nucleosome-free regions ( NFRs ) present throughout the genome , most often at gene promoters and terminators , and part of these RNAs represent by-products of divergent transcription initiation ( Neil et al . , 2009; Xu et al . , 2009 ) . Different mechanisms of RNA quality control prevent their accumulation in wild-type cells . In the nucleus , transcription from bidirectional promoters is restricted by early termination-coupled RNA degradation pathways ( Arigo et al . , 2006; Thiebaut et al . , 2006; Almada et al . , 2013; Ntini et al . , 2013 ) . In the yeast S . cerevisiae , degradation-coupled transcription termination of pervasive transcripts relies on the Nrd1-Nab3-Sen1 ( NNS ) complex , which interacts with the carboxy-terminal domain ( CTD ) of the RNA polymerase II ( RNAPII ) phosphorylated on serine 5 and triggers termination upon recognition of short sequences on the nascent RNA ( Arigo et al . , 2006; Thiebaut et al . , 2006; Gudipati et al . , 2008; Vasiljeva et al . , 2008; Schulz et al . , 2013 ) . Early termination of these transcripts ( named CUTs for cryptic unstable transcripts ) is coupled with the recruitment of the Trf4-Air2-Mtr4 ( TRAMP ) complex and the Rrp6-containing nuclear RNA-exosome for rapid degradation ( Wyers et al . , 2005; Davis and Ares , 2006; Tudek et al . , 2014 ) . However , a significant proportion of the pervasive transcripts can escape this early nuclear quality-control step and be exported to the cytoplasm where they are targeted for degradation by the cytoplasmic 5′-3′ exonuclease Xrn1 ( XUT; van Dijk et al . , 2011 ) . Some pervasive transcripts seem to be immune enough to both surveillance pathways to be detectable in wild-type cells and are called SUTs ( stable unannotated transcripts; Xu et al . , 2009 ) . The distinction between these different classes of transcripts is not very stringent . Their behavior can be similar , as illustrated by the fact that in the absence of Xrn1p the average levels of SUTs and CUTs increase by 7 . 9-fold and 3 . 6-fold respectively ( van Dijk et al . , 2011 ) . What makes pervasive transcripts ( XUTs in the first instance ) highly sensitive to Xrn1p-dependent cytoplasmic degradation remains to be determined but may be linked to the presence of small spurious open reading frames ( ORF ) in these transcripts that will make them substrates for the nonsense-mediated mRNA decay ( NMD ) as has been recently demonstrated for a new class of unannotated transcripts ( Smith et al . , 2014 ) . Originally described more than two decades ago as a quality-control pathway targeting for degradation mRNAs containing a premature termination codon ( Leeds et al . , 1991 ) , NMD is a translation-coupled quality-control pathway affecting cytoplasmic transcripts with restricted coding capacities ( for a review see Kervestin and Jacobson , 2012 ) . NMD targets include mRNAs harboring ‘upstream open reading frames’ ( uORFs ) ( He et al . , 2003; Arribere and Gilbert , 2013 ) and unspliced or incorrectly spliced transcripts exported to the cytoplasm ( Sayani et al . , 2008; Kawashima et al . , 2014 ) accounting for roughly 10–15% of mRNAs in yeast as well as in human cells ( Kervestin and Jacobson , 2012 and references therein ) . This number may be an underestimate since recent studies underscored the structural heterogeneity of most mRNAs at both their 5′ and 3′ -ends ( Ozsolak et al . , 2010; Arribere and Gilbert , 2013; Pelechano et al . , 2013; Waern and Snyder , 2013 ) . Transcript heterogeneity at the 5′-end is particularly relevant for NMD targeting since the use of different transcription start sites ( TSSs ) will generate , for a given gene , mRNAs with different 5′-UTRs that may include uORFs . Although NMD targets were previously analysed genome-wide using NMD-deficient strains of S . cerevisiae , the techniques used to perform these analyses ( He et al . , 2003; Kawashima et al . , 2014; Smith et al . , 2014 ) did not allow monitoring such individual transcript isoforms and were thus susceptible to have missed a large number of transcripts degraded by this pathway . In order to address this question in a global and systematic manner , we used a modified 5′-RACE approach ( Hashimoto et al . , 2009; Arribere and Gilbert , 2013 ) to perform a genome-wide analysis of TSSs in wild-type S . cerevisiae cells as well as in cells deficient for nuclear and NMD RNA quality-control pathways . Minor transcript isoforms targeted for degradation by NMD were identified for almost half of yeast protein-coding genes , underscoring the low specificity of TSS selection by RNAPII . In particular , our study revealed for the majority of protein-coding genes the use of TSSs downstream the ATG start codons . This phenomenon has the potential to generate N-terminally truncated proteins if the first ATG encountered by ribosomes translating this 5′ truncated transcript is in the right reading frame . Such an undesirable outcome seems , however , to be counteracted by the significant depletion of in-frame relative to out-of-frame ATGs at the beginning of protein-coding genes in yeast . This bias increases the probability of such transcripts to have very short coding regions and be efficiently degraded by NMD . Our analysis also showed that NMD restricts the accumulation of cryptic transcripts initiating inside transcribed ORFs as a consequence of altered chromatin structure and provides a fail-safe control mechanism for the removal of pervasive transcripts that escaped degradation by the nuclear quality-control pathway , reminiscent of what has been previously shown for some unspliced pre-mRNAs ( Sayani and Chanfreau , 2012 ) . This includes not only a large fraction of the XUTs and SUTs but also a fraction of the CUTs and previously unannotated transcripts , which accumulate to substantial levels only when both pathways are inactive . To identify TSSs genome-wide and with high specificity we used a modified genomic 5′-RACE approach ( Hashimoto et al . , 2009; Arribere and Gilbert , 2013 ) that we called TSS sequencing , which involves a biotin purification step and allows the selective enrichment of the 5′-ends of capped transcripts ( see Figure 1A and ‘Materials and methods’ ) . We evaluated the ‘false-discovery’ rate of the method for the identification of TSSs ( that is , the proportion of sequencing reads not actually mapping to the 5′-end of capped RNAs ) by comparing libraries made using samples treated or not with the tobacco acid pyrophosphatase ( TAP ) , which is required for efficient and specific ligation of the biotinylated primer to the 5′-end of capped-RNAs . To extend this analysis to transcripts that are unstable in wild-type cells , we prepared libraries using RNAs extracted from upf1∆rrp6∆ double mutant strains . To minimize polymerase chain reaction ( PCR ) amplification biases and provide an internal control , the S . cerevisiae poly ( A ) + RNAs , treated or not with TAP , were mixed with an equal amount of TAP-treated poly ( A ) + RNAs from Schizosaccharomyces pombe just prior to the ligation step . After normalization using the S . pombe sequencing reads and removal of the ribosomal DNA ones , these experiments generated 2 , 844 , 877 reads when TAP was used compared with 72 , 435 reads when TAP was omitted . The ratio between these two numbers provides an upper limit for the false-discovery rate of TSS identification of 2 . 5% . Analysis of the cumulative 5′-end read counts per nucleotide around the start codon for all protein-coding genes showed that , genome-wide , 77% ( 2 , 190 , 211 ) of the reads obtained for TAP-treated samples could be mapped within a 200 nucleotide region upstream of the start codons with a maximum at around 30 nucleotides upstream from ATGs , while only 22% mapped to the same region when this treatment was omitted ( Figure 1B ) . However , even in the latter case , the number of mapped reads also peaked at around 30 nucleotides upstream of the ATGs , suggesting that a substantial proportion of these sequences likely correspond to genuine TSSs even though they were generated in the absence of TAP treatment . These data suggest that the real false-discovery rate for TSS is substantially less than 2 . 5% . 10 . 7554/eLife . 06722 . 003Figure 1 . Transcription start site sequencing . ( A ) Schematic view of the methodology used to produce the transcription start site ( TSS ) sequence-tag libraries ( RNA molecules are in blue , DNA molecules in red ) . ( B ) All protein-coding genes were aligned on the A of their annotated ATG start codon and the distribution of the TSSs read counts was computed for each position in a window from −200 to +200 nucleotides for samples treated ( blue curve ) or not ( red curve ) with tobacco acid pyrophosphatase ( TAP ) . The insert within the figure shows a zoomed view of the −100 to +50 nucleotide region . ( C ) Correlation between two biologically independent replicates ( replicate 1: library L5p_03 . WT , replicate 2: library L5p_04 . WT; see Table 1 ) . The Pearson's correlation value ( ρ ) between the read counts of the 174 , 151 TSSs identified in the two data sets is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 00310 . 7554/eLife . 06722 . 004Figure 1—figure supplement 1 . Reproducibility of transcription start site ( TSS ) sequencing . Correlation between two data sets generated in our laboratory months apart ( A ) and between our data and those published by Pelechano et al . ( 2013 ) ( B ) . The Pearson's correlation value ( ρ ) between the two data sets is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 00410 . 7554/eLife . 06722 . 005Figure 1—figure supplement 2 . Transcription start site ( TSS ) consensus sequences . ( A ) Consensus sequence around all the identified TSSs generated using the Web-LOGO algorithm ( Crooks et al . , 2004 ) . The site of transcription initiation ( the TSS ) is labeled +1 while the preceding nucleotide position is labeled −1 . ( B ) Percentage of mismatches between the actual genomic sequence and the first base ( red squares ) or all the other bases ( blue dots ) of the complementary DNAs ( cDNAs ) after alignment of the TSS reads on the genome . A , G , C , and T on the bottom line correspond to the genomic sequence . A strong percentage of mismatches ( ∼60% ) was observed specifically for the first cDNA position when pyrimidines are encoded on the genome at the corresponding location . In 80% of the cases , these mismatches consisted of a pyrimidine to A mismatch . ( C ) As in ( A ) but taking into account only the TSSs for which a pyrimidine to A mismatch at the 5′-end of the cDNAs was observed . The vertical arrows pointing to an A indicate the fact that the +1 nucleotide position , a pyrimidine in the genome , corresponds to a mismatched A at the first position of the cDNA . Up to 30% of the cDNA 5′-ends mapping on a pyrimidine actually conform to this consensus . Note that the presence of a conserved A at position −7 while it is located at position −8 in the general consensus and the presence of the pyrimidine to A mismatch suggests that transcription may not start at the pyrimidine but rather at the following A . Indeed , if one would remove the base at the 5′-end of the cDNA reads and align these to the genome , they would then perfectly conform to the general consensus , including the A at position −8 , suggesting that an extra A is added at the 5′-end of the transcript during its synthesis . ( D ) Schematic model depicting how a one nucleotide ‘backward-shift’ relative to the template occurring during the transcription of the first three consecutive As might lead to the incorporation of four As at the beginning of the transcript . This scenario is supported by the fact that for TSSs mapped at PyAAA sequences , the corresponding transcripts carried three or four As at about equal frequencies , which suggests that this ‘one nucleotide back-shifting’ phenomenon occurs about half of the time when transcription initiates on this particular motif . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 005 To estimate the repeatability of the experiment , two independent biological replicates were generated in parallel from wild-type cells and sequenced on different lanes of an Illumina HiSeq 2500 sequencer . The number of reads mapping to the same genomic position were highly correlated ( Figure 1C; Pearson's correlation coefficient ( ρ ) = 0 . 89 , see also Table 1 ) . When we compared data sets for the wild-type strain generated months apart in our laboratory ( reproducibility test ) , we observed lower correlation coefficients ( ρ = 0 . 71 ) likely reflecting biological variation inherent to samples prepared at different moments ( Figure 1—figure supplement 1A ) . However , the correlation between data generated within our lab is still higher than the correlation between our data and results previously published by Pelechano et al . ( 2013 ) ( ρ = 0 . 58; Figure 1—figure supplement 1B ) or Arribere and Gilbert ( ρ = 0 . 48; Arribere and Gilbert , 2013 ) as well as between these two sets of data ( ρ = 0 . 55 ) . 10 . 7554/eLife . 06722 . 023Table 1 . Libraries generated and analysed in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 023LibraryGenotypeTotal reads countUnique reads countUnique reads count mapped on S . cerevisiae genomeUnique reads count mapped on S . pombe genomeTSS-sequencingL5p_01WT8 , 650 , 6555 , 862 , 2455 , 631 , 307–L5p_01upf1∆9 , 624 , 6267 , 161 , 8426 , 733 , 121–L5p_01rrp6∆10 , 949 , 6937 , 640 , 7017 , 155 , 101–L5p_01upf1∆rrp6∆10 , 839 , 3117 , 681 , 9107 , 231 , 052–L5p_02WT10 , 139 , 3984 , 207 , 8543 , 949 , 042–L5p_02upf1∆10 , 607 , 4084 , 325 , 3984 , 067 , 154–L5p_02rrp6∆22 , 861 , 39610 , 269 , 8229 , 711 , 133–L5p_02upf1∆rrp6∆13 , 803 , 0206 , 526 , 8056 , 207 , 357–L5p_03WT11 , 525 , 6314 , 670 , 5864 , 343 , 264–L5p_03upf1∆9 , 832 , 5903 , 338 , 6793 , 122 , 942–L5p_03rrp6∆11 , 811 , 8795 , 228 , 4094 , 905 , 881–L5p_03upf1∆rrp6∆17 , 157 , 2447 , 897 , 0307 , 430 , 231–L5p_04WT13 , 282 , 06910 , 019 , 7426 , 032 , 695–L5p_04upf1∆14 , 162 , 74110 , 872 , 4086 , 724 , 910–L5p_04set2∆14 , 714 , 18711 , 048 , 3986 , 591 , 345–L5p_04upf1∆set2∆16 , 958 , 16712 , 121 , 8948 , 081 , 113–L5p_05WT11 , 270 , 8244 , 446 , 9884 , 172 , 618–L5p_05upf1∆12 , 093 , 6314 , 599 , 3074 , 323 , 962–L5p_05set2∆18 , 047 , 1347 , 132 , 5856 , 724 , 933–L5p_05upf1∆set2∆12 , 719 , 5645 , 637 , 3105 , 333 , 480–L5p_06WT12 , 253 , 2534 , 800 , 7994 , 463 , 020–L5p_06upf1∆10 , 481 , 4133 , 402 , 3173 , 181 , 752–L5p_06set2∆12 , 179 , 1904 , 448 , 7624 , 167 , 876–L5p_06upf1∆set2∆14 , 269 , 2285 , 584 , 0195 , 227 , 982–L5p_07upf1∆rrp6∆ + TAP8 , 890 , 2862 , 973 , 6121 , 621 , 7121 , 134 , 673L5p_07upf1∆rrp6∆ − TAP13 , 836 , 1723 , 214 , 470149 , 0922 , 700 , 018L5p_08upf1∆rrp6∆ + TAP9 , 689 , 1882 , 768 , 4181 , 341 , 9751 , 209 , 108L5p_08upf1∆rrp6∆ − TAP9 , 765 , 6722 , 390 , 79482 , 6852 , 074 , 096L5p_09upf1∆set2∆ + TAP11 , 885 , 9383 , 793 , 6681 , 976 , 1611 , 555 , 818L5p_09upf1∆set2∆ − TAP11 , 105 , 5852 , 936 , 749117 , 2022 , 552 , 264L5p_10upf1∆set2∆ + TAP11 , 665 , 9863 , 797 , 7081 , 343 , 0912 , 147 , 100L5p_10upf1∆set2∆ − TAP11 , 000 , 4292 , 476 , 13559 , 4502 , 145 , 439RNAseqLT_01WT8 , 104 , 0477 , 257 , 4236 , 634 , 5221 , 761 , 046LT_01upf1∆11 , 137 , 26910 , 129 , 3159 , 257 , 1582 , 440 , 195LT_01xrn1∆11 , 619 , 21110 , 631 , 1269 , 737 , 9242 , 310 , 274LT_01upf1∆xrn1∆7 , 947 , 6277 , 299 , 1516 , 645 , 3531 , 678 , 281LT_02WT34 , 611 , 00327 , 825 , 11822 , 658 , 5033 , 635 , 624LT_02upf1∆29 , 379 , 23324 , 421 , 71719 , 815 , 9132 , 966 , 963LT_02xrn1∆26 , 816 , 26722 , 686 , 42018 , 245 , 2512 , 493 , 471LT_02upf1∆xrn1∆25 , 466 , 01621 , 800 , 53216 , 996 , 7282 , 436 , 568 The complementary DNA ( cDNA ) sequences were aligned with the genomic sequence ( allowing one mismatch in the seed sequence; see ‘Materials and methods’ ) . Analysis of the TSSs and their surrounding sequences using the Web-LOGO algorithm ( Crooks et al . , 2004 ) identified a consensus sequence around TSSs ( Figure 1—figure supplement 2A ) similar to the previously reported one derived from a smaller data set ( Zhang and Dietrich , 2005 ) . In particular , we observed a very strong bias to start at a purine ( 88% of mapped TSSs ) , usually following a pyrimidine ( 76% of the mapped TSSs ) , and the enrichment for an A at position −8 relative to the TSS ( A ( N ) 6PyPu consensus ) . Surprisingly , 58% of TSS reads starting with a pyrimidine when aligned on the genome ( 12% of the mapped TSSs ) show a mismatch at their first nucleotide , most of the time an A instead of the encoded pyrimidine ( Figure 1—figure supplement 2B ) . Moreover , in 32% of these cases the surrounding genomic sequences exhibited a specific consensus , A ( N ) 6PyAAA ( where the underlined base is the mapped TSS; Figure 1—figure supplement 2C ) . These observations suggest that , in these cases , transcription actually initiates on the A following the pyrimidine and that an additional A is added at the 5′-end of the transcript , possibly by a back-tracking mechanism as proposed in the model described in Figure 1—figure supplement 2D . This may also apply to other TSSs mapped on a pyrimidine and showing a mismatched first nucleotide . It thus appears from this observation that , even though 12% of the TSSs mapped on a pyrimidine , transcription initiation actually occurred on a pyrimidine in less than 5% of cases ( 42% of non-mismatched cDNAs out of 12% of cDNAs aligned on a pyrimidine ) . To assess the relative contributions of the cytoplasmic NMD and nuclear RNA control pathways in shaping the yeast transcriptome , the genes encoding Upf1 , an RNA helicase essential for NMD , or Rrp6 , a nuclear exosome catalytic subunit , were deleted ( He et al . , 2003; Wyers et al . , 2005 ) . Since , for a given gene , several closely spaced TSSs can be identified ( Pelechano et al . , 2013 ) , we used a peak-calling procedure to define , in three biological replicates , TSS clusters corresponding to transcript isoforms with closely spaced 5′-ends ( TSSCs; see ‘Materials and methods’ ) . This analysis allowed us to identify 17 , 812 TSSCs , among which 5927 could be assigned to 5′-ends of mRNAs corresponding to 5231 ORFs ( O category in Supplementary file 1 ) , as previously defined ( Pelechano et al . , 2013 ) . Among the remaining TSSCs , 502 and 3644 were assigned to stable ncRNAs or repeated sequences ( F category in Supplementary file 1 ) and previously described pervasive transcripts respectively ( C , X and S categories for CUTs , XUTs and SUTs in Supplementary file 1; Neil et al . , 2009; Xu et al . , 2009; van Dijk et al . , 2011 ) . In contrast to mRNA-TSSCs , the majority of which were unaffected in the absence of Upf1 , Rrp6 or both ( Figure 2A ) and , as expected ( Wyers et al . , 2005 ) , CUT-TSSCs were strongly stabilized in the absence of Rrp6 ( Figure 2B; 79% stabilized significantly , as determined using the moderated estimation of fold change and dispersion function of DESeq2—Love et al . , 2014; see ‘Materials and methods’ ) . In contrast , only 23% of the CUT-TSSCs increased significantly in the absence of Upf1 . The opposite situation was observed for XUTs and SUTs , which were more sensitive to the cytoplasmic NMD quality-control pathway ( 28% vs 52% significantly stabilized in the absence of Rrp6 or Upf1 , respectively; Figure 2C , D and Supplementary file 1 ) . Collectively , 39% and 54% of the pervasive transcripts ( CUTs , SUTs and XUTs ) were significantly sensitive to the absence of Upf1 in a wild-type or rrp6∆ background , respectively , indicating that NMD targets about half of the pervasive transcripts . The absence of both Rrp6 and Upf1 had an additive effect on the stabilization of pervasive transcripts . In the double mutant , 76% and 92% of the XUTs/SUTs and CUTs , respectively , were significantly stabilized , suggesting that NMD provides a fail-safe control mechanism for pervasive transcripts that escaped degradation in the nucleus , reminiscent of what has been previously shown for some unspliced pre-mRNAs ( Sayani and Chanfreau , 2012 ) . 10 . 7554/eLife . 06722 . 006Figure 2 . Differential effect of UPF1 and/or RRP6 deletion on mRNAs and pervasive transcripts . Frequency distribution of the ratios of transcription start site clusters ( TSSCs ) read counts in upf1∆ ( red ) , rrp6∆ ( blue ) , or upf1∆rrp6∆ ( purple ) compared with wild-type ( WT ) for mRNAs ( A ) , cryptic unstable transcripts ( CUTs ) ( B ) , Xrn1-sensitive transcripts ( XUTs ) ( C ) and stable unannotated transcripts ( SUTs ) ( D ) . The dashed vertical lines mark a twofold increase in TSSC read counts in the various mutants relative to the wild-type . The number of identified TSSCs and of features to which they were assigned is indicated for each transcript class . Note that CUTs , XUTs and SUTs constituting overlapping transcript populations , when a TSSC was assigned to a pervasive transcript annotated in more than one of these classes , we arbitrarily associated the corresponding TSSC in priority to CUTs , then to XUTs and finally to SUTs . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 006 The accumulation of XUTs in the absence of Upf1 suggests that these transcripts , originally described as Xrn1-sensitive ( van Dijk et al . , 2011 ) , may be primarily targeted for degradation by NMD . We tested this hypothesis by performing northern blot hybridization for three XUTs ( Figure 3A ) and genome-wide transcriptome analyses of wild-type , upf1∆ , xrn1∆ , and upf1∆xrn1∆ cells ( Figure 3B–D ) . Since the absence of Xrn1 leads to the accumulation of decapped RNAs ( Hsu and Stevens , 1993 ) , we could not use the TSS sequencing methodology applied for the other mutants and we thus used , for this particular experiment , a classical RNA sequencing approach for transcript quantification ( see ‘Materials and methods’ ) . Moreover , since the absence of Xrn1 affects the overall cellular mRNA content ( Sun et al . , 2013 ) , an aliquot of a S . pombe culture was added to the cell pellet before RNA extraction to provide an independent internal control for normalization of the results . After having verified that the overall quantifications with ‘RNAseq’ gave results similar to those obtained by TSS sequencing for the wild-type and upf1∆ strains ( Figure 3B and Supplementary file 2 ) , we used the former technique to analyse the genome-wide effect of the XRN1 deletion . Consistent with previous reports ( van Dijk et al . , 2011; Sun et al . , 2013 ) , deletion of XRN1 resulted in a global increase of mRNAs , XUTs and SUTs compared with the wild-type ( Figure 3C , D ) . In contrast to mRNAs , the extent to which XUTs and SUTs were stabilized in xrn1∆ was very similar to the one observed in upf1∆ cells ( Figure 3C , D ) . Furthermore , deleting XRN1 in an upf1∆ background had almost no additional stabilizing effect on XUTs and SUTs , in contrast to mRNAs . This epistatic relationship observed between xrn1∆ and upf1∆ for the stabilization of XUTs and SUTs is consistent with these transcripts being primarily targeted by NMD , and with Xrn1 acting as a downstream effector of this pathway . 10 . 7554/eLife . 06722 . 007Figure 3 . XUTs are primarily targeted for degradation by the nonsense-mediated mRNA decay ( NMD ) . ( A ) Northern blot analysis of three different Xrn1-sensitive transcripts ( XUTs ) in wild-type ( WT ) , upf1∆ , xrn1∆ , and upf1∆xrn1∆ cells . ACT1 was used as a loading control . ( B ) Frequency distribution of the ratios of transcription start site clusters ( TSSCs ) read counts between upf1∆ and wild-type cells obtained with two different methods for library preparation and normalization procedures . ‘TSSCs’ refers to data obtained using the protocol developed to identify TSSs and ‘RNAseq’ to the protocol used to obtain the whole transcriptome ( see ‘Materials and methods’ ) . Only reads corresponding to annotated mRNAs , XUTs and SUTs were included in the analysis ( see Supplementary file 2 ) . ( C ) and ( D ) Frequency distribution of the ratios of read counts for upf1∆ ( red ) , xrn1∆ ( blue ) or upf1∆xrn1∆ ( purple ) compared with wild-type for mRNAs and XUTs and SUTs respectively . The vertical dashed lines mark a twofold increase in read counts . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 00710 . 7554/eLife . 06722 . 008Figure 3—figure supplement 1 . The presence of short open reading frames ( ORFs ) and long 3′-UTRs is a hallmark of natural nonsense-mediated mRNA decay ( NMD ) substrates . ( A ) Comparison of ORF and 3′-UTR sizes for mRNAs ( blue ) , stable unannotated transcripts ( SUTs ) ( red ) , and Xrn1-sensitive transcripts ( XUTs ) ( green ) . The 5′- and 3′-ends were determined for all classes of transcripts using previously published data ( Xu et al . , 2009; van Dijk et al . , 2011; Pelechano et al . , 2013 ) . For SUTs and XUTs , we took into account for the analysis the size of the first potential ORF found after the 5′-end and the distance between the end of the transcript and the stop codon of this potential ORF . The median size of these ORFs was 14 amino acids for XUTs and SUTs , compared with 401 for annotated ORFs . The rightmost and uppermost panel displays the frequency distribution of the different features according to the size of their 3′-UTRs and to the size of their ORF respectively . ( B ) Box plots illustrating the distribution of the size of 3′-UTRs for XUTs and SUTs sensitive ( red ) or not ( blue ) to NMD . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 008 In contradiction to the fact that XUTs and SUTs have been designated as ‘non-coding’ , the strong effect NMD inactivation had on these transcripts indicates that they must , at some point , be translated . Indeed , all XUTs and SUTs carry spurious ORFs ( Figure 3—figure supplement 1A ) and some of them have been identified in ribosomal profiling experiments and shown to encode short peptides ( Ingolia et al . , 2009; Smith et al . , 2014 ) . However , the high sensitivity to NMD of XUTs and SUTs compared with mRNAs suggests the presence of specific features . Assuming that the first encountered ORF is translated , XUTs and SUTs carry , on average , much shorter ORFs and longer 3′-UTRs than mRNAs ( Figure 3—figure supplement 1A ) . The presence of both a long 3′-UTR and a short ORF increases NMD efficiency in budding yeast ( Decourty et al . , 2014 ) and thus explain why these transcripts are efficiently targeted for degradation by this pathway . Furthermore , the few XUTs and SUTs found not to be sensitive to NMD have on average significantly shorter 3′-UTRs ( Figure 3—figure supplement 1B ) , which is in agreement with the recently published observation that , in ribosomal profiling experiments , the length of RNA downstream of the ribosome protected region is significantly longer for NMD-sensitive compared with NMD-insensitive unannotated RNA transcripts ( Smith et al . , 2014 ) . Amongst the 17 , 812 TSSCs identified in this study , 7739 could not be assigned to previously annotated transcripts ( ORFs , stable ncRNAs or pervasive transcripts ) and were only detected in mutants cells ( Figure 4 ) . These newly identified transcripts , expressed at low levels ( Figure 4—figure supplement 1 ) , originated from intergenic regions ( 3011 ‘intergenic-TSSCs’ originating from 1843 intergenic regions ) as well as from within mRNA transcribed regions either in sense ( 2978 ‘B-TSSCs’ within 1889 mRNAs ) or in antisense orientation ( 1750 ‘A-TSSCs’ antisense to 1186 mRNAs; Supplementary file 1 ) . While ‘B-TSSCs’ were sensitive to upf1∆ and almost not affected by the deletion of RRP6 ( Figure 4C and Figure 4—figure supplement 1 ) , ‘intergenic’ and ‘A-TSSCs’ were affected by the two mutations ( Figure 4A , B and Figure 4—figure supplement 1 ) . Furthermore , the absence of both Rrp6 and Upf1 had an additive effect on the accumulation of these last two classes , as is the case for previously identified pervasive transcripts . Combining these two mutations might sometimes even have a synergistic and not only an additive effect , since some of these transcripts were readily detectable only in the upf1∆rrp6∆ double mutant ( e . g . , the transcript found antisense to MAL12/32 in Figure 4E ) . Analysis of individual TSSs signatures for these three classes of previously unannotated transcripts using the Web-LOGO algorithm ( Crooks et al . , 2004 ) yielded a consensus sequence almost indistinguishable from the one obtained for individual TSSs assigned to known mRNAs or previously identified pervasive transcripts ( Figure 4—figure supplement 2 ) , suggesting that the underlying DNA sequence plays an important role in TSS selection by the scanning polymerase . 10 . 7554/eLife . 06722 . 009Figure 4 . Novel transcripts revealed upon deletion of UPF1 and/or RRP6 . ( A–C ) Frequency distribution of the ratios of transcription start site clusters ( TSSCs ) read counts in upf1∆ ( red ) , rrp6∆ ( blue ) or upf1∆rrp6∆ ( purple ) compared with wild-type for transcripts initiating within intergenic regions—‘intergenic TSSCs’ ( A ) , from within an mRNA transcribed region but antisense to the mRNA—A-TSSCs ( B ) or within an mRNA transcribed region , but in the sense orientation with respect to the mRNA—B-TSSCs ( C ) . ( D ) Schematic representation of the various classes of TSSCs described above . The small blue and orange vertical bars represent individual TSSs within TSSCs ( dashed lines ) . ( E ) Northern blot analysis of poly ( A ) + RNA from wild-type , upf1∆ , rrp6∆ and upf1∆rrp6∆ . The category to which the transcripts belong is indicated on the right . ACT1 was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 00910 . 7554/eLife . 06722 . 010Figure 4—figure supplement 1 . Distribution of read counts for different classes of transcripts . Box plots illustrating the distribution of read counts for the different classes of transcription start site clusters ( TSSCs ) identified in this study for wild-type ( WT ) ( grey ) , upf1∆ ( red ) , rrp6∆ ( blue ) , and upf1∆rrp6∆ ( purple ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01010 . 7554/eLife . 06722 . 011Figure 4—figure supplement 2 . Transcription start site ( TSS ) consensus sequences for the different classes of transcripts . Consensus sequences around the TSSs generated using the Web-LOGO algorithm ( Crooks et al . , 2004 ) for the different categories of TSS clusters ( TSSCs ) identified in this study: mRNAs ( A ) , cryptic unstable transcripts ( CUTs ) ( B ) , Xrn1-sensitive transcripts ( XUTs ) ( C ) , stable unannotated transcripts ( SUTs ) ( D ) , 'intergenic' transcripts ( E ) , A intragenic ( F ) and B intragenic ( G ) transcripts . The numbers of reads and TSSs are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 011 Even though transcripts initiated inside ORFs are expected to be targeted for degradation by NMD due to the presence of short spurious ORFs , the high number of internally initiated transcripts ( ‘A and B-TSSCs’ ) identified in cells lacking UPF1 was surprising . The synthesis of such transcripts is normally repressed within transcribed regions unless the proper chromatin structure cannot be re-established in the wake of RNAPII ( Smolle and Workman , 2013 ) . In particular , histone methylation by Set2 was shown to be a key determinant of this repression . We therefore analysed the impact of SET2 deletion on ‘A’ and ‘B’ TSSCs identified in the upf1∆ strain . While deletion of SET2 had little effect on the internal TSSCs identified in the single upf1∆ mutant , it revealed new Set2-sensitive ‘A’ and ‘B’ TSSCs ( Figure 5 and Supplementary file 3 ) . As expected and in contrast to A-TSSCs and B-TSSCs , deletion of SET2 had no global effect on ORF TSSCs and only a marginal effect on CUTs , XUTs and SUTs or ‘intergenic’ TSSCs ( Figure 5—figure supplement 1 ) . In agreement with the observed enrichment in Set2-catalyzed H3K36 methylation towards the 3′ end of ORFs ( Pokholok et al . , 2005 ) , the ‘A’ and ‘B’ TSSCs more sensitive to Set2 were located further away from the mRNA 5′-ends ( Figure 6A–C ) . Analysis of the sequence surrounding individual TSSs from these TSSCs identified a pattern almost identical to the one found for ORF-TSSs ( Figure 6—figure supplement 1 ) , confirming that they corresponded to bona fide transcription initiation events . However , unlike the TSSs associated with other features identified in this study , the ‘B’ TSSs were not associated with a strong NFR ( Figure 6—figure supplement 2 ) . Instead , analysing the nucleosome density around these TSSs revealed the presence of a weak NFR surrounded by two well-positioned nucleosomes . This particular pattern might explain the high sensitivity of ‘B’ TSSs to mutations affecting chromatin structure , such as the deletion of the Set2 histone methyl-transferase ( Smolle and Workman , 2013 ) . 10 . 7554/eLife . 06722 . 012Figure 5 . Effect of the absence of SET2 on transcription start sites ( TSSs ) identified inside open reading frames ( ORFs ) . ( A ) Frequency distribution of the ratios of TSS clusters ( TSSCs ) read counts in upf1∆set2∆ compared with upf1∆ cells for the intragenic A-TSSCs . The comparison was performed for all the TSSCs ( blue line ) and for the ones identified in a single upf1∆ mutant ( red line ) . ( B ) As in ( A ) but for B-TSSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01210 . 7554/eLife . 06722 . 013Figure 5—figure supplement 1 . Deletion of SET2 specifically increases the expression level of intragenic transcription start site clusters ( TSSCs ) . ( A–G ) Frequency distribution of the ratios of TSSC read counts in upf1∆ ( red ) , set2∆ ( green ) or upf1∆ set2∆ ( orange ) compared with wild-type for mRNAs ( A ) , cryptic unstable transcripts ( CUTs ) ( B ) , Xrn1-sensitive transcripts ( XUTs ) ( C ) , stable unannotated transcripts ( SUTs ) ( D ) , ‘intergenic’ ( E ) , ‘A-’ ( F ) and B-TSSCs ( G ) . The dashed vertical lines mark a twofold ( log2 = 1 ) increase in TSS counts in the mutants relative to the wild-type . The numbers of identified TSSCs and of features to which they were assigned is indicated for each class . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01310 . 7554/eLife . 06722 . 014Figure 6 . Transcription start site ( TSSs ) identified inside open reading frames ( ORFs ) show a differential sensitivity to the absence of SET2 according to their position along the mRNA . ( A ) Visualization of the TSS reads at the YIL136w and YMR114c loci in the wild-type ( WT; black ) , upf1∆ ( red ) , set2∆ ( green ) , and upf1∆set2∆ ( orange ) cells . The blue arrows represent ORFs and the horizontal red bar the position of the probes used for the Northern blots displayed on the right . Arrowheads in the right panel indicate the position of the full-length and internally initiated ( B1 , B2 , and B ) transcripts . ( B ) and ( C ) Frequency distribution of A-TSSCs and B-TSSCs read counts respectively in upf1∆set2∆ vs upf1∆ according to the distance from their associated mRNA TSS . Blue and red lines are for TSSCs sensitive and insensitive to the deletion of SET2 respectively . ( D ) Frequency distribution of read counts for the A- ( blue ) and B- ( red ) TSSCs in upf1∆set2∆ compared to set2∆ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01410 . 7554/eLife . 06722 . 015Figure 6—figure supplement 1 . Consensus sequences around the transcription start sites ( TSSs ) for ( A ) A-TSSCs and ( B ) B-TSSCs identified in the upf1∆set2∆ mutant generated using the Web-LOGO algorithm ( Crooks et al . , 2004 ) . The numbers of reads and TSSs are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01510 . 7554/eLife . 06722 . 016Figure 6—figure supplement 2 . Transcription start sites ( TSSs ) are associated within nucleosome-free regions . Average nucleosome density retrieved from Kaplan et al . ( 2009 ) plotted as a function of the distance of individual TSSs for TSSs belonging to the different categories of TSS clusters ( TSSCs ) identified in this study: mRNAs ( A ) , cryptic unstable transcripts ( CUTs ) ( B ) , Xrn1-sensitive transcripts ( XUTs ) ( C ) , stable unannotated transcripts ( SUTs ) ( D ) , 'intergenic' transcripts ( E ) , A intragenic ( F ) and B intragenic ( G ) transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 016 Importantly , 55% of the A-TSSCs and 40% of the B-TSSCs , whether repressed by Set2 or not , were sensitive to Upf1 ( i . e . , increased significantly in upf1∆set2∆ compared with set2∆ cells; Figure 6D and Supplementary file 3 ) , revealing NMD to be an important quality-control mechanism to eliminate internally initiated transcripts . In contrast to their strong impact on the steady state levels of a variety of pervasive transcripts , including previously unannotated ones , inactivation of nuclear and/or cytoplasmic RNA quality-control pathways had a relatively minor global effect on the steady state levels of coding transcripts ( Figure 2 ) . Consistent with previously reported data ( He et al . , 2003; Wyers et al . , 2005 ) , the effect on mRNAs was larger upon deletion of UPF1 ( with 17% of the mRNAs-TSSCs showing a significant increase in the absence of Upf1 ) than upon deletion of RRP6 and deletion of both genes simultaneously did not show any additive effect ( Figure 2A ) . Yet , due to the mRNA 5′-end heterogeneity , analysing individual TSSs gave a different picture . Indeed , transcript isoforms in which TSSs where followed by uORFs were , globally , stabilized in the absence of UPF1 , while those in which the annotated start codon was the first ATG downstream the TSS were mostly unaffected ( Figure 7A , solid lines; Figure 7B ) . A large fraction of these NMD-sensitive transcripts corresponded to minor isoforms , explaining why NMD had only a weak effect on the overall mRNA-TSSC levels ( Figure 2A ) . Yet , for 1129 out of the 5231 active genes ( 22% ) , a fraction of their transcript isoforms carried at least one uORF and was significantly sensitive to NMD . Isoforms carrying more than one uORF appeared even more sensitive to NMD ( Figure 7A ) , suggesting that the first AUGs of uORFs containing transcripts are not always efficiently used for translation initiation , likely because they are not in a favorable context ( Arribere and Gilbert , 2013 ) . 10 . 7554/eLife . 06722 . 017Figure 7 . Deletion of UPF1 reveals numerous minor mRNA-associated transcription start sites ( TSSs ) . ( A ) Genes were aligned by their start codon and the log2 of ratios of TSS reads for upf1∆ vs wild-type was plotted for TSSs upstream or downstream ( iTSSs ) the annotated ATG start codons , as depicted in the right panel . The main open reading frames ( ORFs ) are represented by large blue arrows and upstream ORFs ( uORFs ) or small internal out-of-frame ORFs by small orange arrows . The thin blue arrows indicate the TSSs . ( B ) The curve represents the probability at each nucleotide position that the distributions of reads corresponding to the red and blue curves shown in A are the same . The dashed red line marks the 0 . 05 p-value . ( C ) Genes were aligned by their annotated start codons ( A of the ATG at position +1 ) and the cumulative TSS read counts per nucleotide ( smoothed over 11 nucleotides ) was plotted for the wild-type ( black ) and upf1∆ ( red ) cells . Inset: Magnification of the +1 to +50 region . The p-value < 2 . 2 × 10−16 is the probability ( ANOVA test ) that the distributions of the values , per nucleotide , for the red and black curves are the same within the +1 to +50 nucleotides interval . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01710 . 7554/eLife . 06722 . 018Figure 7—figure supplement 1 . Initiation of iTSS at purines reflects a bias in codon composition . ( A ) As for Figure 7C but without smooting . ( B ) Genes were aligned by their annotated start codons and the total number of transcription start sites ( TSSs ) per nucleotide was plotted for the wild-type ( black ) and upf1∆ ( red ) cells . ( C ) Zoom of panel ( B ) for the region −20 to +50 nucleotides around the annotated start codons ( upper panel ) . Note that A and G of the start codon are frequently used for transcription initiation and iTSSs display periodicity of three nucleotides . This periodicity can be explained by the fact that RNA polymerase II transcription initiates preferentially at purines ( R ) ( Zhang and Dietrich , 2005 ) , which are enriched at the first position of codons as previously reported ( Mackiewicz et al . , 1999 ) and depicted in the middle and bottom panels . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 01810 . 7554/eLife . 06722 . 019Figure 7—figure supplement 2 . ( iTSSs ) consensus sequences . Transcription start sites downstream of annotated ATGs ( iConsensus sequences around the iTSSs generated using the Web-LOGO algorithm ( Crooks et al . , 2004 ) . To avoid a potential bias due to the presence of the ATG start codon in all genes , only iTSSs mapping at least four nucleotides from the start codon were used for this analysis . The numbers of reads and TSSs are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 019 Unexpectedly , another important category of transcript isoforms found to be sensitive to NMD corresponded to TSSs mapping downstream to the annotated ORF ATG ( here called iTSSs; Supplementary file 4 ) . Only observed after TAP treatment ( see insert in Figure 1B ) , these iTSSs must correspond to genuine capped-RNAs . We considered genes as having iTSSs ( 3327 genes; 64% of the expressed genes ) if they contained at least four iTSS reads , which accounted for 99% of all iTTS reads . A fraction of transcript isoforms initiated at iTSSs followed by an out-of-frame ATG and significantly stabilized in the absence of Upf1 could be identified for 1821 out of the 5231 expressed genes ( 35% , Figure 7A , B ) . In contrast to the weak overall effects observed for TSSs located directly upstream of the start codon , the mean expression level of iTSSs significantly increased in the upf1∆ strain relative to the wild-type ( Figure 7C , and Figure 7—figure supplement 1 ) . Note that RNAPII transcription initiating preferentially at purines , the A and G of the annotated start codons are quite frequently used as sites of transcription initiation and the first nucleotide of codons are preferential sites of internal initiation because these positions are enriched in purines ( Figure 7—figure supplement 1B , C; Mackiewicz et al . , 1999 ) . The consensus sequence for the iTSSs was not different from the general consensus sequence for RNAPII transcription initiation ( Figure 7—figure supplement 2 ) . Altogether , transcript isoforms significantly stabilized in NMD-deficient cells , either because they carry a uORF or because they initiated at an out-of-frame iTSS , were found in 2437 genes; that is , 47% of the 5231 expressed genes . In transcripts initiated at iTSSs , the first encountered ATGs are most of the time out-of-frame relative to the main ORF ( Figure 8A ) . The first ORFs of the corresponding transcripts are thus short and followed by a long 3′-UTR , making them excellent NMD substrates . This bias towards out-of-frame ATGs was specifically observed at the beginning of genes for which iTSSs were identified ( Figure 8B , C ) . This correlated with a significantly lower frequency of methionine in the N-terminal part of the corresponding proteins ( Figure 8D ) , even though methionines were found to be globally underrepresented at the beginning of all yeast proteins . The frequent use of iTSSs by RNAPII thus does not generally result in the production of truncated proteins and tends to generate transcripts that are sensitive to NMD . 10 . 7554/eLife . 06722 . 020Figure 8 . Distribution of in-frame vs out-of-frame ATG codons and methionines for genes with or without transcription start sites downstream of annotated ATGs ( iTSSs ) . ( A ) Cumulative total number of transcription start sites ( TSSs ) per nucleotide in upf1∆ cells for TSSs directly followed by the annotated ATGs ( blue line ) or a upstream open reading frame ( uORF ) ( red line ) and for TSSs downstream the annotated ATGs ( iTSSs ) and followed by an out-of-frame ATG ( red dashed line ) or by an in-frame ATG ( blue dashed line ) . ( B ) Proportion of in-phase ATGs ( +1 frame ) following an annotated start codon , binned over nine nucleotides , for protein-coding genes with ( red; 3327 genes ) or without ( blue; 1904 genes ) iTSS reads in upf1∆ , as defined in the text . The dashed line indicates the expected value for a random distribution . Genes with iTSSs reads are significantly depleted of in-frame ATGs relative to all ATGs in the first 100 nucleotides when compared with genes without iTSSs reads ( p-value = 3 . 13 10−5; ANOVA ) . ( C ) Proportion ( log2 ) along the ORFs of in-frame codons , normalized over the regions downstream the first 300 nucleotides of genes , for all codons ( binned over nine nucleotides ) for the two sets of genes defined in B . The ATG codon is in red . ( D ) Frequency of methionines per amino acid position along the ORFs for the two sets of genes defined in ( B ) . The p-values ( ANOVA method ) for the difference in methionine composition over the regions between 0–100 and 100–1000 nucleotides are indicated on the figure . DOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 020 We describe here the use of a modified 5′-RACE technique to identify the 5′-ends of capped RNA , which map RNAPII TSSs . The method , which includes a streptavidin-biotin purification step , is highly specific and can be applied to any eukaryotic organism . Our results provide the most comprehensive genome-wide identification of TSSs in budding yeast , not only for mRNAs but also for a wide range of pervasive transcripts , including previously non-annotated ones . This was made possible by combining this highly specific TSS determination technique with the use of mutants affecting both the nuclear exosome and the cytoplasmic NMD pathway . Detailed analyses of the sequencing reads and the corresponding upstream genomic sequences extended and validated the previously identified sequence consensus surrounding TSSs in S . cerevisiae ( A ( N ) 6PyPu; Figure 1—figure supplement 2A and Figure 4—figure supplement 2 , and Zhang and Dietrich 2005 ) . Importantly , this consensus sequence was observed for the TSSs of all the different classes of transcripts identified in this study . Note that 1 , 775 , 474 sites in the genome conform to this consensus but only 85 , 714 ( 4 . 8% ) were used as TSSs . We also identified cases in which transcription started on what appears to be a ‘slippery’ sequence ( PyAAA ) , which resulted in the incorporation of an additional non-encoded A at the 5′-end of the transcripts . The simplest explanation for this observation is that the very short nascent RNA formed after two or three nucleotide incorporation is able to shift back by one nucleotide , probably together with the RNAPII , before transcription resumes ( Figure 1—figure supplement 2D ) , suggesting that the ternary complex formed by the transcribing polymerase , the RNA moiety and chromatin is rather labile at this early stage of transcription . Our analyses of TSSs in NMD-deficient cells revealed the prevalent role of this cytoplasmic RNA quality-control pathway in eliminating pervasive transcripts . We found that 52% of the SUTs and XUTs were significantly stabilized in upf1∆ cells when using the TSSCs from the TSS-sequencing experiments ( see Supplementary file 1 ) . When using the RNAseq approach , this proportion was even higher ( 70%; see Supplementary file 2 ) , possibly because of the overall higher sequence read counts for a given transcript , which provides more power to statistical analyses . Since NMD is a translation-coupled RNA degradation pathway , the stabilizing effect seen upon UPF1 deletion on pervasive transcripts indicated that once these transcripts reached the cytoplasm they associate with the translation machinery . Indeed , once they reached the cytoplasm the pervasive transcripts ( being capped and poly-adenylated ) cannot be distinguished from normal mRNAs and associate with the translation machinery . However , since they usually have short ORFs followed by relatively long 3′ UTR regions ( Figure 3—figure supplement 1 ) , a hallmark of NMD substrates in yeast ( Decourty et al . , 2014 ) , they will be targeted for degradation by this RNA quality-control pathway . This is supported by the identification of pervasive transcripts ( mainly XUTs and SUTs ) in ribosomal profiling experiments performed in wild-type cells and the recent discovery of a new class of unannotated transcripts ( uRNAs ) associated with polyribosomes and encoding short peptides ( Ingolia et al . , 2009; Smith et al . , 2014 ) . The association of so-called ‘non-coding’ transcripts with the translation machinery and their ensuing high sensitivity to NMD has also been observed in multicellular organisms . In mouse embryonic stem cells , many long non-coding RNAs ( lncRNAs ) were found to be associated with translating ribosomes and an estimated 17 . 4% of lncRNAs were found to be upregulated in the absence of UPF1 , compared with only 4% for protein-coding genes ( Ingolia et al . , 2014; Smith et al . , 2014 ) . Our results also showed that XUTs , originally described as Xrn1-sensitive transcripts ( van Dijk et al . , 2011 ) , are in fact primarily targeted by NMD by virtue of their poor coding capacities , with the Xrn1 exonuclease acting as a downstream effector of this pathway . Thus , we propose to redefine the various classes of pervasive transcripts identified in yeast into transcripts primarily degraded in the cytoplasm ( SUTs and XUTs ) and transcripts more sensitive to the nuclear exosome ( CUTs ) even though each of these degradation pathways similarly affect some transcripts and can show a synergic effect ( see below ) . Our analysis also revealed the prevalent role of NMD in the removal of cryptic transcripts initiating from intragenic promoters either in sense ( here called B ) or in antisense orientation ( here called A ) . The strong impact of NMD inactivation on these transcripts indicated that they are exported to the cytoplasm and associate with the translation machinery , as is the case for the pervasive transcripts ( see above ) . Even though we cannot rule out an indirect effect of the UPF1 deletion on chromatin structure , the high number of intragenic transcripts ( whether sense or antisense ) identified in the upf1∆ single mutant suggests that the inhibition of transcription initiation by transcription-coupled chromatin modifications is not fully efficient and thus that some of these transcripts are produced as part of the normal transcription cycle in wild-type cells , with NMD playing an important role to eliminate them . Deletions of UPF1 and RRP6 had an additive effect on the accumulation of CUTs , SUTs and XUTs . However , a number of pervasive transcripts , in particular ‘A’ and ‘intergenic’ ones , were stabilized to substantial levels only when both the nuclear and the cytoplasmic RNA quality-control pathways were compromised , suggesting that they can act synergistically . Two mechanisms of transcription termination have been reported in yeast . One depends on the cleavage and poly-adenylation complex ( CPF-CFI/II ) , which generates the poly-adenylated mRNAs that get exported and translated in the cytoplasm , and another , which involves the NNS complex and is shared by CUTs and sn ( o ) RNAs precursors ( see ‘Introduction’ ) . However , the demarcation between the two modes of termination is far from being strict as some terminators can often be recognized by both pathways depending on their distance from the TSS ( Porrua et al . , 2012 ) . In the early phase of transcription elongation the RNAPII CTD repeats are mainly phosphorylated at Ser5 , which favor recruitment of the NNS complex; while transcription proceeds , Ser2 gets phosphorylated at the expense of Ser5 , promoting the recruitment of the CPF-CFI/II complex ( Ahn et al . , 2004; Kim et al . , 2004 ) . For some pervasive transcripts , these two termination pathways may compete and generate transcripts terminated by the NNS pathways and degraded by the nuclear exosome , as well as transcripts terminated by the CPF-CFI/II pathway and exported to the cytoplasm where they are targeted for degradation by the NMD . The produced RNAs would thus accumulate to detectable levels only when both RNA quality-control pathways are inactivated . Since the nuclear exosome has recently been shown to act with the NNS complex to promote early transcription termination at specific targets in S . cerevisiae ( Fox et al . , 2015 ) , we cannot rule out that in wild-type or upf1∆ cells transcription will normally be terminated by the NNS complex and the synthesized transcripts degraded by the nuclear exosome , while in rrp6∆ cells transcription will proceed until the polymerase encounters the next CPF-CFI/II termination signal giving rise to longer transcripts exported to the cytoplasm and targeted for degradation by the NMD . NMD was previously shown to target a few hundred uORF-containing mRNAs ( He et al . , 2003; Guan et al . , 2006; Johansson et al . , 2007; Arribere and Gilbert , 2013 ) . However , the use of NMD-deficient cells allowed us to identify 22% of the active genes for which a fraction of their transcript isoforms carried at least one uORF and were significantly sensitive to NMD . Moreover , 35% of the expressed genes generated transcript isoforms initiating at iTSSs and significantly stabilized in the absence of Upf1 ( see Figure 7A , B ) . Some transcripts initiated at iTSSs were previously described in wild-type cells and shown to allow the synthesis of N-terminal variants of proteins exhibiting differential stabilities or localizations ( Wu and Tzagoloff , 1987; Gammie et al . , 1999; Arribere and Gilbert , 2013; Pelechano et al . , 2013 ) but their number was vastly underestimated , probably because of their sensitivity to NMD ( see Figure 7C ) . These two observations reveal an important role for NMD in getting rid of numerous undesired transcripts , the majority of which likely resulted from the low specificity of TSS selection by RNAPII . Yet , we cannot rule out that some of these transcripts have a biological function . For example , the use of alternative TSSs giving rise to transcripts with very different sensitivity to NMD could be used for regulatory purposes , in a way similar to that described for some genes of the nucleotide biosynthetic pathway ( Kuehner and Brow , 2008; Thiebaut et al . , 2008 ) . Altogether , transcript isoforms carrying a uORF or starting at an iTSS and targeted by NMD constituted the major transcript isoforms for only 446 mRNAs and minor transcript isoforms for almost half of the expressed genes . Therefore , although not quantitatively impacting the overall mRNA levels substantially ( Figures 2A , 7C ) , NMD affects qualitatively a large fraction of the mRNA transcription units . NMD also has a strong impact on the accumulation of pervasive and cryptic transcripts and thus appears as a major player shaping the yeast transcriptome . In addition , we observed that transcripts initiated at iTSSs are significantly depleted in ATGs in the +1 frame , precluding the synthesis of N-terminally truncated proteins and ensuring their efficient NMD degradation . It thus suggests that this phenomenon might be important enough to have imposed an evolutionary constraint on the methionine content of the N-terminus of the yeast proteins . All the strains are a derivative of BY4741 and were obtained directly from the Euroscarf deletion collection ( http://web . uni-frankfurt . de/fb15/mikro/euroscarf/ ) or generated by crossing with a can1∆ derivative of BY4741 ( LMA1057 , see Table 2 ) . Cells were grown to mid-exponential phase in YPD-rich medium at 30°C in a microturbidostat as previously described ( Decourty et al . , 2008 ) , harvested by centrifugation and the pellet was frozen in liquid nitrogen . 10 . 7554/eLife . 06722 . 021Table 2 . Yeast strains used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 021StrainGenotypeReferenceBY4741Mat a , his3∆1 , ura3∆0 , leu2∆0 , met15∆0 ( Brachmann et al . , 1998 ) LMA1057/3401BY4741 can1∆This studyLMA1774/2759BY4741 can1∆ , upf1∆::HIS3MX6This studyLMA1676/3405BY4741 can1∆ , rrp6∆::hphMX6This studyLMA1772BY4741 can1∆ , upf1∆::KANMX6 , rrp6∆::HPHMX6This studyLMA1790BY4741 can1∆ , upf1∆::KANMX6This studyLMA2758BY4741 can1∆This studyLMA2760BY4741 can1∆ , xrn1∆::KANMX6This studyLMA2762BY4741 can1∆ , xrn1∆::KANMX6 , upf1∆::HIS3MX6This studyLMA2921/3403BY4741 can1∆ , set2∆::KANMX6This studyLMA2922BY4741 can1∆ , set2∆::KANMX6 , upf1∆::HIS3MX6This studyLMA3409BY4741 can1∆ , upf1∆::HIS3MX6 , rrp6∆::HPHMX6This study Total RNA was extracted using the hot acid phenol protocol ( Collart and Oliviero , 2001 ) . Poly ( A ) +-RNA were obtained by two successive rounds of purification using oligo ( dT ) 25 magnetic beads ( New England Biolabs , Ipswich , MA ) following the manufacturer's protocol . Northern blots were carried out on poly ( A ) +-RNA as described in Neil et al . ( 2009 ) using 32P-labeled riboprobes except for ACT1 for which a 32P-labeled oligonucleotide was used . Approximately 500 ng of poly ( A ) +-RNA was mixed with 10 units of Antarctic phosphatase ( New England Biolabs ) in a final volume of 50 μl . After 1 hr at 37°C , the reaction was treated with phenol/chloroform and ethanol precipitated . The RNA pellet was resuspended in 44 μl of water and 10 units ( 1 μl ) of TAP ( Epicentre , Madison , WI ) and 5 μl of 10× TAP buffer was added . The reaction was incubated 1 hr at 37°C followed by phenol/chloroform extraction and ethanol precipitation . The RNA pellet was resuspended in 5 μl of water . Next , the RNAs were ligated overnight at 16°C with 50 pmoles of the biotinylated oligonucleotide 3041 ( Table 3 ) in a 20 μl reaction containing 10 units ( 1 μl ) of T4 RNA ligase I ( New England Biolabs ) and ATP at a final concentration of 1 mM . RNAs were subsequently fragmented by incubation for 10 min at 70°C after addition of 5 μl of a 50 mM ZnCl2 , 50 mM Tris-HCl pH7 . 4 solution . The reaction was stopped by the addition of 1 μl EDTA 0 . 5 M and biotinylated RNA were purified using streptavidin magnetic beads according to the manufacturer's protocol ( Dynabeads , MyOne streptavidin C1 , Life Technologies , Carlsbad , CA ) . After washing , the beads were resuspended in 20 μl of water and the bound RNAs eluted by incubation for 5 min at 90°C . This fraction is enriched for 5′-ends of capped RNA molecules . Note that the supernatant of the first step of the purification procedure containing RNA fragments corresponding to the body and the 3′-end of the gene not attached to the biotinylated oligonucleotide can be recovered and used to prepare independent libraries . This RNA population can be further fractionated using oligo ( dT ) 25 magnetic beads to enrich for 3′-end of RNA molecules . The ∼18 . 5 μl eluate from the streptavidin beads was mixed with 50 pmoles of oligonucleotide 3038 ( see Table 3 ) , heat denatured for 5 min at 70°C and slowly cooled down to 30°C in a Biorad iCycler PCR machine . Once the temperature had reached 30°C , 6 μl of 5× RT buffer , 1 . 5 μl of a 10 mM dNTPs solution , 1 . 5 μl of RNasin ( Promega ) , 180 ng of actinomycin D , 300 units of RevertAid reverse transcriptase ( Thermo Scientific , Waltham , MA ) , and water qsp 30 μl were added . The reaction was incubated 10 min at 30°C , followed by 40 min at 42°C , 10 min at 55°C , 10 min at 60°C and 15 min at 75°C . RNAs were then degraded by incubation for 10 min at 75°C after the addition of 3 μl of 1 N NaOH . The reaction was quenched by the addition of 3 μl of 1 N HCl and precipitated by the addition of 3 volumes of 100% ethanol and 0 . 1 volume of 3 M sodium acetate pH 5 . 2 . The precipitated cDNA was subjected to 5 cycles of PCR amplification in 20 μl with Phusion DNA polymerase ( Thermo Scientific ) using Illumina ( San Diego , CA ) multiplexing PCR primer 1 . 0 and 2 . 0 followed by an additional 6 to 8 cycles of amplification using Illumina multiplexing PCR primer 1 . 0 and one PCR primer index . The reaction was purified with Agencourt AmPure XP beads ( Beckman Coulter , Indianapolis , IN ) following manufacturer's instructions at a 1 . 8 × concentration and eluted in 20 μl water . Libraries were quantified using Qubit ( Life Technologies ) and sequenced on an Illumina HiSeq 2000 or 2500 with 50 or 100 bases single-end reads . To validate our method for the identification of TSSs , we compared libraries made with or without treatment of the RNAs with TAP . 500 ng of poly ( A ) +-RNA obtained by purification with oligo ( dT ) 25 magnetic beads starting from S . cerevisiae or S . pombe ( used as an internal reference for normalization ) were dephosphorylated using Antarctic phosphatase as described . Then , S . pombe RNAs and one half of the RNAs extracted from S . cerevisiae were treated with TAP . The second half of S . cerevisiae RNAs was mock treated . Before ligation with the biotinylated oligonucleotide 3041 , each half of S . cerevisiae RNAs was mixed with an equivalent quantity of calf intestinal alkaline phosphatase ( CIP ) /TAP-treated RNAs from S . pombe . The subsequent steps of the library preparation were identical to the ones described above . 10 . 7554/eLife . 06722 . 022Table 3 . Oligonucleotides used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06722 . 022NameSequence 5′-3′ACT1-1407−ACACTTGTGGTGAACGATAGATGGP32 labelled probeYMR114c-839+ATCGAGGTGTAAAGGGTGSynthesis of probeT7-YMR114C-1068−TAATACGACTCACTATAGGGCCTCTGGAGTCTTTCTGGSynthesis of probeYIL136w-1013+ACTGGTGGTCTGGATGGSynthesis of probeT7-YIL136w ( + ) 115−TAATACGACTCACTATAGGGTGCCACTAATTTACTCCGSynthesis of probeNEL025c-35+AACAAATGCCAAGTCGGGACSynthesis of probeT7-NEL025c-263−TAATACGACTCACTATAGGGAAACGTTTGGTAAGAACTCSynthesis of probeSUT093_fwdGAGTCCAGCGTCTCTACACSynthesis of probeT7-SUT093_revTAATACGACTCACTATAGGGGACTTAATTGTCGTTGCTAGGACSynthesis of probeSUT338_fwdGAAAGACCGAAGGTGAAGAGSynthesis of probeT7-SUT338_revTAATACGACTCACTATAGGGGTGGTACAGCCCTGTGTTCCSynthesis of probeSUT779_fwdAACGAGGGAACTAGCCAGSynthesis of probeT7-SUT779_revTAATACGACTCACTATAGGGCTCTTCATCATCTGTGGAGSynthesis of probeTPO2 ( + ) 131−GTATGTAGAAATGTCCGACGSynthesis of probeT7-TPO2-1798+TAATACGACTCACTATAGGGGTAAGGGCTTGAGACSynthesis of probeMAL12/32-1723−GATTCTACCTTCCCATGGSynthesis of probeT7-MAL12/32-1161+TAATACGACTCACTATAGGGTCAAGGTCAGGAGATAGGSynthesis of probeXUT3F5-fwdAGGAAAATGGGACTACAGSynthesis of probeT7-XUT3F5-revTAATACGACTCACTATAGGGTGTAAAAGGGCACAGTCSynthesis of probe3041†5BioTEG/CTTTCCCTACACGACGCTCTTCCGATCTNNNNCGCGrCrGrNrNLigation with TAP treated RNA3118†5BioTEG/CTTTCCCTACACGACGCTCTTCCGATCTNNNNGCCGrCrGrNrNLigation with fragmented RNA3038†GTTCAGACGTGTGCTCTTCCGATCTNNNNNNReverse transcriptionThe sequence in bold face corresponds to the T7 promoter sequence . †The sequence in bold face corresponds to the tag used to identify the 5′ end of the cDNAs . r stands for ribonucleotide . For analysis of the xrn1∆ mutant compared with the upf1∆ mutant , since the absence of Xrn1 leads to the accumulation of decapped RNAs ( Hsu and Stevens , 1993 ) , we had to modify the protocol used for library preparation . Instead of the CIP/TAP treatment before ligation with the biotinylated oligonucleotide 3118 , RNA was fragmented with ZnCl2 and subsequently phosphorylated with T4 polynucleotide kinase ( Thermo scientific ) . The subsequent steps of library preparation were identical to the ones described above for the 5′-end . Moreover , since the absence of Xrn1 affects the overall mRNA content ( Sun et al . , 2013 ) , an aliquot of a S . pombe culture was added to the cell pellet before RNA extraction for library preparation to provide an independent internal control for normalization .	Eukaryotes such as animals , plants and fungi store their DNA within the nucleus of each of their cells . Genes within this DNA contain the instructions needed to make molecules of RNA; some of which can leave the nucleus and be decoded to build proteins . However , not all of the DNA that is copied into RNA actually codes for proteins . Instead , some RNA molecules are important parts of the cell's protein-making machinery in their own right , and others help to regulate the expression of genes as RNAs or proteins . Nevertheless , many non-coding RNAs don't have such clear roles . Often these RNAs—which are called ‘pervasive transcripts’—are quickly destroyed within the nucleus , but it is likely that some molecules will escape this quality-control mechanism . If the cell's protein-making machinery decodes these RNAs , it could lead to the production of faulty or harmful proteins . Recent research suggested that another quality-control mechanism , which typically eradicates incorrectly processed protein-coding RNAs , could also destroy unneeded or harmful pervasive transcripts . But it was not clear how common it was for this process—called ‘nonsense-mediated decay’—to be used for this purpose . Now Malabat , Feuerbach et al . have engineered yeast cells that lacked either the genes required to carry out nonsense-mediated decay or the ability to destroy RNA molecules in the nucleus . Experiments with these yeast cells revealed that about half of all pervasive transcripts can be destroyed via nonsense-mediated decay; this suggests that this mechanism serves as a fail-safe to prevent the build-up of these potentially harmful molecules . Malabat , Feuerbach et al . also revealed that the enzyme complex that copies gene sequences to make RNA molecules will often also copy some extra DNA sequence from before the start of the gene . On the other hand , it is also common for this enzyme complex to miss the start of the gene and produce an RNA molecule that lacks some of the instructions needed to build the correct protein . Further experiments showed that in yeast these two kinds of incorrectly made protein-coding RNAs could both be identified and destroyed by nonsense-mediated decay as well . The next challenge will be to see to what extent these phenomena are conserved in other eukaryotes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2015	Quality control of transcription start site selection by nonsense-mediated-mRNA decay
The spindle position checkpoint ( SPOC ) is a spindle pole body ( SPB , equivalent of mammalian centrosome ) associated surveillance mechanism that halts mitotic exit upon spindle mis-orientation . Here , we monitored the interaction between SPB proteins and the SPOC component Bfa1 by FRET microscopy . We show that Bfa1 binds to the scaffold-protein Nud1 and the γ-tubulin receptor Spc72 . Spindle misalignment specifically disrupts Bfa1-Spc72 interaction by a mechanism that requires the 14-3-3-family protein Bmh1 and the MARK/PAR-kinase Kin4 . Dissociation of Bfa1 from Spc72 prevents the inhibitory phosphorylation of Bfa1 by the polo-like kinase Cdc5 . We propose Spc72 as a regulatory hub that coordinates the activity of Kin4 and Cdc5 towards Bfa1 . In addition , analysis of spc72∆ cells shows that a mitotic-exit-promoting dominant signal , which is triggered upon elongation of the spindle into the bud , overrides the SPOC . Our data reinforce the importance of daughter-cell-associated factors and centrosome-based regulations in mitotic exit and SPOC control . Alongside their canonical role as microtubule organizing centers , centrosomes of metazoans or spindle pole bodies ( SPBs , the functional equivalent of the centrosome ) of fungi modulate eukaryotic cell division by serving as signaling centers ( Arquint et al . , 2014; Fu et al . , 2015 ) . In budding yeast , the SPB is associated with components of two linked pathways: the mitotic exit network ( MEN ) and the spindle position checkpoint ( SPOC ) . The MEN drives mitotic exit ( transition from M-G1 phase ) after extension of the anaphase spindle into the daughter cell body . The SPOC is a surveillance mechanism that monitors orientation of the mitotic spindle . The SPOC prevents M-G1 transition when the spindle fails to align along the mother to daughter axis and so is unable to deliver one nucleus into the daughter cell ( Bloecher et al . , 2000; Muhua et al . , 1998; Pereira et al . , 2000; Wang et al . , 2000; Yeh et al . , 1995 ) . SPOC activation inhibits the MEN . SPBs are embedded in the nuclear envelope throughout the cell cycle and are composed of three distinct layers ( inner , central and outer plaques ) that are named according to their position with respect to the nuclear envelope ( Jaspersen and Winey , 2004 ) . SPOC and MEN proteins associate with the cytoplasmic side ( outer plaque ) of the SPBs , where cytoplasmic microtubules are nucleated . Within this outer plaque , the transforming acidic coiled coil ( TACC ) family protein Spc72 ( the yeast homolog of CDK5RAP2 ) anchors the γ-tubulin small complexes that nucleate cytoplasmic microtubules ( Knop and Schiebel , 1998; Lin et al . , 2015; Usui et al . , 2003 ) . Nud1 ( the yeast homolog of Centriolin ) links Spc72 to the central layer of the SPB through Cnm67 - another SPB core protein ( Brachat et al . , 1998; Elliott et al . , 1999 ) . Nud1 also serves as a scaffold for the recruitment and activation of MEN kinases ( Gruneberg et al . , 2000; Rock and Amon , 2011; Rock et al . , 2013; Valerio-Santiago and Monje-Casas , 2011 ) . The key element of SPOC is a GTPase activating protein ( GAP ) complex composed of the GAP Bub2 and its binding partner Bfa1 ( Bardin et al . , 2000; Pereira et al . , 2000 ) . The GAP-complex inhibits the GTPase Tem1 , whose activation would otherwise trigger the MEN ( Bardin et al . , 2000; Geymonat et al . , 2002; 2003; Hu et al . , 2001; Pereira et al . , 2000; Scarfone and Piatti , 2015; Wang et al . , 2000 ) . This cell cycle arrest provides extra time to enable the cell to correctly align its spindle along the mother-to-daughter cell axis . Loss of SPOC function in cells with misaligned spindles leads to multi-nucleation and anucleation , as cells undergo mitotic exit and cytokinesis irrespective of whether one nucleus has migrated into the daughter cell or not . The two components of the bi-partite Bfa1-Bub2 GAP complex localize interdependently at the SPB outer plaque throughout the cell cycle ( Lee et al . , 2001; Pereira et al . , 2000; 2001 ) . During normal spindle alignment , the Bfa1-Bub2 complex is unequally distributed between the two SPBs , to predominantly associate with the SPB that is directed towards the daughter cell ( dSPB ) ( asymmetric localization ) ( Fraschini et al . , 2006; Pereira et al . , 2000; 2001 ) . As long as the spindle is correctly positioned , Bfa1-Bub2 remains stably bound to the dSPB ( Caydasi and Pereira , 2009; Monje-Casas and Amon , 2009 ) . As soon as the anaphase spindle moves into the daughter cell compartment , Bfa1 is inactivated by the conserved polo-like kinase Cdc5 that inhibits Bfa1-Bub2 GAP activity ( Geymonat et al . , 2002; 2003; Hu and Elledge , 2002 ) . Upon spindle misalignment , the MARK/PAR family kinase Kin4 is recruited to the Spc72 component of both SPBs ( D'Aquino et al . , 2005; Maekawa et al . , 2007; Pereira and Schiebel , 2005 ) . Phosphorylation of Bfa1 by Kin4 at SPBs converts the asymmetric and stable SPB localization of Bfa1-Bub2 into a symmetric ( same amount on each SPB ) and dynamic association ( Caydasi and Pereira , 2009; Monje-Casas and Amon , 2009; Pereira et al . , 2000 ) . Kin4 also reduces the levels of SPB-bound Bfa1 ( Caydasi and Pereira , 2009 ) . These drastic changes in Bfa1-Bub2 SPB localization are essential for SPOC function and require binding of the 14-3-3 family protein Bmh1 to Bfa1 previously phosphorylated by Kin4 kinase ( Caydasi et al . , 2014 ) . Although , it is known that Kin4/Bmh1 pathway alters Bfa1-Bub2 association with the SPBs to engage the SPOC , some important questions remain unanswered: How is the Bfa1-Bub2 GAP complex recruited to the SPB outer plaque ? How does Bfa1-Bub2 association with the SPB respond to spindle misalignment ? Which SPB pool of Bfa1 is regulated by Kin4 and Bmh1 ? Here , we have used acceptor photobleaching Förster ( fluorescence ) resonance energy transfer ( FRET ) to analyze the interaction between Bfa1-Bub2 and structural components of SPBs . Our data reveal that , in an unperturbed mitosis Bfa1 C-terminus associated with the C-termini of two components of the SPB outer layer , Nud1 and Spc72 . SPOC activation specifically disrupted the interaction of Bfa1 with the microtubule linker Spc72 but not with the MEN-scaffolding protein Nud1 . The remodeling of Bfa1-Spc72 interaction in response to spindle misalignment required Kin4 and Bmh1 . We propose a model in which the Kin4/Bmh1 pathway disturbs the interaction between Bfa1 and Spc72 to prevent the inhibition of the GAP complex by Cdc5 . This step is essential for SPOC function and propagates the SPOC signal throughout both cell compartments . Cells lacking SPC72 were SPOC proficient . However , after prolonged mitotic arrest , we observed that spc72∆ cells frequently became binucleated due to SPOC slippage . In binucleated spc72∆ cells that possessed two misaligned spindles , re-alignment of only one of the two spindles was sufficient to silence the SPOC and promote mitotic exit irrespective of the presence of one mis-placed nucleus . The dominant nature of this impact of one spindle over the other suggests that there is a MEN activating and/or SPOC inhibitory signal that is released upon the passage of one SPB into the bud . These findings highlight both the importance of local regulation at the SPB for SPOC integrity and the key contribution of a daughter-specific mechanism that triggers MEN activation . We employed acceptor photobleaching FRET to assess where Bfa1-Bub2 is located at SPBs . In this technique , protein-protein proximities are determined by the energy transfer between protein pairs labeled with fluorescent FRET donor and acceptor tags . In contrast to the conventional FRET detection , which is based on acceptor emission measurements ( i . e . , sensitized emission FRET ) , the acceptor photobleaching FRET monitors the increase in donor fluorescence upon photobleaching of the acceptor ( i . e . , donor de-quenching ) ( Figure 1—figure supplement 1 ) ( Llopis et al . , 2000; Wouters and Bastiaens , 2001 ) . The proportionate increase in the donor fluorescence intensity after photobleaching of the acceptor directly yields the apparent FRET efficiency ( EFRET ) ( Karpova et al . , 2003; Kentner and Sourjik , 2009 ) , which is further corrected for the unspecific signal observed in the donor-only sample ( Figure 1—figure supplement 1B–D ) . Detection of this type of FRET signal is more straightforward and more robust than the sensitized emission measurements of FRET that demands multiple corrections for the spectral crosstalk between donor and acceptor fluorophores . We optimized the bleaching and imaging parameters for acceptor photobleaching FRET by using resident structural SPB proteins as reference associations . The topology of the core SPB proteins including Spc42 , Cnm67 and Spc110 has been previously defined by sensitized emission FRET ( Muller et al . , 2005 ) . Based on this analysis , we chose the FRET pairs of Spc42-Cnm67 and Spc110-Cnm67 as a molecule pair that interacted and a molecule pair that did not interact , respectively . As fluorophores , we used monomeric Turquoise ( mTUR , donor ) and enhanced YFP ( EYFP , acceptor ) . In each case the fluorophore was fused to the C-terminus of each protein ( Goedhart et al . , 2010 ) . In agreement with published data ( Muller et al . , 2005 ) , the acceptor photobleaching FRET technique yielded a positive FRET interaction between Spc42-mTUR and Cnm67-EYFP ( Figure 1—figure supplement 2A , 14% mean FRET efficiency in the donor-acceptor pair ) , while no FRET was detected for the Spc110-mTUR and Cnm67-EYFP pair ( Figure 1—figure supplement 2A ) . In order to estimate the maximum FRET efficiency that could be obtained for the mTUR-EYFP fluorophore pair at the SPB in our experimental system , we constructed a tandem tag composed of EYFP and mTUR fused to Bfa1 . Bfa1-EYFP-mTUR yielded a FRET value of 26% at SPBs ( Figure 1—figure supplement 2B ) , to define the maximum value that can be measured for this particular donor-acceptor pair ( mTUR-EYFP ) . We could however detect no FRET in the cytoplasm using the chimeric Bfa1-mTUR-YFP tandem pair ( data not shown ) , indicating that the concentration of Bfa1 in the cytoplasm sits below our FRET detection limit . Immuno-electron microscopy has established that Bfa1-Bub2 localize to the outer plaque ( Pereira et al . , 2000 ) . We therefore analyzed the juxtaposition of Bfa1-Bub2 to SPB outer plaque structural proteins Nud1 , Spc72 and Cnm67 . C-terminal tagging of Nud1 , Spc72 and Cnm67 with fluorescent proteins did not affect their functionality in microtubule organization , as no defect in nuclear migration and positioning was observed in strains encoding tagged proteins ( Figure 1—figure supplement 3A , B ) ( Grava et al . , 2006 ) . Next , we constructed chromosomally integrated C- or N-terminal fusions of BFA1 or BUB2 with mTUR or EYFP at their respective endogenous loci . The functionality of these gene fusions was confirmed by their ability to maintain a robust SPOC arrest in a kar9∆ strain background . Deletion of KAR9 causes frequent spindle misalignment at non-permissive temperatures ( Miller and Rose , 1998 ) . In the absence of SPOC function , kar9Δ cultures accumulate multi-nucleated or anucleated cells because cells exit mitosis without having segregated chromosomes into the daughter cell compartment ( Bardin et al . , 2000; Pereira et al . , 2000 ) ( Figure 1—figure supplement 3C ) . kar9∆ cells carrying C-terminally tagged BUB2 or N-terminally tagged BFA1 were SPOC deficient ( Figure 1—figure supplement 3C ) . This indicates that these fusions were not functional and so they were not analyzed further . Cells harboring C-terminal fusions of BFA1 , NUD1 or SPC72 and N-terminal fusions of BUB2 with mTUR or EYFP retained SPOC function ( Figure 1—figure supplement 3C and 3D ) . We analyzed the FRET efficiency of pairings between Bfa1-EYFP and either Nud1-mTUR , Spc72-mTUR or Cnm67-mTUR at the bud-directed SPB in cycling cells ( Figure 1A ) . Pairing Bfa1-EYFP with Nud1-mTUR or Spc72-mTUR yielded a FRET signal , whereas no FRET was detected between Bfa1-EYFP and Cnm67-mTUR ( Figure 1A ) . Similar FRET efficiencies were measured in metaphase- and anaphase-arrested cells ( Figure 2—figure supplement 1A , B ) . Unlike Bfa1 , mTUR-Bub2 did not display any FRET when paired with Nud1-EYFP or Spc72-EYFP ( Figure 2—figure supplement 1C ) . Importantly , the mTUR-Bub2 and Bfa1-EYFP combination generated a FRET signal at SPBs ( Figure 2—figure supplement 1D ) . These data show that the C-terminus of Bfa1 resides in close proximity to the C-termini of both Nud1 and Spc72 at SPBs . The C-terminus of Bfa1 is also positioned in close proximity to the N-terminus of Bub2 , in support of their binding to SPBs as a protein complex ( Pereira et al . , 2000 ) . 10 . 7554/eLife . 14029 . 003Figure 1 . Bfa1 interacts with the SPB outer layer proteins Spc72 and Nud1 . ( A ) Box-whisker plots representing the distributions of FRET efficiency values for Bfa1 ( C-terminally tagged with EYFP ) in pair with Nud1 , Spc72 or Cnm67 ( C-terminally tagged with mTUR ) measured at the dSPB as depicted in the cartoon . The FRET data shown here and in subsequent figures are one out of two biological replicates unless otherwise specified . For Box-Whisker plots representing FRET data , the boxes show the lower and upper quartiles , the whiskers show the minimum and maximal values excluding outliers; outliers ( not shown ) were calculated as values greater or lower than 1 . 5 times the interquartile range; the line inside the box indicates the median . N is the sample size; asterisks show significant difference according to student’s t-test ( p<0 . 01 ) and the exact p-values are indicated in the accompanying data file ( Figure 1—source data 1 ) . ( B ) In vitro binding assay of bead bound recombinant MBP-Bfa1 with bacterially purified GST-Spc72 ( i ) , 6His-Spc72-C ( codons 231–622 ) ( ii ) and 6His-Mlc1 ( iii ) . MBP on beads and 6His-Mlc1 were used as negative controls for in vitro binding reaction . Spc72 was detected with anti-GST antibody; Spc72-C and Mlc1 were detected with anti-6His antibodies after immunoblotting . The Ponceau S stained membrane shows the levels of MBP-Bfa1 and MBP used in the assay . One representative blot out of two independent experiments is shown in each panel . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00310 . 7554/eLife . 14029 . 004Figure 1—source data 1 . Raw data and the calculated FRET efficiencies of Nud1-Bfa1 and Spc72-Bfa1 pairs at SPBs in cycling cells ( source data for Figure 1A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00410 . 7554/eLife . 14029 . 005Figure 1—figure supplement 1 . The basic principles of acceptor photobleaching technique to measure FRET . ( A ) Schematic representation of FRET between D-donor ( mTUR - turquoise fluorescent protein tagged with host protein X ) and A-acceptor ( EYFP - yellow fluorescent protein tagged with host protein Z ) . Note that donor fluorescence increases after photobleaching of the acceptor . ( B ) Experimental setup to measure FRET . The order of image acquisition is indicated from 1 to 4 and the region of photobleaching within the sample field is delineated by grey circles . 515 nm: laser wavelength used for YFP photobleaching . ( C ) FRET for each protein pair tagged with mTUR and EYFP was calculated using mean fluorescence intensities of three samples: ( DA ) – with donor and acceptor tagged protein pair ( i ) , ( D ) - with donor only tag ( ii ) , ( bg ) - no tag ( iii ) . Fluorecence intensities of donor fluorescence were measured from regions around SPBs indicated with red squares . For untagged cells , a region within the cytoplasm was measured ( used for background correction of the signal intensities ) . ( D ) Equations used for quantification of FRET efficiency: ( 1 ) background correction of the mean fluorescence intensities ( FL ) where FL ( DA ) , FL ( D ) , FL ( bg ) are the raw values of mean fluorescence intensities of the donor and acceptor ( DA ) , the donor only ( D ) and untagged ( bg ) samples described in Figure 1—figure supplement 1C . FLcor: background corrected values . ( 2 ) Calculation of percentage increase in FLcor after acceptor photobleaching for the donor and acceptor pair [E ( DA ) ] and for the donor only control [E ( DA ) ] . ( 3 ) Normalization of the E ( DA ) and E ( D ) values to the median of E ( D ) [E ( D ) median] to obtain the 'FRET efficiency%' shown in the box-whisker plots . FRET for each pair was recognized as positive if the 'FRET efficiency% ( DA ) ' was significantly higher than the 'FRET efficiency% ( D ) ' according to student´s t-test with ( p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00510 . 7554/eLife . 14029 . 006Figure 1—figure supplement 2 . Validation of acceptor photobleaching technique with FRET-positive and FRET-negative controls . ( A–B ) Box-Whisker plots representing the distributions of FRET efficiencies measured in pairs of fluorescently tagged proteins: Spc42-Cnm67 ( FRET-positive pair ) and Spc110-Cnm67 ( FRET-negative pair ) ( A ) and in the tandem pair of EYFP and mTUR in fusion with Bfa1 ( B ) . Box-whisker plot is one out of two technical replicates from the same experiment . N: sample size . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data file for exact p-values ( Figure 1—figure supplement 2—source data —1 and 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00610 . 7554/eLife . 14029 . 007Figure 1—figure supplement 2—source data 1 . Raw and calculated FRET efficiencies of Spc42-Cnm67 and Spc110-Cnm67 pairs at SPBs in cycling cells ( source data for Figure 1—figure supplement 2A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00710 . 7554/eLife . 14029 . 008Figure 1—figure supplement 2—source data 2 . Raw and calculated FRET efficiencies of Bfa1-EYFP-mTUR tandem pair at SPBs in cycling cells ( source data for Figure 1—figure supplement 2B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 00810 . 7554/eLife . 14029 . 009Figure 1—figure supplement 3 . Functionality of tagged SPB proteins and SPOC components . ( A ) Fluorescent C-terminal fusions of Nud1 and Spc72 are functional . The indicated cell types were grown at 23°C and then shifted to 30°C for 3 hr before fixation and DAPI staining . Graphs show the percentages of cells with the nuclei positioned as depicted in the cartoon to the left . Light grey bars represent correct nuclear positioning ( at the bud neck region ) . N: 200 cells per strain . ( B ) Cnm67 fusion proteins are functional . Anaphase nuclear positioning was analyzed for wild type ( CNM67 ) and CNM67 tagged strains as depicted . CNM67 deleted cells have a strong defect in anaphase spindle positioning and were used as a positive control . N: 200 cells per strain . Please note that cnm67∆ doesn’t have nuclear positioning defects early in the cell cycle ( Grava et al . , 2006 ) ( our unpublished data ) . ( C–D ) Functionality of SPOC in kar9∆ cells carrying Bfa1 , Bub2 ( C ) and Nud1 , Spc72 ( D ) fused with the indicated fluorophores . The indicated cell types were grown at 23°C and shifted to 30°C for 3 hr to induce the accumulation of cells with misaligned spindles ( blue bars ) . Black bars indicate the SPOC deficient phenotypes as a result of cell cycle progression despite spindle orientation defects . N: 200 cells per strain . One representative graph out of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 009 Recombinant proteins have previously been used to demonstrate the direct physical interaction between Nud1 and Bfa1 ( Gruneberg et al . , 2000 ) . To determine whether Bfa1 also directly interacts with Spc72 , we used bacterially purified Bfa1 fused to maltose binding protein ( MBP-Bfa1 ) , full length Spc72 tagged with glutathione-binding protein ( GST ) ( GST-Spc72 ) and a C-terminal truncated fragment of Spc72 ( codons 231–622 ) tagged with 6 histidines ( 6His-Spc72-C , Figure 1B ) . Full length GST-Spc72 bound to MBP-Bfa1 but not to MBP-beads ( Figure 1B , lanes 1 and 2 ) . The C-terminal 391 amino acids of Spc72 were sufficient to confer this interaction , as 6His-Spc72-C associated with MBP-Bfa1 but not to MBP ( Figure 1B , lanes 3 and 4 ) . Notably , MBP-Bfa1 or MBP did not interact with an unrelated 6His-tagged protein ( 6His-Mlc1 ) to highlight the specific nature of the association with 6His-Spc72-C ( Figure 1B , lanes 5 and 6 ) . Taken together , Bfa1 directly binds to Nud1 and Spc72 in vitro , suggesting that the interactions established by FRET arise from direct physical interactions . In an unperturbed mitosis , more Bfa1 molecules bind to the daughter directed SPB ( dSPB ) than to the mother directed SPB ( mSPB ) ( Pereira et al . , 2000 ) . This behavior is referred to as asymmetric Bfa1 localization . We asked whether differential association of Bfa1 with Nud1 and Spc72 at the two SPBs explains this asymmetric accumulation of Bfa1 . To test this hypothesis , we compared FRET for Bfa1-Nud1 and Bfa1-Spc72 pairs at the dSPB and mSPB in metaphase arrested cells where Bfa1 predominately associates with dSPB and is only weakly detectable on the mSPB ( Figure 2A and B ) ( Pereira et al . , 2000 ) . The metaphase arrest was achieved through the depletion of the anaphase promoting complex regulatory subunit Cdc20 ( Shirayama et al . , 1999 ) . The levels of FRET between Bfa1-Nud1 or Bfa1-Spc72 were similar at the mSPB and dSPB ( Figure 2C and D ) . This indicates that Bfa1 is in close proximity with Nud1 and Spc72 on both SPBs . Thus , asymmetric SPB localization of Bfa1 cannot be dictated by the differential association of Bfa1 with Spc72 or Nud1 . 10 . 7554/eLife . 14029 . 010Figure 2 . Bfa1 interacts with Spc72 and Nud1 at the daughter and mother SPBs . ( A–B ) Representative images of metaphase-arrested cells carrying Nud1-mTUR Bfa1-EYFP ( A ) and Spc72-mTUR Bfa1-EYFP ( B ) . Scale bar: 3 μm . ( C–D ) Box and Whiskers plots showing FRET efficiency for the indicated Nud1-Bfa1 ( C ) and Spc72-Bfa1 ( D ) pairs at the daughter and mother SPBs ( dSPB and mSPB ) in metaphase-arrested cells . N: sample size . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data files ( Figure 2—source data 1 and 2 ) for exact p-values . ( E–F ) Scatter plots of FRET efficiencies for Nud1-Bfa1 ( E ) and Spc72-Bfa1 ( F ) pairs as a function of Bfa1-EYFP fluorescence intensity values obtained from the dSPB and from the mSPB . N: Sample size . R: r-squared value for the best-fit trendline . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01010 . 7554/eLife . 14029 . 011Figure 2—source data 1 . Raw data and the calculated FRET efficiencies of Nud1-Bfa1 and Spc72-Bfa1 pairs at SPBs in metaphase arrested cells ( source data for Figure 2C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01110 . 7554/eLife . 14029 . 012Figure 2—source data 2 . Raw data and the calculated FRET efficiencies of the Spc72-Bfa1 pair at the mother and the daughter SPB in metaphase arrested cells ( source data for Figure 2D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01210 . 7554/eLife . 14029 . 013Figure 2—figure supplement 1 . FRET analysis of Bfa1 and Bub2 . ( A–B ) FRET efficiencies of Bfa1-Nud1 ( A ) and Bfa1-Spc72 ( B ) in metaphase and anaphase arrested cells . Metaphase arrest is achieved by Cdc20 depletion ( Gal1-CDC20 ) and anaphase arrest was induced by overproduction of nondegradable Clb2 ( Gal1-CLB2ΔDB ) . ( C–D ) FRET efficiency of mTUR-Bub2 with Nud1-EYFP and Spc72-EYP ( C ) or with Bfa1 C-terminal fusion ( D ) . Box-whisker plots in C and D are one out of two technical replicates from the same experiment . N: sample size . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data file for exact p-values ( Figure 2—figure supplement 1—source data 1–4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01310 . 7554/eLife . 14029 . 014Figure 2—figure supplement 1—source data 1 . Raw and calculated FRET efficiencies of the Bfa1-Nud1 pair in metaphase and anaphase arrested cells ( source data for Figure 2—figure supplement 1A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01410 . 7554/eLife . 14029 . 015Figure 2—figure supplement 1—source data 2 . Raw and calculated FRET efficiencies of the Bfa1-Spc72 pair in metaphase and anaphase arrested cells ( source data for Figure 2—figure supplement 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01510 . 7554/eLife . 14029 . 016Figure 2—figure supplement 1—source data 3 . Raw and calculated FRET efficiencies of Bub2-Nud1 and Bub2-Spc72 pairs in cycling cells ( source data for Figure 2—figure supplement 1C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01610 . 7554/eLife . 14029 . 017Figure 2—figure supplement 1—source data 4 . Raw and calculated FRET efficiencies of Bub2-Bfa1 pair in cycling cells ( source data for Figure 2—figure supplement 1D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 017 Importantly , FRET efficiency at SPBs was estimated based on the relative change in the fluorescence of the donors Nud1-mTUR or Spc72-mTUR , both of which localize at both SPBs to similar levels ( Erlemann et al . , 2012; Knop and Schiebel , 1997; 1998 ) . However , the acceptor ( Bfa1-EYFP ) is recruited to higher levels at the dSPB than at the mSPB to generate different donor:acceptor ratios on the dSPB and mSPB ( Figure 2C and D ) . To evaluate the effect of different donor:acceptor ratios on FRET efficiency , we compared the FRET efficiency of Bfa1-Nud1 or Bfa1-Spc72 pairs with the fluorescence signal intensity of the acceptor ( Bfa1-EYFP ) before bleaching ( Figure 2E and F ) . The values of FRET efficiency did not correlate with the signal intensity of the acceptor . Thus , changes in the donor:acceptor ratio do not affect FRET measurements in our experimental system . This validates the ability of this approach to compare the association of Bfa1 with Nud1 and with Spc72 on different poles . Phosphorylation of Bfa1 by Kin4 kinase decreases the residence time of Bfa1 at SPBs ( Caydasi and Pereira , 2009 ) . We therefore asked whether Kin4 affected the association of Bfa1 with Nud1 , Spc72 or both partners . To this end , we first analyzed cells overproducing Kin4 ( Figure 3A and B ) . Overexpression of KIN4 activates the SPOC to arrest cells in late anaphase , even when the spindle is correctly aligned ( D'Aquino et al . , 2005 ) . As a control for cells that arrest in late anaphase without SPOC activation , we overexpressed a non-degradable version of the mitotic cyclin CLB2 ( clb2∆DB ) ( Surana et al . , 1993 ) ( Figure 3A and B ) . FRET between Bfa1 and Nud1 was maintained in KIN4 overexpressing cells ( Figure 3A , Figure 3—figure supplement 1 ) . In contrast , the FRET between Bfa1-Spc72 pair disappeared upon KIN4 overexpression ( Figure 3B ) . Thus , Kin4 activity specifically disturbs the interaction between Bfa1 and Spc72 but not that between Bfa1 and Nud1 . 10 . 7554/eLife . 14029 . 018Figure 3 . SPOC activation interferes with Bfa1-Spc72 interaction . ( A–B ) FRET efficiencies for Nud1-Bfa1 ( A ) and Spc72-Bfa1 ( B ) pairs . Cells were arrested in anaphase by clb2△DB overexpression ( Gal1-clb2△DB ) or by KIN4 overexpression ( Gal1-KIN4 ) . ( C–D ) FRET efficiencies for Spc72-Bfa1 ( C ) and Nud1-Bfa1 ( D ) pairs in Gal1-KIN4 ( please note that the data for Gal1-KIN4 cells is identical to Figure 3A ) , Gal1-KIN4 bmh1∆ and bmh1∆ cells grown in galactose medium . ( E–F ) FRET efficiencies for Nud1-Bfa1 ( E ) and Spc72-Bfa1 ( F ) pairs in kar9Δ cells with correct and mis-aligned spindles . ( G–H ) FRET efficiencies of Nud1-Bfa1 ( G ) and Spc72-Bfa1 ( H ) pairs in kar9Δ kin4∆ cells with correct and mis-aligned spindles . Box-whisker plots in G and H are one out of two technical replicates from the same experiment . N: sample size . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data files for exact p-values ( Figure 3—source data 1–8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01810 . 7554/eLife . 14029 . 019Figure 3—source data 1 . Raw data and the calculated FRET efficiencies of Nud1-Bfa1 and Spc72-Bfa1 pairs at SPBs upon KIN4 overexpression ( source data for Figure 3A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 01910 . 7554/eLife . 14029 . 020Figure 3—source data 2 . Raw data and the calculated FRET efficiencies of the Spc72-Bfa1 pair at SPBs upon KIN4 overexpression , and CDC20 depletion ( source data for Figure 3B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02010 . 7554/eLife . 14029 . 021Figure 3—source data 3 . Raw data and the calculated FRET efficiencies of the Spc72-Bfa1 pair in the presence and absence of BMH1 ( source data for Figure 3C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02110 . 7554/eLife . 14029 . 022Figure 3—source data 4 . Raw data and the calculated FRET efficiencies of the Nud1-Bfa1 pair in the presence and absence of BMH1 ( source data for Figure 3D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02210 . 7554/eLife . 14029 . 023Figure 3—source data 5 . Raw data and the calculated FRET efficiencies of the Nud1-Bfa1 pair in kar9∆ cells with normally aligned or misaligned spindles ( source data for Figure 3E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02310 . 7554/eLife . 14029 . 024Figure 3—source data 6 . Raw data and the calculated FRET efficiencies of the Spc72-Bfa1 pair in kar9∆ cells with normally aligned or misaligned spindles ( source data for Figure 3F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02410 . 7554/eLife . 14029 . 025Figure 3—source data 7 . Raw data and the calculated FRET efficiencies of the Nud1-Bfa1 pair in kar9∆ kin4∆ cells with normally aligned or misaligned spindles ( source data for Figure 3G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02510 . 7554/eLife . 14029 . 026Figure 3—source data 8 . Raw data and the calculated FRET efficiencies of the Spc72-Bfa1 pair in kar9∆ kin4∆ cells with normally aligned or misaligned spindles ( source data for Figure 3H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02610 . 7554/eLife . 14029 . 027Figure 3—figure supplement 1 . Controls for Kin4 overproducing cells used in FRET experiments . KIN4 expression levels in samples used for the FRET analysis shown in Figure 3A–D . Total cell lysates from the indicated strains grown in the presence ( + ) or in the absence of galactose ( - ) . Kin4 levels were detected using anti-Kin4 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 027 We previously described how binding of Bmh1 to Kin4-phosphorylated Bfa1 abolishes the stable association of Bfa1 with SPBs ( Caydasi et al . , 2014 ) . In cells lacking BMH1 , Bfa1 remains asymmetrically and stably associated with SPBs , even when Kin4 phosphorylates Bfa1 ( Caydasi et al . , 2014 ) . We therefore proposed that Bmh1 would be required to break the Bfa1-Spc72 interaction . To test this notion , we compared the FRET efficiency between Bfa1 and Spc72 when Kin4 was overproduced in wild type and in bmh1∆ cells ( Figure 3C , Figure 3—figure supplement 1 ) . Kin4 overexpression in the presence of BMH1 greatly decreased the FRET efficiency of the Bfa1-Spc72 pair ( Figure 3C ) . This was not the case in bmh1∆ cells ( Figure 3C ) . Deletion of BMH1 did not influence the FRET efficiency of Nud1–Bfa1 pair ( Figure 3D ) . To determine whether the interaction between Bfa1 and Spc72 is lost upon SPOC activation , we analyzed the FRET efficiencies of Bfa1-Nud1 and Bfa1-Spc72 in cells with correctly and incorrectly positioned anaphase spindles ( Figure 3E and F ) . Spindle position had no impact upon FRET between Bfa1 and Nud1 ( Figure 3E ) . In contrast , spindle misalignment reduced the FRET efficiency of the Bfa1-Spc72 pair to the level seen in the donor-only control ( Figure 3F ) . In the absence of KIN4 , however , the FRET between Bfa1 and Spc72 persisted irrespective of spindle position ( Figure 3G and H ) . Together , these data show that , upon spindle misalignment , Kin4 , with the assistance of Bmh1 , rearranges Bfa1 molecules on the SPB by specifically influencing the Bfa1-Spc72 interaction . Our FRET data shows that SPOC activation specifically interrupts Bfa1 association with Spc72 but not with Nud1 ( Figure 3E and F ) . We therefore asked whether there were two separate pools of Bfa1 , one that binds Spc72 and another Nud1 ( Figure 4A–i ) or whether Bfa1 molecules bind both Spc72 and Nud1 simultaneously to form a single pool ( Figure 4A-ii ) . In the scenario where two individual Bfa1 pools exist ( Figure 4A–i ) , Spc72 should recruit the majority of Bfa1 to SPBs because SPB-bound Bfa1 levels fall drastically upon spindle misalignment ( Caydasi and Pereira , 2009 ) . This coincides well with the loss of Bfa1-Spc72 interaction observed by FRET . Therefore , if two separate Spc72- and Nud1-bound Bfa1 pools exist , less Bfa1 would bind to SPBs in the absence of SPC72 ( Figure 4A–i ) . We therefore analyzed Bfa1 localization in cells lacking SPC72 . For this analysis , we used the W303 strain background in which SPC72 is not essential ( Hoepfner et al . , 2002; Soues and Adams , 1998 ) yet Bfa1-Bub2 SPB binding and phospho-regulation is the same as in S288C background where SPC72 is essential ( Caydasi and Pereira , 2009; D'Aquino et al . , 2005; Monje-Casas and Amon , 2009; Pereira and Schiebel , 2005 ) . Cells lacking SPC72 cannot form long cytoplasmic microtubules; hence logarithmically growing spc72∆ cultures contain cells with misaligned spindles alongside cells with correctly aligned spindles ( Hoepfner et al . , 2002; Soues and Adams , 1998 ) . FRET analysis showed that Bfa1 associated with Nud1 at SPBs of spc72∆ cells regardless of spindle orientation ( Figure 4B and C ) . Interestingly , SPB-bound Bfa1 levels were not reduced in spc72∆ cells ( Figure 4D ) , arguing against the model where there are two pools of Bfa1 that separately bind Spc72 and Nud1 . 10 . 7554/eLife . 14029 . 028Figure 4 . Bfa1-SPB binding dynamics in the absence of Spc72 . ( A ) Schematic representation of two possible mechanisms for Bfa1 binding at the SPB . i: Bfa1 is recruited as two separate pools . ii: Bfa1 associates simultaneously with Nud1 and Spc72 . Dashed arrows indicate dynamic binding . ( B–C ) Representative still images and FRET efficiencies for spc72Δ NUD1-mTUR BFA1-EYFP cells with aligned and misaligned spindle . Box-whisker plot is one out of two technical replicates from the same experiment . N: sample size . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data file for exact p-values ( Figure 4—source data 1 ) . ( D ) Representative still images and mean fluorescence intensities of Bfa1-GFP localized at the dSPB during anaphase in SPC72 ( pRS-SPC72 spc72Δ ) and spc72Δ cells carrying mCherry-TUB1 . Box-whisker plot is one out of two technical replicates from the same experiment . See the source file ( Figure 4—source data 2 ) for the raw data . ( E-F ) FRAP analysis of Bfa1-GFP at the SPBs in cells with correctly aligned ( E ) and misaligned spindles ( F ) . The black line depicts the best-fit single exponential curve for each data set . Data represent the mean of “N” sized sample . t-half: half recovery time . The graphs show the average fluorescence recovery curves for the corresponding strains . See the accompanying data file for individual curves and raw data ( Figure 4—source data 3 and 4 ) . Data represented in E is one out of two biological replicates . Data for spc72∆ in F is one out of two biological replicates . Data for kar9∆ and kin4∆ kar9∆ comes from one experiment , whose results are in concordance with published data ( Caydasi et al . , 2014; Caydasi and Pereira , 2009 ) . ( G ) Representative still images of ( F ) . Photobleached SPB is marked with squares . Time-lapse series show 3-fold enlarged photobleached regions at the indicated time points . Time zero is the first image taken after photobleaching . Scale bar: 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02810 . 7554/eLife . 14029 . 029Figure 4—source data 1 . Raw data and the calculated FRET efficiencies of the Nud1-Bfa1 pair in spc72∆ cells with normally aligned or misaligned spindles ( source data for Figure 4C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 02910 . 7554/eLife . 14029 . 030Figure 4—source data 2 . Raw and normalized mean fluorescence intensities of Bfa1-GFP at SPBs of spc72∆ and SPC72 cells ( source data for Figure 4D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 03010 . 7554/eLife . 14029 . 031Figure 4—source data 3 . Raw and normalized FRAP data of Bfa1-GFP at the SPBs of spc72∆ and kar9∆ cells with normally aligned spindles . FRAP curves for individual cells are also presented ( source data for Figure 4E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 03110 . 7554/eLife . 14029 . 032Figure 4—source data 4 . Raw and normalized FRAP data of Bfa1-GFP at the SPBs of spc72∆ and kar9∆ cells with misaligned spindles . FRAP curves for individual cells are also presented ( source data for Figure 4F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 032 To gain deeper insight into how Bfa1 associates with SPBs in the absence of Spc72 , we used FRAP ( Fluorescence Recovery After Photobleaching ) analysis to compare the dynamics of Bfa1 association with SPBs in spc72∆ cells with normal and mis-aligned spindles . After photobleaching , Bfa1-GFP failed to recover at SPBs in cells with normally aligned anaphase spindles in the presence of SPC72 ( Figure 4E ) . This confirms the previously reported immobile pool of Bfa1 at the dSPB ( Caydasi and Pereira , 2009; Monje-Casas and Amon , 2009 ) . Cells lacking SPC72 also recruited Bfa1-GFP stably to the dSPB when the anaphase spindle was correctly aligned ( Figure 4E ) . Thus , Bfa1 interaction with Nud1 is stable throughout anaphase when spindle is correctly aligned , independently of Spc72 . Next , we examined cells with misaligned spindles . Consistent with previous data , in the presence of SPC72 , Bfa1 was highly dynamic at both SPBs with a half recovery time of ~ 11 s during spindle misalignment ( Figure 4F and 4G ) ( Caydasi and Pereira , 2009 ) . The half recovery time of Bfa1 was 6 times longer in spc72∆ cells than in SPC72 cells ( Figure 4F and 4G ) . These data suggest that Spc72 regulates the mode of Bfa1 association with Nud1 when the spindle is mis-positioned . Together , our data favor a model in which Bfa1 molecules simultaneously associate with Nud1 and Spc72 ( Figure 4A , ii ) . In cells with properly aligned spindles , Nud1 is sufficient to stably recruit Bfa1 to SPBs independently of Spc72 . However , in cells with misaligned spindles , Spc72 becomes indispensable to regulate Bfa1 SPB binding dynamics . Bfa1-SPB binding dynamics in spc72∆ cells during spindle misalignment were reminiscent of those seen in kin4∆ cells ( Figure 4F and G ) . This is consistent with previous reports that Spc72 recruits Kin4 to SPBs and that Kin4 SPB binding is necessary for its function in SPOC ( D'Aquino et al . , 2005; Maekawa et al . , 2007; Pereira and Schiebel , 2005 ) . Indeed , we were unable to detect Kin4 at SPBs in spc72∆ cells by fluorescence microscopy ( Figure 5A and B ) . Neither could phospho-specific antibodies detect Bfa1 phosphorylation at S180 ( one of the two sites phosphorylated by Kin4 ) in spc72∆ cells ( Maekawa et al . , 2007 ) ( Figure 5C ) . Furthermore , the failure of Bfa1 to bind symmetrically ( with same levels ) to SPBs in spc72∆ cells upon spindle misalignment was also reminiscent of the kin4∆ phenotype ( Figure 5D ) . Collectively , these data indicate that Kin4 is unable to phosphorylate Bfa1 and dislodge Bfa1 from SPBs in the absence of Spc72 . Thus , the reduction of Bfa1-SPB binding dynamics in spc72∆ cells with misaligned spindles is likely to arise from a lack of Kin4 phosphorylation of Bfa1 . 10 . 7554/eLife . 14029 . 033Figure 5 . Lack of Spc72 interferes with Kin4 localization and functioning at the SPB . ( A ) SPB localization of Kin4-GFP in SPC72 ( pRS-SPC72 spc72∆ ) and spc72∆ cells carrying the SPB marker SPC42-eqFP . Cells were arrested in metaphase with nocodazole . ( B ) Quantification of ( A ) . ( C ) Immunoblots showing Bfa1 phosphorylation by Kin4 at S180 residue . Bfa1-GFP was immunoprecipitated from indicated strains . Total amount of immunoprecipitated Bfa1-GFP and Bfa1-GFP that is phosphorylated at S180 were detected by anti-GFP and anti-P-S180 antibodies respectively . Bfa1S180A/S150A served as a control for the specificity of anti-P-S180 antibody . A representative blot out of three independent experiments is shown . ( D ) Box and Whisker plots of Bfa1-GFP fluorescence intensity at SPBs in kar9Δ , spc72Δ and kar9Δ kin4Δ cells with correctly and mis-aligned spindles . Within the same cell , the SPB with stronger and weaker Bfa1 fluorescence intensity were classified as SPB1 and SPB2 , respectively . The maximum Bfa1-GFP fluorescence intensity of each data set was normalized to 1 . The boxes show the lower and upper quartiles , the whiskers show the minimum and maximal values excluding outliers; outliers ( shown as dots ) were calculated as values greater or lower than 1 . 5 times the interquartile range; the line inside the box indicates the median . For each box , 33 SPBs were quantified . Asterisks show significant difference according to student’s t-test ( p<0 . 01 ) . See the accompanying data file for exact p-values ( Figure 5—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 03310 . 7554/eLife . 14029 . 034Figure 5—source data 1 . Raw and normalized mean fluorescence intensities of Bfa1-GFP at SPBs of spc72∆ and kar9∆ cells with normally aligned or misaligned spindles ( source data for Figure 5D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 034 We next asked whether the SPOC is functional in spc72∆ cells . For this , we performed live cell imaging of spc72∆ and kar9∆ cells carrying tubulin tagged with GFP ( GFP-TUB1 ) to decorate microtubules ( Figure 6 ) . We followed cells in which the mitotic spindle either elongated along the mother to daughter cell axis ( Figure 6A–C , correct spindle orientation , upper rows ) or within the mother cell compartment ( Figure 6A–C , misaligned spindle , lower rows ) . We defined the duration of anaphase as being the time that elapsed between the onset of rapid spindle elongation phase and spindle breakdown ( which is a consequence of mitotic exit ) ( Figure 6D ) ( Bardin and Amon , 2001 ) . As previously reported ( Bloecher et al . , 2000; Pereira et al . , 2000 ) , kar9∆ cells with mis-oriented spindles failed to exit mitosis and displayed a prolonged anaphase arrest indicative of SPOC proficiency ( Figure 6A and D ) . Deletion of KIN4 in kar9∆ cells induced cells with misaligned and properly aligned spindles to exit mitosis with similar timing ( Figure 6B and D ) , reflecting the SPOC deficiency of kin4∆ cells . To our surprise , spc72∆ cells were able to sustain an anaphase arrest for the duration of the time-lapse movie ( >60 min ) upon spindle mis-orientation ( Figure 6C and D ) . This delay did not arise from a general deficiency that would extend anaphase because cells exited mitosis 23 min after the onset of spindle elongation when their spindle elongated along the correct mother-daughter cell axis ( Figure 6C and D ) . Furthermore , deletion of BFA1 was lethal in spc72∆ cells ( Figure 6—figure supplement 1 ) , indicating that these cells require SPOC function for survival . Thus , despite the loss of Bfa1 phosphorylation by Kin4 , disruption of Bfa1-Spc72 interaction by means of SPC72 deletion did not compromise the ability of cells with misaligned spindles to engage SPOC arrest . 10 . 7554/eLife . 14029 . 035Figure 6 . spc72∆ cells are SPOC proficient ( A–C ) Representative time-lapse images of kar9Δ ( A ) , kar9Δ kin4Δ ( B ) and spc72Δ ( C ) cells carrying GFP-TUB1 . Time is given in minutes from the start of the inspection . Scale bar 3 µm . ( D ) Comparison of anaphase duration in kar9Δ , kar9Δ kin4Δ and spc72Δ cells with aligned and misaligned spindles . N: number of cells observed from each category . Mean anaphase duration is given in minutes . SD: Standard deviation . ( E ) Immunoblot showing Bfa1 mobility shift . Indicated strains were released from G1-block ( alpha-factor arrest ) in nocodazole containing medium . Samples were collected after 3 hr . Bfa1-GFP was detected in the total cell extracts of indicated strains using anti-GFP antibody . The arrow indicates the hyper-phosphorylated form of Bfa1 . Asterisks indicate an unspecific band detected by the anti-GFP antibody . A representative blot out of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 03510 . 7554/eLife . 14029 . 036Figure 6—figure supplement 1 . Effect of Bfa1 on the growth of spc72∆ cells . spc72∆ pRS316-SPC72 and spc72∆ bfa1∆ pRS316-SPC72 cells were patched on 5-FOA’ containing plates to allow selection of cells that lost the URA3-based plasmid carrying wild type SPC72 ( pRS316-SPC72 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 036 In wild type cells , Bfa1 phosphorylation by Kin4 is required to prevent Cdc5 phosphorylation of Bfa1 , which would otherwise inhibit Bfa1-Bub2 GAP activity and so promote mitotic exit ( D'Aquino et al . , 2005; Geymonat et al . , 2003; Pereira and Schiebel , 2005 ) . We therefore asked whether Cdc5 phosphorylation of Bfa1 is blocked in the absence of SPC72 . In order to examine Bfa1 phosphorylation by Cdc5 , we assessed Bfa1 migration on SDS page gels . Bfa1 becomes hyper-phosphorylated by Cdc5 in kin4∆ cells ( Figure 6E ) ( Maekawa et al . , 2007; Pereira and Schiebel , 2005 ) . Levels of the hyper-phosphorylated forms of Bfa1 were strongly reduced in spc72∆ cells ( Figure 6E ) , although Kin4 was not able to phosphorylate Bfa1 in these cells ( Figure 5C ) . Thus , not only Kin4 but also Cdc5 requires Spc72 to efficiently phosphorylate Bfa1 . Our data altogether suggests that Spc72 serves as a platform that integrates Kin4 and Cdc5 counteracting actions on Bfa1 . To this point , we show that Spc72 is necessary for Bfa1 phosphorylation by both Cdc5 and Kin4 . As Spc72-Bfa1 interaction is disrupted during the SPOC arrest , we reasoned that the disassociation of Bfa1 from Spc72 might be required to prevent the inhibitory phosphorylation of Bfa1 by Cdc5 . To test this hypothesis , we constructed cells bearing endogenous SPC72 C-terminally fused with the GFP-binding protein ( GBP ) . GBP efficiently binds GFP or GFP tagged proteins in human and yeast cells ( Bertazzi et al . , 2011; Rothbauer et al . , 2008 ) . Spc72-GBP constitutively recruited Bfa1-GFP to both SPBs regardless of the spindle position ( Figure 7A ) . Emphasizing the importance of Spc72-Bfa1 disengagement for SPOC , coexistence of Bfa1-GFP and Spc72-GBP in the same cell resulted in SPOC failure ( Figure 7B ) . We further analyzed the phosphorylation profile of Bfa1-GFP in these cells . In the presence of Spc72-GBP , Bfa1-GFP accumulated hyper-phosphorylated forms , which disappeared upon depletion of Cdc5 ( Figure 7C , lanes 3 , 4 and 5 ) . Thus , persistence of Spc72-Bfa1 interaction provokes Cdc5 dependent phosphorylation of Bfa1 . Importantly , Kin4 also contributed to the phosphorylation of Bfa1-GFP in Spc72-GBP cells , as deletion of KIN4 reduced the hyper-phosphorylated forms of Bfa1 ( Figure 7C , lanes 4 and 6 ) . Our data suggests a model where Bfa1 dissociates from Spc72 to prevent the inhibitory phosphorylation of Bfa1 by Cdc5 ( Figure 7D ) . 10 . 7554/eLife . 14029 . 037Figure 7 . Bfa1-Spc72 interaction provokes Cdc5 phosphorylation of Bfa1 ( A ) Still images ( right ) and the SPB-bound mean fluorescence intensities ( left ) of Bfa1-GFP in kar9∆ and SPC72-GBP kar9∆ cells . SPB1 indicates the SPB closer to bud where SPB2 is the SPB closer to the mother cell compartment . Yellow or red arrows point the two SPBs in a cell with a normally aligned or misaligned spindle respectively . The numbers 1 and 2 next to the arrows indicate SPB1 and SPB2 respectively . Scale bar: 3 µm . In the Box-Whisker plots showing Bfa1-GFP mean fluorescence intensities , the boxes show the lower and upper quartiles , the whiskers show the minimum and maximal values excluding outliers; circles represent the outliers calculated as values greater or lower than 1 . 5 times the interquartile range; the line inside the box indicates the median . Asterisks show significant difference according to student’s t-test ( p<0 . 001 ) . Sample sizes for kar9∆ cells were 49 and 23 , whereas sample sizes for kar9∆ SPC72-GBP cells were 38 and 14 for each SPB with normal and misaligned spindles respectively . See the accompanying data file for exact p-values ( Figure 7—source data 1 ) . ( B ) SPOC integrity of the indicated cell types . Cells were grown at 23°C and shifted to 30°C for 3 hr to induce the accumulation of cells with misaligned spindles ( blue bars ) . Black bars show the percentage of multi-nucleated phenotypes , which indicates SPOC deficiency N: 100 cells per strain . A representative plot out of three biological replicates is shown . ( C ) Immunoblot showing Bfa1-GFP mobility shift . Indicated strains arrested in G1 using alpha-factor and released in alpha-factor free , nocodazole containing YPDA medium . Samples were collected after 2 . 5 hr . Gal1-CDC5 bearing strains were grown and G1-arrested in raffinose and galactose containing medium . Releasing from G1 block in glucose containing medium repressed the Gal1 promoter to maintain Cdc5 depletion . Bfa1-GFP was detected in the total cell extracts of the indicated strains using anti-GFP antibody . Anti-TAT1 antibody was used to detect tubulin as a loading control . Asterisks indicate the hyper-phosphorylated form of Bfa1 . A representative blot out of three independent experiments is shown . ( D ) Model for SPOC activation and Bfa1-SPB remodeling . When the mitotic spindle is correctly aligned , Bfa1 molecules are stably in contact with Nud1 and Spc72 , where Cdc5 can phosphorylate and thereby inactivate Bfa1 . When the spindle misaligns , Kin4 binds to Spc72 to phosphorylate Bfa1 ( Maekawa et al . , 2007 ) . Kin4 phosphorylated Bfa1 is recognized by Bmh1 ( Caydasi et al . , 2014 ) . Bmh1-bound Bfa1 disconnects from Spc72 but remains associated with Nud1 , although dynamically . Cdc5 cannot phosphorylate Bfa1 when Bfa1 is disconnected from Spc72 . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 03710 . 7554/eLife . 14029 . 038Figure 7—source data 1 . Raw and normalized mean fluorescence intensities of Bfa1-GFP at SPBs of SPC72 kar9∆ and SPC72-GBP kar9∆ cells with normally aligned or misaligned spindles ( source data for Figure 7A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 038 Spindle misalignment is observed in both SPC72 and KAR9 knock-out cells . However , in contrast to kar9∆ cells , SPOC deficient phenotypes ( multiple spindles and/or buds ) accumulate with a high frequency ( > 40% ) in spc72∆ cultures ( Figure 8A ) ( Hoepfner et al . , 2002 ) . This was somehow puzzling considering that the SPOC was functional in spc72∆ cells ( as shown above through the mitotic exit delay upon spindle misalignment ) ( Figure 6 ) . To understand why SPOC deficient phenotypes accumulated in these populations , we analyzed the behavior of spc72∆ GFP-TUB1 cells by live cell imaging in greater detail . We monitored the time from anaphase onset until spindle re-alignment . In a kar9∆ population most of the cells reoriented their spindle in less than 50 min after anaphase onset . In contrast , spindle mis-orientation persisted for longer than 50 min in the majority of spc72∆ cells ( Figure 8B ) . The spindles of spc72∆ cells that remained arrested in anaphase for an extended period ( > 50 min ) broke down as a consequence of mitotic exit ( Figure 8C and 6D ) . These cells re-entered the cell cycle to become binucleate cells ( Figure 8D ) . This indicated that the SPOC was unable to maintain an indefinite cell cycle arrest in this population of cells . Interestingly , in the majority of binucleated cells that entered anaphase ( 10 out of 15 cells ) , one spindle remained in the mother cell compartment while the second entered the bud ( Figure 8D , upper panel ) . In all such cells , the mother-bud oriented spindle and the misoriented spindle disassembled simultaneously , as the cells exited mitosis ( Figure 8D and E ) . This behavior was not due to binucleation per se , because cells harboring two mis-oriented spindles remained arrested in late anaphase with intact anaphase spindles ( Figure 8D , lower panel , and Figure 8E ) . This indicates that satisfaction of the SPOC by the correctly aligned spindle triggers a dominant signal that silences or bypasses the SPOC throughout the cell . This signal is likely to be a diffusible signal as the breakdown of the correctly aligned and misaligned nuclei occurs simultaneously . 10 . 7554/eLife . 14029 . 039Figure 8 . SPOC slippage in spc72∆ cells ( A ) Population analysis of GFP-TUB1 kar9∆ and GFP-TUB1 spc72∆ cells . Percentage of cells with single or multiple spindles is indicated in the left panel . Representative images of GFP-TUB1 spc72∆ cells are shown in the right panel . Fixed samples were used for analysis ( N>100 ) . Note that the presence of more than one GFP-TUB1 signal in cells is an indication of multiple nuclei . ( B ) Duration of spindle misalignment in kar9∆ GFP-TUB1 and spc72∆ GFP-TUB1 cells . Time-lapse series were used for analysis ( N: 99 for kar9∆ , N: 54 for spc72∆ ) . The time elapsed between the anaphase onset and spindle orientation in the mother to daughter direction was recorded as the time spent misaligned ( t ) . T=0 indicates that no spindle misalignment was observed in anaphase . ( C ) Representative images from time-lapse series of a GFP-TUB1 spc72∆ cell with a mis-aligned spindle that exited mitosis after prolonged arrest . Time point one ( min ) is the first time point after the start of the time-lapse . Please note that the cell depicted here had a misaligned anaphase spindle already at time point one . Spindle breakdown ( t=54 ) is an indication of mitotic exit . ( D ) Representative images from time-lapse series of GFP-TUB1 spc72∆ cells with two spindles . In the upper panel , one spindle stays misaligned while the other one re-aligns in the mother-daughter direction . In this cell , both spindles broke at t=19 , indicating mitotic exit . In the lower panel , both spindles stay misaligned during the course of the experiment . Mitotic exit was not observed in this cell . ( E ) Diagram depicting the fate of an spc72∆ cell with a misaligned spindle . Upon prolonged spindle misalignment , mitotic exit occurs ( checkpoint slippage ) . The binucleated cell enters a second cell cycle with two nuclei , each forming a spindle . The two spindles could both misalign ( lower panel ) or correctly align ( middle panel ) . These two events are rare ( graph in the right panel ) . Alternatively , one of the two spindles could stay misaligned while the other is correctly aligned ( upper panel ) . This occurs more frequently ( graph in the right panel ) . The frequencies of each three scenarios plotted on the right were calculated based on time-lapse analysis of GFP-TUB1 spc72∆ cells ( N: 15 cells ) . Arrowheads indicate the point of spindle breakage . Scale bar: 3 µm . Data represented is a collection of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 14029 . 039 In conclusion , inefficiency of spindle re-orientation , together with SPOC leakage after prolonged arrest , leads to the accumulation of binucleated cells in spc72∆ cells . Binucleated cells , on the other hand , have a higher probability of producing bi- or multinucleated cells in the next round of cell division , as one correctly aligned spindle is sufficient to disable SPOC signaling for both itself and its misaligned neighbor . Altogether these data explain why spc72∆ populations accumulate bi- and multiple-nuclei even though they are SPOC proficient . A positive FRET between two fluorophores only occurs when the fluorophores are in close apposition ( < 10 nm ) . This indicates that the fluorophore-fused proteins are in close proximity and likely to physically interact ( Zal and Gascoigne , 2004 ) . Interactions between SPB core components have been previously mapped by conventional FRET based on acceptor emission measurements ( sensitized emission FRET ) ( Muller et al . , 2005 ) . In comparison to sensitized emission FRET , acceptor photobleaching FRET yielded more robust measurements for Bfa1-Bub2 SPB interactions . Our FRET data show that the C-terminal domain of Bfa1 resides in close proximity to the C-termini of the γ-tubulin receptor protein Spc72 and the MEN scaffold protein Nud1 at the SPB outer plaque . Direct physical interaction of Nud1 and Bfa1 was already shown ( Gruneberg et al . , 2000 ) . In vitro binding assays performed here established that Bfa1 also associates with Spc72 . Thus , FRET interactions between Bfa1-Nud1 and Bfa1-Spc72 most likely reflect physical interactions at SPBs in vivo . C-terminally tagged Bfa1 also yielded positive FRET with N-terminally tagged Bub2 at SPBs , which is in agreement with the fact that Bfa1 and Bub2 bind to SPBs as a bipartite complex ( Caydasi and Pereira , 2009; Pereira et al . , 2000 ) . We could not detect FRET between the N-terminally tagged Bub2 and Nud1 ( or Spc72 ) . However we cannot exclude the possibility that Bub2 may interact with Nud1 or Spc72 by other means . We observed loss of Bfa1-Spc72 ( but not Bfa1-Nud1 ) FRET upon SPOC activation . The loss of FRET signal between Bfa1 and Spc72 might indicate the physical separation of the two proteins or only the separation of the fluorophores coupled to Bfa1 and Spc72 C-termini . Forced binding of Bfa1 to Spc72 resulted in loss of SPOC activity , suggesting that dissociation of the two proteins is important for checkpoint functioning . Our data show that Kin4 kinase and the 14-3-3 protein Bmh1 are involved in this process . Loss of Bfa1-Spc72 FRET interaction during spindle misalignment was only observed in KIN4 but not in kin4∆ cells . Furthermore , overexpression of KIN4 abolished the FRET signal between Bfa1 and Spc72 in cells with normal aligned anaphase spindles in a Bmh1-dependent manner . We propose that the remodeling of Bfa1-SPB binding is triggered by Kin4 phosphorylation of Bfa1 . As a direct consequence of this phosphorylation , Bmh1 binds to Bfa1 and breaks Bfa1-Spc72 proximity ( Figure 7D ) ( Caydasi et al . , 2014 ) . The reorganization of Bfa1-SPB binding may result from a possible Bfa1 conformational change triggered by recruitment of Bmh1 to Kin4 phosphorylated Bfa1 . Such regulation may resemble the conformational change that Mad2 undergoes upon activation of the spindle assembly checkpoint ( SAC ) ( Luo et al . , 2002; 2004; Mapelli et al . , 2007; Mapelli and Musacchio , 2007; Sironi et al . , 2002; Yang et al . , 2007 ) . Spc72 has been proposed to constitute a part of the SPOC mechanism that senses cytoplasmic microtubule failure and recruits Kin4 to SPBs in response to this defect ( Maekawa et al . , 2007 ) . Surprisingly , spc72∆ cells were SPOC proficient even though Kin4 neither localized at SPBs nor phosphorylated Bfa1 . However , Cdc5-dependent hyper-phosphorylation of Bfa1 , which inactivates the Bfa1-Bub2 GAP complex ( Geymonat et al . , 2003; Hu et al . , 2001 ) , was greatly reduced in the absence of SPC72 . The loss of Bfa1 phosphorylation by Cdc5 in the absence of Spc72 probably explains why the SPOC is still functional in these cells despite the lack of Kin4 phosphorylation of Bfa1 . Given that Cdc5 interacts with Spc72 and phosphorylates Bfa1 at the SPB ( Archambault and Glover , 2008; Maekawa et al . , 2007; Park et al . , 2004; Snead et al . , 2007 ) , it is tempting to speculate that Cdc5 only phosphorylates Bfa1 when in complex with Spc72 . In this case , disruption of the Bfa1-Spc72 interaction in cells with mis-oriented spindles might be a way to insulate the GAP complex from being inhibited by Cdc5 ( Figure 7D ) . We tested this model using BFA1-GFP SPC72-GBP cells where Spc72-Bfa1 interaction is maintained even during spindle misalignment . In support of our model , Bfa1-Spc72 tethering caused increased phosphorylation of Bfa1 by Cdc5 and hence SPOC failure . Interestingly , phosphorylation of Bfa1 by Kin4 was also observed in these cells . This indicates that Kin4 can only inhibit Bfa1 phosphorylation by Cdc5 if Bfa1 dissociates from Spc72 . This is in line with the observation that Bfa1 phosphorylated by Kin4 is still a substrate of Cdc5 in vitro ( Maekawa et al . , 2007 ) . We therefore propose that Spc72 acts as a scaffold protein that coordinates the regulation of the checkpoint effector Bfa1 by both Kin4 and Cdc5 kinases in cells with mis-aligned spindle . Biochemical and FRET data are consistent with the binding of Bfa1 to Nud1 and Spc72 . This raises the question of how Bfa1 associates with these proteins at the SPB . Bfa1 could bind separately or simultaneously to Nud1 and Spc72 . The analysis of spc72∆ cells exclude the possibility of a stable Bfa1-Spc72 and a dynamic Bfa1-Nud1 pool co-existing at SPBs since Bfa1 still associated with SPBs via Nud1 in a stable manner in spc72∆ cells with correctly aligned spindles . Interestingly , in spc72∆ cells with mis-aligned spindles , Bfa1 became dynamic , yet not as much as in wild type cells but similar to kin4∆ cells . This observation is important two-fold: First , it indicates that the Bfa1 binding site of Nud1 functions efficiently without Spc72 . Second , it suggests that the main function of Spc72 is not in Bfa1 binding to the SPB but instead in the SPB recruitment of Kin4 , which then regulates Bfa1 SPB dynamics through phosphorylation of Bfa1 as discussed above . Simultaneous binding of Bfa1 to Spc72 and Nud1 is a possibility considering the SPB structure , where Nud1 and Spc72 interact through their C-termini with each other and with Bfa1 ( Gruneberg et al . , 2000 ) . If this is the case , the binding affinity of Bfa1 to Nud1 must be higher than to Spc72 . This would explain why the loss of Spc72 does not significantly influence Bfa1 protein levels and binding dynamics at SPBs . However , alternative scenarios might also explain why Bfa1 has the same SPB binding behavior in wild type and spc72∆ cells with properly aligned spindles . For example , an adaptation mechanism that activates an additional SPB binding site for Bfa1 at SPBs may compensate for the loss of Spc72 or Bfa1 might not directly bind to Spc72 in vivo . We consider the latter possibility as unlikely because of the positive Bfa1-Spc72 FRET signal at the SPB , demonstrating the very close neighborhood of both proteins at this organelle . Moreover , we observed changes in Bfa1-Spc72 FRET signal that are consistent with previous dynamic measurements of Bfa1 in response to spindle alignment defects or KIN4 overexpression . In any case , further biochemical and biophysical studies will be necessary to evaluate the affinity of Bfa1 towards Nud1 and Spc72 , and to establish whether the same Bfa1 molecule can bind simultaneously to Nud1 and Spc72 . The Bfa1-Bub2 complex is recruited preferentially to the dSPB ( asymmetric binding ) in cells progressing normally through the cell cycle . How this asymmetry is established is still unclear . SPB duplication generates a newly formed SPB next to the old one ( Byers and Goetsch , 1974; Winey et al . , 1991 ) . Under normal growth conditions , the old SPB moves to the daughter cell body ( Juanes et al . , 2013; Pereira et al . , 2001 ) . Recently , it was shown that the old SPB binds more Spc72 molecules then the new one ( Juanes et al . , 2013 ) , raising the possibility that Spc72 could be the determinant of Bfa1 SPB binding asymmetry . However , our data indicate that Bfa1 asymmetry is maintained even when Spc72 is absent through stable binding to Nud1 . In addition , randomization of SPB inheritance ( through transient microtubule depolymerization ) did not alter the asymmetric association of Bfa1 with the dSPB ( Juanes et al . , 2013; Pereira et al . , 2001 ) . Furthermore , our FRET analyses show that , in cells with normal aligned spindles , Bfa1 is still in close proximity to Spc72 and Nud1 at both SPBs . We thus consider it as unlikely that the asymmetric SPB binding of Bfa1-Bub2 arises from association of Bfa1 with different receptors at mother and daughter SPBs . The asymmetry of Bfa1 at SPBs could arise from differential phospho-regulation at the poles . This regulation could be at the level of Bfa1 , Nud1 or Bub2 , which is also required for Bfa1 SPB asymmetry and subjected to phospho-regulation ( Hu and Elledge , 2002; Maekawa et al . , 2007; Park et al . , 2008; Rock et al . , 2013 ) . Phosphorylation could either stabilize Bfa1-Nud1 interaction at the dSPB or destabilize it at mSPB . Cdc5 has been previously suggested to preferentially target Bfa1 to the dSPB ( Kim et al . , 2012 ) . However , the fact that Bfa1 asymmetry was not disturbed in Cdc5-depleted cells challenges this view ( Caydasi and Pereira , 2009 ) . Existing data indicated that Kin4 does not contribute to the reduced levels of Bfa1 at the mSPB ( Caydasi and Pereira , 2009 ) . In addition , our data now shows that Bfa1 asymmetry is not disturbed in spc72∆ cells , where Cdc5 and Kin4 do not phosphorylate Bfa1 . These observations suggest that kinases other than Kin4 and Cdc5 are involved in Bfa1 asymmetry , if the regulation is at the level of phosphorylation . It has been proposed that Bfa1-Bub2 asymmetry is established through cytoplasmic microtubule-cortex interactions ( Caydasi and Pereira , 2009; Monje-Casas and Amon , 2009; Pereira et al . , 2001 ) or cell polarity determinants ( Monje-Casas and Amon , 2009 ) . Bfa1 asymmetry was maintained in spc72∆ cells , which have very short and unstable cytoplasmic microtubules ( Hoepfner et al . , 2002 ) . This implies that cells do not require an intact cytoplasmic microtubule cytoskeleton to establish and/or maintain Bfa1-Bub2 asymmetry . How cell polarity determinants control Bfa1 asymmetry is unclear . Daughter cell associated factors could stabilize the Bfa1-Nud1 interaction at the dSPB , for example by influencing the post-translational regulation of Bfa1-Bub2 or Nud1 . Our data show that SPOC is not an everlasting checkpoint . Slippage from SPOC can occur after prolonged mitotic arrest . In kar9∆ cells , this is a relatively rare event because the spindle is able to realign quickly in the majority of the cells due to the presence of the alternative dynein-dependent spindle alignment pathway ( Eshel et al . , 1993; Li et al . , 1993 ) . This situation is however different in spc72∆ cells . Lack of functional cytoplasmic microtubules impairs spindle realignment in the majority of the spc72∆ cells . As a consequence of this long cell cycle arrest , we frequently observed SPOC slippage and accumulation of multi-nucleated cells . Our data is consistent with a previous report that analyzed the cell cycle progression of binucleated yeast cells lacking cytoplasmic microtubules ( Sullivan and Huffaker , 1992 ) . By following spc72∆ cells that underwent SPOC slippage and became multinucleated , we observed a novel phenomenon in yeast mitotic exit: In cells with two misaligned anaphase spindles , realignment of one spindle was sufficient to trigger mitotic exit in both misaligned and correctly aligned nuclei . A similar phenomenon was described in binucleated budding yeast cells obtained by other means ( Falk et al . , 2016 ) . This observation indicates that a dominant mitotic exit signal is generated after spindle entry into the bud . This is likely a diffusible signal as it almost simultaneously affects the other nuclei that still reside in the mother cell compartment with a misaligned spindle . The passage of one nucleus into the bud may trigger mitotic exit via several possible SPOC bypassing and/or silencing mechanisms . Daughter cell specific factors , such as the putative Tem1 guanine nucleotide exchange factor Lte1 ( Adames et al . , 2001; Bardin et al . , 2000; Keng et al . , 1994; Pereira et al . , 2000; Shirayama et al . , 1994 ) could trigger the MEN to initiate mitotic exit in the bud . Active GTP-bound Tem1 would then fully activate the phosphatase Cdc14 that in turn counteracts and decreases the activity of mitotic cyclin-dependent kinase ( Cdk ) to promote mitotic exit and cytokinesis ( Bardin and Amon , 2001; Meitinger et al . , 2012 ) . In this case , Tem1 activation in the daughter cell followed by mitotic Cdk1 down-regulation would bypass the SPOC in the mother cell compartment ( SPOC bypassing mechanism ) . Interestingly , Lte1 is also a Kin4 inhibitor ( Bertazzi et al . , 2011; Falk et al . , 2011 ) , implying that bud specific factors may promote MEN activation through inhibition of SPOC components ( SPOC silencing ) . SPOC silencing can also occur at the level of Bfa1-Bub2 . Cdc5 was proposed to inhibit Bfa1-Bub2 GAP activity in cells with normal aligned spindle ( Geymonat et al . , 2002; 2003; Hu and Elledge , 2002 ) . However , it is likely that factors other than Cdc5 inactivate Bfa1-Bub2 in the daughter cell compartment . Two observations support this notion . We found that Bfa1-Bub2 was not efficiently phosphorylated by Cdc5 in spc72∆ cells , yet these cells did not have a delayed mitotic exit when their mitotic spindle was correctly aligned . In addition , a mutant form of Bfa1 ( Bfa1-11A ) , which cannot be phosphorylated by Cdc5 , does not cause a delayed mitotic exit ( Hu et al . , 2001 ) . Mathematical modeling also predicted the existence of an additional Bfa1-Bub2 inhibitory mechanism ( Caydasi et al . , 2012 ) . Indeed , the Cdc42 effector protein Gic1 was shown to directly bind to Bub2 and inhibit its interaction with Tem1 ( Hofken and Schiebel , 2004 ) . Our study thus re-invigorates the concept that cell polarity associated factors play a critical role in SPOC silencing and mitotic exit ( Hofken and Schiebel , 2002; 2004; Monje-Casas and Amon , 2009 ) . The molecular characterization of these factors , which were undermined in the last decade , will be critical to shed light onto the mechanisms controlling mitotic exit and/or SPOC silencing . Of importance , our data indicate that the mitotic exit stop-signal generated by the SPOC at the misaligned spindle cannot sustain the mitotic arrest if the checkpoint is satisfied in the neighboring spindle . Analogously , in mammalian cells with two mitotic spindles , the stop-anaphase signal generated by the SAC at one of the two spindles was not sufficient to sustain the checkpoint arrest if the checkpoint was satisfied at the neighboring spindle ( Rieder et al . , 1997 ) . Similar observations were made for G2-M checkpoint in binucleated plant cells ( del Campo et al . , 1997; Gimenez-Abian et al . , 2001 ) . Therefore , local satisfaction of checkpoints seems to be sufficient for global checkpoint silencing in binucleated cells . Interestingly , cytokinesis defects often generate binucleation , which is associated with aneuploidy and tumor formation . We postulate that misbalanced checkpoint integrity in binucleated cells might be a driver for genome instability and cancer development . All yeast strains and plasmids used in this study are listed in the Supplementary file 1 ( Baudin-Baillieu et al . , 1997; Caydasi et al . , 2010; 2014; Gruneberg et al . , 2000; Khmelinskii et al . , 2007; Knop and Schiebel , 1998; Maekawa et al . , 2007; Pereira et al . , 2001; Sikorski and Hieter , 1989; Surana et al . , 1993 ) . Yeast strains of ESM356-1 or YPH499 background are derivatives of S288C , strains of BMA64 background are isogenic with W303 . Basic yeast methods and growth media were previously described ( Sherman , 1991 ) . For live-cell imaging yeast strains were grown to log-phase in filter sterilized synthetic complete media ( SC ) at 30°C . Plasmid-bearing strains were grown in SC media lacking the selective amino acid . To cure yeast strains of URA3-based plasmids , cells were grown on 5-fluoroorotic acid ( 5FOA , 1 mg/ml ) containing agar plates . Expression from Gal1 promoter was induced by adding galactose ( 2% final concentration ) into log-phase yeast cultures grown in 3% raffinose-containing media . For repression of the Gal1 promoter ( i . e . Gal1-CDC20 and Gal1-CDC5 depletion ) , cells grown in raffinose and galactose containing medium were transferred into glucose containing medium . For depletion of Cdc5 , cells were arrested in G1 using alpha factor prior to their transfer in glucose containing medium . G1-phase arrest was achieved by adding 10 μg/ml synthetic alpha-factor ( Sigma , St . Louis , MO ) to the logarithmically growing ( log-phase ) culture and further incubation of the culture until >90% of cells formed mating projections . To induce microtubule depolymerization , 15 μg/ml nocodazole ( Sigma ) was added to the log-phase yeast cultures grown in yeast/peptone/dextrose medium with 0 . 1 mg/l adenine ( YPDA ) at 30°C until the mid-log phase ( 107 cells/ml ) . For protein immunoprecipitation , cell pellets were collected after 2 . 5 hr of incubation in nocodazole containing medium . kar9Δ cells were grown at 23°C until and shifted to 30°C for 2–4 hr prior the experiment . PCR-based strategy was used for gene deletions , tagging of yeast proteins with fluorophores and construction of Gal1-KIN4 strains ( Janke et al . , 2004; Knop et al . , 1999 ) . Plasmids to generate FRET fusions encoded monomeric versions of YFP and mTUR with A206K mutation ( Zacharias et al . , 2002 ) . mCherry-TUB1 , Gal1-clb2∆DB , GFP-TUB1 were integrated in the yeast genome by homologous recombination using yeast integration plasmids ( Caydasi et al . , 2010 ) . N terminal tagging of proteins with Gal1-mTUR or Gal1-EYFP using the plasmids pYG2 and pYG3 resulted in only moderate expression of the corresponding protein , as the levels of Gal1-mTUR or Gal1-EYFP tagged proteins were only slightly increased in comparison to endogenous levels of the corresponding proteins ( data not shown ) . For FRAP , time lapse microscopy and FRET analysis , cells grown in filter sterilized growth media were imaged after attachment of the cells on glass-bottom dishes ( MatTek , Ashland , MA ) or on 96-well glass-bottom microwell plates ( Matrical Bioscience MGB096-1-2-LG , Spokane , WA ) coated with 6% concanavalin A-Type IV ( Sigma ) . Cells were fixed with 70% Ethanol prior to nuclear staining using 1 µg/ml 4' , 6-diamidino-2-phenylindole ( DAPI , Sigma ) in PBS . FRAP experiments , time lapse analysis of GFP-TUB1 expressing cells and still images of Kin4-GFP and Bfa1-GFP were performed using DeltaVision RT system with softWoRx software ( Applied Precision , Issaquah , WA ) equipped with a camera ( Photometrics CoolSnap HQ; Roper Scientific , Tucson , AZ ) . Images were acquired with 2 x 2 binning on a 100x UPlanSAPO objective with a 1 . 4 NA ( Olympus , Tokyo , Japan ) and a Mercury arc light source . For time-lapse movies , 12 z-stacks of 0 . 3 μm optical section spacing were taken at each time point . Movies were taken for 90 min with 1 min time interval . The z-stacks were sum-projected using SoftWoRx software before image analysis . Images for other microscopy experiments were acquired using a Zeiss Axiophot microscope equipped with a 100x NA 1 . 45 Plan-Fluor oil immersion objective ( Zeiss , Jena , Germany ) , Cascade 1K CCD camera ( Photometrics , Tucson , AZ ) and MetaMorph software ( Universal Imaging Corp . , Chesterfield , PA ) . Fluorescence intensity measurements , FRAP experiments and FRAP data analysis was as described in Caydasi and Pereira ( 2009 ) . The protocol for FRET analysis is explained in the following section thoroughly . Strains for FRET were grown in Synthetic Defined Low Fluorescence media prepared with 6 . 9 g/l yeast nitrogen base lacking folic acid and riboflavin ( SD LoFluo , Formedium , Norfolk , UK ) supplemented with all standard amino acids ( Sherman , 1991 ) and 2% glucose or 3% raffinose and 2% galactose . FRET microscopy was performed using a wide-field microscope ( Olympus ) equipped with a 150 W Mercury-Xenon arc burner . Images were acquired with a 60x UPLFLN air objective ( N/A=0 . 9; Olympus ) and on Hamamatsu C9100 EM-CCD camera ( Hamamatsu , Japan ) . EM gain was set to 165 in all experiments . Emission from mTUR was recorded from one plane after focusing at the DIC channel . Fluorescence of mTUR was detected from 417 nm to 451 nm before and after photobleaching of the EYFP . EYFP photobleaching was induced by a 4 s laser pulse of 100 mW 515 nm solid-state laser ( Cobolt , Sweden ) . Images were acquired in the order as presented in Figure 1—figure supplement 1B . Images were analyzed using ImageJ software by measuring mean fluorescence intensities of mTUR signals around SPBs from the part of the image where the EYFP signal was bleached as depicted in Figure 1—figure supplement 1B . Fluorescence intensities were corrected for the background and the FRET efficiency was calculated as the percentage increase in corrected mTUR signal after photobleaching of EYFP ( Kentner and Sourjik , 2009 ) . FRET efficiency values were finally corrected for the median value of the donor-only control ( corresponding reference strain with mTUR fusion only , Figure 1—figure supplement 1D ) . To determine if FRET efficiency reflects energy transfer , FRET results for the pair and corresponding negative control were compared using the two-tailed Student t-test . p<0 . 01 was considered statistically significant . Yeast protein extracts and immunoblotting were performed as described in Janke et al . , 2004 . Antibodies were rabbit anti-GFP antibody , goat anti-GST ( GE Healthcare , Waukesha , WI ) , mouse anti-His antibody ( GE Healthcare ) , rabbit anti-P-S180 ( Maekawa et al . , 2007 ) , mouse anti-TAT1 ( Sigma ) and rabbit anti-Kin4 ( lab collection , see the next section ) . The anti-GFP antibody used in Figure 7 was a gift from E . Schiebel ( ZMBH , University of Heidelberg , Germany ) whereas the anti-GFP antibody used in other experiments was a gift from J . Lechner ( BZH , University of Heidelberg , Germany ) . Secondary antibodies were goat anti-mouse , goat anti-rabbit or rabbit anti-goat IgGs coupled to horseradish peroxidase ( Jackson ImmunoResearch Laboratories Inc , West Grove , PA ) . Rabbit polyclonal antibodies were generated against bacterially purified 6His-Kin4 lacking the last 50 amino acids ( PSL GmbH , Heidelberg , Germany ) . Antibodies were purified from preabsorbed serum by affinity purification using the immobilized antigen . MBP-Bfa1 and GST-Spc72 were purified from Escherichia coli as described previously ( Geymonat et al . , 2009; Maekawa et al . , 2007 ) . 6His-Spc72231-622 ( Spc72-C ) and 6His-Mlc1 were purified from E . coli ( BL21 ) according to the manufacturer instructions ( EMD Biosciences , San Diego , CA ) . In vitro binding assays were performed as following: Purified GST-Spc72 , 6His-Spc72-C , 6His-Mlc1 were incubated with bacterially purified MBP-Bfa1 or MBP bound amylose beads ( New England Biolabs , Ipswich , MA ) in PBS at 4°C for 45 min . MBP-Bfa1 and MBP–bound beads were then washed four times with PBS containing 0 . 75% Nonidet P-40 , and samples were heated at 65°C for 15 min in HU-DTT buffer ( 200 mM Tris-HCl , pH 6 . 8 , 8 M urea , 5% SDS , 0 . 1 mM EDTA , 0 . 005% bromophenol blue , and 15 mg/ml DTT ) before loading on SDS–PAGE gels . Immuprecipitation of Bfa1-GFP was done using a cell pellet derived from a 200 ml cell culture ( 2 × 107 cells/ml ) treated with nocodazole until >95% of the cells were arrested in metaphase . Cells were lysed using acid-washed glass beads in a FastPrep FP120 Cell Disturber ( MP Biomedicals , Irvine , CA ) . Lysis buffer contained 50 mM Tris-HCl buffer ( pH 7 . 5 ) , 150 mM NaCl , 5% glycerol , 350 μg/ml benzamidine , 100 mM β-glycerophosphate , 50 mM NaF , 5 mM NaVO3 and complete EDTA-free protease inhibitor cocktail ( Roche , Basel , Switzerland ) . Lysates were incubated with Triton X-100 ( 1% final concentration ) for 15 min at 4°C . Total extracts were cleared by centrifugation at 10 , 000 g for 20 min at 4°C . Bfa1-GFP was precipitated from total cell extracts using GBP conjugated with Sepharose 4B ( Lin et al . , 2014 ) ( gift from E . Schiebel ) . Phosphorylation of the immunoprecipitated Bfa1-GFP at S180 residue was detected by immunoblotting using anti-P-S180 antibodies . On the same immunoblot , the level of immunoprecipitated Bfa1-GFP was detected using anti-GFP antibodies .	A cell must duplicate its genetic material and then separate the two copies before it divides . This process is carefully controlled so that each new cell receives an identical set of chromosomes after cell division . In budding yeast , new daughter cells grow as a bud on the side a larger mother cell and are eventually pinched off . A surveillance mechanism in budding yeast monitors the placement of the molecular machine ( called the spindle ) that separates the copies of the chromosomes . This mechanism then stops the cell from dividing if the spindle is not positioned correctly . Many of the components of this surveillance mechanism – which is called the spindle position checkpoint – associate with structures at the ends of the spindle . However , it was not clear how these components do this and how it helps them to check if the spindle is positioned correctly . Now , Gryaznova , Caydasi et al . have used a technique called FRET to answer these questions for an important component of the spindle position checkpoint , a protein called Bfa1 . The main advantage of FRET is that it can be used to monitor changes in protein-protein interactions in living cells . This approach identified two proteins that provide sites for Bfa1 to bind to at the ends of the spindle . The experiments also showed that Bfa1 specifically detaches from one of these proteins ( called Spc72 ) when the spindle position checkpoint is activated . This action keeps Bfa1 ( and therefore the spindle position checkpoint ) active , which in turn stops the cell from starting to divide . Further experiments then showed that Spc72 acts like a regulatory hub that controls Bfa1’s activity . This allows an as-yet unidentified mechanism to coordinate cell division with the position of the spindle . The findings of Gryaznova , Caydasi et al . also suggest that unknown factors switch off the spindle position checkpoint when the spindle is correctly positioned to allow the cell to divide . Future work could now aim to identify the mechanism and the unknown factors . Finally , in a related study , Falk et al . show that the spindle position checkpoint is inactivated when one end of the spindle is moved out of the mother cell and into the bud .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	A FRET-based study reveals site-specific regulation of spindle position checkpoint proteins at yeast centrosomes
Hydroxychloroquine and chloroquine are used extensively in malaria and rheumatological conditions , and now in COVID-19 prevention and treatment . Although generally safe they are potentially lethal in overdose . In-vitro data suggest that high concentrations and thus high doses are needed for COVID-19 infections , but as yet there is no convincing evidence of clinical efficacy . Bayesian regression models were fitted to survival outcomes and electrocardiograph QRS durations from 302 prospectively studied French patients who had taken intentional chloroquine overdoses , of whom 33 died ( 11% ) , and 16 healthy volunteers who took 620 mg base chloroquine single doses . Whole blood concentrations of 13 . 5 µmol/L ( 95% credible interval 10 . 1–17 . 7 ) were associated with 1% mortality . Prolongation of ventricular depolarization is concentration-dependent with a QRS duration >150 msec independently highly predictive of mortality in chloroquine self-poisoning . Pharmacokinetic modeling predicts that most high dose regimens trialled in COVID-19 are unlikely to cause serious cardiovascular toxicity . Chloroquine and hydroxychloroquine are closely related 4-aminoquinoline drugs used for the treatment of malaria , amoebiasis and rheumatological conditions ( World Health Organization , 2015 ) . Until the late 1990s chloroquine was the drug of choice for the treatment of malaria . Hundreds of tons corresponding to over 200 million doses were consumed annually . Since then its use has declined because of widespread resistance in Plasmodium falciparum . Today there is greater use of hydroxychloroquine in rheumatoid arthritis and discoid and systemic lupus erythematosus . Chloroquine and hydroxychloroquine are antivirals with a broad range of activities ( including flaviviruses , retroviruses , and coronaviruses ) ( Savarino et al . , 2003; Liu et al . , 2020; Wang et al . , 2020; Yao et al . , 2020; Mégarbane and Scherrmann , 2020 ) . Their antiviral activity against SARS-CoV2 is expected to be weak as predicted unbound concentrations of hydroxychloroquine and chloroquine in vivo with currently recommended doses are lower than the half-maximal effect concentrations reported in Vero cell cultures ( White et al . , 2020 ) . This has motivated trialling of higher loading and maintenance doses than usually given in malaria or hepatic amoebiasis ( World Health Organization , 2015; Conan , 1949 ) . As of the 11th June 2020 , 77 interventional clinical trials evaluating these drugs in both the prevention and treatment of COVID-19 were registered and recruiting participants on ClinicalTrials . gov . Figure 1 shows a scatter plot of the duration versus total dose in base equivalent for the 55 trials using chloroquine or hydroxychloroquine for the treatment of hospitalised COVID-19 patients for which exact dosing could be extracted . The doses and durations varied greatly . The high doses used by some trials have caused concern over the potential for cardiovascular toxicity . Nevertheless , despite the lack of any convincing evidence of clinical benefit , these drugs are now being used extensively in COVID-19 prevention and treatment across the world . Sadly , patients who need hydroxychloroquine or chloroquine for the treatment of malaria and rheumatological conditions are often having difficulty obtaining them ( Straits Times , 2020 ) . Borba et al . reported results from a randomised trial in Brazil of two dose regimens of chloroquine in COVID-19 treatment ( CloroCovid-19 study , ClinicalTrials . gov NCT04323527 ) ( Borba et al . , 2020 ) . The trial was stopped early after recruiting only 81 patients because of cardiac toxicity and greater mortality in the higher dose group; two of 41 patients given the higher dose regimen developed ventricular tachycardia before death and 7 of 37 patients in this group developed electrocardiograph QTcF intervals over 500 msec compared with 4 of 36 in the lower dose group . The higher dose regimen , comprising 600 mg base chloroquine administered twice daily for ten days , is substantially higher than recommended in malaria or rheumatological conditions ( World Health Organization , 2015 ) and represents the standard malaria loading dose repeated 19 times at 12 hr intervals . Chloroquine is dangerous in overdose and has been used extensively for suicide . High concentrations of chloroquine cause hypotension , arrhythmias , coma , acute respiratory distress syndrome ( ARDS ) , and fatal cardiac arrest ( Riou et al . , 1988; Clemessy et al . , 1995; Clemessy et al . , 1996; Mégarbane et al . , 2010 ) . In Zimbabwe the mortality of chloroquine in overdose was approximately six times higher than for other drugs ( Ball et al . , 2002 ) . In France in the early 1980s there was a suicide epidemic ( Riou et al . , 1988; Clemessy et al . , 1995; Clemessy et al . , 1996; Mégarbane et al . , 2010 ) following the publication of a book entitled Suicide: mode d’emploi ( Suicide: a how-to guide ) ( Guillon and Le Bonniec , 1982 ) . This unfortunate experience allowed characterization of the factors associated with death from chloroquine overdose . Outcome in chloroquine self-poisoning depends on the dose ingested , the delay in reaching hospital and the quality of intensive care support . Using the blood concentration measurements taken from self-poisoning patients managed by experienced intensivists in the French National referral centre allowed development of a pharmacokinetic-pharmacodynamic model to estimate the relationship between chloroquine dosing and a fatal outcome . A pooled analysis using electrocardiograph QRS interval data from healthy volunteers ( Pukrittayakamee et al . , 2014 ) and from the French self-poisoning cohorts helped characterise the contribution of QRS prolongation to the recorded QT prolongation . These models were used to estimate the safety of the main treatment regimens currently used in COVID-19 clinical trials and of the standard malaria treatment regimen for comparison . We pooled individual patient whole blood chloroquine + desethylchloroquine concentrations and outcomes from three large prospectively studied hospital self-poisoning cohorts ( n = 302 , Figure 2 , top panel ) ( Riou et al . , 1988; Clemessy et al . , 1995; Clemessy et al . , 1996; Mégarbane et al . , 2010 ) . All the patients included in this analysis were treated in the same hospital in Paris and all had whole blood concentrations measured on admission . The overall mortality was 11% ( 33 of 302 ) . Of the patients with multiple concentration measurements ( n=173 ) , 61 ( 35% ) reached their peak whole blood chloroquine concentrations after admission . The distribution of differences on the log scale between admission whole blood concentrations and peak concentrations in patients who peaked after hospital admission was used to estimate a correction term for all other patients . Bayesian logistic regression was used to estimate the relationship between mortality and inferred whole blood peak chloroquine concentrations ( Figure 2 , bottom panel ) specifically adjusted for the desethyl metabolite levels and for the non-observed peak concentrations in 241 ( 80% ) of the patients . The model also adjusted for differences in treatment received between cohorts . The whole blood chloroquine concentration associated with 1% mortality was 13 . 5 µmol/L ( 95% C . I . 10 . 1 to 17 . 7 ) . 1% is the lowest mortality for which a reliable estimate can be derived from these data . The 1% threshold value was then used to evaluate the risk of fatal toxicity in chloroquine treatment regimens under evaluation in COVID-19 . We used two pharmacokinetic models to predict peak chloroquine concentrations for six chloroquine regimens . Five of the simulated regimens are potential COVID treatment regimens , and one is the standard malaria treatment regimen . A very wide range of chloroquine and hydroxychloroquine regimens are currently in use for the treatment of COVID-19 ( Figure 1 ) . The COVID-19 regimens we simulated correspond to the highest chloroquine treatment regimen ( in terms of total dose ) registered at ClinicalTrials . gov ( Borba et al . , 2020 ) ; the regimen given in the World Health Organization SOLIDARITY trial ( NCT04330690 ) ; a lower 7 day adaptation of the SOLIDARITY regimen; and weight-optimised equivalent regimens . The pharmacokinetic model estimated from whole blood concentrations ( Höglund et al . , 2016 ) predicted a slightly wider range of values for maximum chloroquine concentrations ( Cmax ) than the plasma-based model ( Appendix 3 shows the predicted Cmax distributions for both models in a 70 kg adult ) . Assuming a whole blood to plasma concentration ratio of 4 , the whole blood-based model and the plasma-based model predicted approximately identical median Cmax values . Thus , the whole blood pharmacokinetic model provides higher estimates for fatal toxicity because it predicts a larger variance for the Cmax distribution and hence more extreme outliers . As a sensitivity analysis , the equivalent outputs for the plasma-based model are shown in Appendix 4 . Based on the whole blood model , the median Cmax following a chloroquine dose of 600 mg base equivalent given twice daily for ten days to a 70 kg adult is 10 . 7 µmol/L ( Figure 3 , panel A ) . This is more than three times higher than the median Cmax for the chloroquine malaria treatment regimen at the same body weight . Approximately 60% of 70 kg adults continuing to receive the twice daily 600 mg dose have blood concentrations which rise into the ‘danger zone’ defined as whole blood concentrations above 10 µmol/L . In comparison , only two in a thousand 70 kg adults receiving the twice daily 310 mg base maintenance dose ( as given in the SOLIDARITY and RECOVERY trials ) for ten days would reach concentrations above 10 µmol/L ( approximately the lower bound for the 1% mortality threshold ) . Taking into account the uncertainty around this threshold value , the model estimates that fewer than one per thousand receiving the twice daily 310 mg dose would have peak concentrations above the 1% mortality threshold ( Figure 3 , panel B ) . Clearly body weight is an important determinant of exposure if doses are not adjusted for weight , as is common with oral administration ( chloroquine and hydroxychloroquine tablets usually contain 155 mg base ) . At most , two in a thousand 40 kg adults receiving the 3 day malaria treatment regimen without weight adjustment ( i . e . an unusually high mg/kg dose ) are predicted to reach peak concentrations above 10 µmol/L . Coupling the pharmacokinetic model with the pharmacodynamic model , we estimated the expected weight-dependent fatality ratios for the five COVID-19 chloroquine treatment regimens , truncating the concentration-fatality curve at the 1% mortality threshold ( only taking into account concentrations going above this threshold ) . Administering 600 mg base equivalent of chloroquine phosphate twice daily for ten days as in the CloroCovid-19 trial is predicted to result in absolute mean fatality ratios ranging between 0 . 05% ( 90 kg , 95% credible interval ( C . I . ) , 0% to 0 . 3% ) and 3 . 5% ( 40 kg , 95% C . I . 1 . 0% to 8 . 0% ) , Figure 3 , panel C . In comparison , only the flat dosing regimens are predicted to result in fatality ratios greater than one per thousand , and then only for weights less than 55 kg: for a 40 kg adult the 10 day SOLIDARITY regimen could result in 0 . 2% mortality ( 95% CI , 0 to 0 . 8 ) . The 10 day flat regimen ( without a loading dose ) of 500 mg chloroquine phosphate salt ( 310 mg base equivalent ) twice daily was recommended by the Health Commission of Guangdong Province for adults weighing more than 50 kg ( Multicenter collaboration group of Department of Science and Technology of Guangdong Province and Health Commission of Guangdong Province for chloroquine in the treatment of novel coronavirus pneumonia , 2020 ) . This is not predicted to incur a significant risk of cardiotoxicity . Overall the model predictions support this body weight threshold for dose reduction . We pooled individual electrocardiograph QRS intervals and chloroquine concentration data from 16 healthy volunteers ( Pukrittayakamee et al . , 2014 , for each individual the QRS interval was measured twice at rest and 12 times in presence of detectable plasma chloroquine concentrations after a single 620 mg base oral dose of chloroquine phosphate ) , and 290 self-poisoning patients ( no QRS data from Riou et al . , and one missing QRS measurement from the Clemessy cohort ) , Figure 4 . We fitted a hierarchical Bayesian Emax sigmoid regression model to the paired concentration-QRS data and the steady state QRS data ( healthy volunteers only ) , a total of n=514 datapoints . A whole blood chloroquine concentration of 3 µmol/L ( usually observed in malaria treatment ) is associated with a slight QRS prolongation of 6 . 7 msec ( 95% credible interval , 5 . 5–7 . 8 ) . Clinically significant intraventricular conduction delay ( QRS prolongation ) , resulting in durations greater than 150 msec , was strongly associated with observed chloroquine concentrations above 10 µmol/L . In self-poisoning , 1 out of 108 patients with admission whole blood concentrations less than 10 µmol/L had a QRS interval longer than 150 msec ( this accounts for the bias term estimated in the Clemessy cohort ) . In comparison , QRS durations longer than 150 msec were recorded in 36 out of 182 ( 20% ) of the patients with whole blood concentrations greater than 10 µmol/L . In self-poisoning patients , QRS duration was an independent predictor of death ( adjusted odds ratio for death of 1 . 3 for each 10 msec increase in QRS , p<10-5 ) . Chloroquine and hydroxychloroquine are already being used extensively , and often in high doses , to prevent and treat COVID-19 despite the current lack of convincing evidence of benefit . Use of these drugs has become extremely politicised . The first large randomised trial in the treatment of hospitalised patients has just reported ( RECOVERY , NCT04381936: press release only at the time of writing , results are not yet peer reviewed ) . This result strongly suggests that hydroxychloroquine ( and also lopinavir-ritonavir ) does not benefit hospitalised patients . The preliminary results of the same trial show that dexamethasone has a major life saving benefit in patients receiving oxygen or being ventilated . This supports a paradigm of early benefit from an effective antiviral and late benefit from an anti-inflammatory drug . Thus there remains a potential role for chloroquine and hydroxychloroquine in prevention and early treatment . However the prevention trials will not report for many months hence . Confusion over the risks associated with chloroquine and hydroxychloroquine has been compounded by a recent high profile claim of increased mortality and high rates of ventricular arrhythmia in a very large observational data set . This data set was almost certainly fabricated ( and the paper was subsequently retracted , Mehra et al . , 2020 ) . Unfortunately over-reactions by some regulatory agencies , unjustified extrapolations of risk from QT measurements , failure to distinguish the effects of the drugs given alone from the combination with azithromycin , and misunderstanding of their clinical pharmacology have conspired to reduce confidence in these drugs and jeopardise the very trials needed to characterise risks and benefits in the prevention and treatment of COVID-19 ( White et al . , 2020 ) . Chloroquine and hydroxychloroquine have unusual pharmacokinetic properties with very large total apparent volumes of distribution ( Vd ) ( chloroquine > hydroxychloroquine ) and very slow terminal elimination rates ( terminal half-lives exceed one month ) which cannot be measured accurately . Thus , distribution processes , rather than elimination , govern the blood concentration profiles in the first days following the start of treatment . Large single doses ( as in self-poisoning ) are very dangerous because they result in high blood concentrations as the drugs distribute out from a central ‘compartment’ that is hundreds of times smaller than the total apparent Vd ( compartmental modeling of chloroquine pharmacokinetics provides only an approximation of the distribution processes ) . These high concentrations can cause potentially lethal cardiovascular and nervous system toxicity . Chloroquine affects cardiac muscle depolarization , repolarization and contractility and also causes vasodilation . There is concentration-dependent electrocardiograph QRS widening and JT prolongation ( Clemessy et al . , 1995; Clemessy et al . , 1996 ) which both contribute to QT prolongation . Death usually results from refractory hypotension or ventricular fibrillation . Self-poisoning with chloroquine has provided an unfortunate opportunity to correlate drug exposure with the risk of iatrogenic death . As chloroquine has relatively weak antiviral activity , if there is benefit in COVID-19 infections , it is likely to require high drug exposures . At the beginning of the COVID-19 pandemic a large number of investigators started studies of chloroquine and hydroxychloroquine . The toxicity thresholds were not well defined and a range of dose regimens were evaluated in the hope that benefits would exceed risks ( Figure 1 ) . The high dose chloroquine regimen evaluated by Borba et al . in Brazil did encounter cardiac toxicity . This study referenced the recommendations by the Health Commission of Guangdong Province based on the Chinese experience , but there may have been confusion between salt and base weights ( Multicenter collaboration group of Department of Science and Technology of Guangdong Province and Health Commission of Guangdong Province for chloroquine in the treatment of novel coronavirus pneumonia , 2020 ) . Whereas the Chinese authorities recommended 500 mg salt twice daily ( two tablets of chloroquine phosphate 250 mg , comprising 155 mg base each ) , the Brazil study ( Borba et al . , 2020 ) gave doses in base equivalent ( 600 mg base 12 hourly ) which were consequently much higher . Importantly the patients in the Brazilian study also received azithromycin ( which also prolongs the QT interval , and this may well have contributed to ventricular arrhythmias , Saleh et al . , 2020; Lane et al . , 2020 ) . This pharmacometric analysis of the French self-poisoning cohorts provides an evidence-based toxicity threshold . It suggests that a continued regimen of 600 mg twice daily can be directly lethal . Lower dose regimens such as those evaluated in the large RECOVERY and SOLIDARITY trials , and the original Guangdong recommendations , are predicted to be safe . Doses of chloroquine alone , resulting in peak concentrations greater than 10 µmol/L are associated with more than 1% risk of fatal toxicity . There are several limitations to this study . It is a retrospective individual patient data analysis . In the suicide attempts , other drugs or alcohol were often taken as well , although none with the acute lethal toxicity of chloroquine . Patients were managed by experienced intensivists on intensive care units where there was close clinical and laboratory monitoring . Mortality might be higher in less well supported settings , or in overloaded hospitals in high-income settings . The age range in self-poisoning is also younger than in the majority of more seriously ill COVID-19 patients . The spectrophotometric assay method does not separate chloroquine from its desethylated metabolite , and is relatively insensitive . Desethychloroquine has generally similar biological properties , and the assay performs well at the high concentrations of relevance to this study . We corrected for the presence of the metabolite under the Bayesian model , but this correction increases uncertainty around the threshold concentration associated with 1% mortality . The predictions of absolute mortality under the regimens simulated in this work are sensitive to the parameterization of the pharmacokinetic model . Cmax is not observed directly but is an output model-based quantity estimated from data . There are no large population pharmacokinetic studies to verify the precision of the Cmax predictions , especially for the critical upper tails of the distribution . The pharmacokinetic-pharmacodynamic model developed here was based on the largest prospective series of chloroquine self-poisonings studied in a single referral centre . However , it was not possible to apply a standard pharmacokinetic model to the observed data in order to estimate the individual Cmax values . Mainly because of vomiting , admission blood concentrations only weakly correlate with the self-administered chloroquine doses ( Clemessy et al . , 1995 ) . Our model was based on a majority of admission whole blood concentrations . The data from patients whose peak drug concentrations occurred after hospital admission were used to approximate the difference between unobserved peak concentrations and admission concentrations in the remaining patients . Overall , the admission concentrations were estimated to be approximately 5% lower on average than the peak concentrations . This corresponds to prior expectations: the mean interval to hospitalization for this large self-poisoning cohort was 4 . 5 hr compared with an average time to peak concentration after oral administration of 3 hr ( range 1–6 hr ) ( Pukrittayakamee et al . , 2014 ) . The outcome of chloroquine poisoning depends on the quality of intensive care support . Over the period of this retrospective review , in addition to the experience gained in this one referral centre , there were significant improvements in intensive care which would have improved the prognosis for a given drug exposure . Sustaining the cardiorespiratory system mechanically with extracorporeal membrane oxygenation ( ECMO ) ( Mégarbane et al . , 2007 ) while toxic concentrations decline as the drug distributes may have made a significant contribution to improved survival . One death occurred in a 32-year-old female who took 2 . 8 grams of chloroquine phosphate and whose admission whole blood chloroquine+desethychloroquine concentration was 9 . 8 µmol/L . She had homozygous sickle cell disease and an admission hemoglobin of 7 . 9 g/dL . She was treated with ECMO and blood transfusion . She died from hemorrhagic ARDS even though her cardiac function improved . It is likely that acute and chronic complications of sickle cell disease contributed significantly to the fatal outcome associated with chloroquine overdose . Our analysis does not make specific predictions for the toxicity of hydroxychloroquine regimens currently in use for the treatment of COVID-19 . The majority of clinical trials are evaluating hydroxychloroquine , not chloroquine , as it is considered to be slightly safer and is more widely available in the countries where most trials are being conducted . Although hydroxychloroquine has a wider therapeutic margin in experimental animals ( McChesney , 1983 ) , it has generally similar pharmacokinetic properties and , although there are limited data , the pharmacodynamic properties are also generally similar to chloroquine . The clinical features of hydroxychloroquine and chloroquine in overdose are also similar . But unlike chloroquine , there are no large cohorts of hydroxychloroquine self-poisoning on which to define toxicity thresholds . Thus , given available toxicity data , it is very unlikely that equivalent hydroxychloroquine concentrations to chloroquine are more dangerous , but rather that the safety margins are wider . Our systematic review of regimens currently being used for the treatment of COVID identified one trial currently recruiting patients ( PATCH , NCT04329923 ) which is administering very high doses of hydroxychloroquine ( Figure 1 ) . This dose of 1200 mg/day ( 930 mg base/day ) for 14 days would be predicted to have a significant risk of incurring dangerous toxicity were hydroxychloroquine and chloroquine to have molar equivalent toxicities . However the majority of regimens currently being tested for COVID-19 treatment are predicted to be safe . Concerns over QT prolongation dominate the reasons for not endorsing or not participating in COVID 19 trials with chloroquine or hydroxychloroquine . Many recent articles on COVID-19 therapeutics caution against the risks of iatrogenic ‘torsade de pointes’ ( TdP ) with these drugs ( Pastick et al . , 2020; Monzani et al . , 2020; Chary et al . , 2020; Guzik et al . , 2020; Guastalegname and Vallone , 2020; van den Broek et al . , 2020; Jeevaratnam , 2020; Sapp et al . , 2020 ) . Both drugs do cause significant QT prolongation and , whilst they block several different cation channels affecting myocardial depolarization ( resulting in QRS prolongation ) and repolarization ( resulting in JT prolongation ) , the multichannel block is considered unbalanced , and both drugs are regarded as ‘torsadagenic’ ( Vicente et al . , 2018 ) . Many of the recent reports rediscovering this well known effect omit description of the contribution of QRS widening to QT prolongation ( and thereby overestimate JT prolongation ) . Approximately 350 tons of hydroxychloroquine are manufactured each year , almost all of which is used for rheumatological conditions , so millions are receiving long term regimens . Many of these patients have systemic lupus erythematosus which is commonly associated with myocarditis , QT prolongation , and a high prevalence of arrhythmias ( Gawałko et al . , 2020; Bourré-Tessier et al . , 2015 ) . Reviews and authoritative websites all state that hydroxychloroquine causes TdP . Yet , despite this enormous use , in a systematic review of the literature we can find only a single reported case of TdP outside of COVID-19 treatment . This was in a 67-year-old Taiwanese female with systemic lupus erythematosus , cirrhosis and a previous myocardial infarct who was receiving hydroxychloroquine ( Chen et al . , 2006 ) . The patient survived . The WHO pharmacovigilance database ( VigiBase ) contains reports of 83 episodes of ventricular tachycardia associated with hydroxychloroquine over a 52 year period of which seven were fatal ( this does not distinguish acute toxicity from the well described chronic toxic cardiomyopathy associated with use over many years ) . Overall , in systemic lupus erythematosus , patients receiving chloroquine have fewer arrhythmias than those on other medicines ( Teixeira et al . , 2014; Wozniacka et al . , 2006 ) . In a very large retrospective observational study of 956 , 374 patients with rheumatoid arthritis starting hydroxychloroquine , there were significantly less arrhythmias in the first month of treatment with hydroxychloroquine compared to patients ( N = 310 , 350 ) started on sulphasalazine ( Lane et al . , 2020 ) . This reflects the under-recognised anti-arrhythmic effects of these drugs ( Harris et al . , 1988 ) . In COVID-19 treatment there have been two reports so far of TdP . An 84-year-old Israeli woman with metastatic breast cancer who was taking bisoprolol , letrozole and memantine and received chloroquine treatment ( Szekely et al . , 2020 ) , and a 68-year-old American male given hydroxychloroquine and azithromycin ( Chorin et al . , 2020 ) . Neither of these episodes were fatal . In Europe and the USA , where chloroquine and hydroxychloroquine are allowed for compassionate use , there have been more than 300 , 000 deaths from COVID-19 . The apparently low rates of reported ventricular arrhythmia are surprising given that evidence of myocarditis in lethal COVID-19 is common ( Kang et al . , 2020; Long et al . , 2020 ) . The two largest randomised trials in hospitalised COVID-19 patients ( RECOVERY and SOLIDARITY ) have together randomised 2473 patients to receive hydroxychloroquine . They do not show increased mortality at the times expected if hydroxychloroquine was causing lethal cardiovascular toxicity ( i . e . on the first and last days of treatment ) . Except in deliberate self-poisoning TdP has not been reported with chloroquine either , despite billions of malaria treatments and extensive use in rheumatological conditions and hepatic amoebiasis for over 70 years . Thus it appears that the individual risk over the short term with currently used doses is very low . Subjects with pre-existing long QT intervals , electrolyte imbalance , concomitant QT prolonging medications ( notably azithromycin in COVID-19 ) and structural heart disease are likely to be at greater individual risk . The very low risks of drug induced arrhythmia over the short term should be distinguished clearly from long term cumulative cardiotoxicity . The main cardiac concerns with long term usage of chloroquine and hydroxychloroquine are conduction defects and cardiomyopathy ( Tönnesmann et al . , 2013 ) . In summary , despite causing significant QT prolongation , there is no evidence that the risks of TdP and sudden death with chloroquine and hydroxychloroquine given alone at currently recommended doses over the short term are increased above those of the background population . Restricting the evaluation of these drugs in COVID-19 for this specific concern therefore seems unwarranted . In assessing the potential toxicity of these drugs QRS widening correlates with severity . COVID-19 ranges from very mild illness to severe disease necessitating intensive care . Case fatality ratios are highly age dependent ( Wu et al . , 2020 ) . The potential toxicity of a treatment regimen has to be balanced against the severity of illness . If chloroquine is to benefit patients with COVID-19 , it is likely to require high exposures . Understanding the concentration-dependent toxicity of chloroquine is essential in assessing the risk-benefit trade-off in COVID-19 clinical trials . This is evidence-based whereas the presumption of risk from lethal TdP derived from the measurement of electrocardiograph QT prolongation ( which dominates the current literature ) is not . From a clinical perspective , there are several practical implications of these observations in self-poisoning . If high doses of chloroquine or hydroxychloroquine are being given to hospitalised patients and blood concentrations cannot be measured rapidly ( which is the usual situation ) , then electrocardiographic monitoring is informative . QRS interval widening can be used as an indicator of toxicity . If the QRS interval is less than 100 msec , then serious cardiovascular toxicity is very unlikely . JT interval prolongation is expected , and should be monitored too , but it is often harder to assess ( as the end of T wave definition is often unclear in chloroquine toxicity ) . Hypokalemia is an important manifestation of chloroquine poisoning and a contributor to tachyarrhythmias ( Clemessy et al . , 1995 ) . With high dose regimens plasma potassium concentrations should be maintained over 4 . 0 mmol/L and plasma magnesium concentrations over 0 . 8 mmol/L . Overall treatment regimens for hospitalised COVID-19 patients which result in whole blood chloroquine concentrations below 10 µmol/L for more than 95% of patients have an acceptable safety margin . The largest prospectively studied cohorts of self-poisoning with chloroquine have all been assembled by the national clinical toxicology unit in Paris , France ( Réanimation Médicale et Toxicologique , Hôpital Lariboisière ) . The clinical and laboratory characteristics of these cohorts have been published in detail previously ( Riou et al . , 1988; Clemessy et al . , 1995; Clemessy et al . , 1996; Mégarbane et al . , 2010 ) . The extensive experience gathered in this clinical toxicology unit established the standard of care for chloroquine poisoning , including mechanical ventilation , appropriate sedation and optimum use of inotropes and vasopressors . Pre-hospital care was provided by emergency physicians in mobile intensive care units . These units could perform 12-lead electrocardiograms as well as advanced life support . Whole blood chloroquine concentrations were measured on admission and at varying times subsequently . The whole blood chloroquine concentrations following self-poisoning were determined using ultraviolet spectrophotometry at a wavelength of 343 nm . This analytical method does not differentiate between chloroquine and the biologically active desethyl metabolite . Original drug concentration measurements , electrocardiograph QRS durations and outcome data were available for all prospectively studied patients from 1987 to 1994 ( Clemessy et al . , 1995; Clemessy et al . , 1996 ) and a later cohort studied from 2003 to 2007 ( Mégarbane et al . , 2010 ) . Patients were included in the prospective studies , if there was a history of attempted suicide and chloroquine was present in the admission whole blood sample . For all patients studied prospectively between February 1987 and January 1992 ( n=167 , Clemessy et al . , 1996 ) multiple whole blood chloroquine measurements were taken during the first 48 hr of hospital stay . The number and timing of samples varied between cases . Original data were not available for Riou et al . , 1988 so concentrations and outcomes were extracted from graphs in the publication ( Figure 3 in Riou et al . , 1988 ) . Whole blood chloroquine concentrations and outcome could be determined approximately for all 102 patients . Data were extracted using the web version of WebPlotDigitizer ( https://automeris . io/WebPlotDigitizer/ ) . Electrocardiograph QRS durations were manually read for all patients in Clemessy et al . , 1995; Clemessy et al . , 1996 . Automated and manual readings were used for patients in Mégarbane et al . , 2010 . We did not have reliable QT interval measurements for the chloroquine self-poisoning patients . It is often impossible to determine accurately the end of the T wave in the presence of the extremely high chloroquine concentrations observed in self-poisoning ( in the 10–100 µmol/L range ) . Several case reports of massive chloroquine overdose show illustrative electrocardiograms in which the QT interval is difficult or impossible to measure accurately , e . g . Gunja et al . , 2009; de Olano et al . , 2019 . An ECG typical of chloroquine self-poisoning , with substantial QRS widening and moderate JT prolongation , is shown in Appendix 6 . To characterise the relationship between chloroquine concentrations and QRS prolongation , we used plasma chloroquine concentration measurements ( from solid phase extraction and high performance liquid chromatography with UV detection ) and electrocardiograph QRS interval measurements from 16 healthy volunteers who took single 620 mg doses of chloroquine base orally ( Pukrittayakamee et al . , 2014 , we note that in the original publication it states that this was a 600 mg single dose , but this is a typographic error ) . The healthy volunteer QRS data consisted of 32 measurements in the absence of drug ( two separate occasions for each volunteer ) and 192 ( 12 per volunteer ) paired chloroquine concentration-QRS measurements . We conducted a literature review of all published case reports and hospital cohorts on chloroquine poisoning . We extracted data from 12 case reports ( 13 patients ) in which whole blood or plasma chloroquine concentrations were reported . We analyzed only case reports in which blood or plasma concentrations were obtained ante-mortem as the post mortem redistribution of chloroquine from tissues to the blood is unknown . However , the data from the case reports exhibited significant bias towards patients with high concentrations who survived ( i . e . unusual cases ) so these were excluded from the final model . To review reports of Torsade de Pointes associated with chloroquine or hydroxychloroquine PubMed and EmBase were searched using the terms ‘HYDROXYCHLOROQUINE’ or ‘CHLOROQUINE’ AND either ‘TORSADE’ , ‘ARRHYTHMIA’ , ‘SUDDEN DEATH’ or ‘CARDIAC ARREST’ . We searched ClinicalTrials . gov on the 11th of June 2020 ( https://clinicaltrials . gov/ct2/home ) for clinical trials using hydroxychloroquine and chloroquine for the treatment of COVID-19 ( search terms: condition = COVID; other terms = hospital; intervention = hydroxychloroquine OR chloroquine; status = recruiting ) . This gave 77 results . After filtering out prevention studies , we could determine the total dose in base equivalent for 55 treatment studies ( these are presented in a Figure 1—source data 1 ) . Chloroquine has complex pharmacokinetic properties characterised by a very large volume of distribution and a terminal elimination half-life of over a month . Both of these parameters are difficult to estimate accurately from pharmacokinetic data ( White et al . , 2020 ) . Whole blood ( the currently preferred matrix ) concentrations are significantly higher than plasma concentrations because of binding to red blood cells , leukocytes and platelets . Two models were used for simulation in NONMEM ( v . 7 . 4 . 3 , Icon Development Solution , Ellicott City , MD; the NONMEM simulation code is available online; see github repository link below ) . The first was a published two-compartment disposition model with a transit compartment absorption model using whole blood measurements from adult vivax malaria patients treated with a standard 25 mg base/kg regimen over three days ( Höglund et al . , 2016 ) ( n=75 , with over 1000 concentration measurements ) . Allometric scaling of body weight , with the exponent fixed to 0 . 75 for clearance parameters and one for volume parameters , was added to the published model to predict concentrations at different body weights . Although a two compartment model underestimates the terminal elimination phase , this has little effect on the blood concentration profile in the first weeks of treatment ( White et al . , 2020 ) . The second was a three-compartment model with the absorption described by a transit compartment model fitted to plasma chloroquine concentrations from healthy adult volunteers who took oral single 620 mg base equivalent doses of chloroquine on two separate occasions with or without primaquine ( Pukrittayakamee et al . , 2014 , a total of 640 concentration measurements: each volunteer was sampled 20 times at two separate dosing occasions; the model is not yet published ) . This model included allometric scaling as described above . Published estimates for the plasma to blood ratio vary considerably ( reviewed in Mégarbane et al . , 2010 ) . We choose a scaling ratio of 4 as this resulted in equal median concentrations for both pharmacokinetic models and is approximately the median value from the review in Mégarbane et al . , 2010 . Five potential chloroquine COVID-19 adult treatment regimens , and one malaria treatment regimen were simulated: Regimens 1–3 and 6 are ‘flat’ dosing regimens ( i . e . not weight-based ) . All regimens were simulated for weights ranging from 40 to 90 kg at intervals of 5 kg . We compared the relationship between admission whole blood chloroquine + desethyl metabolite concentration and death using logistic regression ( maximum likelihood fit ) for data gathered retrospectively ( n=91 , Riou et al . , 1988 ) and data collected prospectively ( n=302 , Riou et al . , 1988; Clemessy et al . , 1995; Clemessy et al . , 1996; Mégarbane et al . , 2010 ) . The retrospective data gave substantially different results with much higher probabilities of death ( approximately 60% versus 5% at 20 µmol/L , see Appendix 1 ) . We therefore excluded the 91 retrospectively studied patients reported in Riou et al . but retained the 11 prospectively studied patients from that publication . This gave a total of 302 unique patient observations . We modeled the probability of death as a function of the log whole blood peak chloroquine concentration ( µmol/L ) using Bayesian logistic regression . The peak concentration is considered a latent variable ( unobserved ) for patients whose blood chloroquine concentrations peaked before hospital admission or for patients who only had their whole blood concentration measured on admission ( n=241 ) . The peak concentration is considered observed for those who had multiple concentration measurements and for whom the peak occurred after hospital admission ( n=61 ) . The model does not attempt to estimate individual peak concentrations for the patients whose peak concentrations were unobserved , but rather an average difference δ , on the log scale , between peak and admission concentrations ( this can be considered as a bias correction term ) . This assumes that the probability of death , conditional on the peak drug concentration , was the same for those whose blood concentrations peaked before and after hospital admission . This assumption is likely to be incorrect . Patients whose levels peaked after hospital admission would have received supportive hospital care more rapidly ( relative to time of peak concentration ) than those who peaked before admission . However , the extent of this bias is unmeasurable . In addition , the model adjusted for differences in hospital treatment received between the three cohorts: cohort 1 was the 11 prospectively studied patients in Riou et al . , 1988; cohort 2 was the 247 patients studied in Clemessy et al . , 1995; Clemessy et al . , 1996 ( named herein ‘the Clemessy cohort’ ) ; cohort 3 was the 44 patients studied in Mégarbane et al . , 2010 . Substantial differences in standard of care occurred between these three cohorts . For example , the use of emergency extracorporeal life support started in the early 2000s and is thought to reduce chloroquine self-poisoning mortality considerably ( Mégarbane et al . , 2007 ) . The prior over the study specific intercept terms was a normal distribution with mean 0 and standard deviation 0 . 5 . We note that the stan code uses the following convention: cohort 1 coded as 0; cohort 2 coded as 1; cohort 3 coded as 2; healthy volunteers coded as 3 . To model the relationship between whole blood chloroquine concentrations and a fatal outcome we use a correction factor γ to account for the desethyl metabolite included in the assay measurement . Between 1 and 6 hr post ingestion , the desethyl metabolite will account for approximately 30% of the total concentration measured ( Pukrittayakamee et al . , 2014 ) . The values of γ and δ are not directly identifiable from the data , but there are good a priori estimates for δ ( Pukrittayakamee et al . , 2014 ) . The data from the individuals whose levels peaked after hospital admission were used to construct an informative prior distribution for γ . In summary , for individuals with an observed peak concentration measurement , the likelihood function given the outcome is Bernouilli with the parameter on the logit scale equal to α+β⁢log⁡ ( γ⁢x ) +ac⁢o⁢h⁢o⁢r⁢t , where x is the observed concentration ( assumed to be the peak concentration ) , α is the global intercept term , ac⁢o⁢h⁢o⁢r⁢t is the hospital cohort random intercept term , and β is the concentration-dependent effect ( slope ) . For individuals who peaked before hospital admission , or who only had one concentration measurement , the likelihood is Bernouilli with parameter on the logit scale equal to α+β⁢ ( log⁡ ( γ⁢x ) +δ ) +ac⁢o⁢h⁢o⁢r⁢t . The Bayesian posterior distribution over the model parameters used informative prior distributions for all four parameters . For the intercept term α , this was a normal distribution with mean −15 ( i . e . 1 in 107 chance of dying at a whole blood concentration of 1µmol/L ) , and standard deviation 1; for the β coefficient on the log concentration , it was a normal distribution with mean 4 and standard deviation 1 . The bias term δ was given an exponential prior with rate 8 ( estimated from the 61 individuals whose levels peaked after hospital admission ) . The metabolite correction factor γ was given a normal prior with mean 0 . 7 and standard deviation 0 . 065 ( the standard deviation was estimated from chloroquine and metabolite measurements in 107 malaria patients from Chu et al . , 2018 ) . For each chloroquine regimen and each weight category considered , we simulated 1000 pharmacokinetic profiles . The simulation was run from the time of the first dose until 7 hr after the last dose ( 7 hr will adequately capture the Cmax value ) . Each simulated pharmacokinetic profile was summarised by the peak concentration denoted Cmax . The set of Cmax values was then used to predict mean fatality ratios using the concentration-fatality model estimated from the chloroquine self-poisoning data ( herein referred to as the pharmacodynamic model ) . We propagated uncertainty from the pharmacodynamic model by estimating desired quantities using 4000 random draws from the posterior distribution . For the estimation of mortality under the different regimens , we truncated the concentration-dependent mortality prediction at 1% ( the concentration at which this truncation occurs will vary depending on the draw from the posterior distribution ) . This is because estimating mortality ratios below 1% is unreliable given a sample size of a few hundred and will be highly dependent on the prior distribution chosen and the parametric form of the concentration-fatality curve . For example , the posterior predictive model estimates a mortality of the order of 1 per 100 , 000 at 2µmol/L . This is approximately the background rate of sudden unexplained death in a young adult population ( Chan et al . , 2018 ) , therefore rendering the prediction essentially unverifiable ( non-scientific ) . We estimated the relationship between chloroquine concentrations and QRS interval duration by fitting a Bayesian sigmoid Emax model . The likelihood function of the regression model was specified as: ( 1 ) QRSi∼Normal ( Emax−Emax−Emin1+ek ( log10xi−E50 ) , ϵi ( j ) ) , j={1 , 2 , 3}where QRSi is the observed QRS duration and xi is the chloroquine concentration for individual i . This concentration is simultaneously measured ( plasma converted to whole blood ) for the healthy volunteers , and is the admission chloroquine concentration ( 70% of measured whole blood concentration to remove the metabolite contribution ) for the self-poisoning patients . The standard error ϵ is modeled separately for the self-poisoning patients and the healthy volunteers . This helps to adjust partially for the concentration-dependent heteroskedasticity in QRS values and the differences in measurement error between pooled datasets ( manually read ECGs in the Clemessy cohort and greater difficulty in accurate interval measurement at the high concentrations seen in self-poisoning ( cohorts 2 and 3 ) versus automated reading in the healthy volunteer data ) . The data from the self-poisoning patients are considered independent and identically distributed conditional on the observed concentration and the study cohort . The prior for the error term ϵi ( 1 ) for the Clemessy cohort was a normal distribution with mean 25 and standard deviation 5 . The prior for the error term ϵi ( 2 ) for cohort 3 was a normal distribution with mean 15 and standard deviation 5 . The data from the healthy volunteers are analyzed with a hierarchical error model ( also known as a mixed effects model ) . We estimate an individual intercept term for each volunteer ( the prior distribution over intercept terms was a normal with mean 0 and standard deviation σH⁢V; the prior over σH⁢V was an exponential distribution with rate parameter 0 . 2 corresponding to a standard deviation of 5 msec in normal QRS values between individuals ) . The measurement error model , ϵi ( 3 ) , was then given a normal prior with mean 2 and standard deviation 1 . The QRS interval data from the healthy volunteers measured in the absence of drug were used to estimate the mean increase in QRS at detectable drug levels ( the difference between Emin and the mean QRS when no drug is detectable ) . From a visual check of the raw QRS data ( Appendix 5 ) , the QRS values from the Clemessy series appear to be slightly biased downwards , and so we modeled this with an additional bias term ( prior is a normal distribution with mean −20 and standard deviation 5 ) . Emax is the maximum mean QRS duration ( prior is a normal distribution with mean 180 and standard deviation 10 ) ; Emin is the minimum mean QRS duration ( prior is a normal distribution with mean 90 and standard deviation 4 ) ; E50 is log10 chloroquine concentration that results in a mean QRS duration of half the maximal increase ( prior is a normal distribution with mean 1 . 3 and standard deviation 1 ) ; the slope parameter k is modeled on the log scale ( prior is a normal distribution with mean 1 and standard deviation 1 ) . The prior distribution over the hierarchical intercept terms for the healthy volunteers was a normal distribution with mean 0 and standard deviation σi which was given an exponential prior with rate 0 . 2 . Both Bayesian regression models were fitted to data using stan ( Carpenter et al . , 2017 ) , implemented in R . Uncertainty around each fit was reported as centred 95% credible intervals . For each model 105 posterior samples were drawn from eight independent chains , the first half were discarded for burn-in and second half thinned every 100 samples . This resulted in 4000 posterior samples used to characterise the posterior distributions . Mixing was assessed by agreement between chains and traceplots . Comparison of prior and posterior distributions are given in Appendix 2 . All code ( including the NONMEM simulation scripts ) and data can be found on github at https://github . com/jwatowatson/Chloroquine-concentration-fatality ( Watson , 2020; copy archived at https://github . com/elifesciences-publications/Chloroquine-concentration-fatality ) .	Hydroxychloroquine and chloroquine are closely-related drugs used for the treatment of malaria and rheumatological conditions , such as lupus . Laboratory tests have indicated that these drugs could also be used against the virus that causes COVID-19 . Given the urgent need , these drugs have been fast-tracked into large-scale clinical trials , bypassing the usual stages that would provide estimates for suitable dosage . The dosage is a critical factor in a clinical trial: too low and the drug will not have an effect , too high and the side effects may counteract any potential benefits . Laboratory tests suggest that higher doses of chloroquine or hydroxychloroquine are needed for treating COVID-19 compared to malaria or lupus . However , there are concerns about the high doses used in some trials , as the drugs can have lethal side effects . Indeed , chloroquine has been used extensively in suicide attempts , particularly in France . To address these concerns , Watson et al . set out to determine the highest dosage of chloroquine ( and thus of hydroxychloroquine , approximately ) that does not cause unacceptable side effects . First , data was analysed regarding the concentration of chloroquine in the blood of 302 patients who had intentionally overdosed on the drug , since this concentration is tightly correlated with their risk of death . Watson et al . used a statistical model to calculate the maximal chloroquine concentration in a person’s blood associated with a one per cent risk of death . This is taken to be the threshold above which any potential benefit of chloroquine treatment would be outweighed by the possibility of lethal toxicity . Watson et al . also estimated the relationship between chloroquine concentrations and changes in electrocardiogram patterns , which record the electrical activity of the heart . This makes it possible to determine whether a high dose of chloroquine has led to dangerous levels in the blood . Using a mathematical model of how chloroquine is metabolised , Watson et al . predicted that most of the trials that tested chloroquine as a treatment for COVID-19 did not reach the calculated threshold concentration . An exception was the CloroCovid-19 trial in Brazil , which was stopped early because people in the higher dosage group suffered more heart problems and died in greater numbers than those in the lower dosage group . Two large randomised trials , RECOVERY and SOLIDARITY , have shown no benefit of hydroxychloroquine or chloroquine in the treatment of COVID-19 , changing clinical practice worldwide . Both of these trials used high doses resulting in higher hydroxychloroquine or chloroquine concentrations than normally observed in the treatment of malaria or rheumatological conditions . The results from Watson et al demonstrate that the lack of benefit seen in these two large clinical trials is not due to the drug dosage being too high .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine" ]	2020	Concentration-dependent mortality of chloroquine in overdose
The Ciona notochord displays planar cell polarity ( PCP ) , with anterior localization of Prickle ( Pk ) and Strabismus ( Stbm ) . We report that a myosin is polarized anteriorly in these cells and strongly colocalizes with Stbm . Disruption of the actin/myosin machinery with cytochalasin or blebbistatin disrupts polarization of Pk and Stbm , but not of myosin complexes , suggesting a PCP-independent aspect of myosin localization . Wash out of cytochalasin restored Pk polarization , but not if done in the presence of blebbistatin , suggesting an active role for myosin in core PCP protein localization . On the other hand , in the pk mutant line , aimless , myosin polarization is disrupted in approximately one third of the cells , indicating a reciprocal action of core PCP signaling on myosin localization . Our results indicate a complex relationship between the actomyosin cytoskeleton and core PCP components in which myosin is not simply a downstream target of PCP signaling , but also required for PCP protein localization . Despite the importance of the planar cell polarity ( PCP ) pathway in ensuring the proper orientation of cells in a range of embryonic and adult tissues , the molecular mechanisms of this pathway are not fully understood . The PCP pathway is ancient in metazoans , and its requirement for proper development and physiology has been well described in insects ( e . g . , Drosophila ) and chordates . The function of the PCP pathway can be seen in the coordinated polarization of subcellular components such as cilia and multi-celled structures such as bristles in epithelial sheets and organs like the Drosophila wing , vertebrate inner ear , and kidney , as well as in dynamic processes , such as cell migration , convergence and extension of mesoderm , neural tube closure , and axonal guidance ( Ybot-Gonzalez et al . , 2007; Wallingford , 2012; Tissir and Goffinet , 2013; Papakrivopoulou et al . , 2014 ) . One of the complicating issues of PCP signaling in Drosophila , where it has been most extensively addressed , is that it appears to consist of several semi-independent pathways that may act either in parallel or in sequence ( Gray et al . , 2011; Lawrence and Casal , 2013 ) . The best characterized of these pathways is the core PCP pathway , which acts locally to coordinate polarity between neighboring cells . The key components of the core pathway include the transmembrane proteins Strabismus/Van Gogh and Frizzled , and the cytoplasmic proteins Prickle and Dishevelled . Some components of this pathway are shared with the wnt/β-catenin signaling pathway , although unlike the Wnt/β-catenin pathway the core PCP pathway is thought to act primarily by directing cytoskeletal organization , rather than by regulating transcription . While the core PCP pathway coordinates local polarity in groups of cells , there appears to be a requirement for an additional signaling pathway ( s ) that brings this local coordination into register with the axes of the organ or embryo . The precise nature of this so-called ‘global’ signaling pathway has been elusive , although several candidate pathways are supported experimentally . One model for the Drosophila global polarizer stresses the importance of mechanical forces generated during morphogenesis leading to the alignment of cells ( Eaton and Jülicher , 2011 ) . Two signaling pathways have also been proposed to act as global polarizers—the Wg/Wnt4 pathway ( Wu et al . , 2013 ) , and the Dachsous ( Ds ) /Fat/Four-jointed ( Fj ) pathway ( Casal et al . , 2006; Thomas and Strutt , 2012; Matis et al . , 2014; Olofsson et al . , 2014 ) . The Ds/Fat/Fj module has been most extensively characterized and appears to signal via formation of heterodimers with the extracellular domains of Ds and Fat ( which are atypical cadherins ) , putatively in a gradient across the tissue ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . Most recently , it was reported that the Ds/Fat/Fj module affects polarity through microtubule orientation , which in turn directs core PCP polarization ( Matis et al . , 2014 ) . Fat and Ds orthologs also play roles in vertebrate development , but their precise roles in regulating aspects of PCP remain to be clarified ( Mao et al . , 2011; Saburi et al . , 2012 ) . One final complicating factor in assessing the role of the Fat/Ds/Fj pathway in planar polarity is the apparent overlap with the growth-stimulating Hippo pathway ( Lawrence and Casal , 2013 ) . The notochord of the ascidian Ciona provides a particularly tractable model for the study of the PCP pathway in tissue morphogenesis ( Kourakis et al . , 2014 ) . Ascidians are invertebrate chordates , and as members of the chordate subphylum Tunicata they belong to the group of animals that are the closest extant relatives of the vertebrates ( Delsuc et al . , 2006 ) . While the presence of the notochord—a stiff axial rod of mesodermal cells lying under the nerve cord—is a uniting feature of the Chordata , tunicate notochords are much simpler than those of vertebrates , and in Ciona the fully formed notochord consists of only 40 cells arranged in a stack one-cell wide ( Jiang et al . , 2005; Kourakis et al . , 2014 ) . We have described two discrete developmental phases in notochord morphogenesis that shows polarized cell behavior . Initially , the notochord precursor cells undergo a mediolaterally oriented intercalation behavior , which forms the notochord column along the AP axis . The role of the core PCP pathway in the convergent extension of the Ciona notochord is seen in the pk mutant aimless ( aim ) of Ciona savignyi , which has defects in notochord cell intercalation behavior , morphology , and extracellular matrix deposition ( Jiang et al . , 2005; Veeman et al . , 2008 ) . Following the completion of intercalation , the individual notochord cells initially assume a flattened-disk shape ( Jiang and Smith , 2007 ) . Over the next approximately 3–4 hr ( corresponding to stages 21–24 [Hotta et al . , 2007] ) , the cells undergo a dramatic elongation in the A/P axis , which results in the continued elongation of the entire tail . During the process of elongation , a second polarity becomes evident in the A/P axis ( at a right angle to the mediolateral polarity seen earlier at intercalation ) . The first sign of A/P polarity can be seen shortly after the completion of intercalation ( stage 22 ) in the localization of Pk protein to the anterior poles of the cells ( Kourakis et al . , 2014 ) . As elongation continues the core PCP protein Strabismus ( Stbm ) also localizes to the anterior pole , while nuclei move to the posterior poles of the cells where they take on a flattened shape ( Figure 1 ) ( Jiang et al . , 2005; Kourakis et al . , 2014 ) . The exception to this pattern is the posterior-most notochord cell whose nucleus is usually found in reverse orientation relative to the other cells ( Jiang et al . , 2005; Kourakis et al . , 2014 ) . In aim embryos , the nuclei of intercalated cells are still polarized , but they are randomly oriented to either the anterior or posterior pole of each cell ( Kourakis et al . , 2014 ) . Following elongation , the notochord enters a new phase in its development ( from stage 24 onward ) . A matrix is secreted into extracellular pockets that form between A and P faces of the cells ( Dong et al . , 2009; Denker and Jiang , 2012; Deng et al . , 2013 ) . As this process continues , the pockets of matrix between the cells expand and then fuse to make a single , uninterrupted lumen along the length of the notochord . 10 . 7554/eLife . 05361 . 003Figure 1 . Ciona intestinalis late-tailbud embryo ( stage 23 ) expressing an electroporated Histone 2A/Red Fluorescent Protein ( H2A-RFP ) in the notochord . Insets show two cells to illustrate the polarization of the nuclei to the posterior of the cells . Scale bar is 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 003 In this manuscript , we examine the relationship between core PCP signaling and the actin/myosin network . We report here that the notochord cells have anteriorly polarized myosin machinery . While existing models depict the polarization of myosin as downstream of PCP signaling , we instead present evidence for a more complex series of interactions in which the core PCP components and the myosin machinery act in a reciprocal fashion to promote cell polarization . The Ciona intestinalis genome is predicted to contain six members of the non-muscle myosin II family ( Chiba et al . , 2003 ) . One of these , myosin10/11/14/9 , is reported to be expressed exclusively in the notochord ( Satou et al . , 2001 ) and is found in a two-gene operon with a myosin regulatory light chain ( MRLC ) ( Satou et al . , 2008 ) . This MRLC is one of five found in the C . intestinalis genome and has previously been used to mark myosin assemblies in notochord cells ( Dong et al . , 2011 ) . By convention , Ciona genes are named according to their vertebrate orthologs ( Stolfi et al . , 2014 ) . This particular Ciona MRLC shares orthology to vertebrate MRLCs 9 and 12 ( phylogeny not shown ) and will be referred to here as MRLC9/12 . To visualize myosin assemblies in developing notochord cells , we constructed fusion proteins of MRLC9/12 and Myosin10/11/14/9 with Venus fluorescent protein and myc antigen , respectively . The resulting cDNAs were fused to the Brachyury cis-regulatory region to drive notochord expression ( Corbo et al . , 1997 ) , and then introduced to embryos by electroporation . Low doses of plasmid were used to encourage highly mosaic expression . This allows the visualization of the exogenous proteins in cells isolated from the expression of neighboring cells , which would otherwise make the determination of anterior vs posterior protein localization difficult ( Jiang et al . , 2005; Kourakis et al . , 2014 ) . At the earliest developmental stage analyzed , stage 19—when the notochord cells are mid-intercalation , both Myosin10/11/14/9-myc and MRLC9/12-Venus were observed throughout the cells , with the highest localization on the perinotochordal surfaces where cells contact the basement membrane that surrounds the developing notochord ( white arrows , stage 19; Figure 2A; see also Veeman et al . , 2008 ) . At the immediate completion of intercalation ( stage 21 , Figure 2B ) , the localization of both fusion proteins was still the strongest at the basal/lateral edges . However , less than an hour later , at stage 22 , both proteins were found localized to the anterior pole of each notochord cell ( arrowhead , Figure 2C ) , as well as continuing to be found at the basal/lateral edges . Quantification of MRLC9/12-Venus fluorescence signal along the anterior/posterior axis of notochord cells confirms a distinct transition in MRLC9/12-Venus localization between stages 21 and 22 ( insets , Figure 2B , C ) . During notochord cell elongation , and at its completion ( stage 24 ) , the anterior localization of both proteins persists ( Figure 2D ) . The onset of anterior/posterior localization of myosin assemblies closely matches the localization of Pk-myc , with A/P localization of Pk-myc starting at stage 22 ( Kourakis et al . , 2014 ) . However , the Pk-myc and MRLC9/12 proteins do not appear to completely colocalize spatially . At stage 22 , Pk-myc , Myosin10/11/14/9-myc , and MRLC9/12-Venus are each found to be enriched on the anterior side of the cells . However , as elongation progresses , Pk-myc localization becomes restricted to a small region within the center of this domain ( Jiang et al . , 2005; Kourakis et al . , 2014 ) , while Myosin10/11/14/9-myc and MRLC9/12-Venus remain more dispersed ( Figure 2 ) . 10 . 7554/eLife . 05361 . 004Figure 2 . Time course for the subcellular localization of myc-tagged myosin10/11/14/9-tagged myosin regulatory light chain ( MRLC ) and Venus-tagged myosin regulatory light chain ( MRLC9/12-Venus ) at the stages indicated . Arrows in A indicate contacts of cells with the basement membrane , and arrowheads in C indicate anteriorly localized protein . Densitometry readings across the center of four cells demonstrate the transition in MRLC9/12 localization from stage 21 to stage 22 ( charts in B and C ) . ( D ) At stage 24 , the anterior localization of both proteins persists . Anterior is to the right in all panels . Scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 004 To examine the localization of active endogenous myosin complexes in the notochord , embryos were stained with an antibody to mono-phosphorylated ( ser19 ) MRLC ( pMRLC ) ( Ikebe and Hartshorne , 1985 ) . Specific staining was observed along the cortices of the cells ( white arrowheads , Figure 3A ) . Because the endogenous protein is not mosaically expressed , it is not possible to definitively resolve the expression to either the anterior or posterior face of the cells . Co-staining for electroporated Pk-myc and pMRLC shows relatively little overlap between the two proteins , with Pk-myc concentrated to a smaller domain at the center of the cell , as described above , while pMRLC distributed along the face of the cell . In contrast , a much higher degree of colocalization was observed for pMRLC and Stbm-Venus ( Figure 3B , yellow arrowheads ) . As was reported previously ( Jiang et al . , 2005 ) , Stbm is distributed evenly along the anterior face of the cell , rather than being concentrated centrally like Pk . 10 . 7554/eLife . 05361 . 005Figure 3 . Subcellular localization of endogenous phospho-Myosin Regulatory Light Chain ( pMRLC ) and electroporated myc-tagged Prickle ( Pk-myc ) ( A ) , and Venus-tagged strabismus ( Stbm-Venus ) ( B ) . White arrowheads in A indicate anti-pMRLC staining , while yellow arrowheads in B indicate punctae of strong pMRLC and Stbm-Venus colocalization . Scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 005 To examine possible relationships between polarization of the myosin machinery , core PCP ( Pk-myc and Stbm-Venus ) and nuclear polarity , embryos were treated with chemical inhibitors that disrupt the actin cytoskeleton ( cytochalasin B and latrunculin ) or myosin II motor activity ( blebbistatin ) . Treatments were performed both to embryos with fully polarized notochord cells ( stage 23 ) , as well as to younger embryos ( stage 21 ) before polarity is evident . At stage 23 , cytochalasin B caused notochord nuclei to detach from posterior membranes and drift anteriorly toward the middle of the cell ( Figure 4B , E; and Video 1 ) . In these same cells , we observed a loss of polarization of both Pk-myc ( Figure 4A , B ) and Stbm-Venus ( Figure 4D , E ) . Embryos were scored for cytochalasin B treatment at both mid-stage 23 and late-stage 23 , and only 6% ( n = 35 ) and 5% ( n = 20 ) of cells showed Pk-myc polarization , respectively . Even in cells scored as having polarized Pk-myc , the tight localization seen at the center of the anterior membrane was lost , replaced by labeling throughout the anterior membrane . By contrast , cells from embryos treated with DMSO at mid-stage 23 and late-stage 23 had 81% ( n = 59 ) and 100% ( n = 5 ) of cells with Pk-myc properly polarized , respectively . Similar results were seen for Stbm-Venus localization . In embryos treated with cytochalasin B at stage 23 , only 6% ( n = 50 ) had detectable Stbm-Venus localization vs 75% ( n = 52 ) of control DMSO-treated cells . Latrunculin B , which inhibits actin polymerization by a different mechanism than cytochalasin B ( Morton et al . , 2000 ) , also caused a loss of nuclear , Stbm-Venus , and Pk-myc polarization ( Figure 4C , F ) . Only 5% and 4% of latrunculin B treated embryos showed localized Stbm-Venus and Pk-myc , respectively , vs 92% of controls ( n = 57 , 25 , and 25 ) . The stage 23 treatments demonstrated that cytochalasin B could disrupt the notochord cell polarity once it was established . To examine the effect of cytochalasin B on the establishment of polarity , embryos were treated at stage 21 and scored at stage 23 . In these embryos , no polarization of Pk-myc was evident in any of the cells scored ( n = 5 ) vs 92% of cells from vehicle ( DMSO ) -treated embryos ( n = 12 ) having normal anterior polarization of Pk-myc . 10 . 7554/eLife . 05361 . 006Figure 4 . Loss of planar cell polarity protein localization following actin/myosin disruptions . ( A–G ) Single notochord cells from embryos expressing H2A-RFP ( nuclear RFP ) , and either a Pk-myc or Stbm-Venus fusion protein ( as indicated ) . Arrowheads indicate anteriorly localized reporter protein . Embryos were treated with either DMSO ( vehicle ) , cytochalasin B , latrunculin B , or blebbistatin ( also as indicated ) . ( H , I , and J ) Cortical actin staining of notochord cells with phalloidin in embryos treated with vehicle only ( DMSO ) , cytochalasin B , or latrunculin B , as indicated . Arrows in H indicate the equatorial contractile ring of actin . ( H′ , I′ , and J′ ) Co-staining of samples for the cortical protein atypical protein kinase C ( aPKC ) . Arrowhead in H′ indicates concentrated staining at center of cell cortex . Scale bar in A is 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 00610 . 7554/eLife . 05361 . 007Video 1 . Detachment of nuclei during cytochalasin B treatment . The video represents 30 min in real time , beginning approximately 10 min after the addition of cytochalasin B . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 007 Blebbistatin , which blocks myosin motor function , was also able to cause nuclear and Pk-myc depolarization ( Figure 4G ) , but with varying effectiveness . In one set of experiments , we observed Pk-myc polarization in only 2 of 18 cells scored , while in second set , we observed Pk-myc polarization in 31 of 43 cells . The DMSO-treated controls showed 34/41 and 12/13 cells with proper Pk-myc polarity , respectively . Blebbistatin-treated cells scored as having Pk-myc polarization often lacked the restricted domain of labeling at the center of the membrane seen in DMSO-treated controls , instead showing a more even distribution throughout the anterior membrane . As expected , treatment of the embryos with either cytochalasin B or latrunculin B profoundly disrupted the actin network in notochord cells . Staining of control embryos with phalloidin shows strong cortical actin in the notochord cells ( Figure 4H ) . Cytochalasin B treatment resulted in fragmentation of the actin network ( Figure 4I ) , with apparent loss of the equatorial contractile ring ( indicated by arrows in 4H ) ( Sehring et al . , 2014 ) , while latrunculin B resulted in a complete loss of phalloidin staining ( Figure 4J ) . The embryos shown in Figure 4H–J were also stained for the protein atypical protein kinase C ( aPKC ) , which is localized to the cell cortex of notochord cells , but unlike Pk and Stbm is found at both the anterior and posterior poles of the cells ( Denker et al . , 2013 ) ( Figure 4H′ , arrowhead ) . Surprisingly , a stronger disruption in the localization of aPKC was seen with cytochalasin B treatment than with latrunculin B ( Figure 4H′ , J′ ) . We speculate that aPKC may associate with the fragmented actin filaments seen with cytochalasin treatment , resulting in it spreading across the face of the cells , while in the complete absence of filamentous actin ( latrunculin B ) the aPKC is not displaced and is able to remain concentrated at the center of the cell . While we observed that cytochalasin , latrunculin , and blebbistatin treatments all lead to loss of core PCP protein and nuclei localization , the same is not true for myosin localization . Using MRLC9/12-Venus as a marker , we observed that treatment of embryos with cytochalasin B or latrunculin B from stage 21–23 did not disrupt MRLC9/12-Venus polarization ( Figure 5A–D ) , although the cells failed to extend to the same extent as the DMSO-treated controls , and the nuclei were displaced to the center of the cells . Additionally , in those cells treated with cytochalasin B , but not latrunculin B , MRLC9/12-Venus staining was punctate , albeit still anteriorly localized . 10 . 7554/eLife . 05361 . 008Figure 5 . MRLC9/12 localization is resistant to cytochalasin B and blebbistatin . ( A ) Control , vehicle-treated embryo shows polarization of MRLC9/12 to the anterior membrane of notochord cells . ( B and C ) MRLC9/12 polarization is resistant to cytochalasin B and latrunculin B treatments . MRLC9/12-Venus is labeled green , and H2A-RFP ( nuclei ) is labeled red . ( D ) Percentage of notochord cells showing polarized MRLC9/12 for three treatment regimes . The number of cells scored is indicated at the bottom of the columns . e . st . , early stage; l . st . , late stage . ( E and F ) Persistence of MRLC9/12-Venus polarization ( white ) in embryos treated with blebbistatin or vehicle ( DMSO ) from stage 21 to stage 23 . Arrowheads indicate anteriorly localized proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 008 In older embryos , cytochalasin B appeared to be more disruptive . While treatment of embryos at early stage 23 to the start of stage 24 only resulted in a modest disruption of MRLC9/12-Venus polarization ( Figure 5D ) , with 80% of treated cells retaining polarity vs 90% of control cells , treatment of embryos with cytochalasin B at late-stage 23 had a stronger effect , with 57% of cells scored ( n = 44 ) having polarized MRLC9/12-Venus ( vs 100% of cells in DMSO-treated embryos ( n = 13 ) ; Figure 5D ) . The transition to cytochalasin B dependence from early to late-stage 23 may be tied to dramatic changes in cell architecture as the notochord begins the process of lumen formation , first visible at stage 24 , a period during which the orderly A/P arrangement and polarity of cells changes dramatically ( Dong et al . , 2009 ) . The dependence of the anterior polarization of MRLC9/12-Venus on myosin activity was also examined using blebbistatin . Because blebbistatin inhibits myosin II Mg-ATPase activity without disassembling actin/myosin complexes ( Straight et al . , 2003 ) , we would expect to observe a disruption in polarization of MRLC9/12-Venus only if the polarization required myosin motor activity . In embryos treated with 10 μM blebbistatin from stage 21 through stage 23 , a dramatic block in cell elongation was observed , although there was no apparent disruption of MRLC9/12-Venus polarization indicating that MRLC9/12-Venus polarization does not require myosin activity ( Figure 5E , F ) . In summary , the core PCP proteins and myosins respond very differently to cytoskeletal inhibitors . The initiation and maintenance of Stbm-Venus and Pk-myc localization depends on an intact actin cytoskeleton , and addition of the myosin inhibitor blebbistatin leads to loss of Pk-myc and Stbm-Venus polarization . These same treatments also disrupted nuclei , which normally polarize posteriorly in a PCP-dependent manner ( Jiang et al . , 2005 ) , indicating an extensive disruption to the normal polarity of notochord cells . In contrast , the anterior polarization of MRLC9/12-Venus is largely resistant to these treatments , indicating that the polarization of the myosin machinery in Ciona notochord cells can occur independently of the localization of the core PCP proteins Pk and Stbm . Despite the loss of A/P polarization of nuclei , Pk-myc , and Stbm-Venus in cytochalasin B-treated cells , we observed that polarity could be restored when the cytochalasin B was washed out . Within approximately 60 min of placing the embryos in cytochalasin B-free seawater , nuclei were observed to move back to the posterior membranes and resume their flattened shapes ( Figure 6A; Video 2 ) . Likewise , Pk-myc was observed to relocalize to anterior membranes upon wash out of cytochalasin B ( Figure 6B ) . In assessing Pk-myc repolarization , a sample of embryos was first collected at the end of the cytochalasin B treatment to assure the efficacy of the treatment . In cells from this first group of embryos , anterior localization of Pk-myc was observed in only one of 33 cells scored . 1 hr after cytochalasin B was washed out , 13 out of 18 cells scored had re-established anterior localization of Pk-myc ( Figure 6B ) . However , when blebbistatin was added during the recovery from cytochalasin B treatment , the ability of the cells to properly repolarize was greatly reduced ( Figure 6D ) . Only 14 of 45 cells ( 31% ) scored from 19 embryos showed proper repolarization of Pk-myc in the presence of blebbistatin vs 20 of 24 cells ( 83% ) from 15 embryos treated with vehicle ( DMSO ) . In contrast to cytochalasin B-treated embryos , embryos treated with latrunculin B were unable to recover either nuclear or Pk-myc polarization in the notochord , even after extensive washing . 10 . 7554/eLife . 05361 . 009Figure 6 . Recovery of polarity following cytochalasin B treatment . ( A ) Time lapse of nuclear repolarization following removal of embryos from cytochalasin B . Times are from end of cytochalasin B treatment . ( B ) Recovery of Pk-myc polarization following cytochalasin B treatment and wash out . Anterior is to the right in all panels . ( C and D ) Blebbistatin blocks recovery from cytochalsin B treatment . Following removal of embryos from cytochalsin B , embryos were either treated with vehicle ( DMSO ) ( C ) or 10 μm blebbistatin ( D ) . Arrowheads indicate anterior localized Pk . Scale bar in A is 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 00910 . 7554/eLife . 05361 . 010Video 2 . Movement of nuclei to the posterior poles of notochord cells following removal in cytochalasin B . The video represents 28 min in real time , beginning approximately 10 min after removal from cytochalasin B . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 010 Cytochalasin B causes extensive disruption of notochord cells . The cells undergo a dramatic shape change , the polarization of Pk-myc , Stbm-Venus , and nuclei is lost , and the localization of aPKC is disrupted . Despite this , polarization of MRLC9/12-Venus persists , and much of the actin network remains cortical , although fragmented ( Figure 4I ) . Inhibition of myosin motor activity with blebbistatin gives similar effects on Pk-myc and nuclei . Thus , in blebbistatin-treated cells or cytochalasin B-treated cells , while the actin and myosin machinery are inactivated , they appear to remain polarized and cortical . Removal of embryos from cytochalasin B allows the recovery of Pk-myc and nuclei , but not in the presence of blebbistatin . Together , these results support an active role for actin/myosin machinery in repolarization of core PCP proteins and nuclei . Consistent with this , latrunculin treatment , while not disrupting MRLC9/12-Venus localization , caused a much more extensive disruption the cortical actin network than did cytochalasin B ( Figure 4J ) . These cells do not recover nuclear or Pk-myc polarization when removed from latrunculin , presumably because of the more extensive loss of the actin network . In summary , despite the loss of Stbm and Pk localization in cytochalasin-treated cells a polarity persists that can restore proper core PCP and nuclei polarity . This persistent polarity appears to be the actin/myosin machinery . Previous reports in Drosophila have shown an important role for the microtubule network in PCP function ( Shimada et al . , 2006; Harumoto et al . , 2010; Olofsson et al . , 2014 ) . The microtubule network in the Ciona notochord can be readily visualized with a GFP-labeled version of the microtubule-binding protein Ensconsin ( Figure 7A; [Roure et al . , 2007] ) , and addition of 10 μm nocodazole results in a complete disruption of the network ( Figure 7B ) . To test for a role of microtubules in the establishment of polarity , embryos were treated with nocodazole at stage 21 . At stage 21 , notochord intercalation is complete , but no signs of A/P polarity are evident . The resulting embryos observed at stage 23 showed relatively normal notochord development with no perturbations in either nuclear position ( Figure 7C , D ) or in Pk-myc localization ( Figure 7E ) . Of 26 nocodazole-treated cells examined , 21 ( 81% ) had normal Pk-myc localization vs 23 of 25 cells ( 92% ) for a DMSO-treated control . The addition of vinblastine , another microtubule inhibitor , at stage 21 is consistent with our nocodazole results; after vinblastine treatment at stage 21 , 210 of 211 cells ( 99 . 5% ) showed proper nuclear polarity , compared to 227 of 228 ( 99 . 6% ) for the DMSO control ( data not shown ) . Moreover , the maintenance of Pk-myc polarization was also not disrupted by the addition of nocodazole at stage 23 . In embryos treated with 10 μm nocodazole for 1 hr starting at stage 23 , we observed that 42 of 45 cells ( 93% ) scored had anterior-localized Pk-myc ( vs 32 of 36 cells ( 89% ) from DMSO-treated embryos ) . 10 . 7554/eLife . 05361 . 011Figure 7 . Notochord A/P polarity does not require microtubular network . ( A ) Visualization of the notochord microtubule network with GFP-tagged ensconsin , and B , disruption of the network with 10 μm nocodazole . ( C–E ) Nocodazole treatment from stage 21 disrupts neither nuclear ( D ) nor Pk-myc ( E ) polarity . Embryos/cells are shown at stage 23 . Similarly , nocodazole treatment at stage 21 does not disrupt MRLC9/12 polarity ( control , F; treated , G ) . ( H ) Cell nuclei properly repolarized in the presence of nocodazole following cytochalasin treatment . Arrowheads indicate anterior localized MRLC9/12 . Scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 011 Disruption of microtubules by the addition of nocodazole also had little effect on the anterior localization of MRLC9/12 in notochord cells . To test for a role in the establishment of polarity , nocodazole was added for 1 hr at late-stage 20 , when notochord cell intercalation is nearly complete . In those cells scored , 90 of 102 had properly polarized MRLC9/12 ( 88 . 2% ) , while all DMSO-only treated notochord cells scored ( 111/111 , 100% ) showed anterior MRLC9/12 ( Figure 7F , G ) . When treated later , at stages 22–23 ( well after intercalation is complete ) polarization of MRLC9/12-Venus was nearly unaffected in nocodazole-treated embryos; 166 of 167 ( 99% ) of notochord cells counted showed anterior MRLC . Control animals showed localization in 100% of cells observed ( 61/61 ) . Finally , addition of nocodazole did not disrupt recovery of polarity following cytochalasin B treatment . We observed that nuclei were able to properly repolarize ( 65/66 , 98 . 5% ) following cytochalasin wash out in the presence of 10 μm nocodazole . ( Figure 7H ) . Our results do not support a role for microtubules in the polarization of nuclei , Pk , or MRLC9/12 . When nocodazole is added at a developmental stage immediately preceding that in which polarization of Pk-myc and MRLC9/12 is first evident , very little disruption of the polarity of these molecules is observed later . It is unknown what mechanism leads to the polarization of these molecules between stages 21 and 22 , and an earlier , but yet unknown , polarity cue may be present . We have observed that the addition of nocodazole earlier interferes with intercalation , and that the onset of A/P polarization and intercalation appear to be linked , leaving open the possibility of an earlier role of microtubules in A/P polarity . Nevertheless , the role of microtubules in the localization of core PCP proteins in the Ciona notochord differs significantly from the localization of these proteins in other models for PCP ( Shimada et al . , 2006; Harumoto et al . , 2010; Olofsson et al . , 2014 ) . The functional link between myosin polarization and the core PCP pathway was further investigated by examining Myosin10/11/14/9-myc and MRLC9/12-Venus localization in the C . savignyi aim line , which carries a null mutation in the pk gene ( Jiang et al . , 2005 ) . Despite the absence of Pk activity , a few notochord cells in aim embryos are still able to intercalate . These notochord cells , which are usually but not exclusively in the posterior notochord , individually display a polarity , as is evident by the localization of the nuclei to either the anterior or posterior poles of intercalated cells . However , the polarity is not coordinated between cells , and they show no bias with respect to the A/P axis ( Kourakis et al . , 2014 ) . We assayed the localization of Myosin10/11/14/9-myc or MRLC9/12-Venus in 34 cells from 23 aim embryos , 25 having posterior nuclei and 9 having anterior nuclei ( Figure 8A ) . We observed two distinct phenotypes for myosin localization . For those cells with posterior nuclei ( i . e . , the same as observed in wild-type embryos ) , we observed that both Myosin10/11/14/9-myc and MRLC9/12-Venus were polarized anteriorly ( Figure 8B , D ) as normal . By contrast , for those cells with anterior nuclei , we were unable to detect either posterior or anterior enrichment of myosin ( with one exception ) , Figure 8A , C , E . 10 . 7554/eLife . 05361 . 012Figure 8 . Myosin10/11/14/9 and MRLC9/12 localization in homozygous aim embryos . ( A ) Distribution of nuclear and Myosin10/11/14/9 ( green ) or MRLC9/12 ( orange ) localization phenotypes ( ant . , anterior; post . , posterior; unloc . , unlocalized ) in aimless embryos . Each dot represents a single scored cell . ( B and C ) MRLC9/12-Venus ( green ) localization in cells with posterior and anterior nuclei ( red ) , respectively . ( D and E ) Myosin10/11/14/9-myc ( green ) localization in cells with posterior and anterior nuclei ( red ) , respectively . Anterior is to the right in all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 05361 . 012 Thus , despite disruption of core PCP protein localization using cytoskeletal inhibitors , or by loss of Pk activity as is the case with the aim mutant , myosin can show a wild-type polarization pattern . Nevertheless , we do observe a requirement for Pk activity in properly polarizing Myosin10/11/14/9-myc and MRLC9/12-Venus in a subset of notochord cells in aim embryos . The results presented here demonstrate a dynamic interaction between PCP components and the actin/myosin cytoskeleton in establishing and maintaining notochord cell AP polarity . A linear pathway from the activation of the core PCP components Fz/Dsh to polarization of the actin cytoskeleton via Daam1 and RhoA has been well described , as reviewed by Wallingford ( 2012 ) . PCP-directed polarization of actin is also seen in the Drosophila wing , where the actin-rich prehair is targeted to the distal side , opposite where pk and stbm are enriched ( Wong and Adler , 1993 ) . Our results in the Ciona notochord indicate a more complex interaction in which these components are mutually reinforcing rather than being in a simple linear pathway . Additionally , and in contrast to results in other experimental systems ( Shimada et al . , 2006; Harumoto et al . , 2010; Vladar et al . , 2012; Matis et al . , 2014 ) , we find no evidence of a role for the microtubule network in the localization of either PCP components Pk and Stbm , or MRLC9/12 . We report here for the first time that myosin machinery ( Myosin10/11/14/9 and MRLC9/12 ) is anteriorly polarized in C . intestinalis notochord cells and shows extensive colocalization with the core PCP protein Stbm , suggesting a functional interaction between them . This is further supported by the fact that the disruption of the actin/myosin cytoskeleton with cytochalasin , latrunculin , or blebbistatin all result in the loss of polarization of Pk and Stbm . Conversely , these same treatments leave the polarization of MRLC9/12 largely intact . Thus , Myosin10/11/14/9 in the notochord appears to localize in the absence of core PCP protein localization , similar to the situation for the atypical myosin Dachs in Drosophila , which is also polarized independently of the core PCP pathway ( Brittle et al . , 2012 ) . It should be noted , however , that no direct ortholog of Dachs is found in chordate genomes ( Mao et al . , 2006 ) . Moreover , the persistence of MRLC9/12 localization when Pk and Stbm are delocalized ( e . g . , cytochalasin treatment ) , and the requirement for myosin activity during repolarization after cytochalasin treatment , suggest that myosin polarization in Ciona notochord cells could act upstream of core PCP localization . However , it is important to note that in the developing notochord , Pk , Myosin10/11/14/9 , and MRLC9/12 all appear to simultaneously localize in the A/P axis between stages 21 and 22 ( Figure 2 and [Kourakis et al . , 2014] ) . Thus , the timing of events would argue against a simple model in which myosin localization precedes and directs Pk polarization . Moreover , it has also been observed that polarization of Stbm lags behind that of Pk ( and thus also of myosin ) ( Jiang et al . , 2005 ) , indicating that the core PCP machinery is still assembling at the time we first see polarized myosin . The mechanism and identity of the polarizing cue that leads to the simultaneous polarization of Pk , Myosin10/11/14/9 , and MRLC9/12 remains mysterious . The observations presented here in a Pk null mutant line ( aim ) provides clues to the nature of this signal . Myosin polarization was disrupted in approximately one third of the notochord cells scored in aim embryos . While the pk allele in aim contains an ∼200 base-pair deletion that spans one intron and an adjacent exon , creating a premature stop codon and a likely null allele ( Jiang et al . , 2005 ) , and the Ciona genome contains only a single pk gene , one possible explanation for the variability between cells is that residual Pk-independent PCP activity remains in a subset of the notochord cells in aim embryos . However , by other measures of cell polarity , such as extension and protrusive activity along the mediolaterial axis during intercalation , and the membrane localization of Dishevelled , we see no evidence for residual PCP activity in aim embryos , and moreover , all notochord cells appear to be uniformly disrupted ( Jiang et al . , 2005; Veeman et al . , 2008; Kourakis et al . , 2014 ) . Thus , we surmise that many cells in the notochord are able to properly polarize myosin without input from the core PCP pathway . The notochord cells in aim embryos show a highly consistent pattern in which those having posterior localized nuclei ( i . e . , the wild-type pattern ) have anterior polarized myosin , while those with inverted nuclei polarity ( anterior ) lacked detectable myosin polarity , rather than having inverted myosin polarity . We speculate that those cells in aim embryos showing wild-type nuclear polarity are correctly aligned with a global polarizing cue , and can polarize their myosin complex , while those cells showing anterior nuclei are not in alignment , and myosin does not polarize properly . The role of the core PCP pathway , which is lacking in the aim cells , is thus to act locally to bring the cells into proper alignment . Despite this observation , disruption of Pk and Stbm localization with cytochalasin C , latrunculin , or blebbistatin has relatively little effect on MRLC9/12 localization . One potentially important difference between these two examples is that the aim mutants are null for pk from fertilization , while the cytoskeletal inhibitors are added post-intercalation . Thus , while we see no evidence of A/P polarization until approximately 1 hr after the completion of intercalation ( e . g . , Figure 2 ) , earlier events during intercalation may be required for the establishment of A/P polarity . The nature of these events and the mechanism that switches notochord polarity from mediolateral to A/P remain unknown . Adult C . intestinalis were collected from Santa Barbara Harbor or purchased from M-REP ( Carlsbad , CA , USA ) and kept at a facility supplied with raw seawater at the University of California , Santa Barbara . Gametes were isolated , mixed , and dechorionated as previously described . Anatomical landmarks were used to stage animal development ( Hotta et al . , 2007 ) . Plasmid constructs were electroporated into one-cell stage embryos ( Christiaen et al . , 2009 ) . The electroporation procedure typically results in mosaic expression , allowing us to score expressing cells in a background of non-expressing neighbors . Nuclei were visualized either using an existing C . intestinalis stable transgenic line expressing GFP under control of the brachyury promoter ( Joly et al . , 2007 ) , or by electroporation with Bra > GFP::H2A::RFP ( Kourakis et al . , 2014 ) . Cytochalasin B ( C2743 , Sigma , St . Louis , MO ) was provided at 10 mg/ml in DMSO and diluted in seawater to 10 µg/ml . Blebbistatin ( B0560 , Sigma ) was prepared as 10 mM in DMSO and diluted in seawater to a final concentration of 10 µM . Nocodazole ( M1404 , Sigma ) was diluted to 10 mM in DMSO and used at a final concentration of 10 µM in seawater . Latrunculin B ( L5288 , Sigma ) was prepared as a 10 mg/ml stock solution and diluted to a working concentration of 2 µg/ml or 10 µg/ml in sea water; both concentrations yielded similar phenotypes . Vinblastine ( V1377 , Sigma ) was added to a final concentration of ( 6 µg/ml ) . For each of these drugs , DMSO alone in seawater served as a negative control . An additional positive control was performed for working concentrations of nocodazole; for any trial using the microtubule inhibitor , we also confirmed that exposure to it resulted in cleavage arrest in embryos . A C . intestinalis MRLC cDNA ( KH2012:KH . C11 . 143; [Satou et al . , 2005] ) was PCR amplified ( forward primer: ATGTCGAGCCGACGAACTAAAAA; reverse primer: AATGTCATCTTTTTCTTTAGCGCCAT ) and cloned into pDONR221 . This was recombined with the Brachyury promoter ( forward primer: TAACGACGATTGTTCCGTCA; reverse primer: TATAGGTTTGTAACTCGCACTGAGC ) donor construct into pSP72-R3-ccdBCmR-R5-Rfa-Venus ( Roure et al . , 2007 ) to generate Bra > MRLC-Venus . The C . intestinalis Stbm cDNA ( KH2012:kh . C4 . 173 [Satou et al . , 2005] ) generated previously ( Jiang et al . , 2005 ) was cloned into pDONR221 , and recombined with the Brachyury promoter donor construct ( above ) into pSP72-R3-ccdBCmR-R5-Venus-Rfa ( Roure et al . , 2007 ) to generate Bra > Venus-Stbm . The Bra promoter sequence was cloned into pSP72BSSPE-SwaI-Rfa and recombined with an entry vector encoding amino acids 18–283 of human ensconsin ( pEntr > Ensconsin-3xGFP ) ( Roure et al . , 2007 ) to generate Bra > Ensconsin-3xGFP . A gBlock fragment ( IDT , Coralville , IA ) containing approximately 1 kb of the C-terminal portion of Myosin 10/11/14/9 ( Kh . C11 . 456; [Satou et al . , 2005] ) cDNA was generated to include a myc tag ( GCATCAATGCAGAAGCTGATCTCAGAGGAGGACCTG ) inserted immediately 5′ of the stop codon . This fragment was cloned into the full-ORF clone of Myosin 10/11/14/9 ( cima8817 ) at the SacII and Nde1 sites . This results in a full-length Myosin 10/11/14/9 cDNA with a C-terminal myc tag , including 5′ and 3′ UTRs . This donor construct was recombined with a Bra promoter containing destination vector to generate Bra > Myosin 10/11/14/9-myc . Embryos were fixed in 2% paraformaldehyde in seawater for 1 hr then washed 4 times in PBST , 10 min each . Embryos were incubated in PBST + 5% goat serum for 1 hr at room temperature , or overnight at 4°C , with the primary antibody diluted 1:1000 ( anti-GFP , anti-myc , anti-RFP; Invitrogen , Grand Island , NY ) , 1:650 ( anti-HA; Millipore , Billerica , MA ) , 1:300 ( anti-PKC ζ; Santa Cruz Biotechnology , Dallas , TX ) , or 1:250 ( anti-phospho-Myosin Light Chain 2 [Ser19]; Cell Signaling Technology , Danvers , MA ) . Animals were washed 5× for 10 min in PBST and then placed in an appropriate secondary antibody , and incubated as described for the primary antibody . Secondary antibodies used were anti-mouse or anti-rabbit Alexa Fluor-labeled antibodies ( Invitrogen ) with a range of excitation/emission spectra , depending on the experiment . Samples were washed 4 to 10× in PBST following secondary incubation . For microscopy , labeled embryos were immobilized onto cover slips coated with 0 . 08% Poly-L-lysine . Fixed embryos were cleared in 80% glycerol or in an isopropyl alcohol series followed by 2:1 benzyl alcohol:benzoyl benzoate ( BABB ) . Embryos were collected from natural spawning of aimless mutants , dechorionated , and injected as described previously ( Kourakis et al . , 2014 ) . An injection solution containing 0 . 5 μg of Myosin10/11/14/9-myc or MRLC9/12-Venus , 0 . 25 μg of Bra > H2A::RFP , and 0 . 5 mg/ml Fast Green ( Sigma ) was injected into one cell of a two or four-cell embryo . Embryos were fixed and immunostained ( as described above ) when they reached stage 23 .	Animal cells that form flat layers of a tissue , such as the skin or the lining of internal cavities , are often orientated in the same direction . The same is true for structures such as hairs or feathers , which are attached to the skin . This phenomenon is known as ‘planar cell polarity’ ( or ‘PCP’ for short ) . Many different organisms use similar mechanisms to establish this kind of tissue pattern . The best-studied mechanism involves the so-called ‘core PCP pathway’ . Signaling proteins in this pathway coordinate the polarity of neighboring cells . Other ‘global signaling pathways’ are thought to first ensure that tissues are correctly orientated within the embryo as a whole , and to do this , the global pathways are thought to align a network of filament-like structures within the cells in a particular direction . Once correctly orientated , these filaments—known as microtubules—have been proposed to help position the components of the core PCP pathway such that they can correctly orientate the rest of the cell . Now Newman-Smith , Kourakis et al . have identified another network of filaments within cells that interacts with components of the core PCP pathway in a sea squirt called Ciona savignyi . This organism begins life as a tadpole-like larva that has a flexible rod-shaped structure , called a ‘notochord’ , running along the length of its body . The cells of the notochord become polarized as they develop . When microtubules are disrupted , their planar polarity remains unaffected . However , when another network of filaments—called the actomyosin network––is chemically disrupted , the polarity of certain core PCP components is lost . The findings of Newman-Smith , Kourakis et al . reveal that the core PCP components and the actomyosin network in this sea squirt reinforce each other's polarity . This represents an alternative to the previously described models of planar polarity in which the core PCP components are thought to drive the polarization of the actomyosin network . Whether this model extends to planar cell polarity mechanisms in other organisms , such humans and other animals with backbones , remains a question for future work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Reciprocal and dynamic polarization of planar cell polarity core components and myosin
The ventricular-subventricular zone ( V-SVZ ) , on the walls of the lateral ventricles , harbors the largest neurogenic niche in the adult mouse brain . Previous work has shown that neural stem/progenitor cells ( NSPCs ) in different locations within the V-SVZ produce different subtypes of new neurons for the olfactory bulb . The molecular signatures that underlie this regional heterogeneity remain largely unknown . Here , we present a single-cell RNA-sequencing dataset of the adult mouse V-SVZ revealing two populations of NSPCs that reside in largely non-overlapping domains in either the dorsal or ventral V-SVZ . These regional differences in gene expression were further validated using a single-nucleus RNA-sequencing reference dataset of regionally microdissected domains of the V-SVZ and by immunocytochemistry and RNAscope localization . We also identify two subpopulations of young neurons that have gene expression profiles consistent with a dorsal or ventral origin . Interestingly , a subset of genes are dynamically expressed , but maintained , in the ventral or dorsal lineages . The study provides novel markers and territories to understand the region-specific regulation of adult neurogenesis . Neural stem/progenitor cells ( NSPC ) persist in the adult mouse brain in the walls of the forebrain ventricles . This neurogenic niche includes the ventricular-subventricular zones in the walls of the lateral ventricles ( V-SVZ ) , home to a subpopulation of astrocytes ( B cells ) that function as the NSPCs ( Chaker et al . , 2016; Doetsch et al . , 1997; Ihrie and Alvarez-Buylla , 2011; Lim and Alvarez-Buylla , 2014; Mirzadeh et al . , 2008 ) . This neurogenic region has also been referred to as the SVZ or the subependymal zone ( Kazanis et al . , 2017 ) . B cells generate intermediate progenitors ( C cells ) that , in turn , give rise to neuroblasts ( A cells ) that migrate to the olfactory bulb ( OB ) ( Obernier et al . , 2018; Ponti et al . , 2013 ) . A subpopulation of B cells also generate oligodendrocytes ( Figueres-Oñate et al . , 2019; Gonzalez-Perez , 2014; Kazanis et al . , 2017; Menn et al . , 2006; Nait-Oumesmar et al . , 1999; Picard-Riera et al . , 2002 ) . From the initial interpretation that adult NSPCs are multipotent and able to generate a wide range of neural cell types ( Morshead et al . , 1994; Reynolds and Weiss , 1992; van der Kooy and Weiss , 2000 ) , more recent work suggests that the adult NSPCs are heterogeneous and specialized , depending on their location , for the generation of specific types of neurons , and possibly glia ( Chaker et al . , 2016; Delgado et al . , 2020; Fiorelli et al . , 2015; Merkle et al . , 2014 , Merkle et al . , 2007; Tsai et al . , 2012 ) . Previous single-cell sequencing experiments in the V-SVZ have described the many broad classes of cells that reside in the niche . For example , transcriptional analyses after cell sorting have identified stages in the B-C-A cell lineage ( Borrett et al . , 2020; Codega et al . , 2014; Dulken et al . , 2017; Xie et al . , 2020 ) , as well as populations of NSPCs that appear to activate after injury ( Llorens-Bobadilla et al . , 2015 ) . Profiling of the entire niche has highlighted differences between quiescent and activated B cells ( Mizrak et al . , 2020; Zywitza et al . , 2018 ) . However , the differences among B cells of equivalent activation state ( e . g . quiescent , primed , or activated ) or the B cell heterogeneity that leads to the generation of diverse neuronal subtypes remain poorly understood . NSPC heterogeneity , interestingly , is largely driven by their location within the adult V-SVZ . This concept explains why young neurons in the OB originate over such a wide territory in the walls of the lateral ventricles . Multiple studies have begun to identify regional differences in gene expression among the lateral , septal , and subcallosal walls of the lateral ventricles ( Delgado et al . , 2021 ) . For example , differences in gene expression of B cells from the septal and lateral walls of the lateral ventricles have been recently observed ( Mizrak et al . , 2019 ) . Other studies have shown that Pax6 and Hopx-expressing cells correspond to dorsal V-SVZ progenitors ( Hack et al . , 2005; Kohwi et al . , 2005; Zweifel et al . , 2018 ) , and Vax1-expressing young neuroblasts are derived from ventral progenitors ( Coré et al . , 2020 ) . Spatially defined lineage-tracing studies using microinjections of viruses have identified subdomains of the V-SVZ that give rise to specific subtypes of OB neurons ( Merkle et al . , 2014; Merkle et al . , 2007; Ventura and Goldman , 2007 ) . Lineage-tracing studies have demonstrated that these regional subdomains largely follow the territories defined by developmentally regulated transcription factors including Pax6 , Nkx2 . 1 , Nkx6 . 2 , and Emx1 ( Delgado et al . , 2020; Delgado and Lim , 2015; Kohwi et al . , 2007 , Kohwi et al . , 2005; Merkle et al . , 2014; Willaime-Morawek et al . , 2006; Young et al . , 2007 ) , but the molecular differences among B cells underlying their regionally-restricted potential are largely unknown . Here , we have undertaken single-cell and single-nucleus RNA sequencing of the microdissected V-SVZ to gain insight into these important questions regarding NSPC heterogeneity and their developmental potential . Clustering analysis reveals strong dorso-ventral differences in lateral wall B cells . Validation of these differential gene expression patterns has revealed the anatomical boundary that separates these dorsal and ventral B cell domains . Additionally , our analysis identifies subpopulations of A cells defined by maturation state and dorso-ventral origin . We also find that a subset of dorso-ventral B cell transcriptional differences are retained through the C and A cell stages of the lineage . These new data advance our molecular understanding of how major region-specific neural lineages emerge in the adult V-SVZ and begin to delineate major functional subclasses of adult-born young neurons . For whole single-cell RNA sequencing ( scRNA-Seq ) , we dissected the lateral wall of the lateral ventricle from hGFAP:GFP mice at postnatal day ( P ) 29–35 ( n=8 , Figure 1A; Figure 1—figure supplement 1A ) . To determine possible sex differences in downstream analyses , two male and two female samples ( n=four samples total ) were dissociated and multiplexed by labeling cells with sample-specific MULTI-seq barcodes ( McGinnis et al . , 2019a ) . Multiplexed samples were then pooled for the remainder of the single-cell isolation protocol . Two technical replicates of pooled samples were loaded in separate lanes of the Chromium Controller chip ( 10x Genomics ) for single-cell barcoding and downstream mRNA library preparation and sequencing ( Figure 1A ) . Cells carrying multiple barcodes or a high number of mRNA reads ( 4128 out of 35 , 025 cells , 11 . 7% ) were considered doublets and were eliminated from downstream analysis . Data from each technical replicate were integrated for batch correction ( Stuart et al . , 2019 ) . We then performed unbiased clustering of cell profiles and calculated UMAP coordinates for data visualization ( Figure 1A ) . The clustering of lateral wall V-SVZ cells was not driven by sample , technical replicate , or sex ( Figure 1—figure supplement 1B–J ) . Cell cluster identities were annotated based on the detection of known cell type markers ( Figure 1B–C ) . We identified 37 clusters , with 14 clusters corresponding to cell types within the neurogenic lineage: NSPCs ( B cells ) , intermediate progenitors ( C cells ) , and neuroblasts ( A cells ) ( Doetsch et al . , 1999; Obernier et al . , 2018 ) . In addition , our analysis identified cell clusters corresponding to parenchymal astrocytes , ependymal cells , neurons , oligodendroglia , microglia , pericytes , vascular smooth muscle cells , and endothelial cells ( Figure 1B–C ) . NSPCs in the V-SVZ correspond to a subpopulation of astrocytes ( B cells ) derived from radial glia ( Doetsch et al . , 1999; Laywell et al . , 2000; Merkle et al . , 2004 ) . B cells have ultrastructure and markers of astrocytes ( Borrett et al . , 2020; Codega et al . , 2014 ) . Therefore , identifying markers that distinguish parenchymal astrocytes from B cells has been a challenge in the field . A fraction of both populations expressed Gfap: 51 . 85% of B cells ( clusters 5 , 13 , 14 , and 22 ) and 24 . 37% of parenchymal astrocytes ( clusters 21 , 26 , and 29 ) . This is consistent with previous reports ( Chai et al . , 2017; Ponti et al . , 2013; Xie et al . , 2020 ) . Note that across all cells captured in our scRNAseq analysis , only B cells , parenchymal astrocytes or ependymal cells expressed high levels of Gfap . Furthermore , among these three cell types , B cells had the highest average expression of Gfap ( 4 . 41 for B cells , 1 . 00 for astrocytes , 0 . 298 for ependymal cells , values in SCT corrected counts ) . Other markers , like S100a6 ( Kjell et al . , 2020 ) ( 88 . 9% of B cells; 54% of parenchymal astrocytes , and 80% of ependymal cells ) and Thbs4 ( Zywitza et al . , 2018 ) ( 45% of B cells; 28 . 77% in parenchymal astrocytes , 2 . 88% in ependymal cells ) are also expressed preferentially in B cells , but they alone do not distinguish these two cell populations ( Figure 1C , Figure 1—figure supplement 2A , B ) . We performed differential gene expression analysis to further distinguish B cells from parenchymal astrocytes . We found that B cells had a higher expression of Maff , Zfp36 , Bex4 , Lgals3 , and Anxa2 compared to parenchymal astrocytes , which in turn were enriched for Clmn , Atp13a4 , Eps8 , Pcdh7 , and Syne1 ( Figure 1—figure supplement 2C , Supplementary file 1 ) . To better understand the biological differences between parenchymal astrocytes and B cells , we performed gene ontology ( GO ) enrichment on the differentially expressed genes . Genes associated with synapse regulation ( GO:0051965 and 0051963 ) , macroautophagy ( GO: 0016241 ) , and dendrite development and morphogenesis ( GO: 0016358 and 0048813 ) , among others , were overrepresented in parenchymal astrocytes compared to B cells ( Figure 1—figure supplement 2D ) . In contrast , B cells were enriched in terms associated with RNA regulation ( GO:0000463 , 0034471 , 0000956 , and 0000966 ) and mitochondrial regulation ( GO:006626 , 0073655 , 0090201 , and 0010823 ) ( Figure 1—figure supplement 2D ) . These differentially represented GO terms support the known neuro-regulatory function of parenchymal astrocytes , as well as the increased transcriptional regulation that has been associated with the transition of NSPCs from quiescent to activated states ( Dulken et al . , 2017; Llorens-Bobadilla et al . , 2015 ) . Using this cell type classification , we included B cell clusters , but not striatal astrocytes , in our downstream analysis of the neurogenic lineage . In our scRNA-Seq dataset , we found that the majority of cells were part of the neurogenic lineage , which is composed of primary progenitors , intermediate progenitors , and young neurons ( the B-C-A cell lineage ) . These neurogenic lineage clusters were in the center of the UMAP plot . With this analysis , the data had a ‘hummingbird-like’ shape with B cells ( clusters 5 , 13 , 14 , and 22 ) at the top in the bird’s head and neck , proliferating cells ( clusters 8 , 10 , 16 , and 17 ) and C cells ( cluster 12 ) resembled the bird’s body and wing , and A cells ( clusters 0 , 1 , 4 , 6 , and 15 ) formed a tail ( Figure 2A ) . As expected , all B cell clusters expressed Gfap , GFP , and S100a6 ( Figure 2B–D ) . We identified a subpopulation of B cells as the quiescent B cells ( Codega et al . , 2014; Llorens-Bobadilla et al . , 2015 ) ( clusters 5 , 14 , and 22 ) characterized by high expression of Thbs4 and Gfap , no Egfr , and low Ascl1 expression ( Figure 2B , E , F , G ) . In contrast , cluster 13 ( the neck region of the ‘hummingbird’ ) corresponded to activated B cells with lower expression of Gfap , GFP , and S100a6 , but high Egfr and Ascl1 expression ( Codega et al . , 2014; Figure 2B–D , F–G ) . Cluster 13 cells also expressed Notum , a marker recently associated with activating qNSPCs ( Mizrak et al . , 2020; Figure 2H ) . Bordering cluster 13 in the chest region of the ‘bird’ was cluster 12 , identified as the C cell cluster , which had low expression of astrocytic markers Gfap and GFP ( Figure 2B–C ) , but high expression of Egfr and Ascl1 ( Figure 2F–G ) . The wing area of the ‘bird’ contained Mki67+ proliferating cells ( Figure 2I ) , and in the tail area of the ‘bird’ below , clusters corresponding to neuroblasts were characterized by high expression of Dcx+ ( Figure 2J ) . We used cell-cycle scoring to classify cells by their G2M , S , or G1 phase ( Tirosh et al . , 2016 ) and found that a high number of cells in cluster 12 were in the S phase ( 691/992 , 69 . 6% , Supplementary file 2 ) , consistent with cluster 12 corresponding to intermediate progenitors ( C cells ) ( Figure 2K ) . The unbiased clustering analysis pooled dividing cells into the wing region of the ‘hummingbird’ . A closer look at these clusters of mitotic cells in G2 or metaphase ( G2M; clusters 10 , 16 , 17 , and 8 ) showed that their gene expression pattern overlaps with that of the non-dividing neurogenic lineage cell progression: Cluster 10 expresses markers of B cells ( GFP , Notum ) , cluster 16 had markers of C cells ( Ascl1 ) , and cluster 8 expresses markers of A cells ( Dcx ) ( Figure 2B–K ) . This suggests that these clusters correspond to dividing B , C , and A cells , respectively , which have all been previously observed by electron and confocal microscopy in vivo ( Doetsch et al . , 1997 ) . Identified by Dcx expression , the largest number of cells within the neurogenic lineage corresponded to neuroblasts and young neurons ( A cells ) ( clusters 0 , 1 , 4 , 6 , and 15; the tail region of the hummingbird ) ( Figure 2A , J ) . Interestingly , unbiased clustering subdivided A cells into five subclusters with different gene expression profiles . Consistent with previous work showing that a subpopulation of newly generated neurons continues to divide ( Lois and Alvarez-Buylla , 1993; Menezes et al . , 1995 ) , clusters 6 and 15 contain Dcx+ A cells with proliferative markers ( e . g . Mki67 ) ( Figure 2I–K ) . Genes that distinguished clusters 0 , 1 , 4 , 6 , and 15 are discussed below . The overall progression from B-C-A cells described above is supported by RNA-velocity lineage trajectory reconstruction ( La Manno et al . , 2018 ) , in which genes defining B , C , and A cells are expressed sequentially in distinct phases in pseudotime ( Figure 2L–M ) . Overall , our single-cell dataset recapitulates the known B-C-A cell progression through the neurogenic lineage . Intriguingly , our analysis also reveals heterogeneity among B , C , and A cells , in which each cell type is subdivided into multiple distinct clusters . Among C cells , the heterogeneity was mostly driven by different stages of the cell cycle ( Figure 2F–G , I , K ) . What drives heterogeneity among B and A cell clusters ? In our scRNA-Seq dataset , we found that quiescent B cells ( the head of the ‘bird’; Figure 2A ) were subdivided into three clusters: B cell cluster 5 ( B ( 5 ) ) , B ( 14 ) , and B ( 22 ) ( Figure 3—figure supplement 1A ) . To understand their molecular differences , we conducted differential expression analysis to identify significantly upregulated genes in each of the three B cell clusters ( Figure 3—figure supplement 1B ( i ) ) and candidate cluster-specific marker genes ( Figure 3—figure supplement 1B ( iI ) ) ( Supplementary file 3 ) . When we examined the top ten candidate markers for each cluster , we found genes corresponding to known markers of dorsal and ventral B cell identity ( Figure 3—figure supplement 1C ) . For example , Nkx6 . 2 , a transcription factor expressed in the ventral embryonic and postnatal ventricular zone ( Merkle et al . , 2014; Moreno-Bravo et al . , 2010 ) , is enriched in cluster B ( 14 ) , as are Notum and Lmo1 ( Figure 3A , Figure 3—figure supplement 1C; Borrett et al . , 2020; Mizrak et al . , 2020 ) . Similarly , Gsx2 , a marker of dorsal B cells , is a marker of cluster B ( 5 ) ( Figure 3B , Figure 3—figure supplement 1C ) . Another known dorsal marker , Emx1 , is significantly upregulated in cluster B ( 22 ) ( Figure 3C , Supplementary file 3 ) . When we overlay the expression of these cluster markers on the neurogenic lineage UMAP plot , we find that their expression is largely restricted to cells within a single B cell cluster , and in the case of Gsx2 , is retained in C cells and early-stage A cells ( Figure 3A–C ) . Activated B cells in cluster B ( 13 ) also showed the expression of these regional markers but did not separate into multiple clusters at this resolution ( Figure 2A ) . The distinct regional signature that drives clustering for quiescent B cells seems to be overshadowed by activation and proliferation-related genes as these primary progenitors progress into the neurogenic lineage . To confirm regional organization as a major source of heterogeneity in our scRNA-seq dataset , we turned to a previously generated single-nucleus RNA sequencing ( sNucRNA-Seq ) dataset we had derived from regionally microdissected V-SVZ . We performed single-nucleus sequencing ( sNucRNA-Seq ) from microdissected V-SVZ subregions of P35 CD1 mice ( n=eight males , nine females ) . We isolated single nuclei from four microdissected quadrants of the V-SVZ: the anterior-dorsal ( AD ) , posterior-dorsal ( PD ) , anterior-ventral ( AV ) , and posterior-ventral ( PV ) regions ( Figure 3—figure supplement 2A; Mirzadeh et al . , 2008 ) . The four region samples ( AD , PD , AV , and PV ) were then processed in parallel for sNucRNA-Seq ( Figure 3—figure supplement 2B ) . Our sNucRNA-Seq dataset contains 45 , 820 nucleus profiles . The four region samples underwent quality control steps of filtering out low-quality cells and putative doublets ( see Materials and methods ) . Data from each sample were combined and integrated ( Seurat v3 IntegrateData ) ( Stuart et al . , 2019 ) , then clustered as the scRNA-Seq dataset above . The AD and PD samples had a lower number of cells , but higher sequencing depth compared to the ventral samples ( Figure 3—figure supplement 2I ) . Cell identities were annotated based on the detection of previously described cell type markers ( Märtin et al . , 2019; Zeisel et al . , 2018; Figure 3—figure supplement 2C–E ) . In the sNucRNA-Seq data , we identified 42 clusters , including those corresponding to cell types within the neurogenic lineage: NSPCs ( B cells ) , mitotic intermediate progenitors ( C cells ) , and neuroblasts ( A cells ) ( Figure 3—figure supplement 2C–E ) . Based on the B cell- or astrocyte-specific markers identified in the scRNA-Seq data above , we also identify a parenchymal astrocyte cluster , as well as ependymal cells , striatal neurons , oligodendroglia , microglia , pericytes and vascular smooth muscle cells , endothelial cells , and leptomeningeal cells ( Figure 3—figure supplement 2D–E ) . We also found that all four regions contributed to most clusters ( Figure 3—figure supplement 2F–H ) . To test the hypothesis that B cells are dorso-ventrally organized in the V-SVZ , we took advantage of the region-specific microdissection of the sNucRNA-Seq cells ( Figure 3—figure supplement 2 ) and metadata Label Transfer to predict scRNA-Seq B cell region identity ( Stuart et al . , 2019 ) . Each B cell was assigned both a dorsal and ventral ‘predicted identity’ score based on their similarity to dorsal and ventral nuclei ( Figure 3D ) . We then calculated the difference between dorsal and ventral scores for each scRNA-Seq B cell . We found that cells within each cluster were strongly dorsal-scoring ( green ) or ventral-scoring ( magenta ) , with relatively few cells having similar dorsal and ventral prediction scores ( gray ) ( Figure 3E ) . We found that cluster B ( 14 ) scored more highly for ventral identity on average , while clusters B ( 5 ) and B ( 22 ) scored more highly for dorsal identity ( Figure 3—figure supplement 1D ) . To investigate the potential dorso-ventral spatial organization of dorsal-scoring clusters B ( 5 ) and B ( 22 ) in vivo , we performed RNAscope in situ hybridization for the differentially expressed transcripts Small Nucleolar RNA Host Gene 15 ( Snhg15 , a lncRNA ) , and Contactin-Associated Protein 2 ( Cntnap2 ) , which are upregulated in clusters B ( 5 ) and B ( 22 ) , respectively ( Figure 3—figure supplement 1E–L ) . Overall , we found that Snhg15 and Cntnap2 probes did not exclusively localize in B cells , but were also expressed in subsets of A cells , C cells , and ependymal cells , making it difficult to confirm their B cell expression along the dorsal-ventral axis of the V-SVZ ( Figure 3—figure supplement 1F–H , J–L ) . To identify markers more specific to B cells within the V-SVZ , and test the hypothesis that B ( 5 ) and B ( 22 ) both correspond to dorsally localized B cells , we combined them into a single cluster , B ( 5+22 ) . Looking at this new cluster’s predicted dorsal and ventral identity scores , we found that it had a much higher average dorsal predicted identity score , as well as a lower average ventral predicted identity score than when scored separately , nearly the inverse of cluster B ( 14 ) ’s prediction scores ( Figure 3F ) . This unsupervised , unbiased prediction of region identity at the single-cell level , based on sNucRNA-Seq region-specific microdissection , reinforces our observation that scRNA-Seq B cell clusters have strong gene expression signatures of dorsal or ventral V-SVZ identity . We then asked what genes were differentially expressed between the putative dorsal cluster B ( 5+22 ) and the putative ventral cluster B ( 14 ) ( Figure 3G–I ) . Among the differentially expressed genes were other known markers of dorsal B cells , such as Pax6 and Hopx ( Figure 3Hi , Supplementary file 3 ) , as well as novel dorsal-domain marker candidates such as Urah and Dio2 ( Figure 3Hii , I ) . We found that Urah and Dio2 were highly expressed among GFAP+ cells in the V-SVZ dorsal ‘wedge’ ( the dorso-lateral corner of the V-SVZ enriched in B , C , and A cells ) , while lower expression was observed in the intermediate and ventral regions ( Figure 3J–S ) . Density plots of RNAscope spot number along the ventricular wall showed a higher number of RNAscope spots in the dorsal V-SVZ for Urah and Dio2 ( Figure 3M , R ) . This pattern of expression of these dorsal markers was maintained from anterior to posterior sections ( Figure 3M , N , R , S ) . To determine if the dorsal Urah+ and Dio2+ domains overlap with the dorsal Hopx+ B cell population ( Zweifel et al . , 2018 ) , we performed RNAscope in situ hybridization for Hopx . Hopx was expressed in B cells in the dorsal wedge with no expression in the intermediate and ventral V-SVZ regions ( Figure 3—figure supplement 1M–Q ) . Interestingly , low and high Hopx+ cells were observed in the wedge . Hopxhigh cells showed a subcallosal localization forming a band that extended laterally from the septal corner of the V-SVZ , where Hopx+ cells have been previously described ( Zweifel et al . , 2018; Figure 3—figure supplement 1N–O , Q ) . We confirmed that HOPX protein followed the expression pattern of its mRNA by immunostaining ( Figure 3—figure supplement 1R ) . Cluster B ( 14 ) marker Crym was expressed in intermediate and ventral V-SVZ GFAP+ cells ( Figure 3T–X ) , with very little expression in dorsal V-SVZ and wedge regions ( Figure 3T–X ) . Consistent with our single-cell data , A cells identified by DCX expression were Crym-negative ( Figure 3T–V ) , but a subpopulation of cells in the striatum expressed Crym ( Figure 3U–V ) , which is consistent with previous work ( Chai et al . , 2017; Mizrak et al . , 2019 ) . To determine if CRYM protein in the V-SVZ was also expressed in a regional pattern , we used antibody labeling to study its expression in B cells using P28 hGFAP:GFP mice . Consistent with the RNAscope analysis , ventral SVZ GFP+ B cells were CRYM+ , while dorsal wedge GFP+ cells were negative ( Figure 4A–C ) . We defined two domains in our sections ( see Materials and methods , Figure 3—figure supplement 1S ) and quantified the percent of GFP+ B cells that were CRYM+ . There was a sharp difference with only 2 . 47% GFP+ cells being CRYM+ in the dorsal domain , while 97 . 67% of the ventral GFP+ B cells were CRYM+ . Consistent with our observations using RNAscope ( Figure 3U ) , whole-mount analysis ( Mirzadeh et al . , 2008 ) , and CRYM antibody labeling showed how the domain of CRYM expression increased in size from the caudal to rostral V-SVZ ( Figure 4D–G ) . Using the hGFAP-GFP mice and staining for ß-Catenin to reveal the apical domain of B1 cells and their pinwheel organization ( Mirzadeh et al . , 2008 ) , we confirmed that CRYM was expressed in GFP+ ventral B1 cells , but was largely absent from B1 cells dorsally ( Figure 4H–I ) . Quantifications in these whole mounts showed that 95 . 11% ± 2 . 65 ( SD ) of the GFP+ B1 cells were CRYM+ in the ventral domain of the V-SVZ . In contrast , only 4 . 71% ± 1 . 38 ( SD ) of the GFP+ B1 cells in the dorsal region were CRYM+ ( n=3; T-test , p<0 . 0001 ) . To clearly show at a higher resolution and in sections that B1 cells and their apical process has CRYM expression , we used transmission electron microscopy . Immunogold staining showed that B1 cells in the ventral domain express CRYM ( Figure 4 . J-K ) . Taken together , the above analysis showed that Crym transcript and protein expression defined a wide ventral territory that decreased caudally , and was largely absent from the dorsal V-SVZ including the wedge region . We then conducted GO analysis on differentially expressed genes between dorsal B ( 5+22 ) and ventral B ( 14 ) cells and found that differentially expressed genes in ventral B cells had an overrepresentation of genes involved in the response to growth hormone ( GO:0060416 ) and retinal ganglion cell axon guidance ( GO:0031290 ) , while dorsal B cells were associated with oligodendrocyte differentiation ( GO:0048709 ) , forebrain generation of neurons ( GO:0021872 ) , and central nervous system neuron axonogenesis ( GO:0021955 ) ( Supplementary file 3 ) . To understand the relationships between differentially expressed genes within dorsal and ventral B cell populations , we used gene regulatory network ( GRN ) analysis . We constructed GRNs based on predicted interactions between the top 10 markers in dorsal B ( 5+22 ) cells or ventral B ( 14 ) cells and all other genes expressed in each cluster . Dorsal marker genes and their interaction partners formed one large network of 146 genes and three smaller networks with 11–20 genes each ( Figure 3—figure supplement 3A ) , while ventral markers formed two large networks ( 118 and 91 genes ) and three smaller ones containing 20–32 genes each ( Figure 3—figure supplement 3B ) . The dorsal network was much more highly interconnected than the ventral network ( dorsal clustering coefficient = 0 . 121 , ventral clustering coefficient = 0 . 005 ) . The central dorsal network was enriched for genes associated with regulation of apoptosis ( GO:0042981 ) and central nervous system axonogenesis ( GO:0021955 ) , while ventral networks were enriched for genes associated with cell fate commitment ( GO:0045165 ) , regulation of GABAergic synaptic transmission ( GO:0032228 ) , and regulation of transcription ( GO:0045449 ) . Notably , the dorsal network contained all three genes that make up nuclear receptor subfamily 4A ( Nr4a1 , Nr4a2 , and Nr4a3 ) , members of the steroid-thyroid hormone receptor superfamily that are linked to neuronal specification , axon guidance , and neurotransmission ( Jeanneteau et al . , 2018; Luo et al . , 2008; Soldati et al . , 2012 ) . Additionally , phospholipase A2 group VII ( Pla2g7 ) is predicted to regulate the ventral marker Crym in dorsal cells , likely in a negative direction . To uncover the regulatory context of Crym and Urah , we built GRNs based on their predicted relationships with genes in putative ventral and dorsal B cells , respectively . We identified two modules related to Urah expression in dorsal B ( 5+22 ) cells ( Figure 3—figure supplement 4A ) and six related to Crym expression in ventral B ( 14 ) cells ( Figure 3—figure supplement 4B ) . Urah-associated modules contained genes related to neurogenesis ( GO:0022008 ) , glial cell differentiation ( GO:0010001 ) , and hormone metabolic processes ( GO:0042445 ) , and Crym-associated modules contained genes related to developmental growth ( GO:0048589 ) , lipid catabolic processes ( GO:0044242 ) , and amine metabolism ( GO:0009308 ) . These sets of interacting genes provide clues toward the ways Urah and Crym may contribute to regional identity . Together , these analyses suggest that differentially expressed genes between these two cell populations are functionally related , working to produce unique cell behaviors that may underlie important differences between B cells in dorsal and ventral domains . We found that A cells ( Figure 5A ) were separated into two main sets of transcriptionally-related clusters: clusters A ( 15 ) and A ( 6 ) corresponded to A cells with a strong expression of genes associated with mitosis and cell cycle regulation ( such as Pclaf , Hmgb2 , Rrm2 , Mki67 , Top2a , and the Mcm gene family ) ( Figure 5B , Supplementary file 4 ) . We calculated the ‘area under the curve’ ( AUC ) scores ( Aibar et al . , 2017 ) for sets of genes corresponding to key GO terms in each A cell . The combined expression of genes in both the GO categories mitotic DNA replication ( GO:1902969 ) and mitosis DNA replication initiation ( GO:1902975 ) were highly upregulated in clusters A ( 15 ) and A ( 6 ) ( Figure 5—figure supplement 1A , Supplementary file 4 ) . These clusters are also enriched in cells in S and G2M phases ( Figure 2K ) , indicating that these clusters correspond to dividing neuroblasts/early A cells ( Lois and Alvarez-Buylla , 1993; Menezes et al . , 1995 ) . Clusters A ( 1 ) , A ( 0 ) , and A ( 4 ) corresponded to a second set of A cells expressing high levels of genes involved in cell migration , such as Dab1 and Slit2 ( Figure 5B; Supplementary file 4 ) . We found that neuron migration ( GO:0001764 ) and spontaneous synaptic transmission ( GO: 0098814 ) categories were more strongly present in clusters A ( 0 ) , A ( 1 ) , and A ( 4 ) ( Figure 5—figure supplement 1A ) , indicating that these clusters likely correspond to migrating young neurons . This analysis of the GO terms enriched in A cells is consistent with the pseudotime analysis ( Figure 2L–M ) , indicating that A cells are organized in a continuum of maturation in UMAP space , with dividing neuroblasts at the top of the A cell cluster group and migrating young neurons at the bottom . Interestingly , the combined expression of genes in dorsoventral axonal guidance ( GO:0033563 ) and cerebral cortex regionalization ( GO:0021796 ) , terms associated with the regional specification of the brain , was high in clusters A ( 1 ) and A ( 0 ) , despite individual GO terms not being statistically enriched in these clusters ( Figure 5—figure supplement 1A; Supplementary file 4 ) . Additionally , Runx1t1 , a transcription factor expressed in young neurons from the medial ganglionic eminence ( Chen et al . , 2017 ) , and Nxph1 , expressed in young migrating neurons from subpallial germinal zones ( Batista-Brito et al . , 2008 ) , were among the most differentially expressed genes in cluster A ( 1 ) ( Figure 5B–C , Supplementary file 4 ) . Conversely , Pax6 , a transcription factor that is highly expressed in the pallium and in the dorsal lateral ganglionic eminence ( Ypsilanti and Rubenstein , 2016 ) , was the most differentially expressed gene for cluster A ( 0 ) ( Figure 5B–C , Supplementary file 4 ) . Importantly , Vax1 and Pax6 , which have been previously found to be differentially expressed by ventrally- and dorsally-born A cells ( Coré et al . , 2020 ) , were highly expressed in clusters A ( 1 ) and A ( 0 ) , respectively ( Figure 5B–C ) . Taken together , gene expression suggests that A cells in cluster A ( 1 ) originate from ventral progenitors and A cells in cluster A ( 0 ) originate from dorsal progenitors . To further support this interpretation , we took advantage of the regional microdissections from the sNucRNA-Seq dataset and scored A cells based on their similarity to A cells from ventral and dorsal microdissections ( Figure 5D ) . Among the more mature A cell clusters , we found that cluster A ( 0 ) was enriched in cells that showed high correspondence with A cells from the dorsal dissection , with relatively few cells that had high correspondence with A cells of the ventral dissection ( 1340 predicted dorsal dissection; 399 predicted ventral dissection ) . In contrast , cluster A ( 1 ) had cells with high correspondence to A cells found in both the ventral and dorsal dissections ( 925 predicted dorsal dissection; 761 predicted ventral dissection ) ( Figure 5D , Figure 5—figure supplement 1D ) . This is consistent with the inferred patterns of migrations observed in the walls of the lateral ventricles with many A cells from ventral and dorsal origins accumulating anteriorly to join the RMS and a smaller number of cells migrating from dorsal to ventral locations ( Sawamoto et al . , 2006 ) . We do not see a similar mixing of cells in the non-migratory B cell population ( Figure 3E–F ) . In order to confirm that A cells in A ( 0 ) and A ( 1 ) correspond to dorsal and ventral young neurons , respectively , we looked for markers of A ( 1 ) and A ( 0 ) that were minimally present in the other cluster . Trhde and Ntng1 were expressed in subpopulations of A cells with high dorsal and ventral dissection scores , respectively ( Figure 5E ) . We used RNAscope in situ hybridization with DCX immunolabeling to visualize the localization of transcripts in A cells in vivo . As suggested by the expression pattern in the scRNA-Seq clusters , Ntng1 was not uniformly expressed in all A cells , but in a subset of them ( Figure 5F–H ) , in both dorsal and ventral V-SVZ ( Figure 5F–H ) . Trhde puncta were present in the V-SVZ , along the ventricular lining , and in areas immediately lateral to the DAPI-dense band of V-SVZ B , C , and A cells ( Figure 5—figure supplement 1B ) . This expression pattern is consistent with Trhde expression in both the ependymal and striatal neuron scRNA-Seq clusters ( Figure 5—figure supplement 1C ) . We found few Trhde+ DCX+ cells , in either the dorsal or ventral V-SVZ , where each positive A cell had one or two puncta that did not co-localize with the nucleus ( Figure 5—figure supplement 1B ) . In addition to Ntng1 and Trhde being found in DCX+ A cells dispersed along the full dorso-ventral axis of the V-SVZ ( Figure 5F–H; Figure 5—figure supplement 1B ) , in the DCX-dense RMS corridor , we found relatively small subsets of A cells that were Ntng1+ or Trhde+ , with only one or two mRNA puncta per cell ( Figure 5I–J ) . Given the intermixing of A cells as they undergo tangential chain migration in the V-SVZ , the dorso-ventral sites of A cell origins cannot be visualized and validated by immunostaining or RNAscope ( Figure 5F–H , Figure 5—figure supplement 1B ) . This is consistent with the expected spatial migration patterns of ventrally- vs . dorsally-born A cells ( Fiorelli et al . , 2015; Sawamoto et al . , 2006 ) . Overall , we found that gene expression patterns , as well as independent , unbiased cell identity prediction provided additional support to the hypothesis that Pax6high;Rlbp1high cluster A ( 0 ) represents primarily dorsally born A cells , while Runx1t1high;Vax1high cluster A ( 1 ) represents primarily ventrally born A cells ( Figure 5B–D ) . To better understand the differences between clusters A ( 0 ) and A ( 1 ) , we performed a gene ontology ( GO ) analysis in the differentially expressed genes between these clusters . Differentially expressed genes in A ( 1 ) were enriched for genes associated with particular neurite outgrowth and migratory programs , such as regulation of negative chemotaxis ( GO:0050923 ) , dorsal/ventral axon guidance ( GO:0033563 ) , and V-SVZ-to-OB migration ( GO:0022028 ) , suggesting unique and salient features of the migration and guidance preferentially used by ventrally derived A cells . In particular , cluster A ( 1 ) was enriched in Slit1 and Slit2 , genes that encode guidance cues . Among the differentially expressed genes in cluster A ( 0 ) , we found an overrepresentation of genes involved in presynaptic membrane assembly ( GO:0097105 ) , postsynaptic density assembly ( GO:0097107 ) , and synaptic vesicle clustering ( GO:0097091 ) such as Nrxn1 and Nlgn1 that were upregulated in dorsally derived A cells . Interestingly , a different set of genes involved in axon guidance ( GO:0007411 ) was upregulated in cluster A ( 0 ) , including DCC , Efna5 , and Sema6d . ( Supplementary file 4 ) . Together these data suggest that A cells are a molecularly heterogeneous population and that heterogeneity is indicative of an A cell’s region of origin within the V-SVZ . Interestingly , A ( 1 ) marker Slit2 ( Supplementary file 4 ) and A ( 0 ) marker Pax6 ( Figure 5B–C ) were also highly expressed in clusters corresponding to ventral and dorsal B cells , respectively ( Figure 3I; Supplementary file 3; see below ) . In addition , Rlbp1 , one of the main markers of dorsal B cells ( Figure 3I ) , was also strongly expressed in cluster A ( 0 ) ( Figure 5C ) . To further determine if each dorsal/ventral domain has a specific gene expression signature that persists throughout the neurogenic lineage , we asked which B and A cell subpopulation marker genes were commonly expressed between the ventral B and ventral A cell clusters ( B ( 14 ) and A ( 1 ) ) , and which were commonly expressed between the dorsal B ( 5+22 ) and dorsal A ( 0 ) clusters ( Figure 6A–B ) . We identified four genes that were expressed throughout the dorsal lineage ( Rlbp1 , Gm29260 , Pax6 , and Dcc ) and five genes expressed through the ventral lineage ( Adgrl3 , Slit2 , Ptprn2 , Rbms1 , Sntb1 ) ( Figure 6C ) . The converse comparison , however , of ventral B ( 14 ) and dorsal A ( 0 ) , and dorsal B ( 5+22 ) and ventral A ( 1 ) clusters , yielded only one or two potential lineage marker genes , respectively ( Figure 6—figure supplement 1A–B ) . This suggests that B ( 14 ) and A ( 1 ) , and B ( 5+22 ) and A ( 0 ) had a higher degree of transcriptional overlap , and correspond to ventral and dorsal lineages , respectively . To understand the molecular differences between the putative dorsal and ventral lineages , we used the regional gene sets we identified above to calculate a composite AUC score ( Aibar et al . , 2017 ) for both the dorsal and ventral gene expression signatures ( Figure 6D , Figure 6—figure supplement 1C ) . We found that cells that scored highly for the dorsal genes were largely located on the right side of the ‘bird’ , and the highest ventral-scoring cells were on its left side ( Figure 6D , Figure 6—figure supplement 1C–E ) . In order to understand the functional differences between cells in dorsal and ventral lineages , we normalized the dorsal and ventral scores ( see Methods ) and selected the top-scoring quartile for each lineage ( High Score lineages ) to compare their gene expression ( Figure 6E ) . We found that genes enriched in dorsal or ventral lineages have dynamic expression patterns along pseudotime . For example , Dcc , a netrin-1 receptor , has its peak of expression in dorsal A cells; curiously Slit2 , another guidance molecule , had its peak expression in activated B cells . ( Figure 6C , F , G , Figure 6—figure supplement 1F–G ) . However , the scores of the combined expression for dorsal and ventral lineage signatures were relatively constant throughout the B-C-A cell lineage progression ( Figure 6D ) . We identified 257 significantly differentially expressed genes between the High Score lineages: 108 dorsal markers and 149 ventral markers ( Figure 6H; Supplementary file 5 ) . We then conducted GO analysis on differentially expressed genes between High Score dorsal and ventral cells from clusters B ( 5+22 ) and B ( 14 ) , respectively; and between the High Score dorsal and ventral A cells from clusters A ( 0 ) and A ( 1 ) ( Figure 6A , Figure 6—figure supplement 1 ) . Among the differentially expressed genes in the High Score dorsal B cells , we found an overrepresentation of the thyroid-stimulating hormone secretion ( GO:0070460 ) with higher expression of Dio2 and Slc16a2 . DIO2 converts T4 into the bioactive thyroid hormone T3 and Slc16a2 encoding MCT8 , a major thyroid hormone transporter ( Bernal et al . , 2015 ) . Rostrocaudal neural tube patterning ( GO:0021903 ) , and positive regulation of notch signaling pathway ( GO:0045747 ) categories were also overrepresented . Hes1 and Hes5 , commonly used as readouts of Notch activation ( Ohtsuka et al . , 1999 ) , were upregulated in the High Score dorsal B cells . The High Score ventral B cells had an enrichment in genes involved in dorsal/ventral pattern formation ( GO:0009953 ) , response to laminar fluid shear stress ( GO:0034616 ) , adherens junction assembly ( GO:0034333 ) , and cerebrospinal fluid circulation ( GO:0090660 ) . We also found an enrichment for signaling pathways in the High Score ventral B cells: PDGFR-signaling pathway ( GO:0048008 ) , cellular response to epidermal growth factor stimulus ( GO:0071364 ) , and insulin receptor signaling pathway ( GO:0008286 ) . Dorsal A cells had a higher expression of Nlgn1 and Nrxn1 , genes in the NMDA glutamate receptor clustering category , and also had presynaptic membrane assembly ( GO:0097105 ) and semaphorin-plexin signaling ( GO:0071526 ) categories overrepresented . High Score ventral A cells were strongly enriched in Slit1 , Slit2 , and Robo1 , genes involved in the regulation of negative chemotaxis ( GO:0050923 ) . These cells were also enriched in Cacna1c and Slc8a1 , genes involved in calcium ion import across the plasma membrane ( GO:0098703 ) and calcium ion import into the cytosol ( GO:1902656 ) ( Figure 6—figure supplement 1H; Supplementary file 5 ) . To validate RNA expression of putative dorsal and ventral lineage markers in vivo , we combined RNAscope labeling with GFAP and DCX immunostaining in coronal sections of the V-SVZ . We found that the putative dorsal lineage genes Pax6 and Rlbp1 were highly enriched in the dorsal region of the V-SVZ , with particularly enriched expression in the ‘wedge’ region ( Figure 6I–L , O; Figure 6—figure supplement 1I–L ) . Density plots of RNAscope spot number along the ventricular wall showed a higher number of RNAscope spots in the dorsal V-SVZ for Rlbp1+DCX- and Rlbp1+DCX+ cells ( Figure 6M–N ) . Conversely , the putative ventral markers Adgrl3 and Slit2 had a higher expression in the ventral domain of the V-SVZ ( Figure 6P–S , V; Figure 6—figure supplement 1L–O ) . Interestingly , density plots of RNAscope spots for Slit2+DCX- cells showed higher numbers of spots ventrally , while Slit2+DCX+ were equally distributed along the ventricular wall . Consistent with our scRNA-Seq data , these genes were expressed in both B and A cells in the neurogenic lineage ( Figure 6J , L , Q , S ) . During brain development , regional allocations of the neuroepithelium give NSPCs different neurogenic properties . The adult V-SVZ neurogenic niche retains regionally specified NSPCs that generate different subtypes of neurons destined for the OB . A molecular understanding of what makes adult NSPCs different between regions is largely lacking . Our scRNA-Seq and sNucRNA-Seq datasets provide new information about the diverse cell types that populate the V-SVZ . Our lineage analysis reveals parallel pathways of neurogenesis initiated by different populations of B cells . Interestingly , these differences in B cell identity correlate with unique regional patterns of gene expression , which we validated using reference-based metadata label transfer from the second dataset of regionally dissected single V-SVZ nuclei . We confirmed the regional expression of marker genes by immunostaining and RNAscope analysis . Regional differences in NSPCs potential were demonstrated using restricted viral labeling of non-overlapping territories of the V-SVZ ( Merkle et al . , 2014; Merkle et al . , 2007; Ventura and Goldman , 2007 ) . Labeled ventral B cells produced deep layer granule neurons , calbindin-positive periglomerular cells , and type 1–4 cells in the OB , while dorsal B cells produced superficial granule neurons and tyrosine hydroxylase-positive periglomerular cells ( Merkle et al . , 2014; Merkle et al . , 2007 ) . Similarly , genetic lineage tracing from territories expressing regionally restricted transcription factors also indicates that NSPCs in dorsal and ventral territories generate superficial and deep-layer neurons for the OB , respectively ( Kohwi et al . , 2007; Kohwi et al . , 2005; Merkle et al . , 2014; Young et al . , 2007 ) . It has been suggested that in the adult V-SVZ , a more primitive population of Oct4+/GFAP- NSCs may be present and that these cells may be earlier in the lineage from the ‘definitive’ GFAP+ B cells ( Reeve et al . , 2017 ) . However , regionally specified NSPCs can be lineage traced to the embryo ( Fuentealba et al . , 2015; Furutachi et al . , 2015 ) , and we could not detect a population of Oct4+ cells in our datasets . We , however , cannot exclude that rare primitive OCT4+ NSPCs were not captured in our analysis for technical reasons . More recently , genetic labeling of the most ventral domain of the V-SVZ showed the specific contribution of Nkx2 . 1-expressing B cells to deep layer granule cell neurons in the OB ( Delgado and Lim , 2015 ) . The ventral gene expression program is maintained by an epigenetic mechanism across cell divisions . In the absence of myeloid/lymphoid or mixed-lineage leukemia protein 1 ( MLL1 ) -dependent epigenetic maintenance , the neurogenic lineage shifts to produce aberrantly ‘dorsalized’ OB neuronal subtypes ( Delgado et al . , 2020 ) . This underscores the early embryonic regional specification of adult V-SVZ NSPCs and how these primary progenitors maintain a memory of their regions of origin . Regional genes maintained through the neurogenic lineage could help us understand how NSC identities are maintained to ensure the production of the specific subtypes of interneurons in the OB . Our dataset is restricted to the V-SVZ , a region where A cells ( young neurons ) have begun their differentiation , but remain migratory and immature . As A cells move into the OB they complete their differentiation and begin expressing mature neuronal markers like tyrosine hydroxylase and calbindin in different subdomains of the OB . Our study reveals previously unknown markers of young immature type A cells . Previous work has already shown that Pax6 , which is associated with a subpopulation of superficial granule neurons and of periglomerular cells , is expressed by a subpopulation of young migrating A cells ( Coré et al . , 2020; Kohwi et al . , 2005 ) . Consistent with these findings , we found Pax6 expression largely restricted in the dorsal lineage of A cells ( Figure 5 ) . This further validates the regional heterogeneity we find among A cells . The new set of genes associated with young A neurons derived from dorsal or ventral territories should help future studies determine the early programs in the differentiation of specific subtypes of OB neurons . Particularly interesting are sets of chemotaxis genes that are differentially expressed by ventral and dorsal A cells ( e . g Slit2 expression in ventral A cells ( Figure 6 ) and Dab1 in dorsal A cells ( Figure 5 ) ) . These guidance genes may be linked to the different migratory destinies that young neurons need to adopt once they arrive in the OB . Our dataset provides sets of genes that are differentially expressed in dorsal and ventral B cells . Among these genes , we found well-known regionally expressed transcription factors such as Pax6 , Hopx , Nkx6 . 2 , Gsx2 , and Vax1 ( Coré et al . , 2020; Delgado and Lim , 2015; Hack et al . , 2005; Kohwi et al . , 2005; Merkle et al . , 2014; Taglialatela et al . , 2004; Zweifel et al . , 2018 ) . We also identified Urah , Dio2 , and Crym as novel markers that define largely non-overlapping domains of the V-SVZ . Urah and Dio2 define a dorsal domain that includes the wedge and subcallosal roof of the V-SVZ , and Crym defines a wide ventral domain . Consistent with the above observations , Crym expression has been described in a subpopulation of qNSC in the early postnatal V-SVZ derived from the Nkx2 . 1 domain ( Borrett et al . , 2020 ) . Our dorsal domain overlaps with that defined by Pax6 ( Hack et al . , 2005; Kohwi et al . , 2005 ) and the ventral domain with that defined by Vax1 ( Coré et al . , 2020 ) . Pax6 and Vax1 are transcription factors that link these territories to well-defined embryonic domains involved with the generation of different subsets of neurons in the cortex and striatum . Similarly , Gsx2 is expressed in a gradient in the embryo , with its highest expression in the dorsal lateral ganglionic eminence ( Corbin et al . , 2000; López-Juárez et al . , 2013; Taglialatela et al . , 2004; Waclaw et al . , 2009; Young et al . , 2007 ) . Consistent with the dorsal expression pattern , in our dataset Gsx2 was highly enriched in cluster B ( 5 ) ( Figure 3B ) . The dorsal region was also enriched in Ptprz1 , Hopx , Dio2 , Tnc , and Moxd1 , which are also markers of outer radial glia , a subpopulation of human neural stem cells that continue to generate neurons for the cortex after detaching from the pallial wall of the lateral ventricles during prenatal development ( Nowakowski et al . , 2016; Pollen et al . , 2015 ) . The dorsal V-SVZ domain is likely further subdivided into multiple subdomains . In the current analysis , we pooled together clusters B ( 5 ) and B ( 22 ) as dorsal . However , the largely pallial marker Emx1 and dorsal lateral ganglionic eminence marker Gsx2 were differentially enriched in clusters B ( 22 ) and B ( 5 ) , respectively , suggesting that these two clusters may also represent different sets of regionally specified B cells with distinct embryonic origins . These regions become blurred by cells intermixing in the formation of the wedge region in the postnatal V-SVZ making it difficult to confirm their origin based on expression patterns . In addition to pallial and dorsal subpallial markers , this wedge region likely also includes what has been termed the ventral pallium ( Puelles et al . , 2016 ) , which is characterized in the embryo by the expression of Dbx1 . Unfortunately , our scRNA-Seq analysis did not detect this marker . Further lineage tracing experiments will help determine the precise embryonic origin and nature of the dorsal V-SVZ , including the wedge region . We confirmed that B cells defined as dorsal or ventral in our scRNA-Seq were predicted to correspond to dorsal and ventral microdissections using unsupervised label transfer of cell identity from the sNucRNA-Seq data ( Figure 3D–F ) . We also found that 59 genes in our scRNA-Seq analysis were highly expressed in sNucRNA-Seq B cells from the dorsal microdissection , including Pax6 . Similarly , seven genes were highly expressed in the ventral microdissected sNucRNA-Seq B cells that were also highly expressed in the ventral scRNA-Seq B ( 14 ) cluster , including Slit2 ( Supplementary file 6 ) . However , the identification of single genes in this type of comparison has three limitations: ( 1 ) The dorsal and ventral territories ( Figure 4; see below ) did not precisely correspond to the microdissected areas used for sNucRNA-Seq . For example , regions that we considered dorsal in our microdissection ( ventral to the wedge ) , contained part of the ventral domain highlighted by Crym expression ( Figure 4 ) . ( 2 ) The dorsal samples in the sNucRNA-Seq analysis had fewer cells and higher sequencing depth . ( 3 ) mRNA was likely differentially represented in nuclear or whole-cell preps . To overcome these technical issues , we used the unsupervised Label Transfer algorithm to independently identify closely related B cells between the scRNA-Seq and sNucRNA-Seq datasets . We found that cells of cluster B ( 14 ) were almost all predicted to correspond to the ventral dissection , and the vast majority of cluster B ( 5+22 ) cells were predicted to correspond to cells from the dorsal dissection ( Figure 3E–F ) . The above limitations of the label transfer analysis could also affect our observations for A cells ( Figure 5D ) . The border between the Crym+ and the Urah/Dio2+ territories had not been previously determined and lies in an anatomically undefined location ( Figure 4 ) . Interestingly , the dorsal domain defined by Dio2 and Urah largely overlaps with a region of high Gli1 expression during neonatal and early postnatal stages ( Tong et al . , 2015 ) . This dorsal Sonic Hedgehog-regulated domain in early postnatal life has been linked to oligodendrogenesis . Whether the adult domain we now unravel is developmentally linked to this early oligodendrogenic domain remains to be determined using lineage analysis . However , it is tempting to speculate that these territories are functionally linked , a hypothesis supported by the enrichment of glial-development-associated genes among dorsal markers ( Figure 6G ) . Thyroid hormone signaling has been shown to regulate V-SVZ neurogenesis ( Lemkine et al . , 2005; López-Juárez et al . , 2012; Luongo et al . , 2021 ) . Our analysis shows that Crym , Urah , and Dio2 are differentially expressed by B cells according to region . Crym , Urah , and Dio2 are all associated with the thyroid hormone signaling pathway; Dio2 catalyzes thyroid hormone activation , Crym has been described to bind thyroid hormone , and Urah belongs to a family of thyroid hormone transporters ( Luongo et al . , 2019; Rudqvist et al . , 2012; Vié et al . , 1997 ) . Furthermore , steroid-thyroid hormone family members Nr4a1 , Nr4a2 , and Nr4a3 were enriched in dorsal B cells and were predicted to form part of the core regulatory network of dorsal marker genes . Hormone signaling , and in particular thyroid hormones , may differentially affect B cells in a regional manner and possibly modify the balance of neuronal types produced in the V-SVZ and destined for the OB . Region-specific regulation of V-SVZ stem cells has been previously suggested for the anterior-ventral V-SVZ , where a subpopulation of proopiomelanocortin hypothalamic projections activates Nkx2 . 1+ progenitors ( Paul et al . , 2017 ) . The combination of region-specific innervation and molecular composition of B cells could form the basis for a system of hormonal regulation underlying circuit changes in the OB . In summary , we present a large-scale single-cell description of dorso-ventral identity in the lateral wall of the V-SVZ . Not only do we recapitulate known divisions between dorsal and ventral B cells , but also we identify novel regional B cell markers and uncover gene expression programs that appear to persist throughout lineage transitions ( Figure 7 ) . These data form a basis for future investigation of NSPCs identity , lineage commitment , and embryonic origin , providing clues to help us understand how molecularly defined stem cell territories are spatially organized , and what distinguishes V-SVZ regions from one another . Mice were housed on a 12 hr day-night cycle with free access to water and food in a specific pathogen-free facility in social cages ( up to five mice/cage ) and treated according to the guidelines from the UCSF . Institutional Animal Care and Use Committee ( IACUC ) and NIH . All mice used in this study were healthy and immuno-competent , and did not undergo previous procedures unrelated to the experiment . CD1-elite mice ( Charles River Laboratories ) and hGFAP::GFP ( FVB/N-Tg ( GFAPGFP ) 14Mes/J , The Jackson Laboratory ( 003257 ) ) ( Zhuo et al . , 1997 ) lines were used . Sample sizes were chosen to generate sufficient numbers of high-quality single cells for RNA sequencing , including variables such as sex , and identifying potential batch effects . Biological and technical replicates for each experiment are described in the relevant subsections below . Mice received intraperitoneal administration of 2 . 5% Avertin followed by decapitation . Brains were extracted and 1 mm slices were obtained with an adult mouse brain slicer ( Steel Brain Matrix - Coronal 0 . 5 mm , Alto ) . Four samples were processed: sample 1: two males P35; sample 2: two males P35; sample 3: two females P29; and sample 4: two females P29 . The lateral ventricle walls were microdissected in L-15 medium on ice and the tissue was transferred to Papain-EBSS ( LK003150 , Worthington ) . Tissue was digested for 30 mins at 37°C in a thermomixer at 900 RPM . Mechanical dissociation with a P1000 pipette tip ( 20 s ) , then fire-polished pasteur pipette was performed for 5 min . Tissue was digested for 10 more min at 37°C , and dissociated with the pasteur pipette for another 2 min . Cells were centrifuged for 5 min , 300 RCF at room temp , and the pellet was resuspended with DNAase/ovomucoid inhibitor according to manufacturer's protocol ( Worthington ) . Cells were incubated in Red blood cell lysis buffer ( 420301 , Biolegend ) 3–4 min at 4°C . For MULTI-seq barcoding , cells were suspended with Anchor:Barcode solution ( every sample was labeled with a unique barcode: sample 1 Barcode: TGAGACCT ( ‘A3’ ) ; sample two barcode GCACACGC ( ‘A4’ ) ; sample three barcode AGAGAGAG ( ‘A5’ ) ; and sample four barcode TCACAGCA ( ‘A6’ ) ) for 5 min at 4°C . A Co-Anchor solution was added and incubated for 5 min ( McGinnis et al . , 2019a ) . Samples were combined and filtered with a FlowMi 40 µm filter ( BAH136800040-50EA , Sigma ) . To remove myelin , the cell suspension was incubated with Myelin Removal Beads ( 130-096-733 , Miltenyi Biotec ) ( 6 μl/brain ) for 15 min at 2–8°C . Cells were washed with 0 . 5% BSA-PBS and transferred to MACS columns ( 30-042-401 and QuadroMACS Separator 130-090-976 , Miltenyi Biotec ) . The cell suspension was preincubated with TruStain FcX Plus Antibody ( BioLegend , Key resources table ) on ice for 10 min , then incubated with oligonucleotide-tagged anti-VCAM1 and anti-CD24 antibodies ( BioLegend , Key resources table ) on ice for 30 min , then washed twice with 0 . 5% BSA-PBS by centrifugation ( 5 min , 4°C , 350 RCF ) and filtered with a FlowMi 40 µm filter . The effluent was collected and cell density was counted . Cells were loaded into two wells of a 10x Genomics Chromium Single Cell Controller . We used the 10x Genomics Chromium Single Cell 3’ Library and Gel Bead Kit v3 to generate cDNA libraries for sequencing according to manufacturer’s protocols . GFP expression of isolated cells was observed under an epifluorescence microscope . MULTI-seq and antibody TotalSeq barcode libraries were assembled as previously described ( McGinnis et al . , 2019a ) . Briefly , a MULTI-seq primer is added to the cDNA amplification mix . Afterwards , in the first clean-up step using SPRI beads ( 0 . 6x ) of the standard 10x library prep workflow , the supernatant is saved , transferred to a new tube and a cleanup step using SPRI ( 1 . 6x ) is performed to eliminate larger molecules . A library preparation PCR is also performed for the MULTI-seq barcodes . The barcode library is analyzed using a Bioanalyzer High Sensitivity DNA system and then sequenced . The code for demultiplexing samples and detecting doublets can be found at https://github . com/chris-mcginnis-ucsf/MULTI-seq , McGinnis , 2019b . We pooled gene expression and barcode cDNA libraries from each 10x Genomics Single Cell Controller well ( technical replicates , ‘Lane’ ) and sequenced them at the UCSF Center for Advanced Technology on one lane of an Illumina Novaseq 6000 machine . A total of 2 , 892 , 555 , 503 reads were aligned using CellRanger 3 . 0 . 2-v3 . 2 . 0 ( 10x Genomics ) to a custom version of the mouse reference genome GRCm38 that included the GFP gene ( GFP sequence: Supplementary file 7 ) . Reads corresponding to oligonucleotide-tagged TotalSeq antibodies were assigned to cells in CellRanger according to manufacturer instructions . To identify cell barcodes that most likely corresponded to viable cells , we performed quality control and filtering steps . We excluded cells outside of the following thresholds: UMI count depth: 5th and 95th percentiles; number of genes per cell: below 5th percentile; percentage of mitochondrial gene reads per cell: greater than 10% . We classified cells into sample groups and identified doublets using MULTI-seq barcode abundances ( McGinnis et al . , 2019a ) . We used Seurat Integration ( Seurat 3 ) canonical correlation analysis ( CCA ) to reduce data dimensionality and align the data from technical replicates ( Lane 1 and Lane 2 ) ( Stuart et al . , 2019 ) . Brains were extracted and 0 . 5 mm slices were obtained . We microdissected the anterior ventral , anterior-dorsal , posterior-ventral and dorsal V-SVZ regions of 17 P35 CD1 male ( 8 ) and female ( 9 ) mice . Briefly , we used a brain matrix to cut one millimeter thick coronal slabs of the mouse forebrain and used histological landmarks to identify each sampling area ( e . g . anterior region landmarks: septum; posterior regions: hippocampus ) . Regions were dissected under a microscope to reduce the amount of underlying striatum in each sample . Each micro-dissected V-SVZ region was processed in parallel as a distinct sample . We processed tissue samples for nucleus isolation and sNucRNA-Seq as previously described ( Velmeshev et al . , 2019 ) . Briefly , we generated a single nucleus suspension using a tissue douncer ( Thomas Scientific , Cat # 3431D76 ) in nucleus isolation medium ( 0 . 32M sucrose , 5 mM CaCl2 , 3 mM MgAc2 , 0 . 1 mM EDTA , 10 mM Tris-HCl , 1 mM DTT , 0 . 1% Triton X-100 in DEPC-treated water ) . Debris was removed via ultracentrifugation on a sucrose cushion ( 1 . 8M sucrose , 3 mM MgAc2 , 1 mM DTT , 10 mM Tris-HCl in DEPC-treated water ) in a thick-walled ultracentrifuge tube ( Beckman Coulter , Cat # 355631 ) and spun at 107 , 000 RCF , 4°C for 150 min . The pelleted nuclei were incubated in 250 µL PBS made with DEPC-treated water on ice for 20 min . The resuspended pellet was filtered twice through a 30 µm cell strainer . We counted nuclei with a hemocytometer to determine nucleus density , and loaded approximately 12 , 000 nuclei from each sample into its own well/lane of a 10x Genomics Chromium Single Cell Controller microfluidics instrument . We used the 10x Genomics Chromium Single Cell 3’ Library and Gel Bead Kit v2 to generate cDNA libraries for sequencing according to manufacturers’ protocols . We measured cDNA library fragment size and concentration with a Bioanalyzer ( Agilent Genomics ) . We pooled the gene expression cDNA libraries from each single nucleus sample and sequenced them on one lane of an Illumina HiSeq 4000 at the UCSF Center for Advanced Technology . The PV sample was further sequenced to increase sequencing depth . A total of 1 , 340 , 031 , 643 reads were aligned using CellRanger 2 . 1 . 0–2 . 3 . 0 ( AV , AD , PD samples ) ; 3 . 0 . 2-v3 . 2 . 0 ( PV sample ) ( 10x Genomics ) to a custom mouse reference genome that includes unspliced ‘pre-mRNA’ ( GRCm38 ) , which we expect to be present in cell nuclei ( Velmeshev et al . , 2019 ) . To identify cell barcodes that most likely corresponded to viable nuclei , we performed quality control and filtering steps . For each region sample , we excluded nuclei outside of the following thresholds: UMI count depth: 5th to 95th percentiles; number of genes per cell: below 5th percentile; fraction of mitochondrial gene reads per cell ( <10% ) . We used Seurat Integration ( Seurat 3 ) canonical correlation analysis ( CCA ) to reduce data dimensionality and align the data from each region ( Stuart et al . , 2019 ) . We used Seurat 3 ( Stuart et al . , 2019 ) to analyze both the Whole Cell and Single Nucleus datasets: for each dataset , cells or nuclei from each 10x Chromium Controller Lane ( scRNA-Seq: Lanes 1 and 2; sNucRNA-Seq: AD , AV , PD , PV lanes ) were integrated using IntegrateData and normalized using regularized negative binomial regression ( SCTransform ) ( Hafemeister and Satija , 2019 ) . We calculated 100 principal components ( PCs ) per dataset , and used 50 ( scRNA-Seq ) or 100 ( sNucRNA-Seq ) to calculate cell cluster identities at five distinct resolutions ( 0 . 5 , 0 . 8 , 1 . 0 , 1 . 5 , and 2 . 0 ) and UMAP coordinates . The cell cluster identities presented in this manuscript correspond to resolution 1 . 5 ( scRNA-Seq metadata column integrated_snn_res . 1 . 5 ) or 2 ( sNucRNA-Seq metadata column integrated_snn_res . 2 ) , and were chosen based on visual correspondence with the expression of known neurogenic lineage markers . Sequenced antibody tags in the scRNA-Seq dataset were separately normalized using NormalizeData ( method: CLR ) and ScaleData , and are included in the Seurat object ‘Protein’ assay as ‘VCAM1-TotalA’ and ‘CD24-TotalA’ . Mouse brains ( n=3 , P30 ) were serially sectioned using a Leica cryostat ( 10-µm-thick sections in Superfrost Plus slides ) . Sections were incubated 10 min with 4% PFA and washed 3x10 min with phosphate-buffered saline ( PBS ) to remove OCT . Slides were incubated with ACD hydrogen peroxide for 10 min , treated in 1x target retrieval buffer ( ACD ) for 5 min ( at 96–100°C ) and rinsed in water and 100% ethanol . Samples were air dried at 60°C during 15 min and kept at room temperature overnight . The day after , samples were treated with Protease Plus for 30 min at 40°C in the RNAscope oven . Hybridization of probes and amplification solutions was performed according to the manufacturer’s instructions . Amplification and detection steps were performed using the RNAscope 2 . 5 HD Red Detection Kit ( ACD , 320497 ) and RNAscope 2 . 5 HD Duplex Reagent Kit ( ACD , 322430 ) . RNAscope probes used: Mm-Lphn3 ( also named Adgrl3 ) ( cat . # 317481 ) , Mm-Rlbp1 ( cat . # 468161 ) , Mm-Crym ( cat . # 466131 ) , Mm-Pax6 ( cat . # 412821 ) , Mm-Slit2 ( cat . # 449691 ) , Mm-Cnatnp2 ( cat . # 449381 ) , Mm-Urah-C2 ( cat . # 525331-C2 ) , Mm-Dio2 ( cat . # 479331 ) , Mm-Hopx ( cat . # 405161 ) , Mm-Ntng1-C2 ( cat . # 488871-C2 ) , Mm-Trhde ( cat . # 450781 ) . Mm-Snhg15 was custom made ( NPR-0009896 , cat . # 889191 ) . DapB mRNA probe ( cat . # 310043 ) was used as negative and Mm-PPIB ( cat . # 313911 ) as positive control . RNAscope assay was directly followed by antibody staining for chicken anti-GFAP ( Abcam , ab4674 , 1:500 ) and rabbit anti-DCX ( Cell signaling , 4604S , 1:200 ) or rabbit anti-S100 ( Dako , Z033 , 1:100 , discontinued ) ( Key Resources table ) . Samples were blocked with TNB solution ( 0 . 1 M Tris–HCl , pH 7 . 5 , 0 . 15 M NaCl , 0 . 5% PerkinElmer TSA blocking reagent ) 30 min and incubated in primary antibodies overnight . Samples were washed with PBS-Tx0 . 1% and incubated with secondary antibodies Donkey anti-Chicken Alexa 647 ( Jackson ImmunoResearch , 703-605-155 , 1:500 ) and Donkey anti-Rabbit biotinylated ( Jackson ImmunoResearch , 711-065-152 , 1:400 ) in TNB buffer for 1 . 5 hr . Samples were washed and incubated with Streptavidin HRP ( 1:200 in TNB solution ) for 30 min . Washed 3x5 min and incubated with Fluorescein Tyramide 5 min ( 1:50 in amplification diluent ) rinsed and incubated with DAPI 10 min . Sections were mounted with Prolong glass Antifade Mountant ( Invitrogen , P36980 ) . Coronal sections ( n=four mice , P30 ) and whole mounts ( n=three mice , P28 ) were incubated with Immunosaver ( 1:200; EMS , Fort Washington , PA ) for 20 min at 60°C , and then 15 min at RT . Tissue was then incubated in blocking solution ( 10% donkey serum and 0 . 2% Triton X-100 in 0 . 1 M PBS ) for 1 hr . followed by overnight incubation at 4°C with the primary antibodies: mouse anti-CRYM ( Santa Cruz , sc-376687 , 1:100 ) , rabbit anti-BETA-CATENIN ( Sigma , C2206 , 1:250 ) , Chicken anti-GFP ( Aves labs , GFP1020 , 1:400 ) , Rabbit anti-HOPX ( Proteintech , 11419–1-AP , 1:500 ) . On the next day , sections were rinsed and incubated with Alexa Fluor secondary antibodies . Samples were mounted with Aqua-poly/mount ( Polysciences Inc , 18606–20 ) . Confocal images were acquired using the Leica Sp8 confocal microscope . Samples processed for RNAscope and immunohistochemistry were imaged at ×20 ( low magnification ) and ×63 ( High magnification ) . For high-magnification RNAscope images , 10–15 optical sections were acquired sequentially using Leica Application Suite X ( LAS X ) software . For quantifications of RNAscope puncta , the V-SVZ of one coronal section hemisphere was tile-scanned per mouse ( n=3 ) . Optical sections were taken through the entire thickness of the section at 0 . 3-micometer intervals using a 63x objective . The tile-scans were imported into Imaris Image Analysis software ( v9 . 7 , Bitplane ) to detect RNAscope puncta using the Spots tool . Automatic spot detection was manually adjusted , and spots were categorized according to colocalization with DAPI and S100-Beta or DCX immunolabeling . Spots outside of the lateral wall V-SVZ were manually removed from the dataset , as were spots in z-planes that lacked antibody labeling ( e . g . the antibody did not fully penetrate the section ) . Finally , a line was drawn using the Measurement tool from the ventral-most point in the lateral wall V-SVZ , through the V-SVZ to the dorso-lateral most extent of the wedge ( raw data available at Figure 3 and Figure 6—source data 1 ) . These data were exported from Imaris and the density of spots along the length of the V-SVZ was quantified using the R sp package ( v . 1 . 4–5 ) . To determine the proportion of B cells ( identified as GFP+ cells ) expressing CRYM in coronal sections , tile-scans of the entire V-SVZ ( n=4 ) were acquired . Coronal sections ( 1 . 42–0 . 98 mm anterior to bregma ) were utilized . Cells were manually counted using Imaris Image Analysis software ( v9 . 7 , Bitplane ) . The length of the V-SVZ , including the wedge and the lateral wall of the lateral ventricle , was divided in 3 . The dorsal domain was defined as the most dorsal third ( V-SVZ wedge and the adjacent lateral wall ) ; the ventral domain encompassed the ventral two thirds ( Figure 3—figure supplement 1S ) . For quantifications of B1 cells expressing CRYM , V-SVZ whole mounts from hGFAP:GFP mice ( n=3 ) were immunostained for GFP , ß-Catenin and CRYM . Images of the apical surface of dorsal and ventral regions ( 8–10 fields/region ) of the lateral wall were acquired by confocal microscopy ( Leica SP8 ) . B1 cells were identified by their GFP+ small apical endings demarcated by ß-Catenin . B1 cells expressing CRYM were manually counted using Imaris Image Analysis software . Note that within the wedge region , in the most dorsal domain , there is no ventricle and this region is not visible in whole mount preparation; this region was included in the above quantification in sections , but it is not included in the quantifications from the whole mounts . B1 cell whole mount quantifications are expressed as mean ± SD ( standard deviation ) . Student’s test ( Excel , Microsoft ) was used for pairwise comparison between two groups . For CRYM pre-embedding immunogold staining , mice ( n=2 ) were perfused with 4% paraformaldehyde ( PFA ) / 0 . 5% glutaraldehyde . Brains were cut into 50 μm coronal sections on a vibratome . Floating sections were incubated with 1% Sodium borohydride in phosphate buffer ( 0 . 1M ) for aldehyde inactivation , cryoprotected with 25% sucrose and permeabilized by freezing and thawing ( 5x ) in methylbutane on dry ice . Sections were blocked with 0 . 3% BSAc ( Aurion ) in PB 0 . 1M for 1 hr at RT and incubated with mouse anti-CRYM ( Santa Cruz , sc-376687 , 1:100 ) 72 hr , 4°C . Sections were rinsed and incubated with goat anti-mouse conjugated to colloidal gold ( 1:50 , UltraSmall , Aurion #25120 ) for 24 hr at 4°C . Silver enhancement and Gold toning were performed as previously described ( Sirerol-Piquer et al . , 2012 ) . Sections were postfixed with 1% osmium-7% glucose in phosphate buffer 0 . 1M , dehydrated and embedded in Durcupan ( Fluka ) . Ultrathin sections ( 70 nm ) were cut , stained with lead citrate and examined under TEM ( Tecnai Spirit G2 , FEI ) . We used AUCell to score cells based on the expression of sets of genes ( Aibar et al . , 2017 ) . For creating the dorsal and ventral scores , we used the dorsal and ventral signature genes ( Figure 6 and Figure 6—figure supplement 1 ) . To identify the High Score dorsal and ventral cells , we normalized both scores to values between 0 ( min ) and 1 ( max ) . We subtracted the ventral score from the dorsal score and the cells in the top quartile corresponded to High Score dosal cells and on the bottom quartile corresponded to the High Score ventral cells . We used Seurat three functions FindMarkers ( two groups ) or FindAllMarkers ( more than two groups ) to identify differentially expressed genes among groups of single cells ( p_val_adj < 0 . 05 ) using a Wilcoxon rank sum test . For single nuclei , Seurat four function FindMarkers was used to identify differentially expressed genes ( p_val < 0 . 05 ) . For detailed parameters see available code ( below ) . Selection of genes from the resulting lists for further analysis are described in the text . Gene Ontology analyses of differentially expressed genes were performed using a binomial test and comparing differentially expressed genes against the whole mouse genome using Panther v . 16 ( Mi et al . , 2021 ) . GO terms with a false discovery rate lower than 5% were considered statistically significant . To construct Gene Regulatory Networks for dorsal and ventral B cells we used gene network inference with ensemble of trees ( GENIE3 ) , a tree regression based method that employs Random Forests to rank regulatory links between genes ( Huynh-Thu et al . , 2010 ) . After calculating link lists from expression matrices in putative dorsal or ventral B cells , we used the top 10 differentially enriched genes in each population to build networks . We selected the top 300 links among genes predicted to be regulators or targets of these markers . To identify regulatory connections of Urah in dorsal cells or Crym in ventral cells , we identified the top 20 predicted regulator or target genes of Crym or Urah , then expanded our networks to include regulators and targets of those genes . Node tables containing expression data were generated and imported to Cytoscape for visualization . GO analysis was completed in Cytoscape using the BiNGO app . Networks are available . B cell label transfer: First , we subsetted quiescent B cells from both the scRNA-Seq and sNucRNA-Seq datasets ( clusters B ( 5 ) , B ( 14 ) , B ( 22 ) , and sNucRNA-Seq cluster 7 ) . We generated the reference sNucRNA-Seq B cell dataset that consisted of equal numbers of B cells per region , randomly selected from the middle 50% of cells by number of genes identified per cell ( 25th-75th percentile of SCT_snn_nFeature ) . This prevented the region with the most nuclei from dominating the prediction scores , and filtering cells by nFeature prior to downsampling resulted in reproducible prediction scoring , likely due to exclusion of low-quality B cells and doublets not rejected in the full dataset quality control steps . Anterior dorsal and posterior dorsal regions were combined to create the Dorsal reference cell set , and the anterior ventral and posterior ventral were combined to create the Ventral reference cell set . Subsetted scRNA-Seq B cells and filtered sNucRNA-Seq B cell sets were individually normalized using SCTransform . We then ran FindTransferAnchors with the following settings: reference . assay = ‘SCT’ , query . assay = ‘SCT’ , normalization . method = ‘SCT’ , npcs=30 , project . query=T , and dims = 1:30 . We then calculated Dorsal and Ventral predicted identity scores for each scRNA-Seq B cell using TransferData ( Stuart et al . , 2019 ) . A cell label transfer: The same method as above was applied to scRNA-Seq A cell clusters A ( 0 ) , A ( 1 ) , A ( 4 ) , A ( 6 ) , and A ( 15 ) , and sNucRNA-Seq clusters 12 and 29 . RNA Velocity in the neurogenic lineage was calculated using scvelo ( Bergen et al . , 2020 ) , using 2000 genes per cell . Moments were calculated using 30 PCs and 30 neighbors . Velocity was estimated using the stochastic model . Pseudotime was plotted using the original UMAP coordinates . The RNA sequencing datasets generated for this manuscript are deposited in the following locations: scRNA-Seq and sNucRNA-Seq GEO Data Series: GSE165555 . Processed data ( CellRanger output . mtx and . tsv files , and Seurat Object . rds files ) are available as supplementary files within the scRNA-Seq ( GSE165554 ) or sNucRNA-Seq ( GSE165551 ) data series or individual sample entries listed within each data series . Web-based , interactive versions of the scRNA-Seq and sNucRNA-Seq datasets are available from the University of California Santa Cruz Cell Browser: https://svzneurogeniclineage . cells . ucsc . edu . The code used to analyze the datasets and generate the figures are available at the following location: https://github . com/AlvarezBuyllaLab/SVZSingleCell ( copy archived at swh:1:rev:37402d867adce3ff4295f06e8fd6289e1d3ba075 ) , Cebrian-Silla et al . , 2021 .	Nerve cells , or neurons , are the central building blocks of brain circuits . Their damage , death or loss of function leads to cognitive decline . Neural stem/progenitor cells ( NSPCs ) first appear during embryo development , generating most of the neurons found in the nervous system . However , the adult brain retains a small subpopulation of NSPCs , which in some species are an important source of new neurons throughout life . In the adult mouse brain the largest population of NSPCs , known as B cells , is found in an area called the ventricular-subventricular zone ( V-SVZ ) . These V-SVZ B cells have properties of specialized support cells known as astrocytes , but they can also divide and generate intermediate ‘progenitor cells’ called C cells . These , in turn , divide to generate large numbers of young ‘A cells’ neurons that undertake a long and complex migration from V-SVZ to the olfactory bulb , the first relay in the central nervous system for the processing of smells . Depending on their location in the V-SVZ , B cells can generate different kinds of neurons , leading to at least ten subtypes of neurons . Why this is the case is still poorly understood . To examine this question , Cebrián-Silla , Nascimento , Redmond , Mansky et al . determined which genes were expressed in B , C and A cells from different parts of the V-SVZ . While cells within each of these populations had different expression patterns , those that originated in the same V-SVZ locations shared a set of genes , many of which associated with regional specification in the developing brain . Some , however , were intriguingly linked to hormonal regulation . Salient differences between B cells depended on whether the cells originated closer to the top ( ‘dorsal’ position ) or to the bottom of the brain ( ‘ventral’ position ) . This information was used to stain slices of mouse brains for the RNA and proteins produced by these genes in different regions . These experiments revealed dorsal and ventral territories containing B cells with distinct ‘gene expression’ . This study highlights the heterogeneity of NSPCs , revealing key molecular differences among B cells in dorsal and ventral areas of the V-SVZ and reinforcing the concept that the location of NSPCs determines the types of neuron they generate . Furthermore , the birth of specific types of neurons from B cells that are so strictly localized highlights the importance of neuronal migration to ensure that young neurons with specific properties reach their appropriate destination in the olfactory bulb . The work by Cebrián-Silla , Nascimento , Redmond , Mansky et al . has identified sets of genes that are differentially expressed in dorsal and ventral regions which may contribute to regional regulation . Furthering the understanding of how adult NSPCs differ according to their location will help determine how various neuron types emerge in the adult brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "neuroscience" ]	2021	Single-cell analysis of the ventricular-subventricular zone reveals signatures of dorsal and ventral adult neurogenesis
Cancer develops and progresses often by inactivating p53 . Here , we unveil nerve growth factor receptor ( NGFR , p75NTR or CD271 ) as a novel p53 inactivator . p53 activates NGFR transcription , whereas NGFR inactivates p53 by promoting its MDM2-mediated ubiquitin-dependent proteolysis and by directly binding to its central DNA binding domain and preventing its DNA-binding activity . Inversely , NGFR ablation activates p53 , consequently inducing apoptosis , attenuating survival , and reducing clonogenic capability of cancer cells , as well as sensitizing human cancer cells to chemotherapeutic agents that induce p53 and suppressing mouse xenograft tumor growth . NGFR is highly expressed in human glioblastomas , and its gene is often amplified in breast cancers with wild type p53 . Altogether , our results demonstrate that cancers hijack NGFR as an oncogenic inhibitor of p53 . Tumorigenesis is highly associated with inactivation of the tumor suppressor p53 , as it is mutated in ~50% of all types of human cancers , and its functions are impaired through various mechanisms in the rest of human cancers ( Levine and Oren , 2009 ) . p53 executes its tumor suppressive function mainly by inducing the expression of a large number of genes involved in cell cycle control , senescence , apoptosis , ferroptosis , autophagy , and metabolism ( Jiang et al . , 2015; Kruiswijk et al . , 2015 ) . Because of the detrimental effects of p53 on normal cells , the cells have developed mechanisms to monitor its activity , which are often hijacked by cancer cells . One key monitor of p53 is MDM2 ( HDM2 in human ) , an oncoprotein encoded by a p53 transcriptional target gene that is amplified or overexpressed in several human tumors ( Fakharzadeh et al . , 1991; Momand et al . , 1992; Oliner et al . , 1992 , 1993; Wu et al . , 1993 ) . MDM2 inhibits p53 activity primarily by binding to and concealing the N-terminal transcriptional activation ( TA ) domain of p53 ( Oliner et al . , 1993 ) and by mediating its poly-ubiquitination and proteolysis ( Fuchs et al . , 1998; Haupt et al . , 1997; Kubbutat et al . , 1997 ) . Genetically , disruption of the TP53 gene completely rescues the lethal phenotype of Mdm2 knockout mice ( Jones et al . , 1995; Montes de Oca Luna et al . , 1995 ) . A myriad of stresses can orchestrate this MDM2-p53 feedback loop . The ARF tumor suppressor directly associates with MDM2 and inhibits MDM2-mediated p53 ubiquitination and degradation upon oncogenic stress ( Palmero et al . , 1998; Zhang et al . , 1998; Zindy et al . , 1998 ) . Also , several ribosomal proteins boost p53 activation by untying the MDM2-p53 loop in response to ribosomal or nucleolar stress ( Zhang and Lu , 2009; Zhou et al . , 2012 , 2015a ) . But , oncogenic proteins can enhance MDM2 E3 ligase activity towards p53 . MDMX ( also called MDM4 ) , the MDM2 homologue , can enhance MDM2-mediated p53 proteasomal degradation by binding to MDM2 , besides directly interacting with p53 and repressing its activity ( Shvarts et al . , 1996 ) . High expression of MDM2 and MDMX in several cancers , such as breast cancer and melanoma , is often considered as the reason why these cancers sustain wild type ( wt ) p53 ( Wade et al . , 2013 ) , but this could only account for a portion of wt p53-harboring cancers . Thus , it is still unknown if there are other proteins that can also suppress p53 function in the remaining cancers . In this study , we revealed a novel feedback regulation of p53 by nerve growth factor receptor ( NGFR , also called p75NTR or CD271 ) . NGFR is a 75 kD single-transmembrane protein without kinase activity and widely expressed in the central and peripheral nervous system ( Barker , 2004 ) . Often partnering with other receptors , such as TrkA , it is involved in a multitude of processes during neurogenesis , such as neural cell death , neuronal differentiation , neurite growth , and synaptic plasticity ( Barker , 2004 ) . Also , the NGF-NGFR cascade activates NF-κB , leading to inhibition of apoptosis ( Carter et al . , 1996 ) and increased survival of schwannoma ( Ahmad et al . , 2014; Gentry et al . , 2000 ) and breast cancer cells ( Descamps et al . , 2001 ) . In addition , overexpression of NGFR observed in many metastatic cancers promotes tumor migration and invasion ( Boiko et al . , 2010; Civenni et al . , 2011; Johnston et al . , 2007 ) . But , in prostate and bladder cancers , NGFR appears to suppress tumor growth and/or metastasis ( Krygier and Djakiew , 2002; Tabassum et al . , 2003 ) . It remains largely elusive why and how NGFR plays opposite roles in the context of different cancers . These studies together with our initial findings that p53 binds to the NGFR promoter and induces its expression in cancer cells motivated us to further explore the functional interplay between NGFR and p53 , and its role in cancer development . As detailed below , we surprisingly found that NGFR inactivates p53 by directly binding to its central DNA-binding domain and preventing its association with its target promoters and by enhancing its MDM2-mediated ubiquitination and proteolysis . This function is ligand-independent because it occurred in the nucleus and without ligand treatment of cancer cells . Biologically , cancer cells hijack the negative feedback regulation of p53 by NGFR to their growth advantage , as down regulation of NGFR induced p53-dependent apoptosis and cell growth arrest as well as suppressed tumor growth . Furthermore , NGFR was found to be highly expressed in 68 . 75% ( 33/48 ) of human gliomas examined . Consistently , NGFR is amplified in breast cancers that harbor wt TP53 based on the TCGA database ( Cerami et al . , 2012; Gao et al . , 2013 ) . Hence , our discovery of NGFR as another feedback suppressor of p53 could explain why some cancers sustain wt p53 and also suggest NGFR as a potential target for the development of new anti-cancer therapy . From our previous studies to assess the global effects of Inauhzin ( INZ ) on p53 pathway in cancer cells ( Zhang et al . , 2012 , 2014; Liao et al . , 2012 ) , we identified NGFR as a potential p53-regulated gene . To confirm this result , we treated three types of p53-containing cancer cell lines ( HCT116p53+/+ , H460 and HepG2 ) with INZ , Doxorubicin ( Dox ) and 5-Fluorouracil ( 5-FU ) . The expression of NGFR mRNA was drastically elevated by all the three agents ( Figure 1A , B and C ) . Consistently , NGFR protein level increased in response to Dox or 5-FU treatment in p53-intact , but not p53-null ( HCT116p53-/- ) or mutated ( PCL/PRF/5 ) , cancer cells ( Figure 1D and E ) . Consistently , ectopic wt , but not mutant , p53 induced NGFR mRNA expression in p53-deficient H1299 and HCT116p53-/- cells ( Figure 1F and G ) . Conversely , knockdown of p53 markedly reduced NGFR mRNA level ( Figure 1H and I ) . These results demonstrate that anti-cancer drug-induced NGFR expression in the cells is p53-dependent . 10 . 7554/eLife . 15099 . 003Figure 1 . p53 transcriptionally induces NGFR expression in cancer cells . ( A , B , C ) NGFR mRNA expression is elevated by p53-inducing agents . HCT116 p53+/+ ( A ) H460 ( B ) and HepG2 ( C ) cells were treated with Inauhzin , Doxorubicin or 5-Fluorouracil for 15 h , and NGFR expression was determined by q-PCR ( Mean ± SEM , n = 3 ) . Three biological replicates ( independent experiments ) and a two-tailed t-test were used for P value calculation , p<0 . 01 . Most q-PCR were performed by three biological replicates , each including three technical replicates . ( D ) NGFR protein expression is elevated by p53-inducing agents in colon cancer cell lines . HCT116 p53+/+ and HCT116 p53-/- cells were treated with Doxorubicin or 5-Fluorouracil for 15 hr followed by IB using antibodies as indicated . ( E ) NGFR protein expression is elevated by p53-inducing agents in liver cancer cell lines . HepG2 and PLC/PRF/5 cells were treated with Doxorubicin or 5-Fluorouracil for 15 hr followed by IB using antibodies as indicated . ( F , G ) NGFR mRNA expression is induced by ectopic wild-type , but not mutant , p53 . HCT116 p53-/- ( F ) and H1299 ( G ) cells were transfected with wild-type or mutant p53 for 30 hr and NGFR expression was determined by q-PCR ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 01 . ( H , I ) NGFR mRNA expression is inhibited by p53 knockdown . HCT116 p53+/+ ( H ) and H460 ( I ) cells were transfected with p53 or control siRNA for 72 hr , and Doxorubicin or 5-Fluorouracil was supplemented 15 hr before the cells were harvested for q-PCR ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 05 . ( J ) A schematic of predicted p53 responsive elements in the NGFR promoter . ( K , L ) p53 induces luciferase activity through RE1 . Luciferase assay was performed using H1299 and U2OS cells as described in Materials and methods ( Mean ± SEM , n = 3 biological replicates ) . ( M , N ) p53 is associated with the NGFR promoter . HCT116 p53+/+ cells were treated with or without Doxorubicin for 15 hr followed by ChIP assay using anti-p53 or mouse IgG . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 003 Next , we identified two potential p53 responsive DNA elements ( RE ) in the NGFR promoter by bioinformatics analysis , RE1 and RE2 , at −4826 bp and –6516 bp upstream from the transcriptional initiation site , respectively ( Figure 1J ) . The RE1 was responsive to p53 , as ectopic p53 markedly activated luciferase reporter expression through RE1 , but not RE2 , in H1299 ( Figure 1K ) and U2OS ( Figure 1L ) cells . This result was validated by chromatin-associated immunoprecipitation ( ChIP ) assays after treating HCT116 p53+/+ cells with or without Dox . Endogenous p53 specifically associated with the RE1-containing NGFR promoter , but not a non-related promoter; this association was enhanced by Dox treatment ( Figure 1M and N ) . Thus , these results demonstrate that NGFR is a bona fide transcriptional target of p53 . Previous studies showed that NGFR induces apoptosis in some cancer cells , but promotes cell survival and growth in other cancer cells ( Molloy et al . , 2011 ) . Thus , we re-determined the role of NGFR in cancer cell growth and proliferation and whether this role is p53-dependent . Surprisingly , knockdown of NGFR elicited significant apoptosis of H460 ( Figure 2A and B ) and HCT116 p53+/+ ( Figure 2C and D ) cells as measured by flow cytometric analyses , consequently reducing cell viability during the 4-day culture ( Figure 2E and F ) . To verify if NGFR supports tumor cell growth and proliferation and to see if this role is p53-dependent , we performed colony formation assays by knocking down NGFR in p53-containing H460 and p53-deficient H1299 cancer cells ( Figure 2—figure supplement 1A ) . NGFR ablation more dramatically repressed the colony formation of H460 cells than that of H1299 cells ( Figure 2G ) . This result suggests that NGFR may support cancer cell survival through both p53-dependent and independent mechanisms . Since NGFR was found to be required for nerve growth factor-mediated NF-κB activation ( Carter et al . , 1996 ) , knockdown of NGFR in H1299 may impair NF-κB activation , thus leading to partially restrained cell viability and reduced colonies . Additionally , immunohistochemical ( IHC ) staining of human gliomas and their adjacent normal tissues exhibited hyperexpression of NGFR in the tumor tissues ( Figure 2H and Figure 2—figure supplement 1B ) . Consistently , NGFR was more highly expressed in human glioma tissues compared to adjacent tissues as determined by immunoblotting ( IB ) ( Figure 2I ) . Taken together , these results suggest that NGFR promotes cancer cell growth and proliferation , likely in part by inactivating p53 . 10 . 7554/eLife . 15099 . 004Figure 2 . NGFR is required for cancer cell survival and clonogenicity , and highly expressed in glioma . ( A ) NGFR knockdown induces apoptosis of H460 cells . Cells were transfected with NGFR or control siRNA for 72 to 96 hr , and Doxorubicin was supplemented 15 hr before the cells were harvested for flow cytometry analyses . ( B ) Quantification of Sub-G1 population in ( A ) ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 05 . ( C ) NGFR knockdown induces apoptosis of HCT116 p53+/+cells . Cells were transfected with NGFR or control siRNA for 72 to 96 hr , and Doxorubicin was supplemented 15 hr before the cells were harvested for flow cytometry analyses . ( D ) Quantification of Sub-G1 population in ( C ) ( Mean ± SEM , n = 2 ) . Three biological replicates were used for p value , p<0 . 05 . ( E , F ) NGFR knockdown suppresses cell survival . H460 ( E ) and HCT116 p53+/+ ( F ) cells were transfected with NGFR or control siRNA and seeded in 96-well plate the next day ( Day 1 ) . Cell viability was evaluated every 24 hr ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 05 . ( G ) NGFR knockdown inhibits clonogenicity . H460 and H1299 cells were transfected with NGFR or control siRNA and seeded in 10-cm plates the next day . Colonies were fixed by methonal and stained with crystal violet solution ( left panel ) . Quantification of colonies is shown in the right panel ( Mean ± SEM , n = 2 ) . Two biological replicates were used for p value , p<0 . 05 . ( H ) IHC analyses of human glioma and the adjacent noncancerous tissues . ( I ) IB analyses of human glioma and the adjacent noncancerous tissues ( left panel ) . Quantification of NGFR expression ( right panel ) ( Mean ± SEM , n = 48 , p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 00410 . 7554/eLife . 15099 . 005Figure 2—figure supplement 1 . Representative expression of NGFR in lung cancer cell lines by siRNA knockdown and in glioma tissues . ( A ) NGFR knockdown validation in the colony formation experiment . ( B ) NGFR is highly expressed in human glioma samples . IHC analyses of human glioma and the adjacent noncancerous tissues reveals that NGFR is overexpressed in glimoas . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 005 To test if NGFR negates p53 activity , we first checked if NGFR affects p53 protein level in H460 cells that were treated with Dox or 5-FU . Indeed , p53 level induced by Dox or 5-FU was markedly reduced by ectopic NGFR ( Figure 3A ) . Inversely , knockdown of NGFR strikingly elevated the level of endogenous p53 and that of its target genes p21 and PUMA in human p53-containing H460 , HepG2 , neuroblastoma SK-N-SH , melanoma SK-MEL-103 and SK-MEL-147 and HCT116 p53+/+ cell lines ( Figure 3B ) , whereas did not affect PUMA expression in HCT116 p53-/- cells ( Figure 3—figure supplement 1 ) . Consistently , knockdown of NGFR led to the global expression of a group of known p53 target genes , as measured by RNA-seq analysis in H460 cells ( Figure 3C ) . This result was further confirmed by utilizing a second siRNA ( siNGFR#2 ) , as NGFR knockdown by this siRNA also induced the protein level of p53 and the mRNA levels of its target genes , including BAX , BTG2 , MDM2 , p21 , and PUMA ( Figure 3D ) . Also , ablation of NGFR prolonged p53’s half-life ( Figure 3E ) , suggesting that NGFR might regulate the stability of p53 . Next , we tested if NGFR affects MDM2-induced p53 ubiquitination by conducting a p53 ubiquitination assay in H1299 cells with ectopic proteins . Surprisingly , ectopic NGFR enhanced MDM2-mediated p53 ubiquitination in a dose-dependent manner ( Figure 3F ) , although this protein was found to mainly reside in the cytoplasmic membrane ( Barker , 2004 ) . Consistently , NGFR also enhanced MDM2-mediated p53 proteasomal degradation , which was abolished by the proteasome inhibitor MG132 ( Figure 3G ) . Without MDM2 , NGFR was unable to alter p53 protein level in cells ( Figure 4 ) . Altogether , these results indicate that NGFR reduces p53 stability by enhancing its MDM2-mediated ubiquitination and degradation . 10 . 7554/eLife . 15099 . 006Figure 3 . NGFR suppresses p53 activity by enhancing MDM2-mediated ubiquitination and proteasomal degradation . ( A ) NGFR inhibits p53 activation by Doxorubicin or 5-Fluorouracil . H460 cells were transfected with NGFR for 30 hr and treated with Doxorubicin or 5-Fluorouracil for 15 hr before harvested for IB using antibodies as indicated . ( B ) NGFR knockdown induces p53 expression and activity . A panel of cancer cell lines were transfected with NGFR or control siRNA followed by IB using antibodies as indicated . The values in the rightmost panel indicate the p53/β-actin ratios . ( C ) NGFR knockdown induces p53 target gene expression . H460 cells were transfected with NGFR or control siRNA followed by RNA-sequencing analyses . Genes with over 1 . 5-fold increase in expression were shown ( Three biological replicates were used for p value , p<0 . 05 . ( D ) Two siRNAs targeting different sequences were used for knocking down NGFR in H460 cells . The expression of p53 target genes was assessed by q-PCR ( Mean ± SEM , n = 2 biological replicates ) , while p53 expression was detected by IB . Validation of NGFR knockdown by the two siRNAs is shown in the right corner . ( E ) NGFR knockdown prolongs p53’s half-life . H460 cells transfected with NGFR or control siRNA for 72 hr were treated with 100 µg/ml of CHX and harvested at the time points as indicated . IB was performed using antibodies as indicated ( upper panel ) and quantification of p53/β-actin ratio is shown in the lower panel . ( F ) NGFR promotes MDM2-induced p53 ubiquitination . H1299 cells were transfected with combinations of plasmids encoding p53 , HA-MDM2 , Myc-NGFR or His-Ub and treated with MG132 6 hr before harvested for in vivo ubiquitination assay . Bound proteins and inputs were detected by IB using antibodies as indicated . ( G ) NGFR enhancesMDM2-mediated p53 proteasomal degradation . H1299 cells were transfected with combinations of plasmids encoding p53 , HA-MDM2 or RFP-NGFR followed by IB using antibodies as indicated . MG132 was supplemented to the medium for 6 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 00610 . 7554/eLife . 15099 . 007Figure 3—figure supplement 1 . Knockdown of NGFR does not affect PUMA expression in the p53-null HCT116 cells . HCT116 p53-/- cells were transfected with control or NGFR siRNA for 72 hr followed by IB using antibodies as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 00710 . 7554/eLife . 15099 . 008Figure 4 . NGFR inactivates p53 independently of MDM2 . ( A ) NGFR interacts with p53 in the absence of MDM2 . MEF p53-/-;Mdm2-/- cells were transfected with plasmids encoding Myc-NGFR or p53 followed by co-IP-IB assays using antibodies as indicated . ( B ) NGFR does not affect p53 protein expression in the absence of MDM2 . MEF p53-/-;Mdm2-/- cells were transfected with combinations of plasmids encoding Myc-NGFR , HA-MDM2 or p53 followed by IB using antibodies as indicated . ( C ) NGFR represses p53-induced luciferase activity independently of MDM2 . MEF p53-/-;Mdm2-/- cells were transfected with plasmids as indicated in the figure and luciferase assay was performed as described in Materials and methods ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 05 . ( D ) NGFR inhibits p53-induced target gene expression independently of MDM2 . MEF p53-/-;Mdm2-/- cells were transfected with plasmids as indicated in the figure followed by q-PCR analysis ( Mean ± SD , n = 3 ) . Three technical replicates were used for p value , p<0 . 05 . ( E ) NGFR impedes p53 association with the p21 promoter in H1299 cells . Cells were transfected with plasmids as indicated and treated with MG132 for 6 hr before harvested for ChIP-PCR analyses . ( F ) NGFR impedes p53 association with the p21 and Bax promoters independently of MDM2 . MEF p53-/-;Mdm2-/- cells were transfected with plasmids as indicated followed by ChIP-PCR analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 00810 . 7554/eLife . 15099 . 009Figure 4—figure supplement 1 . MDMX , though binds to NGFR , is not required for NGFR-mediated p53 inactivation . ( A ) MDMX binds to NGFR . H1299 cells were transfected with HA-MDMX and Myc-NGFR followed by co-IP-IB assays using antibodies as indicated . ( B ) NGFR inactivates p53 independently of MDM2 and MDMX . MEF p53-/-; Mdm2-/-; Mdmx-/- cells were transfected with plasmids encoding each of Myc-NGFR and p53 follow by IB analysis using antibodies as indicated ( left panel ) and Q-PCR analysis of p21 expression ( right panel ) ( Mean ± SD , n = 3 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 009 Next , we tested if NGFR interacts with MDM2 by conducting a set of reciprocal co-immunoprecipitation ( co-IP ) assays . Indeed , ectopic NGFR bound to ectopic MDM2 , and vice versa ( Figure 5A ) . We then mapped their binding domains by performing a set of co-IP and GST-pull down assays . NGFR was specifically pulled down with the C-terminal aa 284–491 fragment , but not N-terminal fragments , of MDM2 , which encompasses the zinc-finger , acidic , and ring-finger domains , in H1299 cells ( Figure 5C ) . This result was verified by in vitro GST-pull down assays ( Figure 5D ) . Using the same approach , we also mapped the MDM2-binding domain of NGFR . Unexpectedly , MDM2 interacted with the N-terminal extracellular and transmembrane domain , but not the C-terminal cytoplasmic domain , of NGFR ( Figure 5E ) . Also , we validated the interaction between endogenous NGFR and MDM2 in human neuroblastoma SK-N-SH cells that sustains high level of NGFR ( Figure 5H ) . To determine if this interaction takes place in the nucleus , we used H460 cells stably expressing NGFR , HCT116 p53+/+ and SK-MEL-147 cells , fractionated them , and conducted IB assays . Indeed , full-length NGFR was present in both cytoplasm and nucleus ( Figure 5—figure supplement 1A–C ) . Consistently , confocal microscopy analysis revealed that NGFR is present in all of the cellular compartments of SK-MEL-147 cells , albeit with less intensity in the nucleus than that in the cytoplasm ( Figure 5—figure supplement 1D ) . Of note , the nuclear and cytoplasmic levels of NGFR increased upon Dox treatment likely due to p53 activation ( Figure 5—figure supplement 1D ) . Also , NGFR was co-immunoprecipitated with MDM2 from the nuclear extracts of NGFR-stably expressed H460 ( Figure 5I ) and SK-MEL-147 cells in a reciprocal pattern ( Figure 5J ) . Hence , these results demonstrate that the N-terminus of NGFR binds to the C-terminus of MDM2 , and this interaction occurs in the nucleus . 10 . 7554/eLife . 15099 . 010Figure 5 . NGFR interacts with MDM2 in the nucleus . ( A , B ) NGFR interacts with MDM2 . H1299 cells were transfected with plasmids encoding Myc-NGFR , Flag-MDM2 or HA-MDM2 followed by co-IP-IB assays using antibodies as indicated . ( C ) Mapping the NGFR binding domain on MDM2 by co-IP-IB assays . H1299 cells were transfected with the plasmid encoding V5-tagged MDM2 fragment , aa 1–150 , aa 1–301 or aa 284–491 , along with the Myc-NGFR-encoded plasmid . Co-IP-IB assays were performed using antibodies as indicated . ( D ) Mapping the NGFR binding domain on MDM2 by GST-pull down assay . The prokaryotic expressed GST-tagged MDM2 fragment , aa 1–150 , aa 1–301 or aa 284–491 , or GST protein alone was incubated with cell lysates overexpressing Myc-NGFR . Bound proteins were detected by IB using anti-NGFR or coomassie staining . ( E ) Mapping the MDM2 binding domain on NGFR . H1299 cells were transfected with the plasmid encoding Myc-tagged NGFR fragment , aa 1–272 or aa 273–427 , along with the Flag-MDM2-encoded plasmid . Co-IP-IB assays were performed using antibodies as indicated . ( F ) A schematic of NGFR binding region on MDM2 . ( G ) A schematic of MDM2 binding region on NGFR . ( H ) Endogenous interaction of NGFR and MDM2 in SK-N-SH cells . Cells treated with Doxorubicin for 12 hr and MG132 for 6 hr were harvested for co-IP-IB assays using antibodies as indicated . ( I , J , K ) NGFR interacts with both MDM2 and p53 in the nucleus . Nuclear fractions from NGFR-stably expressed H460 ( I ) and SK-MEL-147 cells ( J , K ) were subjected to co-IP-IB assays using antibodies as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 01010 . 7554/eLife . 15099 . 011Figure 5—figure supplement 1 . Full-length of NGFR localizes in both nucleus and cytoplasm . ( A , B , C ) NGFR-stably expressed H460 ( A ) , SK-MEL-147 ( B ) and HCT116 p53+/+ cells ( C ) were fractionated into nuclear and cytoplasmic/membrane fractions which were subsequently subjected to IB using antibodies as indicated . ( D ) SK-MEL-147 cells were treated with or without Dox for 20 hr and fixed in −20°C for overnight . The fixed cells were subjected to immunostaining using antibodies as indicated . The images were acquired by a confocal microscope . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 011 While testing the endogenous NGFR-MDM2 binding , we also detected p53 in the NGFR-MDM2 complex ( Figure 5H–K ) . Thus , we tested if p53 binds to NGFR directly or indirectly through MDM2 . First , we conducted co-IP-IB and GST-Pull down assays . Indeed , NGFR and p53 co-immunoprecipitated with each other in reciprocal co-IP-IB assays ( Figure 6A–B ) . Further confirming this interaction was the mapping of the NGFR-binding domain to the central DNA-binding domain of p53 using a similar co-IP-IB assay by transfecting plasmids encoding individual p53 fragments together with NGFR plasmid into H1299 cells ( Figure 6C ) . This mapping was validated by a GST-pull down assay using purified GST-p53 fusion proteins with different p53 fragments . NGFR specifically bound to the full-length p53 and its central DNA-binding domain ( aa 101–300 ) ( Figure 6D ) . This finding explained why NGFR , MDM2 and p53 could form a ternary complex ( Figure 5H–k ) . We also mapped the p53-binding domain of NGFR by conducting a set of co-IP-IB assays . Like MDM2 , p53 also bound to the N-terminal region of NGFR ( Figure 6E ) . Together with the result showing that NGFR binds to p53 in the nucleus ( Figure 5I–K ) , these results demonstrate that the N-terminus of NGFR binds to the central DNA-binding domain of p53 , and this interaction also occurs in the nucleus . 10 . 7554/eLife . 15099 . 012Figure 6 . NGFR interacts with p53 in the nucleus . ( A , B ) NGFR interacts with p53 . H1299 cells were transfected with plasmids encoding Myc-NGFR , Flag-p53 or p53 followed by co-IP-IB assays using antibodies as indicated . ( C ) Mapping the NGFR binding domain on p53 by co-IP-IB assays . H1299 cells were transfected with the plasmid encoding flag-tagged p53 fragment , aa 1–300 , aa 101–300 , aa 101–393 or aa 301–393 , along with the Myc-NGFR-encoded plasmid . Co-IP-IB assays were performed using antibodies as indicated . ( D ) Mapping the NGFR binding domain on p53 by GST-pull down assay . The prokaryotic expressed GST-tagged full-length p53 or p53 fragment , aa 1–73 , aa 100–290 or aa 291–393 , or GST protein alone was incubated with cell lysates overexpressing Myc-NGFR . Bound proteins were detected by IB using anti-NGFR or coomassie staining . ( E ) Mapping the p53-binding domain on NGFR . H1299 cells were transfected with the plasmid encoding Myc-tagged NGFR fragment , aa 1–272 or aa 273–427 , along with the V5-p53-encoded plasmid . Co-IP-IB assays were performed using antibodies as indicated . ( F ) A schematic of NGFR binding region on p53 . ( G ) A schematic of p53 binding region on NGFR . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 012 The finding that NGFR binds to the central DNA-binding domain of p53 suggests that this protein might directly regulate p53 transcriptional activity independently of MDM2 . To test this idea , we first determined if NGFR can bind to p53 in the absence of MDM2 by introducing their expression plasmids alone or together into MEFMdm2-/-;p53-/- cells followed by co-IP-IB assays . Indeed , NGFR was co-immunoprecipitated with p53 in these cells ( Figure 4A ) . Interestingly and as also mentioned above , ectopic NGFR at different doses failed to decrease the level of ectopic p53 in MEFMdm2-/-;p53-/- cells ( Figure 4B ) , suggesting that NGFR cannot directly degrade p53 without MDM2 . Then , we tested if NGFR could regulate p53 transcriptional activity in the absence of MDM2 by performing luciferase reporter assays . Interestingly , ectopic NGFR inhibited p53-induced luciferase activity driven by the miR-34a promoter in MEFMdm2-/-;p53-/- cells in a dose dependent fashion ( Figure 4C ) . Also , NGFR repressed ectopic p53-induced expression of p21 and Puma in MEFMdm2-/-;p53-/- cells ( Figure 4D ) . Consistently , ectopic NGFR markedly reduced the association of p53 with the endogenous p21 and Bax promoters in these cells as analyzed by ChIP assays in H1299 or MEFMdm2-/-;p53-/- cells ( Figure 4E ) . Since MDMX , the partner of MDM2 , also bound to NGFR ( Figure 4—figure supplement 1A ) , MEFMdm2-/-; Mdmx-/-; p53-/- cells were used to determine if MDMX is required for p53 inactivation by NGFR . NGFR still inhibited p53 activity as measured by p21 and Puma expression ( Figure 4—figure supplement 1B ) . Collectively , these results demonstrate that NGFR also inhibits p53 by directly binding to its central DNA-binding domain and preventing its association with its target promoter . The discovery that NGFR is an inhibitor of p53 in cancer cells suggested the possibility of that depletion of this protein could sensitize cancer cells to chemotherapeutic agents . We tested this idea by treating H460 and SK-N-SH cells with different doses of Dox and Cisplatin after introducing control or NGFR siRNAs into the cells . Indeed , both of the chemotherapeutic drugs induced p53 and its target PUMA or p21 more apparently when the cells were transfected with NGFR siRNA compared to those with control siRNA in a dose-dependent fashion ( Figure 7A–D ) . Also , this more robust p53 induction led to more cytotoxicity , as the IC50 of Cisplatin in NGFR siRNA-transfected H460 was 2 . 2-fold less than that in control siRNA-transfected H460 cells ( Figure 7E ) . The result was repeated in HCT116 p53+/+ cells , in which the IC50 of Cisplatin dropped by 1 . 6 folds in response to NGFR knockdown ( Figure 7G ) . This enhanced cytotoxicity was p53-dependent , as less or no significant difference in cell viability was observed when p53-null H1299 or HCT116 p53-/- cells were used for the same experiment ( Figure 7F and H ) . These results suggest that NGFR confers resistance of cancer cells to chemotherapeutic-triggered p53 activation and cytotoxicity , and thus , inactivation of NGFR can sensitize the cells to chemotherapy . 10 . 7554/eLife . 15099 . 013Figure 7 . NGFR confers cancer cells resistance to chemotherapeutic agents . ( A , B ) NGFR knockdown enhances Cisplatin or Doxorubicin-triggered p53 activation in H460 cells . Cells were transfected with NGFR or control siRNA for 72 hr and Cisplatin ( A ) or Doxorubicin ( B ) was supplemented in the medium 12 hr before the cells were harvested for IB using antibodies as indicated . ( C , D ) NGFR knockdown enhances Cisplatin ( C ) or Doxorubicin ( D ) -triggered p53 activation in SK-N-SH cells . The same experiments as shown in ( A , B ) were performed except that the SK-N-SH cell line was used instead . ( E , F ) NGFR knockdown sensitizes H460 ( E ) but not H1299 cells ( F ) to Cisplatin treatment . Cells were transfected with NGFR or control siRNA and seeded in 96-well plates the next day . Cisplatin was supplemented 48 hr before cell viability was determined using CCK-8 as described in Materials and methods . ( G , H ) NGFR knockdown sensitizes HCT116 p53+/+ ( G ) but not HCT116 p53-/- cells ( H ) to Cisplatin treatment . The same experiments described in ( E , F ) were performed except for using different cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 013 To further translate the above findings into biological significance , we then determined if NGFR is required for tumor growth in mice by generating a xenograft tumor model using severe combined immunodeficiency ( SCID ) mice and H460 cells . NGFR shRNA or control shRNA-packaged lentivirus was produced to infect H460 and SK-MEL-147 cells . As expected ( Figure 3B ) , the expression of p53 and its target genes p21 , PUMA , and MDM2 was induced by knocking down NGFR using the shRNA-expressing lentiviruses ( Figure 8A and Figure 8—figure supplement 1A ) . Correspondingly , H460 cells with deficient NGFR grew slowly ( Figure 8B ) and formed less colonies ( Figure 8C ) . In line with this result , the xenograft tumors derived from NGFR-shRNA H460 cells grew remarkably slower than those from control-shRNAH460 cells in the SCID mice ( Figure 8D ) . Also , the average tumor weight from control cells was ~ten-fold heavier than that from NGFR-shRNA cells by the end of the experiment ( Figure 8E ) . Tumors derived from NGFR-shRNA H460 cells are much smaller than those derived from control-shRNA H460 cells ( Figure 8F and Figure 8—figure supplement 1B ) . Because of the diminutive size of the tumors generated from the NGFR knockdown group , we were unable to collect sufficient amounts of tumor tissues for WB and IHC staining . But , by performing q-PCR analysis , we found that the expression of p21 , PUMA , and MDM2 is significantly elevated in response to NGFR depletion in the xenograft tumors ( Figure 8G ) , consistent with the in vitro results ( Figure 3C and D ) . Thus , these results demonstrate that NGFR is required for tumor growth in vivo . 10 . 7554/eLife . 15099 . 014Figure 8 . NGFR is required for tumor growth in vivo . ( A ) Lentiviral-based knockdown of NGFR induces p53 pathway . H460 cells were transduced with lentivirus expressing NGFR or control shRNA for 72 hr followed by IB using antibodies as indicated . ( B ) H460 cells stably expressing NGFR shRNA show suppressed cell survival . H460 cells stably expressing NGFR or control shRNA were seeded in 96-well plates and cell viability was evaluated every 24 hr ( Mean ± SEM , n = 3 ) . Three biological replicates were used for p value , p<0 . 01 . ( C ) H460 cells stably expressing NGFR shRNA exhibit restrained clonogenic capacity . H460 cells stably expressing NGFR or control shRNA were seeded on 10-cm plates and colonies were fixed by methonal and stained with crystal violet solution ( upper panel ) . Quantification of colonies is shown in the lower panel . ( D ) H460 cells stably expressing NGFR shRNA reveal less xenograft tumor volume in average ( Mean ± SEM , n = 6 ) . ( E ) H460 cells stably expressing NGFR shRNA reveal less xenograft tumor weight in average ( Mean ± SEM , n = 6 ) . Six pairs of tumors were used for p value , p<0 . 01 . ( F ) Representative xenograft tumors reveal that NGFR knockdown dramatically suppressed tumor growth in vivo . ( G ) The p53 pathway is activated by NGFR knockdown in the xenograft tumors . The expression of p21 , PUMA and MDM2 was determined by q-PCR ( Mean ± SEM , n = 3 ) . Three pairs of tumors were used for p value , p*<0 . 01 . ( H ) A model for NGFR regulation of the MDM2-p53 loop in cancer . NGFR inactivates p53 through two mechanisms: 1 ) by directly associating with MDM2 and enhancing MDM2-mediated p53 ubiquitination and proteasomal degradation , and 2 ) by repressing p53 transcriptional activity through direct interaction with its DNA binding domain , consequently leading to tumorigenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 01410 . 7554/eLife . 15099 . 015Figure 8—figure supplement 1 . Lentivirus-mediated knockdown of NGFR activates p53 in a melanoma cell line and suppresses in vivo tumor growth . ( A ) p53 pathway is activated in NGFR-shRNA stably expressed SK-MEL-147 cells . SK-MEL-147 cells infected with lentivirus expressing scrambled or NGFR shRNA were subjected to IB using antibodies as indicated . ( B ) Representative xenograft tumors in mice . NGFR knockdown suppresses tumor growth in vivo , original image related to Figure 8F . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 01510 . 7554/eLife . 15099 . 016Figure 8—figure supplement 2 . NGFR is apt to be amplified in breast cancers sustaining wild-type p53 . TCGA database was searched and the data were modified from the cBioPortal for Cancer Genomics ( http://www . cbioportal . org/; Gao et al . , 2013; Cerami et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15099 . 016 First , we demonstrate that NGFR is a bona fide p53 target gene . The expression of NGFR at both mRNA and protein levels was induced by INZ , Dox , and 5-FU in several p53-positive , but not in p53-null , cancer cells ( Figure 1A–E ) . Also , overexpression of wt , but not mutant , p53 stimulated the expression of NGFR in p53-null cancer cells ( Figure 1F and G ) . In addition , knockdown of p53 in HCT116 p53+/+ and H460 impaired the induction of NGFR mRNA by Dox and 5-FU ( Figure 1H and I ) . Furthermore , p53 stimulated the expression of luciferase gene driven by the p53RE sequence derived from the NGFR promoter ( Figure 1J–L ) . Finally , in response to Dox , p53 was induced to associate with the p53RE-containing NGFR promoter in H460 cells . Undoubtedly , NGFR is an authentic p53 target gene , which is also in line with a recent finding that NGFR is a transcriptional target for the p53 analog , p73 , in the mouse nerve system ( Niklison-Chirou et al . , 2013 ) . However , NGFR surprisingly counteracted p53 function , as knockdown of this protein in p53-containing HCT116 or H460 cells led to drastic apoptosis and significant growth arrest ( Figure 2A–F ) . Also , knockdown of NGFR led to a more significant decrease of colony formation in p53-containing H460 cells compared to p53-deficient H1299 cells ( Figure 2G ) . Of note , the suppressive effect of NGFR knockdown on H1299 colony formation might be due to the oncogenic function of this protein via other oncoproteins , such as NF-κB ( Carter et al . , 1996 ) . But , this result does not affect our conclusion that negation of p53 function is one of the oncogenic functions of NGFR , as it is strongly supported by the evidence that knockdown of NGFR leads to activation of the p53 pathway in H460 , HepG2 , SK-N-SH , HCT116 p53+/+ and melanoma SK-MEL cells ( Figure 3B–D ) . Remarkably , a number of p53 responsive target genes were induced to different degrees when NGFR was knocked down in H460 cells ( Figure 3C ) . The NGFR inhibition of p53 is further supported by the fact that it inactivates p53 by two distinct mechanisms as detailed below . One mechanism for NGFR inactivation of p53 is by working with MDM2 . First , ectopic NGFR enhanced MDM2-mediated ubiquitination and proteosomal degradation of p53 ( Figure 3F and G ) . Inversely , knockdown of NGFR led to p53 stabilization ( Figure 3E ) . These actions appeared to be implemented through direct interaction of NGFR with MDM2 , as NGFR interacted with MDM2 both in vitro ( Figure 5D ) and in the nucleus ( Figure 5I–K ) via its N-terminal region ( Figure 5E ) . Although stress signals ( Palmero et al . , 1998; Shieh et al . , 1997; Tang et al . , 2008; Zhang et al . , 1998; Zindy et al . , 1998 ) ( Zhang and Lu , 2009; Zhou et al . , 2012 , 2015a ) have been shown to compromise MDM2-induced ubiquitination and proteolysis of p53 and thus activate p53 in normal cells , cancer cells employ oncogenic molecules , such as MDMX ( Gembarska et al . , 2012; Lam et al . , 2010; Slack et al . , 2005; Wade et al . , 2013 ) , which strengthen MDM2-mediated inhibition of p53 and thus induce cell resistance to p53 activation . Our data indicates that NGFR is one of these molecules as knockdown of NGFR enhances chemotherapeutic agents-induced p53 activation ( Figure 7A–D ) and cell growth inhibition ( Figure 7E– H ) . Surprisingly , NGFR can also inactivate p53 independent of MDM2 . First , NGFR binds to p53 in cells and in vitro ( Figure 6A–E ) . Interestingly , NGFR bound to the central DNA-binding domain of p53 ( Figure 6D ) , and reduced the binding of p53 to the p21 and BAX promoters ( Figure 4E–F ) . All of these activities of NGFR are MDM2-independent , as they were all detected in MEFp53-/-/MDM2-/- cells ( Figure 4A–6D ) . However , without MDM2 , ectopic NGFR was unable to mediate p53 degradation ( Figure 4B ) . These results demonstrate that NGFR can directly bind to the central DNA-binding domain of p53 and impair its ability to bind to its target promoters in addition to facilitating MDM2-dependent p53 ubqiutination and degradation . Hence , our study illustrates two novel mechanisms by which NGFR inactivates p53 , which are ligand-independent as further described below . Our findings that NGFR can directly and indirectly suppress p53 activity were unexpected , as NGFR was originally found to be a transmembrane pan-receptor involved in the initiation , development , and maintenance of the nervous system and human cancers ( Lee et al . , 2001; Molloy et al . , 2011; Patapoutian and Reichardt , 2001 ) , which binds with low affinity to all mature neurotrophins , including nerve growth factor ( NGF ) , brain-derived neutrophic factor ( BDGF ) , neurotrophin 3 ( NT-3 ) , and neurotrophin 4/5 ( NT-4/5 ) , as well as their precursors , pro-neurotrophins , with high affinity ( Barker , 2004 ) . Although a previous study showed that in response to neurotrophin signaling , NGFR might mediate p53-dependent neuron cell death by activating the JNK pathway ( Aloyz et al . , 1998 ) , our studies as presented here unveil a novel nuclear function of this membrane receptor , i . e . , to promote cancer cell proliferation and survival by directly inhibiting p53 transcriptional activity and indirectly destabilizing p53 protein via MDM2 . Convincingly , we showed that NGFR binds to MDM2 in the nucleus and assists this E3 ligase to ubiquitinate p53 and promote its degradation ( Figures 3 , 5 ) . Also , independently of MDM2 , NGFR directly bound to the central domain of p53 in the nucleus and prevented p53 from binding to its target promoters ( Figures 4 , 6 ) . These inhibitory activities of NGFR toward p53 are ligand-independent , as they occurred in the nucleus and NGF treatment did not appear to affect p53 level and activity in H460 , HCT116 , and HepG2 cancer cells tested ( Data not shown ) . Hence , our results for the first time uncover a ligand-independent and nuclear p53 inhibitory function of this membrane neurotrophic receptor by acting on the MDM2-p53 loop . Although it remains to be studied if NGFR can also regulate p53 stability and activity in a negative feedback fashion in normal nerve systems , our findings demonstrate that cancer cells hijack this anti-p53 activity of NGFR toward their growth advantage in a way similar to that for MDM2 and MDMX inactivation of p53 in cancers ( Momand et al . , 1992; Oliner et al . , 1992 , 1993; Shvarts et al . , 1996; Wu et al . , 1993 ) , as knockdown of NGFR markedly induced p53-dependent apoptosis and cytotoxicity as well as eliminated xenograft tumor growth ( Figures 2–8 ) . In line with our findings , a growing body of evidence has identified NGFR as a robust cell surface biomarker not only for neural crest stem cells , but also cancer initiating or stem-like cells ( Tomellini et al . , 2014 ) . Also , NGFR was highly expressed in a number of cancers , including the cancer initiating cells of melanoma ( Boiko et al . , 2010 ) , squamous cell carcinomas ( Murillo-Sauca et al . , 2014 ) , osteosarcoma ( Tian et al . , 2014 ) , brain cancer ( Biagiotti et al . , 2006 ) , breast cancer ( Kim et al . , 2012 ) , and neuroblastoma ( Biagiotti et al . , 2006 ) . Particularly , patient derived NGFR-positive , but not NGFR-negative , melanoma cells were remarkably capable of generating tumors , promoting metastasis , and maintaining self-renewal ( Boiko et al . , 2010 ) . Also , we detected the high expression of NGFR in several melanoma cell lines ( Figure 3B ) and primary human gliomas ( Figure 2H , I and Figure 2—figure supplement 1B ) . Several of these cancers , such as breast cancer , melanoma , osteocarcoma , and neuroblastoma , often display less or no TP53 mutations ( Petitjean et al . , 2007; Soussi et al . , 2005 ) . Although MDM2 and MDMX have been shown to play a role in suppressing p53 in these cancers ( Gembarska et al . , 2012; Lam et al . , 2010; Slack et al . , 2005; Wade et al . , 2013 ) , it is highly likely that NGFR might also contribute to the reason why these cancers harbor wt p53 . Our findings as demonstrated here further support this likelihood and provide novel mechanisms underlying NGFR inactivation of p53 in these cancers or cancer stem cells . Together with the aforementioned studies ( Boiko et al . , 2010; Tomellini et al . , 2014 ) , our findings also suggest that NGFR might play a role in maintaining the renewal and proliferating capabilities of stem cells or cancer stem cells by inactivating p53 , as p53 has been shown to be crucial for stem cell differentiation and apoptosis ( Krizhanovsky and Lowe , 2009 ) and to be a major roadblock for these stem cells to renew and proliferate ( Cicalese et al . , 2009; Hong et al . , 2009; Kawamura et al . , 2009; Marion et al . , 2009; Utikal et al . , 2009 ) . In contrast to our findings and the oncogenic role as described above , NGFR has also been shown to exert a tumor suppressive function by repressing tumor growth in several cancers ( Molloy et al . , 2011 ) , which has been largely attributed to its receptor activity ( Barker , 2004; Bredesen and Rabizadeh , 1997; Carter et al . , 1996; El Yazidi-Belkoura et al . , 2003; Khursigara et al . , 2001; Nykjaer et al . , 2004; Tomellini et al . , 2014 ) . One previous study showed that the pro-neurotrophin-NGFR signaling can induce neuron death through the JNK-p53 pathway ( Aloyz et al . , 1998 ) , but little is known about if this is also true in cancer or if NGFR can activate p53 independent of its ligands . All these studies did not notice the receptor-independent intracellular function of NGFR . Thus , our study not only offers new insights into the mechanism of intracellular NGFR-mediated tumor promotion , but also demonstrates that this oncogenic function is p53-dependent , which could explain why several human cancers that express high levels of NGFR harbor wt p53 . Finally , our study suggests NGFR as a potential therapeutic target for cancers that harbor wt p53 and high levels of NGFR . The RFP-tagged plasmid pDSRed-NGFR was purchased from Addgene created by Dr . Moses Chao . The Myc-tagged NGFR expression plasmid was generated by inserting the full-length cDNA amplified by PCR from pDSRed-NGFR into the pcDNA3 . 1/Myc-His vector , using the following primers , 5-CCGGAATTCATGGGGGCAGGTGCCACC-3 and 5-CGCGGATCCCACCGGGGATGTGGCAGT-3 . The Myc-tagged plasmids expressing NGFR fragment , aa 1–272 or aa 273–427 , were generated by the same approach using the corresponding primers . The pGL3-RE1 and RE2 plasmids were generated by inserting the genomic DNA covering p53 RE1 or RE2 into the pGL3-promoter vector using the following primers , 5-CGGGGTACCTTCTACTGTCATGTCAAAGGAA-3 and 5-CCGCTCGAGCCCTCCAGCTACTACTCAGAC-3 for RE1; 5-CGGGGTACCGGCAAGTGGCATTGGTGGTA-3 and 5-CCGCTCGAGTCGTTTGTAAAGTGGGCATAA-3 for RE2 . The lentiviral-based plasmid NGFR shRNA-1 was generated by inserting the following sequence , 5-CCGGCCGAGCACATAGACTCCTTTACTCGAGTAAAGGAGTCTATGTGCTCGGTTTTTG-3 into pLKO . 1 vector . The plasmids NGFR shRNA-2 and −3 were purchased ( Sigma-Aldrich , St . Louis , MO , USA ) . The plasmids encoding HA-MDM2 , Flag-MDM2 , V5-MDM2 fragments , GST-MDM2 fragments , HA-MDMX , p53 , Flag-p53 , His-Ub and pGL4-miR-34a-luciferase were described previously ( Dai and Lu , 2004; Dai et al . , 2004; Zhou et al . , 2015b ) . Plasmids encoding GST-tagged p53 fragments and Flag-tagged p53 fragments were gifts from Wei Gu and Mushui Dai , respectively . V5-tagged p53 was purchased from Addgene ( Junk et al . , 2008 ) . Anti-Flag ( Sigma-Aldrich , St . Louis , MO , USA ) , anti-Myc ( 9E10 , Santa Cruz Biotechnology , Santa Cruz , CA , USA ) , anti-V5 ( E10 , Thermo Scientific , Waltham , MA , USA ) , anti-GFP ( B-2 , Santa Cruz Biotechnology ) , anti-NGFR ( Millipore , Billerica , MA , USA; EP1039Y , GeneTex , Irvine , CA , USA and D4B3 , Cell signaling Technology , Danvers , MA , USA ) , anti-p53 ( DO-1 , Santa Cruz Biotechnology ) , anti-p21 ( CP74 , Neomarkers , Fremont , CA , USA ) , anti-PUMA ( H-136 , Santa Cruz Biotechnology ) and anti-β-actin ( C4 , Santa Cruz Biotechnology ) were commercially purchased . Antibodies against MDM2 ( 2A9 , 2A10 and 4B11 ) were previously described ( Dai and Lu , 2004; Dai et al . , 2004 ) . Human cancer cell lines H460 , H1299 , HCT116p53+/+ , HCT116p53-/- , HepG2 , PCL/PRF/5 , U2OS , SK-MEL-103 , SK-MEL-147 , MEFp53-/-;Mdm2-/- and MEFp53-/-;Mdm2-/-; Mdmx-/- cells were cultured in Dulbecco’s modiﬁed Eagle’s medium ( DMEM ) supplemented with 10% fetal bovine serum , 50 U/ml penicillin and 0 . 1 mg/ml streptomycin . The human neuroblastoma cell line SK-N-SH was cultured in RPMI 1640 medium supplemented with 10–15% fetal bovine serum , 50 U/ml penicillin and 0 . 1 mg/ml streptomycin . All cells were maintained at 37°C in a 5% CO2 humidiﬁed atmosphere . Cells seeded on the plate overnight were transfected with plasmids as indicated in ﬁgure legends using TurboFect transfection reagent following the manufacturer’s protocol ( Thermo Scientific ) . Cells were harvested at 30–48 hr post-transfection for future experiments . GST-tagged MDM2 fragments or p53 fragments were expressed in E . coli and conjugated with glutathione-Sepharose 4B beads ( Sigma-Aldrich ) . Protein-protein interaction assays were conducted by using cell lysates with mammalian-expressed Myc-NGFR . Briefly , the cell lysates were incubated and gently rotated with the glutathione-Sepharose 4B beads containing 500 ng of GST-MDM2 fragments , GST-p53 fragments or GST only at 4°C for 4 hr . The mixtures were washed three times with GST lysis buffer ( 50 mM Tris/HCl pH 8 . 0 , 0 . 5% NP-40 , 1 mM EDTA , 150 mM NaCl , 10% glycerol ) . Bound proteins were analyzed by IB with the antibodies as indicated in the figure legends . Chromatin immunoprecipitation ( ChIP ) assay was performed using antibodies as indicated in the figure legends and described previously ( Liao and Lu , 2013 ) . The reverse cross-linked immuoprecipitated DNA fragments were purified using GeneJET gel extraction kit ( Thermo Scientific ) followed by PCR analyses for the p53-responsive DNA elements on the promoters of human NGFR and p21 and mouse p21 using the following primers , 5-GACTCCAACCTTGCTAATTCCT-3 and 5-TGACCTTCACCAGTTCTCACT-3 for human NGFR , 5-GCTCCCTCATGGGCAAACTCACT-3 and 5-TGGCTGGTCTACCTGGCTCCTCT-3 for human p21 , and 5-CCTTTCTATCAGCCCCAGAGGATACC-3 and 5-GACCCCAAAATGACAAAGTGACAA-3 for mouse p21 . Total RNA was isolated from cells using Trizol ( Invitrogen , Carlsbad , CA , USA ) following the manufacturer’s protocol . Total RNAs of 0 . 5 to 1 µg were used as templates for reverse transcription using poly- ( T ) 20 primers and M-MLV reverse transcriptase ( Promega , Madison , WI , USA ) . Quantitative PCR ( qPCR ) was conducted using SYBR Green Mix according to the manufacturer’s protocol ( BioRad , Hercules , CA , USA ) . The primers for human NGFR , mouse p21 and Puma are as follows , 5-CCTGGACAGCGTGACGTTC-3 and 5-CCCAGTCGTCTCATCCTGGT-3 for NGFR , 5-CCAGCAGAATAAAAGGTGCCACAGG-3 and 5-GCATCGCAATCACGGCGCAA-3 for mouse p21 , and 5-ACGACCTCAACGCGCAGTACG-3 and 5-GAGGAGTCCCATGAAGAGATTG-3 for mouse Puma . The primers for human BTG2 , BAX , MDM2 , p21 , PUMA and GAPDH were previously described ( Sun et al . , 2010; Zhou et al . , 2015b ) . Cells transfected with siRNAs as indicated in the figure legends were fixed with ethanol overnight and stained in 500 ml of propidium iodide ( Sigma-Aldrich ) stain buffer ( 50 µg/ml PI , 200 µg/ml RNase A , 0 . 1% Triton X-100 in phosphate-buffered saline ) at 37°C for 30 min . The cells were then analyzed for DNA content using a BD Biosciences FACScan flow cytometer ( BD Biosciences , San Jose , CA , USA ) . Data were analyzed using the CellQuest ( BD Biosciences ) and Modfit ( Verity , Topsham , ME , USA ) software programs . Sub-G1 as an indicator of apoptosis was measured by determining the number of events in channels 0 to 40 in a 256 channel histogram . To assess the long term cell survival , the Cell Counting Kit-8 ( CCK-8 ) ( Dojindo Molecular Technologies , Rockville , MD , USA ) was used according to the manufacturer’s instructions . Cell suspensions were seeded at 2000 cells per well in 96-well culture plates at 12 hr post-transfection . Cell viability was determined by adding WST-8 at a final concentration of 10% to each well , and the absorbance of the samples was measured at 450 nm using a Microplate Reader ( Molecular Device , SpecrtraMax M5e , Sunnyvale , CA , USA ) every 24 hr for 4 days . The RNA-sequencing service was provided by the Genomics and Biostatistics Core at the Tulane Center for Aging and the RNA-sequencing data were analyzed by the Cancer Crusaders Next Generation Sequence Analysis Core of the Tulane Cancer Center . Experiments were triplicate and genes with over 1 . 5-fold increase in expression ( p<0 . 05 ) were shown in the study . Cells were trypsinized and seeded with the same amount on 10-cm plates following siRNA transfection for 12 to 18 hr . The medium was changed every 3 days until the colonies were visible . Puromycin was added in the medium when the stable cell lines were used in the experiment . Cells were then fixed by methonal and stained by crystal violet solution at RT for 30 min . ImageJ was used for quantification of the colonies . Cells were harvested and lysed in lysis buffer consisting of 50 mM Tris/HCl ( pH7 . 5 ) , 0 . 5% Nonidet P-40 ( NP-40 ) , 1 mM EDTA , 150 mM NaCl , 1 mM dithiothreitol ( DTT ) , 0 . 2 mM phenylmethylsulfonyl ﬂuoride ( PMSF ) , 10 µM pepstatin A and 1 mM leupeptin . Equal amounts of clear cell lysate ( 20–80 µg ) were used for immunoblotting ( IB ) analyses as described previously ( Zhou et al . , 2013 ) . Cells were transfected with pCMV-β-galactoside together with the plasmids as indicated in the figures . Luciferase activity was determined and normalized by a factor of β-Gal activity in the same assay as described previously ( Liu et al . , 2014; Zhou et al . , 2015b ) . H1299 cells were transfected with plasmids encoding p53 , HA-MDM2 , His-Ub or Myc-NGFR as indicated in the figure legends . At 48 hr after transfection , cells were harvested and split into two aliquots , one for IB and the other for ubiquitination assays . In vivo ubiquitination assays were conducted as previously described ( Zhou et al . , 2013 ) . Briefly , cell lysates were incubated with Ni-NTA beads that capture His-tagged proteins/complex at RT for 4 hr . The captured proteins were eluted and analyzed by IB with the indicated antibodies . Immunoprecipitation ( IP ) was conducted using antibodies as indicated in the ﬁgure legends and described previously ( Zhou et al . , 2013 ) . Briefly , ~500 to 1000 μg of proteins were incubated with the indicated antibody at 4°C for 4 hr or overnight . Protein A or G beads ( Santa Cruz Biotechnology ) were then added and the mixture was left to incubate at 4°C for additional 1 to 2 hr . The beads were washed at least three times with lysis buffer . Bound proteins were detected by IB with antibodies as indicated in the ﬁgure legends . Cells were fixed with methanol in −20°C for overnight . The fixed cells were washed by PBS and blocked with 8% BSA in PBS for 1 hr followed by incubation with primary antibodies ( D4B3 for NGFR , 1:200 dilution; DO-1 for p53 , 1:100 dilution ) in 2% BSA in 4°C for overnight . The cells were then washed and incubated with the corresponding secondary antibodies and DAPI . Images were acquired with a confocal microscope ( Olympus FV1000 ) . Glioma and adjacent normal tissue samples were collected and archived at the First Affiliated Hospital of Nanchang University , Jiangxi , China . Due to the statistical consideration of small sample size ( ≥40 ) , we collected 48 pairs of Glioma and matched adjacent normal tissues . Fresh glioma tissues and paired noncancerous tissues were immediately snap-frozen in liquid nitrogen and stored at −80°C until their use in immunoblotting . Immunohistochemical staining was performed as previously described ( Zeng et al . , 2014 ) . All patients provided written informed consent to participate in the study and all primary glioma samples without preoperative radiotherapy were included and confirmed by pathologists . The siRNAs against NGFR and p53 ( Life Technologies , Carlsbad , CA , USA ) were commercially purchased . 40~60 nM of siRNAs were introduced into cells using TurboFect transfection reagent following the manufacturer’s protocol . Cells were harvested ~72 hr after transfection for IB or qPCR . Lentiviral plasmids based on pLKO . 1 were packaged with the 2nd Generation Packaging System . Briefly , pLKO . 1 plasmids containing scrambled or NGFR shRNAs , along with the packaging plasmids pMD2 . G and pCMV-dR8 . 2 , were transfected into 293T cells . The cells were maintained at 37°C in a 5% CO2 humidiﬁed atmosphere for 72 hr and the supernatant was harvested to infect H460 or SK-MEL-147 cells . The medium was changed by overnight infection and cells were split to two aliquots , one for IB analysis and the other for selection with 2 µg/ml puromycin . Seven-week-old female NOD/SCID mice were purchased from Jackson Laboratories . Mice were subcutaneously inoculated with 3 x 106 H460 cells infected with lentivirus encoding control shRNA or NGFR shRNA in the right and left flanks , respectively . Tumor growth was monitored every other day with electronic digital calipers ( Thermo Scientific ) in two dimensions . Tumor volume was calculated with the formula: tumor volume ( mm3 ) = ( length x width2 ) /2 ( Zhang et al . , 2012 ) . Mice were sacrificed by euthanasia and tumors were harvested and weighed . Mice with the largest or smallest tumor size were excluded from our study . To detect p53 activation and apoptotic signals in vivo , the tumors were disrupted in Trizol and subjected to RT-qPCR analyses .	Cancer often develops as a result of alterations to “tumor suppressor” genes within cells . This results in the cells growing and dividing too much , which causes a tumor to form . One of the most important tumor suppressor genes produces a protein called p53 , which is lost or mutated in roughly half of all human cancers . In the other half of cancers p53 itself is normal , but is often disabled by proteins that promote tumor growth . One of the remaining challenges in the field of cancer research is to identify which proteins inhibit p53 directly . Identifying these proteins would help clarify why many human cancers , such as some brain cancers , breast and skin cancers , often maintain a normal form of the p53 tumor suppressor protein . Zhou et al . now provide evidence that shows that a protein called nerve growth factor receptor ( NGFR ) is one such protein . NGFR was known to be important for the healthy development of the brain and nervous system . Unexpectedly , however , Zhou et al . found that NGFR binds directly to p53 and disables it in several types of human cancer cells . This finding is likely to be important because NGFR is produced in abnormally high amounts in several human cancer types , including skin , breast , bone and some brain cancers . Reducing the levels of NGFR in cancer cells caused the cells to become more sensitive to some anti-cancer drugs . Overall , the results presented by Zhou et al . suggest that developing new drugs that target NGFR could produce new treatments for human cancers that have a normal form of the gene that produces p53 . More experiments are also needed to find out whether NGFR has other ways of promoting the development of cancerous tumors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2016	Nerve growth factor receptor negates the tumor suppressor p53 as a feedback regulator
The HIV-1 protein Rev controls a critical step in viral replication by mediating the nuclear export of unspliced and singly-spliced viral mRNAs . Multiple Rev subunits assemble on the Rev Response Element ( RRE ) , a structured region present in these RNAs , and direct their export through the Crm1 pathway . Rev-RRE assembly occurs via several Rev oligomerization and RNA-binding steps , but how these steps are coordinated to form an export–competent complex is unclear . Here , we report the first crystal structure of a Rev dimer-RRE complex , revealing a dramatic rearrangement of the Rev-dimer upon RRE binding through re-packing of its hydrophobic protein–protein interface . Rev-RNA recognition relies on sequence-specific contacts at the well-characterized IIB site and local RNA architecture at the second site . The structure supports a model in which the RRE utilizes the inherent plasticity of Rev subunit interfaces to guide the formation of a functional complex . Retroviruses such as HIV have small genomes and utilize multiple strategies such as overlapping reading frames and alternative splicing , to encode their repertoire of proteins . The ∼9 kb HIV-1 genome codes for 15 proteins expressed from a set of unspliced and singly-spliced mRNAs in addition to the fully spliced messages ( Frankel and Young , 1998 ) . These unspliced and singly-spliced species encode the viral structural and accessory proteins and genomic RNA needed to assemble new virions , but because they contain introns are typically retained in the nucleus for splicing . To export these RNAs to the cytoplasm , the viral protein Rev ( Figure 1A ) , expressed from a fully spliced message , translocates into the nucleus , forms an oligomeric complex on a ∼350 nucleotide , highly structured intronic RNA element , the Rev Response Element ( RRE ) ( Figure 1B ) , and directs their export through the Crm1 nuclear export pathway ( Fornerod et al . , 1997; Pollard and Malim , 1998 ) . 10 . 7554/eLife . 04120 . 003Figure 1 . Overall organization and structure of the Rev dimer-RRE complex . ( A ) Domain organization of full-length Rev . The protein used to co-crystallize the complex contains residues 1–70 and contains the higher-order oligomerization disrupting mutations , L12S and L60R , and the surface-entropy reducing mutation , E47A . ( B ) Secondary structure of the RRE ( Watts et al . , 2009; Bai et al . , 2014 ) with the region used for the co–crystal structure shown in red . ( C ) Sequence of IIB40 RNA with the region corresponding to the RRE shown in red . The bases are numbered according to older studies ( Battiste et al . , 1996 ) for consistency . The RNA contains a terminal UUCG tetra-loop to enhance stability and favor crystallization . See also Figure 1—figure supplement 1 ( D ) Overall arrangement of the Rev dimer-RNA complex: The RNA ( red ) is held between the first Rev molecule ( blue ) bound at IIB and the second Rev molecule ( green ) bound at the junction site . The N-terminal 11 residues and C-terminal 7 residues of Rev 1–70 are not visible in the structure . See also Figure 1—figure supplement 2 , 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 00310 . 7554/eLife . 04120 . 004Figure 1—figure supplement 1 . Rev and RRE constructs . ( A ) Rev dimer structure ( Daugherty et al . , 2010b ) with the higher-order oligomerization disrupting mutations , L12S and L60R shown in red . Also shown is the surface-entropy reducing mutation , E47A in blue or green . ( B ) RRE IIB40 is a derivative of the stem II three-helix junction and contains two adjacent Rev-binding sites ( red ) . ( C ) Gel-shift assays comparing binding of 32P-labeled RRE-stem II or IIB40 to Rev . Free , F , monomer , M and dimer , D complexes are indicated . A doublet/smeared band is observed for the monomer complex with both RNAs and is indicative of conformational heterogeneity . ( D ) Binding curves calculated from gel-shift assays in ( C ) . Apparent dissociation constants , Kd and Hill coefficient ( n ) were determined using the equation: Fraction of RNA bound = [Rev]n/ ( Kdn + [Rev]n ) as mean ± s . d . of two replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 00410 . 7554/eLife . 04120 . 005Figure 1—figure supplement 2 . Representative electron density maps . ( A ) The final model fit into 5 . 5 Å density-modified SAD electron density map ( contoured at 2 . 0 σ ) . ( B ) 2Fo − Fc electron density map at 3 . 2 Å after molecular replacement ( contoured at 2 . 0 σ ) . ( C and D ) Stereo view of 2Fo − Fc electron density maps located around ( C ) nucleotides that form the core structure of the junction site ( contoured at 2 . 0 σ ) ( D ) Gln51 which hydrogen bonds across the dimer interface in the RNA-bound form ( contoured at 1 . 0 σ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 00510 . 7554/eLife . 04120 . 006Figure 1—figure supplement 3 . Comparison of Rev monomers . Superimposition of Rev monomers corresponding to ( A ) the first Rev molecule in the unbound dimer ( grey ) and in the dimer-RNA complex ( blue ) ( backbone RMSD = 1 . 47 Å ) ( B ) the second Rev molecule in the unbound dimer ( grey ) and in the dimer-RNA complex ( green ) ( backbone RMSD = 1 . 41 Å ) ( C ) the first ( blue ) and second Rev ( green ) molecules from the dimer-RNA structure ( backbone RMSD = 0 . 87 Å ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 006 Rev cooperatively assembles on the RNA using its oligomerization and RNA-binding domains to form a Rev hexamer on a shorter , but functional ∼240 nucleotide RRE ( Malim and Cullen , 1991; Zapp et al . , 1991; Daugherty et al . , 2008 , 2010a ) . This assembled RNP then presents Leu-rich nuclear export sequence ( NES ) peptides ( Figure 1A ) to interact with the host export receptor , Crm1 , targeting these RRE-containing mRNAs for export ( Fornerod et al . , 1997 ) . In this paper we address the structural principles that govern assembly of the Rev oligomer onto the RNA and its consequences for Rev function . The binding of Rev to the RRE is nucleated at the stem IIB hairpin , where the α-helical arginine-rich motif ( ARM ) of Rev ( Figure 1A , B ) binds in the major groove of the RNA and makes an extensive set of sequence-specific and electrostatic contacts ( Malim et al . , 1990; Cook et al . , 1991; Heaphy et al . , 1991; Kjems et al . , 1991; Battiste et al . , 1996 ) . The oligomeric complex proceeds to form by the sequential addition of Rev subunits ( Pond et al . , 2009 ) . Crystal structures of Rev oligomerization surfaces depict a central hydrophobic core that mediates interactions between Rev subunits and positions the ARMs on one side of the Rev oligomer to bind RNA and the NESs on the other side to recruit Crm1 ( DiMattia et al . , 2010; Daugherty et al . , 2010b ) , highlighting the modular nature of the Rev domains . In addition to the modular architecture of Rev , the RNA scaffold also plays a key role in defining the organization of the complex . Biochemical studies have shown that the RRE controls the oligomeric state of Rev ( Daugherty et al . , 2010a ) , and a recent SAXS structure shows that the RRE adopts an ‘A’ shaped architecture poised to recruit Rev ( Fang et al . , 2013 ) . It remains unknown how the RNA structure is matched to the arrangement of the Rev oligomer to form the export–active complex . One particularly fascinating question is how the individual subunits of Rev recognize the RRE . An NMR structure of a single α-helical ARM bound to the IIB RNA hairpin has been known for quite some time ( Battiste et al . , 1996 ) , and a second binding site in stem IA was identified that utilizes a different surface of the helical ARM from another subunit to recognize the RNA ( Daugherty et al . , 2008 ) . Other binding sites in the RRE may use yet other recognition strategies that have not been defined . As a step towards understanding how individual binding sites are organized on the RRE , how their arrangement relates to the architecture of the Rev oligomer , and how the different sites are recognized by Rev , we solved the structure of a Rev dimer bound to an RRE fragment ( IIB40 ) containing two Rev-binding sites ( Daugherty et al . , 2008 ) ( Figure 1C ) . This first high-resolution structure of a Rev-RRE complex ( Figure 1D ) uncovers yet another RNA recognition mode for Rev and shows how the remarkable plasticity of the Rev dimer allows it to adapt to the RNA framework . To better understand how Rev assembles an oligomeric complex on the RRE , we sought to determine the crystal structure of a Rev-RNA complex . Rev has a strong tendency to aggregate and fold incorrectly when not bound to RNA ( Daugherty et al . , 2010b ) . We circumvented these problems by making RNA complexes with Rev 1–70 harboring the oligomerization-disrupting mutations , Leu12Ser and Leu60Arg ( Jain and Belasco , 2001 ) , which forms a stable Rev dimer ( Daugherty et al . , 2010b ) , and a Glu47Ala surface-entropy reducing mutation ( Goldschmidt et al . , 2007 ) , which aided in crystallization ( Figure 1A , Figure 1—figure supplement 1 ) . For the RNA , we utilized variants of RRE IIB40 , which has the high-affinity IIB site positioned adjacent to a second site ( referred to as the junction site ) that mimics the stem II junction ( Daugherty et al . , 2008 ) ( Figure 1C , Figure 1—figure supplement 1 ) , and variants of a 68-nucleotide fragment comprising stems IIA , IIB and IIC . Previous biochemical studies have shown that Rev assembly initiates at stem IIB and proceeds through stem II along the length of stem I with the second Rev molecule binding at the stem II junction ( Mann et al . , 1994; Charpentier et al . , 1997 ) . The junction site was designed based on those studies ( Daugherty et al . , 2008 ) and also on the observation that Rev binding melts a A-U base-pair in stem IIA at the junction ( Charpentier et al . , 1997 ) , a finding recently confirmed by SHAPE-seq experiments ( Bai et al . , 2014 ) . Gel shift assays show that Rev binds to RRE stem II and to IIB40 with comparable affinity and cooperativity ( Figure 1—figure supplement 1 ) , suggesting that IIB40 recapitulates Rev assembly at the entire stem II junction . We attempted co-crystallizations using ∼35–40 RNA variants with altered stem lengths and tetraloop sequences and obtained well-diffracting crystals with only a single Rev dimer-IIB40 complex . We solved its structure using single-wavelength anomalous dispersion to 5 . 5 Å to generate models for molecular replacement at 3 . 2 Å , resulting in the first high-resolution view of a Rev-RRE complex ( Figure 1D , Figure 1—figure supplement 2 , Table 1 ) . 10 . 7554/eLife . 04120 . 007Table 1 . Diffraction data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 007NativeTungsten derivativeData Collection Space groupP 41 3 2P 41 21 2 Cell Dimensions a , b , c ( Å ) 165 . 3 , 165 . 3 , 165 . 3147 . 22 , 147 . 22 , 199 . 44 α , β , γ ( ° ) 90 , 90 , 9090 , 90 , 90 Resolution ( Å ) 45 . 85 − 3 . 2 ( 3 . 314 − 3 . 2 ) 49 . 34 − 5 . 55 ( 5 . 75 − 5 . 55 ) Redundancy25 . 9 ( 15 . 9 ) 15 . 2 ( 13 . 6 ) Completeness ( % ) 99 . 72 ( 98 . 36 ) 99 . 6 ( 96 . 5 ) I/σI21 . 56 ( 1 . 74 ) 13 ( 1 . 8 ) R-meas0 . 138 ( 2 ) 0 . 151 ( 1 . 52 ) Refinement Resolution ( Å ) 45 . 85 − 3 . 2 ( 3 . 31 − 3 . 2 ) No . reflections342771 ( 20153 ) Rwork/Rfree19 . 3 ( 27 . 7 ) /21 . 1 ( 30 . 7 ) No . atoms RNA849 Protein862 Ligand/Ion5 Water1 B-factors RNA113 . 9 Protein115 . 3 Ion153 . 8 Water112 . 7 RMS deviations Bond lengths ( Å ) 0 . 002 Bond angles ( ° ) 0 . 39Statistics for the highest-resolution shell are shown in parentheses . The Rev dimer-RRE structure depicts essential early steps in the assembly of the larger oligomeric complex . The overall architecture of the complex shows the RNA held between two Rev molecules arranged slightly asymmetrically in a V-shaped topology , with the two Rev ARMs buried in distorted major grooves positioned on opposite sides of the RNA ( Figure 1D ) . The structures of the individual Rev monomers are nearly identical to each other and to the RNA-free form ( Figure 1—figure supplement 3 ) . Comparison of the structures of the free and RNA-bound Rev dimers , described below , illustrates how the homo-oligomer adapts to bind RNA cooperatively . The Rev dimer in its RNA-free state also displays a V-shaped topology ( DiMattia et al . , 2010; Daugherty et al . , 2010b ) but its conformation changes dramatically upon RNA binding , with the two RNA-binding helices coming much closer together in the RNA-bound complex ( Figure 2A ) . The Rev subunit bound to the junction site forms an extensive surface with the RNA that causes the dimer interface to pivot around a single residue , Ile55 ( Figure 2—figure supplement 1 ) , known to be critical for dimer integrity and function ( Jain and Belasco , 2001; Edgcomb et al . , 2008 ) . This altered conformer results in complete repacking of hydrophobic residues at the dimer interface ( Figure 2B ) , formed by residues Leu18 and Leu22 from the first helix and Ile55 and Ile59 from the second helix . Phe21 , which formed a significant part of the RNA-free interface , is largely excluded . 10 . 7554/eLife . 04120 . 008Figure 2 . Reorganization of the Rev dimer interface upon RNA binding . ( A ) Rev dimer crossing angles differ significantly between RNA-bound ( top ) and RNA-free ( bottom ) states ( B ) Packing of hydrophobic residues at the dimer interface in the RNA-bound ( top ) and RNA-free ( bottom ) states . Phe21 is largely excluded from the interface upon RNA binding and , in general , the interface is more loosely packed ( Figure 2—figure supplement 1 ) . ( C ) Gln51 hydrogen bonds across the dimer interface in the RNA-bound conformation ( top ) but is unable to interact in the free state ( bottom ) . ( D ) Gel shift assays with 32P-labeled IIB40 RNA and Rev 1–70 dimer show that a monomeric Rev-RNA complex accumulates when Gln51 is mutated to Ala and the dimer affinity is reduced ∼30-fold ( Figure 4—figure supplement 1 ) . F corresponds to free RNA , M to the Rev monomer-RNA complex , and D to the Rev dimer-RNA complex . ( E ) Gel shift assays with the 234-nt RRE and full-length Rev , visualized using SYBR Green II staining , show reduced accumulation of dimer species and a modest loss of binding affinity with the Gln51Ala mutant ( Figure 2—figure supplement 1 ) , quantified in panel ( F ) . Mu corresponds to Rev multimer-RNA complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 00810 . 7554/eLife . 04120 . 009Figure 2—figure supplement 1 . Rev interactions in the free and RNA-bound structures . ( A ) Model illustrating the 70° rotation that one subunit of the unbound Rev undergoes in order to contact the RNA at the junction site , pivoting around Ile55 ( yellow ) ( B ) View of the dimer interface from the top ( RNA-binding side ) showing that the second Rev molecule in the RNA-bound dimer is situated ∼3 . 2 Å further away from the first Rev subunit than in the RNA-free arrangement ( C ) Gel-shift assay using 32P-labeled full-length RRE ( 234 nucleotides ) and full-length Rev illustrating the modest loss of binding affinity upon mutating Gln51 to Ala . Apparent Kd and n are: wild-type Rev Kd = 145 pM , n = 2 . 1; Q51A Rev Kd = 280 pM , n = 1 . 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 009 Interestingly , the bound conformation buries less surface area at the dimer interface than the free form ( 1000 Å2 vs 1500 Å2 ) and also shows looser hydrophobic packing ( the two Rev subunits are ∼3 . 2 Å further apart ) ( Figure 2—figure supplement 1 ) , suggesting that this bound dimer is energetically less favorable but is compensated by the energy of RNA binding of the second Rev subunit . The weaker interface may be further compensated by hydrogen bonding between the well-conserved Gln51 residues symmetrically located between the two subunits and observed only in the bound form ( Figure 2C ) . Thus , mutation of Gln51 to alanine results in accumulation of Rev monomer-RNA complexes and ∼30-fold reduced affinity ( Figure 2D ) . In the context of the full-length hexameric Rev-RRE complex , the effect of the Gln51Ala mutation was modest but reproducible , showing a twofold reduction in affinity in gel shift assays using a vast excess of protein over RNA ( Figure 2—figure supplement 1 ) and reduced accumulation of dimer species in assays with stoichiometric amounts of Rev and RRE ( Figure 2E , F ) . The modest effect of the mutation in the full-length context suggests that the Gln51 interaction may be less crucial during later stages of Rev-RRE assembly and the energetic contributions to cooperative binding may shift to other interactions , possibly coupled to further rearrangements of the Rev subunit interfaces . Alternatively , there may be multiple pathways towards cooperative assembly such that if one pathway is blocked , another may be used . The pliability of the Rev dimer interface contributes to RNA-binding specificity by allowing the Rev subunits to orient properly to multiple binding sites arrayed on the RRE , with binding at the primary IIB RNA site nucleating assembly . Comparison of our Rev dimer-IIB40 structure to the NMR structure of a Rev ARM peptide-IIB complex ( Battiste et al . , 1996 ) shows that both the Rev arginine-rich-helix and IIB RNA site are remarkably similar in both the structures ( Figure 3A ) , although with subtle changes to some Rev-IIB contacts ( Figure 3B , Figure 3—figure supplement 1 ) . For example , in the NMR structure , Asn40 hydrogen bonds to a G47-A73 base pair in IIB RNA , but is in a different plane and contacts G47 and G71 from adjacent base pairs in the dimer-RNA complex ( Figure 3B ) . Interestingly , the Asn40-IIB interaction observed in the crystal structure is similar to that seen in an NMR structure of a Rev-ARM peptide-aptamer complex ( Ye et al . , 1996 ) , suggesting that optimal Rev-RNA contacts can be selected from several favorable conformers . The subtle differences between the NMR and crystal structures likely reflect a lack of precision of the early NMR data , with no violations observed to the current structure . The similarity in Rev-IIB recognition between the NMR structure , which depicts the first step in Rev-RRE assembly , and the crystal structure , which captures the progression into cooperative assembly , reiterates the importance of preserving this highly sequence-specific interface to support a specific and productive assembly . 10 . 7554/eLife . 04120 . 014Figure 3 . Features of Rev-RNA recognition . ( A ) The ARM peptide-IIB NMR structure ( yellow ) ( Battiste et al . , 1996 ) and the Rev dimer-IIB40 structure are nearly superimposable ( backbone RMSD = 1 . 27 Å ) . ( B ) Contacts made by Asn40 from the first Rev subunit at the IIB RNA site are different in the Rev dimer-IIB40 ( top panel ) and ARM peptide-IIB complexes ( bottom panel ) . See also Figure 3—figure supplement 1 . ( C ) RNA structure at the second junction site shows a G-A base-pair and A42-U43 bulge that widens the major groove to accommodate the RNA-binding helix of the second Rev subunit . See also Figure 3—figure supplement 2 , 3 . ( D ) Rev- RNA recognition at the three known binding sites in the RRE , shown in the three top panels as views down the helical axis of the Rev ARM at the IIB site , junction site , and stem IA . The binding residues for stem IA are inferred from alanine mutants ( in red ) ( Daugherty et al . , 2008 ) . The two bottom panels show side views from the co–crystal structure indicating that turns 1–3 of the ARM helix contact IIB while turns 2–4 contact the junction site . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01410 . 7554/eLife . 04120 . 015Figure 3—figure supplement 1 . Schematic representation of Rev-RNA contacts at IIB and junction sites . Schematic representation of Rev-RNA contacts at ( A ) IIB and ( B ) the junction site , generated using Nucplot ( Luscombe et al . , 1997 ) . Red lines indicate hydrogen bonds to the phosphate backbone or sugar , green lines indicate hydrogen bonds to the bases and blue lines indicate Van der Waals contacts . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01510 . 7554/eLife . 04120 . 016Figure 3—figure supplement 2 . Rev-RNA recognition at the junction site . ( A ) Model of IIB40 with a deleted AU bulge from the junction site ( orange ) superimposed with IIB40 showing that deleting the bulge would result in a steric clash between the RNA in A-form geometry and the second Rev subunit . ( B ) Gel-shift assay using 32P-labled IIB40 or bulge-deleted RNA and Rev showing an absence of the Rev-dimer-RNA complex with the bulge deleted . ( C ) Simulated annealing Fo − Fc omit map in green ( contoured at 3 . 0 σ ) around the water molecule bridging Arg44 from the second Rev subunit and the RNA junction site . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01610 . 7554/eLife . 04120 . 017Figure 3—figure supplement 3 . Model for RNA-directed cooperative assembly . ( A ) An A-form helix was placed near the junction site to represent the stem IIA helix not present in the crystallized RNA . The second Rev subunit from the crystal structure ( green ) is oriented to use its higher-order oligomerization surface to recruit a third Rev subunit ( yellow ) ( B ) The model generated in ( A ) was placed within the RRE SAXS envelope ( Fang et al . , 2013 ) , based on the positioning of sites from the SAXS model . ( C ) RRE secondary structure from Figure 1B placed within the envelope , indicating how other binding sites might be placed , including the IA site . Figures were generated using UCSF Chimera ( Pettersen et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 017 The junction site in the RNA forms a structural scaffold that helps orient the Rev subunits . It is formed by an A44-G76 base-pair , identified earlier by in vitro selection studies ( Bartel et al . , 1991 ) , and a A42-U43 bulge that widens an otherwise narrow A-form major groove to accommodate the α-helix of the second Rev subunit ( Figure 3C , Figure 3—figure supplement 2 ) . Both A42 and U43 are unpaired and A42 is co-axially stacked on stem IIB , consistent with biochemical mapping experiments showing increased accessibility of these nucleotides upon Rev binding ( Charpentier et al . , 1997; Bai et al . , 2014 ) . Neither base is directly contacted by Rev , suggesting that their primary role is to create the junction site architecture . Consistently , deleting the bulge results in exclusively monomeric binding ( Figure 3—figure supplement 2 ) , and previous studies showed that simply introducing bulges with variable sequences juxtaposed to IIB is sufficient to mediate co-operative Rev binding ( Zemmel et al . , 1996 ) . Most contacts at the junction site are to the phosphate backbone with the exception of Arg 43 and Arg44 ( Figure 3—figure supplement 1 ) . Notably , Arg44 from both Rev subunits contacts the two binding sites on the RNA , in one case using water-mediated interactions not previously observed for RRE recognition ( Figure 3—figure supplement 1 , 2 ) . The junction site in the full-length RRE is formed by three stems ( IIA , IIB , and IIC ) , with only IIB and IIC present in the crystallized IIB40 RNA . Stem IIA likely forms an A-form helix that merges into the junction to determine the relative placement of the remaining RRE . The second Rev subunit is positioned in the crystal structure to use its higher order oligomerization surface to recruit a third Rev subunit with its ARM oriented towards stem IIA for sequential Rev assembly . Presumably , the overall ‘A’-shaped RRE topology ( Fang et al . , 2013 ) and arrangement of binding sites dictates the positioning of newly recruited subunits ( Figure 3—figure supplement 3 ) . The specific model for Rev-RRE assembly proposed by Fang et al . ( 2013 ) portrays an initial Rev dimer bridging the RRE IIB and IA sites , but our structure and other biochemical studies of Rev-RRE assembly ( Charpentier et al . , 1997; Bai et al . , 2014 ) suggest that the first Rev dimer binds to stem II before additional subunits are added . Rev dimers formed later during assembly are likely positioned to bridge the two arms of the ‘A’-shaped RRE , presumably involving the IA site . The diversity of RNA recognition by the Rev ARM is quite striking , now with examples of how Rev binds to three sites in the RRE , and yet other binding modes observed with in vitro selected nucleic acids ( Xu and Ellington , 1996; Ye et al . , 1999; Bayer et al . , 2005 ) . The current crystal structure shows that each of the two ARMs of the Rev dimer deeply insert into a distorted RNA major groove , each burying ∼1500 Å2 of surface area , and that the same face of the ARM is used to contact the RNA in both cases ( Figure 3D ) . However their helical registers differ and recognition of IIB is highly sequence-specific compared to the junction site . Furthermore , at another characterized Rev-binding site in the RRE , stem IA , yet a different binding mode seems to be utilized , with mutational studies indicating a different face of the ARM used to bind the RNA ( Figure 3D ) ( Daugherty et al . , 2008 ) . The Rev-RNA interface at the junction site illustrates that protein contacts at individual RRE sites can be largely sequence-nonspecific if the ARM is optimally presented to the RNA , especially from the context of additional bound Rev subunits . While binding to some sites such as stem IIB or IA can show substantial specificity even in their isolated contexts , this does not seem to be the case for the remaining Rev binding sites ( Daugherty et al . , 2008 ) . Binding at these sites might require a more extensive RRE framework and neighboring Rev subunits and may differ from the configuration seen at the junction site , potentially adding further diversity to the recognition modes utilized by the ARM . To test the functional implications of the observed interactions in the crystal structure , we monitored the effects of key mutations in Rev on RRE binding , viral RNA export , and viral replication . A previous study identified Asn40 as the most critical residue for IIB recognition , followed by Arg44 , Arg38 and Arg39 ( Tan et al . , 1993 ) . In the dimer-RNA complex however , alanine substitution of Arg44 , which contacts both RNA sites , was most deleterious ( 250-fold reduced affinity ) followed by Asn40 ( 100-fold ) , Arg39 ( 30–50-fold ) , and Arg38 ( 10-fold ) ( Figure 4A , Figure 4—figure supplement 1 ) . These mutations , as well as mutation of Gln51 which hydrogen bonds across the dimer interface , were all defective in viral RNA export and translation in Rev-RRE-dependent Gag-Pol reporter assays ( Figure 4B ) . When engineered into HIV-1 , all mutations , except Gln51 , also showed attenuated viral replication in tissue culture ( Figure 4C , Figure 4—figure supplement 2 ) . As noted above , Gln51 is important at the level of dimer formation but may be less critical in the full Rev-RRE context . RNA fluorescence in situ hybridization ( FISH ) of infected cells indicated that these Rev mutations caused retention of unspliced HIV-1 RNAs in the nucleus ( Figure 4D ) , correlating with reduced Gag protein production ( Figure 4—figure supplement 3 ) . Thus , key interactions observed in the Rev-RNA crystal structure have important functional roles in HIV RNA export and viral replication . 10 . 7554/eLife . 04120 . 010Figure 4 . Role of RNA-binding and dimer interface residues in Rev function . ( A ) Representative binding curves with Rev 1–70 dimer mutations at positions observed to contact the RNA , calculated from gel shift assays using 32P-labeled IIB40 . Apparent dissociation constants , Kd and Hill coefficient ( n ) are reported in Figure 4—figure supplement 1 . ( B ) p24 production from a Rev-dependent , transiently transfected pCMV GagPol-RRE reporter assay complemented in trans with full length Rev mutants , quantified by ELISA and normalized to both Rev and GAPDH expression levels ( representative western blots below show that the mutations do not affect Rev steady state expression levels or stability ) . Data points are mean ± s . d . of biological triplicates . ( C ) Viral replication kinetics for HIV-1NL4-3 containing Rev mutants , monitored by p24 release into the culture supernatant . Data points are mean ± s . d . of biological triplicates . See also Figure 4—figure supplement 2 ( D ) FISH of total HIV RNA or unspliced Gag RNA in the presence of the indicated Rev mutation . Nuclei were stained using DAPI ( blue ) . Nef::Rev3xF denotes infection with virus containing c-terminally 3× flag-tagged Rev in the nef locus . See also Figure 4—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01010 . 7554/eLife . 04120 . 011Figure 4—figure supplement 1 . Gel-shift assays with the indicated Rev mutants . ( A ) Representative gel-shift assays using 32P-labeled IIB40 and the indicated Rev mutants ( B ) Apparent binding constants ( Kd ) and Hill coefficients ( n ) calculated from fraction of total RNA bound from gel-shift assays using 32P-labeled IIB40 and the indicated Rev mutants . Values reported are mean ± s . d . from two replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01110 . 7554/eLife . 04120 . 012Figure 4—figure supplement 2 . Design of virus constructs . The endogenous rev locus was deleted by mutating the start codon ( ATG ) to ACG and introducing a termination codon at residue 23 . Both mutations are silent in the overlapping tat open reading frame . The nef locus , dispensable for virus replication ex vivo , was disrupted by introducing SacII and Xba1 restriction endonuclease sites in place of the start ATG codon . Codon-optimized Rev DNA was synthesized ( to avoid the introduction of direct nucleotide repeats within the HIV genome that typically results in recombination ) with the endogenous Kozak sequence and flanking SacII and XbaI restriction sites to facilitate cloning within the repurposed nef locus . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 01210 . 7554/eLife . 04120 . 013Figure 4—figure supplement 3 . RNA FISH and protein immunofluorescence studies . ( A ) FISH of total HIV RNA or unspliced Gag RNA showing that there is no observable phenotypic difference between rev in the endogenous locus and rev engineered into the nef locus . ( B ) Immunofluorescence staining for 3× flag-tagged Rev and HIV p24 capsid protein , showing that although Rev is expressed and localized in the nuclei , p24 production is attenuated to different degrees in the presence of Rev mutants . Nuclei were stained using DAPI ( blue ) . Nef::Rev3xF denotes infection with virus containing c-terminally 3× flag-tagged Rev in the nef locus . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 013 Previous crystal structures of Rev dimers and the NMR structure of a Rev ARM-IIB complex defined key building blocks of Rev-RRE assembly , but it has been unclear how they are employed to guide cooperative RNA binding of the oligomer to assemble a fully functional complex . Our structure of the Rev-RRE complex illustrates a direct physical coupling between RNA binding and protein oligomerization , where RNA binding drastically alters the conformational state of the Rev dimer . The architecture of the RRE and positioning of individual Rev-binding sites configures the predominantly hydrophobic protein–protein interfaces to form a cooperative complex . This is reminiscent of the DNA-induced hetero-dimerization of nuclear receptors , where DNA sequence and spacing between protein-binding half-sites configures the dimer interface and dictates the choice of dimerization partners ( Meijsing et al . , 2009; Rastinejad et al . , 2013 ) . The use of non-polar residues at protein–protein interfaces has been postulated to be an attractive choice for evolutionary change as it can readily accommodate structural perturbations ( Lesk and Chothia , 1980 ) . Structural studies of the HIV-1 capsid protein clearly illustrate the role of plasticity at interfaces to assemble complex structures such as fullerene cones ( Byeon et al . , 2009; Pornillos et al . , 2009 , 2011 ) . The simple modular structure of Rev , including its pliable protein–protein interface and diverse RNA-binding domain , highlights how a virus with limited coding capacity can build a large , asymmetric RNP using a small , homo-oligomeric protein to achieve remarkable structural and functional complexity . Our structure expands the repertoire of known Rev dimer conformers to three , each with different crossing angles between the RNA-binding helices . Given the hydrophobic nature of the interfaces , other conformers are possible , especially when placed in the context of host–protein complexes . These conformers can be combined in multiple arrangements to generate oligomers with diverse architectures ( Figure 5A ) . Since Rev likely interacts with other host proteins to transport and package viral RNAs during the virus life cycle ( Jager et al . , 2012; Naji et al . , 2012 ) , different quaternary structures of Rev may generate conformational states adapted to different host complexes at various stages of viral replication . This is highly reminiscent of the Ebola virus protein VP40 , whose dimeric , hexameric , and octameric forms each have different roles in the viral life cycle ( Bornholdt et al . , 2013 ) , again highlighting the value of adaptable homo-oligomers in small RNA viruses . Recent crystal structures of HIV Tat ( Tahirov et al . , 2010 ) and Vif ( Guo et al . , 2014 ) with host protein complexes illustrate the importance of scaffolding structures to enable viral proteins to adopt ordered , functional states . It is remarkable that for Rev , it is not the folding of a single polypeptide chain that is dictated by its scaffolding partner ( s ) but rather its oligomeric architecture . 10 . 7554/eLife . 04120 . 018Figure 5 . Potential diversity of Rev oligomeric structures and functional implications . Three types of Rev–Rev interactions observed by crystallography , Rev dimers in the RNA-free ( Daugherty et al . , 2010b ) or RNA-bound states ( the current structure ) and a Rev dimer using the higher-order oligomer interface ( DiMattia et al . , 2010 ) are shown within the circle with Rev in grey and the Rev-ARMs in blue . ( A ) The higher-order oligomer interface was used to combine Rev dimers in the RNA-free state or RNA-bound state in various arrangements to generate examples of hexamers with different architectures . ( B ) Models illustrating how changes to the RRE structure ( red ) can alter the architecture of Rev oligomer . Such changes can alter the Rev-RRE ‘jellyfish’ architecture and spatial distribution of NESs , potentially changing their local effective concentration and avidity for the Crm1 dimer ( in grey , with RanGTP in light brown and NES-binding sites in yellow ) , thereby tuning nuclear export activity . Coordinates of the Crm1-RanGTP dimer are from Booth et al . , 2014 . DOI: http://dx . doi . org/10 . 7554/eLife . 04120 . 018 The function of Rev in RNA export is well defined , and the arrangement of Crm1 export complexes provides one clear example in which RNA organizes the Rev oligomer into a functionally important state . The ‘jellyfish’ model depicts Rev hexamers to be associated with the RNA on one side of the protein and NESs that recruit Crm1 export receptors on the other side ( Daugherty et al . , 2010b ) . Recent structural studies show that the Rev-RRE complex binds a Crm1 dimer utilizing two NES-binding pockets positioned for Rev recognition ( Booth et al . , 2014 ) . This study also shows that the RRE plays an important role in assembly of the export complex by enhancing the affinity of the Rev oligomer for Crm1 , probably by using the RNA scaffold to organize the Rev subunits and increase the local concentration of NES peptides for Crm1 binding . Additionally , SAXS studies of the RRE ( Fang et al . , 2013 ) and SHAPE-seq analysis of Rev-RRE assembly ( Bai et al . , 2014 ) reveal that the RNA is compact and pre-ordered to bind Rev . Consequently , it may be inferred that changes to the RRE scaffold can affect the overall arrangement of Rev subunits and redistribute the spatial availability of NESs for Crm1 recruitment and nuclear export ( Figure 5B ) . Multiple studies support this hypothesis: ( 1 ) Resistance to a dominant negative mutant of Rev with a defective NES arose through two mutations in the RRE that changed the RNA structure but not Rev multimerization ( Legiewicz et al . , 2008 ) , consistent with an RNA-guided reorganization of Rev oligomers . ( 2 ) Conversely , Rev complexes formed with simple repeats of the IIB hairpin displayed high RNA-binding affinity but did not recapitulate full export activity ( Symensma et al . , 1999 ) , indicating that the specific RRE architecture properly arranges the Rev oligomer for function . ( 3 ) A study of cognate Rev-RRE pairs from HIV-infected patients found that mutations in the RRE were responsible for changes in activity as the pairs evolved during the course of infection ( Sloan et al . , 2013 ) , suggesting that small changes in RRE sequence can alter the architecture of the Rev-RRE complex to tune its functional output . Previous studies likened the RRE to a molecular rheostat , where it is sensitive to the intracellular Rev concentrations and export activity adjusts accordingly ( Mann et al . , 1994 ) . It appears now that RRE can also perform as a structural/evolutionary rheostat , where changes to RRE structure during the course of infection can rearrange the Rev oligomer and tune the levels of nuclear export . It will be interesting to determine if interactions with other host proteins exploit the Rev-RRE oligomer in other ways . We used a Rev 1–70 dimer construct harboring the higher-order oligomerization-disrupting mutations , Leu12Ser and Leu60Arg ( Daugherty et al . , 2010b ) , and a Glu47Ala surface-entropy reducing mutant ( Goldschmidt et al . , 2007 ) for initial crystallization screens . We note that position 47 in most HIV-1 Rev isolates already is alanine , unlike the Glu47 found in HIV strain HXB3 . Rev protein was expressed in Escherichia coli BL21 ( DE3 ) with an N-terminal hexa-histidine tag and GB1 solubility-enhancing domain and purified using Ni affinity chromatography essentially as described ( Daugherty et al . , 2010a ) . However , instead of using RNAses to eliminate contaminating endogenous E . coli RNA that co-purifies with Rev , we added NaCl to 2 M and urea to 1 M to the cleared lysate ( Marenchino et al . , 2009 ) and incubated for 1–2 hr at 4°C while binding in batch to the Ni-NTA resin . Washes and elutions were performed as described ( Daugherty et al . , 2010a ) . Fractions of pure protein were pooled , concentrated to ∼0 . 2 mM and purified by size-exclusion using a Superdex 75 column equilibrated in 40 mM Tris pH 8 . 0 , 0 . 2 M NaCl , 0 . 1 M Na2SO4 . To remove the His-GB1 tag , protein fractions after size-exclusion were pooled and cleaved with Tev protease after adding ammonium sulfate to 0 . 4 M . The pure , tag-free Rev obtained was concentrated to 40–50 μM and stored at 4°C prior to complex formation with RNA . RNAs were synthesized by T7 in vitro run-off transcription from synthetic DNA templates as described ( Daugherty et al . , 2008 ) . Purified and annealed RNAs were lyophilized and stored at −20°C . Lyophilized RNAs were resuspended in water at 0 . 4–0 . 5 mM . Protein and RNA were incubated together at 1 . 9:1 ratio at room temperature for 30 min , concentrated to 5–6 mg/ml using an Amicon ultrafree 4 3 kDa cut-off concentrator ( EMD Millipore , Billerica , MA ) , exchanged into crystallization buffer ( 10 mM HEPES pH 6 . 5 , 50 mM KCl , 2 mM MgCl2 ) using zeba desalting columns ( Thermo Fisher Scientific , Rockford , IL ) and concentrated to 10–12 mg/ml . Initial screens were performed using a Mosquito robot with ProComplex and Nucleix suites ( Qiagen Inc , Valencia , CA ) . After screening ∼35–40 Rev dimer-RNA complexes , one complex comprising the Rev dimer Glu47Ala mutant and an RNA derived from IIB40 , containing a UUCG tetraloop , formed cubic crystals in several conditions with PEG 400 , PEG 4000 and PEG 8000 . Optimized crystals were grown using hanging drop vapor diffusion by mixing the complex at 1:1 with reservoir solution containing 50 mM MES 6 . 0 , 50 mM KCl 1–4% PEG 4000 ( for native diffraction data collection ) or 50 mM sodium cacodylate pH 6 . 0 , 5 mM MgCl2 , 8–14% PEG 400 ( for heavy atom soaking ) . Crystals appeared in 1–2 days and grew to their full size of 100–200 microns in 3–4 days . After 1 week , native crystals were transferred into a drop containing well solution supplemented with 5% PEG 4000 . PEG 400 and PEG 4000 were added directly to the drop in small increments to a final concentration of 25% PEG 400 and 14% PEG 4000 after which the crystals were flash frozen in liquid nitrogen . Crystals grown in MES buffer failed to survive heavy atom soaks but crystals grown in 50 mM cacodylate pH 6 . 0 , 5 mM MgCl2 , 8–14% PEG 400 were more tolerant to heavy atom soaks although they did not diffract beyond 4 . 5 Å . Crystals for heavy atom soaking were transferred to a fresh drop of well solution supplemented with 5% PEG 4000 and PEG 400–25% was added to the drop in small increments . For anomalous data collection , crystals were soaked in 7 . 1 mM ammonium tetrathio tungstate ( Hampton Research , Aliso Viejo , CA , Heavy Atom Screen M2 ) for 15 min before flash freezing in liquid nitrogen . Data sets were collected at the Advanced Light Source beamline 8 . 3 . 1 at 100 K . Crystals belonged to either a tetragonal ( P41212 ) spacegroup or a cubic spacegroup ( P4132 ) . Diffraction data was processed and scaled using XDS ( Kabsch , 2010 ) . Attempts with molecular replacement using the unbound Rev dimer structure or models of Rev-IIB RNA complexes as starting models with native diffraction datasets ( collected at 1 . 116 Å ) were unsuccessful . Using single wavelength anomalous dispersion ( SAD ) data to 5 . 5 Å on a W-soaked crystal ( collected at 1 . 2146 Å ) , tungstate ions were located using ShelxD , phased and density-modified using ShelxE ( Sheldrick , 2010 ) . The crystal belonged to P41212 space group with three copies of the Rev dimer-RNA complex and one W site per asymmetric unit . Density for the Rev helices and RNA was very clear and Rev molecules from the Rev-dimer structure ( Daugherty et al . , 2010b ) and an RNA fragment from IIB RNA NMR structure ( Battiste et al . , 1996 ) were fit into the protein and RNA densities , respectively , using rigid-body fit in Coot ( Emsley et al . , 2010 ) . The resulting model was used for molecular replacement into the 3 . 2 Å native data in the cubic space group ( P4132 ) in Phaser ( McCoy et al . , 2007 ) . Initial solutions scored poorly but the appearance of positive density near the RNA fragment at the second Rev-binding site suggested that the solution was correct . Following iterative rounds of model building of the RNA using RCrane ( Keating and Pyle , 2010 ) in Coot and refinement in Phenix ( Adams et al . , 2010 ) , a second round of molecular replacement was carried out using the newly built RNA alone as the search model . This time , the solution was clear with positive density for both Rev subunits . Rev subunits were then built into the density with iterative model building and refinement . The final model had good stereochemistry and showed no Ramchandran outliers ( Chen et al . , 2010 ) . While density for most water molecules or ions was not apparent at this resolution , we could place one water molecule bridging a Rev-RNA contact . Solvent-accessible surface areas were calculated using PISA ( Krissinel and Henrick , 2007 ) and figures were generated using PyMOL ( Schrodinger , 2010 ) . Electrophoretic mobility shift assays were performed in gel shift buffer ( 10 mM HEPES pH 7 . 5 , 0 . 1 M KCl , 1 mM MgCl2 , 0 . 5 mM EDTA , 2 mM DTT , 10% glycerol , 50 μg/ml yeast tRNA and 0 . 2 mg/ml BSA ) . Rev protein stock for RNA binding was prepared in gel shift buffer supplemented with fivefold molar excess of yeast tRNA . Rev was diluted in gel shift buffer and mixed with an equal volume of <25 pM 32P-RNA . Reactions were incubated at room temperature for 15 min and loaded onto continuously running 6% ( for 234 nucleotide RRE ) or 10% polyacrylamide gels ( 0 . 5× TBE ) . Gels were run at room temperature for 60–90 min , dried and exposed to a phosphorimaging screen for >12 hr . Bands were quantified using ImageJ ( Schneider et al . , 2012 ) and data were fit using MS Excel . For gel shift assays using RRE and detection with SYBR Green II staining ( Invitrogen ) , gel shift buffer contained 10 mM HEPES pH 7 . 5 , 0 . 3 M KCl , 1 mM MgCl2 , 0 . 5 mM EDTA , 2 mM DTT , 10% glycerol and 0 . 2 mg/ml BSA ( Fang et al . , 2013 ) . Rev was serially diluted and mixed with an equal volume of RRE at 125 nM , incubated for 15 min and loaded onto continuously running 6% polyacrylamide gels ( 0 . 5× TBE ) . Gels were run at room temperature for 1–2 hr , stained with SYBR Green II and visualized under UV light . Bands were quantified using ImageJ ( Schneider et al . , 2012 ) . The pCMV-GagPol-RRE reporter construct ( Srinivasakumar et al . , 1997 ) was modified to contain the HIV-1 SF2 RRE sequence ( used in biochemical assays ) and was complemented in trans with pcDNA4TO-Rev-3xFlag ( Invitrogen mammalian expression vector ) in 293T cells by transient transfection . The optimal ratio of reporter to Rev expression plasmids was first determined to establish maximal signal under non-saturating conditions for the assay . Subsequently , 50 , 000 HEK 293T cells , each in 96-well microtiter plate wells , were transfected with 90 ng of reporter plasmid , 0 . 3 ng of Rev expression plasmid , 5 ng of an mCherry fluorochrome expression vector ( to visualize transfection efficiency ) and 4 . 7 ng of pBS carrier DNA . 48 hr after transfection , the culture supernatant was removed , the cells were lysed and intracellular p24 was quantified by ELISA . mCherry expression was consistent across all transfections at the time of cell lysis . The expression of Rev-3xFlag and GAPDH was monitored by western blot analyses using mouse α-Flag ( Sigma-Aldrich Corp . St . Louis , MO ) and mouse α-GAPDH ( Abcam , Cambridge , MA ) antibodies , respectively , followed by goat α-mouse IgG-HRP ( Santa Cruz Biotechnology , Dallas , TX ) . HRP-bound antigens were then detected by chemi-luminescence and imaged using a BioRad ChemiDoc MP imager ( Bio-Rad Laboratories , Inc . Hercules , CA ) . Expression levels for Rev mutants were within twofold of wild-type Rev , when normalized to GAPDH loading control , and were used to normalize the quantified p24 levels ( Figure 4B ) . HeLa cells ( 150 , 000 ) cultured on 12 mm poly-L-lysine coated glass coverslips were acutely infected with 100 ng p24 of HIV pseudotyped with VsV-G envelope glycoprotein and containing 3x-flag epitope-tagged Rev in the nef locus in DMEM media containing 8 μg/ml polybrene . These inoculation conditions equate to an effective MOI of 1 . 0–0 . 5 as determined previously via TCID50 determination in HeLa cells . 24 hr post inoculation , the cells were washed twice with PBS , fixed in 4% paraformaldehyde/PBS for 10 min at RT , washed in 0 . 15 M glycine/PBS , permeabilized with 0 . 5% Triton X-100/PBS , and then blocked with 1% bovine serum albumin ( BSA ) /PBS/0 . 1% Tween-20 at 37°C for 1 hr . Antigens were then probed with a 1:250 dilution ( vol/vol ) of rabbit polyclonal α-HIV p24 ( NIH AIDS Reagent Progam ) and a 1:500 dilution of mouse α-flag ( Sigma ) in 1% BSA/PBS/0 . 1% Tween-20 at 37°C for 1 hr . The cells were washed twice with PBS/0 . 1% Tween-20 and then probed with a 1:250 dilution of goat α-rabbit IgG-rhodamine ( MP Biomedicals , LLC , Santa Ana , CA ) , a 1:500 dilution of goat α-mouse IgG-FITC ( MP Biomedicals ) and 100 ng/ml DAPI nuclear stain . The cells were washed three times with PBS/0 . 1% Tween-20 and the coverslips were mounted to glass slides using Vectashield hardset mounting media for fluorescence . Cells were imaged using a 63× , 1 . 3 NA Leica objective and a Hamamatsu C4742-12 cooled CCD camera on a Leica DMI6000B stand . 25 z-stack images were taken using exposures normalized to the brightest sample , background-subtracted , and then merged to a single image using the brightest point projection method in the Volocity imaging software suite ( Perkin Elmer , Waltham , MA ) . For RNA subcellular localization by FISH , acutely infected HeLa cells were fixed with methanol for 10 min at RT and rehydrated in 2× SCC/10% formamide at 37°C for 10 min . They were then probed using Stellaris probe sets ( Biosearch ) targeting HIV-1 gag nt790–2292 conjugated to Quasar570 dye and HIV-1 nef nt8787–9407 conjugated to Quasar670 dye at 125 nM in 2× SCC/10% formamide/100 mg/ml dextran sulfate for 2 hr at 37°C . The total HIV-1 RNA probe set included 39 unique DNA oligonucleotides and the HIV-1 gag RNA probe set included 48 unique DNA oligonucleotides each computationally predicted to bind specifically to their target sequence ( Biosearch Stellaris FISH Probe Designer web-based program ) . Following incubation , 2 ml of 2× SCC/10% formamide was added and incubated for 30 min at 37°C to dissociate nonspecifically bound probe . The wash was aspirated , and 1 ml of 2× SCC/10% formamide containing 100 ng DAPI nuclear stain was added and incubated at 37°C for 30 min . The cells were then washed twice with 2× SCC , coverslips mounted to glass slides , and cells were imaged with brightness and contrast adjusted using the same linear parameters for all images . Atomic coordinates and structure factors for the Rev dimer-RRE crystal structure have been deposited in the Protein Data Bank under the accession code 4PMI .	To be able to multiply , viruses have to first infect a host cell and then hijack the host's molecular machinery to make viral proteins . One stage of this process takes place in the nucleus of the host cell and involves the viral DNA being transcribed to make RNA molecules . These RNA molecules must then be exported from the nucleus to the cytoplasm , where the viral proteins are made . In the case of HIV-1 , a protein called Rev has an important role in the export process . The Rev protein , which is supplied by the virus , binds to a region on the viral RNA molecules called the Rev Response Element . The Rev protein then binds to a group of host proteins called the Crm1 export complex to send the viral RNA molecules to the cytoplasm . Jayaraman et al . now provide the first in-depth 3D structure of two Rev molecules bound to a fragment of the Rev Response Element . The Rev molecules change shape when they bind to the element , and specific regions of the element were found to be important for this . The experiments suggest that the Rev Response Element directs the positioning of the Rev proteins on itself to match the shape needed to bind to Crm1 export complex . In parallel work from the same laboratory , Booth et al . have produced a 3D structure of the whole complex . Both structures shed new light on how the HIV-1 virus is able to multiply in its host , which may aid future efforts to develop new treatments for the disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2014	RNA-directed remodeling of the HIV-1 protein Rev orchestrates assembly of the Rev–Rev response element complex
Replicate populations of natural and experimental organisms often show evidence of parallel genetic evolution , but the causes are unclear . The wrinkly spreader morph of Pseudomonas fluorescens arises repeatedly during experimental evolution . The mutational causes reside exclusively within three pathways . By eliminating these , 13 new mutational pathways were discovered with the newly arising WS types having fitnesses similar to those arising from the commonly passaged routes . Our findings show that parallel genetic evolution is strongly biased by constraints and we reveal the genetic bases . From such knowledge , and in instances where new phenotypes arise via gene activation , we suggest a set of principles: evolution proceeds firstly via pathways subject to negative regulation , then via promoter mutations and gene fusions , and finally via activation by intragenic gain-of-function mutations . These principles inform evolutionary forecasting and have relevance to interpreting the diverse array of mutations associated with clinically identical instances of disease in humans . Prediction of evolutionary change from a set of first principles , even in the most elementary of biological systems , has proven difficult ( de Visser and Krug , 2014 ) . This is due in part to the stochastic nature of mutation , but also to lack of understanding of the molecular properties of gene products and their interactions , that is , the processes underpinning development of phenotypes—including genetic and developmental constraints—which are themselves a product of the genotype-to-phenotype map ( Pigliucci , 2010 ) . The ubiquity of parallel genetic evolution , observed both in nature and in laboratory experiments ( Flowers et al . , 2009; Gerstein et al . , 2012; Meyer et al . , 2012; Zhen et al . , 2012; Herron and Doebeli , 2013; Stern , 2013 ) shows that evolution can be highly reproducible . While repeated evolution of similar traits is considered evidence of adaptive evolution , it also suggests the possibility that evolution may be governed by rules that if understood , would lead to a more predictive science ( Hansen , 2006; Bull and Molineux , 2008; Stern and Orgogozo , 2009; de Visser and Krug , 2014; Neher et al . , 2014 ) . Not all explanations for parallel evolution necessitate the existence of underlying rules . If evolution proceeds via a single route—because there is no other—then there is no reason to suppose that evolution is anything other than idiosyncratic ( Jost et al . , 2008; Zhen et al . , 2012; Vogwill et al . , 2014 ) . Should evolution proceed along a single pathway when multiple are available and yet the fitness of the phenotype from the common path is superior , then—other than the pleasure of discovery—there is no dilemma to solve . If , however , evolution proceeds along a single pathway and yet that pathway is just one of a number of possible routes to a range of phenotypes with equivalent fitness , then determining the underlying causes becomes a matter of interest . Unfortunately opportunities for discovery of such pathways are limited ( Gompel and Prud'homme , 2009; Stern , 2013 ) ( Figure 1A ) . 10 . 7554/eLife . 07074 . 003Figure 1 . Determining the causes of parallel evolution . ( A ) A selective challenge repeatedly leads to a phenotypic adaptation with the same mutational solution represented by red circles . If only one solution is ever observed ( the red circles ) , how can it be proven that alternative pathways exist ( the unrealized [shaded] mutational pathways marked with ‘ ? ’ ) ? ( B ) In an experimental system where common genetic pathways to an adaptive phenotype can be removed ( the red circle with a cross ) , alternative pathways ( blue and yellow ) can be revealed and the underlying causes of the parallel evolution determined . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 003 One way to proceed is via model populations amenable to replication . The outcome of evolution in each population—propagated under identical conditions—can then be determined . Discovery of a novel solution would indicate the existence of at least two pathways—moreover , the relative fitness of types can be determined . While in principle straightforward , progress requires the analysis of many thousands of replicate populations ( Desai , 2013 ) —or molecules ( Ellington , 1994 ) —and even if possible , failure to find an alternate route would not mean that one does not exist . An alternative approach is to take an experimental system where the most common genetic ( mutational ) pathways to a particular adaptive phenotype can be identified and then eliminated without deleterious effects on fitness of the ancestral type ( Heineman et al . , 2009 ) . Thus evolution can be re-run from different starting genotypes that differ in the spectrum of pathways available to evolution ( Figure 1B ) . If the same phenotypic solution can arise from a genotype devoid of the typically used genetic route , then it is clear that evolution has more options at its disposal than are typically realized . Precisely this approach was taken previously using experimental populations of Pseudomonas . McDonald et al . ( 2009 ) revealed the existence of three commonly used mutational routes to a single adaptive ‘wrinkly spreader’ ( WS ) phenotype , but showed that additional , less frequently utilized pathways existed ( the nature of these pathways was not determined ) . Here , beginning with the ancestral type devoid of the three known routes to WS , we propagated multiple independent populations and identified 91 new WS mutants with similar fitness to the common WS types . A combination of genetics and genome sequencing revealed ten new single mutational step routes and three additional paths requiring two or more mutations . Our data provide an explanation for why the newly discovered pathways are rarely followed , provide a set of hierarchical principles , and show how genetic constraints can bias the outcome of evolution . The Pseudomonas fluorescens SBW25 ( Silby et al . , 2009 ) , hereafter ‘SBW25’ , experimental system of adaptive radiation has been extensively studied ( Rainey and Travisano , 1998; Spiers et al . , 2002; Bantinaki et al . , 2007; McDonald et al . , 2009 ) and has several features that makes it ideal for addressing the question of bias in evolutionary pathways . Every time the ancestral SM genotype of SBW25 is placed in a nutrient-rich static microcosm , metabolism-driven depletion of oxygen imposes strong selection for mutants that colonize the oxygen replete air–liquid interface ( Figure 2A ) . The most successful of various mat-forming types ( Ferguson et al . , 2013 ) display a wrinkled morphology on agar plates ( Figure 2A ) and are known collectively as wrinkly spreaders ( WS ) . In a previous experiment , the mutational origins of 26 independent WS genotypes were unravelled ( McDonald et al . , 2009 ) . All mutations resided in one of three pathways ( Wsp , Aws , and Mws ) . Each pathway harbours a di-guanylate cyclase ( DGC ) responsible for production of cyclic-di-GMP ( c-di-GMP ) . When the cognate DGC is constitutively activated , cells over-produce an acetylated cellulose polymer ( Spiers et al . , 2002 , 2003 ) —the proximate cause of the WS phenotype ( Figure 2B ) ( McDonald et al . , 2009 ) . That evolution followed just three pathways was unexpected given that the SBW25 genome carries 39 putative DGCs ( McDonald et al . , 2009 ) . The majority of mutations were loss-of-function changes in negative regulators of the wsp , aws , and mws-encoded DGCs ( McDonald et al . , 2009 ) . Such a spectrum of mutations made sense given that loss-of-function mutations in each negative regulator resulted in constitutive activation of the DGC ( Goymer et al . , 2006; Malone et al . , 2007 ) , with ensuing downstream effects ( Figure 2B ) ( McDonald et al . , 2009 ) . No evidence of mutational hotspots—a possible cause of bias—was obtained . 10 . 7554/eLife . 07074 . 004Figure 2 . The Pseudomonas fluorescens SBW25 wrinkly spreader model . ( A ) The ancestral smooth ( SM ) strain evolves to colonize the air–liquid interface of a static microcosm . The ability to make a mat at the air–liquid interface is dependent on mutational activation of the gene products of the wss operon , which encodes the biosynthetic machinery for production of cellulose that function as an extracellular glue ( Spiers et al . , 2003 ) . These mat-forming types display a wrinkled morphology on agar plates and are referred to as wrinkly spreaders ( WS ) . Over-activation of cellulose production is caused by an increase in the second messenger c-di-GMP—the product of diguanylate cyclases ( DGCs ) ( Goymer et al . , 2006; Malone et al . , 2007 ) . ( B ) All WS mutants described in previous work have mutations in one of three loci ( wsp , aws , and mws ) , all involving DGCs under negative regulation ( McDonald et al . , 2009 ) . There are 36 additional DGCs in the genome , and the ability to remove the three commonly followed pathways makes it possible to determine whether evolution can follow alternate routes to the WS phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 004 A unique feature of the experimental Pseudomonas model is the ability to remove the common pathways from the ancestral SM type without deleterious effect on fitness ( McDonald et al . , 2009 ) . This allows a test of the hypothesis that there exist alternative evolutionary pathways to WS that are not typically followed , either because the WS types that are generated have low fitness or because properties of the genotype-to-phenotype map make alternate routes unlikely . 200 independent glass microcosms were inoculated with the smooth ( SM ) ΔwspΔawsΔmws mutant . After 6 days growth in static microcosms , dilutions were plated on KB agar and the resulting colonies screened for types exhibiting the WS morphology . WS types were found in 91 microcosms , in contrast , when the founding genotype is ancestral SM SBW25 , all microcosms harbour WS types after just 3 days of propagation ( McDonald et al . , 2009 ) . The mutational causes of WS types were sought by suppressor analysis , using a transposon mutagenesis screen to find candidate loci for targeted Sanger sequencing ( Giddens et al . , 2007 ) or by genome re-sequencing . We found single mutations in 86 of the mutants in ten different loci , all encoding a protein with a putative DGC domain , similar to the three previously known pathways . Four of the remaining WS types had double mutations and one genotype had three mutations . The mutational targets are summarized in Figure 3 and Table 1 and full details are available in Figure 3—source data 1 . 10 . 7554/eLife . 07074 . 005Figure 3 . Mutational routes to WS . Numbers in parentheses are the number of independent mutants found out of 91 . Putative functional effects of mutations are based on mutational patterns . Classification of mutational types has been colour coded: Red , extragenic negative regulators; orange , intragenic negative regulators; yellow , promoters ( activating mutations ) ; green , gene fusion/promoter capture; blue , intragenic ( activating mutations ) ; and purple , double and triple mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 00510 . 7554/eLife . 07074 . 006Figure 3—source data 1 . Full genetic data for all new WS mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 00610 . 7554/eLife . 07074 . 007Table 1 . Summary of mutations in different categoriesDOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 007CategoryGene locusTypes of mutationNumber of mutantsExtragenic negative regulatorwspF ( PFLU1224 ) deletion , insertion , stop , frameshift , amino acid substitutionprevious workawsX ( PFLU5211 ) deletion , insertion , stop , frameshift , amino acid substitutionprevious workIntragenic negative regulatormwsR ( PFLU5329 ) deletion , insertion , amino acid substitutionprevious workPFLU0085deletion , insertion , amino acid substitution43Promoter activatingPFLU0956substitutions , small deletion and insertions9PFLU5698substitutions , small insertion7PFLU1349deletion , small insertion2Promoter capture/gene fusionPFLU0183deletion8PFLU4306deletion8PFLU4308deletion1Intragenic activatingPFLU0956amino acid substitution1PFLU3448amino acid substitution2PFLU3571amino acid substitution3PFLU5960amino acid substitution2Double and triple mutantsPFLU0458amino acid substitution , stop codon , frameshift3PFLU4744amino acid substitution , frame shiftPFLU0458amino acid substitution1PFLU0621substitution ( promoter ) PFLU4414frame shift1PFLU4443stopPFLU4744amino acid substitutionMutational targets described in McDonald et al . ( 2009 ) . The most common of the previously unseen routes to WS ( 43/91 ) involved mutation in PFLU0085 , a putative DGC lacking other annotated domains . All 22 ( 7 unique ) base pair substitutions ( BPSs ) were clustered in the region 1223–1340 of the open reading frame upstream of the DGC domain . Further disruptive mutations within PFLU0085 were identified including in-frame deletions ( 20 mutants , 3–477 bp ) and a 141-bp duplication—all in the same 1223–1340 region—suggesting that a wide variety of disruptive mutations within a small window can produce WS . This implicates 1223–1340 as a negative regulator of the downstream DGC domain . Four other genes harboured intragenic mutations , but these were less common , suggesting a smaller target size less consistent with a negative regulatory role . One mutation ( R321L ) was found in PFLU0956 close to the third predicted transmembrane helix preceding the DGC domain . A similar pattern was evident at PFLU3448 where two mutations affecting the same amino acid ( A200V and A200T ) were identified . This residue resides close to the last of seven predicted transmembrane helices . At PFLU3571 , two mutations were identified—both with W13R substitutions in the first of two transmembrane helices . A third mutation in PFLU3571 was found in the C-terminal end of the HAMP domain close to the N-terminal part of the DGC domain . The relative rarity of these mutations compared to those in PFLU0085 suggests a reduced target size that is again inconsistent with a negative regulatory role . More likely , these mutations bring about changes in protein–protein interactions , changes in localization of diguanylate cyclases in the cell , or alterations in the relative orientation of domains . All such alterations could reasonably activate production of c-di-GMP ( and thus cellulose ) without increasing catalytic rate . The fourth gene—that encoding PFLU5960—contained a mutation ( D160G ) in the DGC domain itself . The affected amino acid is close to the active site ( amino acids 200–205 ) based on a Phyre2 structure prediction model ( Kelley and Sternberg , 2009 ) of the protein ( Supplementary file 1 ) . Such a mutation likely increases catalytic activity of the imperfect GSDEF site . Mutations were found in the upstream region of three DGC-encoding genes ( Figure 3 , Figure 3—source data 1 ) . The most common was associated with PFLU0956 where six BPSs and three indels in the −54 to −59 region relative to the start codon were identified . Six BPSs and one 14 bp duplication were found upstream of PFLU5698 and two indels were found upstream of PFLU1349 . In all cases , the position of the mutation indicates no impact on the ribosomal binding site . The effect of a subset of these promoter mutations on transcription is described below . For three loci , deletion mutations caused fusions between an open reading frame encoding a DGC domain and an upstream gene . Such mutations could lead to an increase in transcription by promoter capture , or may change the functional or regulatory connections of the protein resulting in activation of the associated DGC . For PFLU0183 , eight deletions were identified: each generating an in-frame fusion with a putative fatty acid desaturase ( PFLU0184 ) located upstream . Eight deletions generating in-frame fusions were also found between PFLU4306 and PFLU4305 . PFLU4306 encodes a DGC domain protein and the upstream gene encodes a putative L-lactate dehydrogenase . A third example was defined by a single mutation that fused the GGDEF domain protein PFLU4308 to PFLU4313—the latter located upstream of the DGC and encoding a hypothetical protein . This last fusion involved deletion of four intervening genes . For two of the loci WS-generating mutations arose at frequencies similar to those in promoters , suggesting that gene fusions and promoter capture play a major evolutionary role in gene and/or protein evolution . Remarkably , double mutations in the same two genes , PFLU0458 and PFLU4744 were found in three independent WS mutants . Both contain early frame shift mutations or stop codons suggesting these are loss-of-function mutations ( Figure 3 , Figure 3—source data 1 ) . PFLU0458 harbours a DGC domain but without a catalytic motif ( suggesting loss of enzymatic activity ) and an EAL domain typically involved in c-di-GMP degradation . Loss-of-function mutations suggest a role for PFLU0458 as a c-di-GMP degrading enzyme , which is also supported by data from Pseudomonas aeruginosa , where the orthologue PA5017 ( dipA , pch ) lacks DGC activity , but is proficient for phosphodiesterase activity ( Roy et al . , 2012 ) . PFLU4744 encodes the alginate biosynthesis transcriptional activator algZ/amrZ that has a previously known role in cellulose expression and biofilm formation ( Giddens et al . , 2007 ) . The remaining double mutant harboured a mutation in PFLU0458 , as above , and a second mutation upstream ( −51 ) of the DGC-encoding protein PFLU0621 . The single triple mutant harboured a mutation in PFLU4744 that was found in combination with PFLU0458 described above , but in this instance was combined with mutations in two other regulatory proteins: PFLU4414 ( cheA , chemotaxis histidine kinase ) and PFLU4443 ( adnA/fleQ , flagella activator , negative regulator of algR ) . A connection between PFLU4744/PFLU4443 and the cellulose biosynthetic ( wss ) operon is known ( Giddens et al . , 2007 ) . Representative intragenic mutations , promoter mutations , and gene fusions ( Figure 4 ) were reconstructed in the ancestral SM SBW25 background to confirm that they are the sole cause of the WS phenotype and also to rule out any dependency on the ΔwspΔawsΔmws genetic background . 10 . 7554/eLife . 07074 . 008Figure 4 . Fitness effects of WS types arising from single mutations . ( A ) Fitness in a 1:100 invasion assay against the ancestral SM genotypes . ( B ) Fitness in a 1:1 competitive assay against the high fitness LSWS ( WspF S301R ) genotype . Error bars represent SD ( n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 00810 . 7554/eLife . 07074 . 009Figure 4—figure supplement 1 . Colony morphology of WS types arising from single mutations on KB agar with and without Congo Red . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 00910 . 7554/eLife . 07074 . 010Figure 4—figure supplement 2 . Microcosms of WS types arising from single mutations after 24 hr and 72 hr static growth . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 010 To confirm that the phenotypic basis of the selective advantage is similar to the previously described WS mutants ( where activation of a DGC results in overproduction of cellulose [McDonald et al . , 2009] ) reconstructed mutants were stained with calcofluor and Congo red and their colony morphology examined . All single mutants were positive for calcofluor staining and colony morphologies with Congo red staining showed that all stained to a higher degree than the SM ancestor , but there were differences ( Figure 4—figure supplement 1 ) . All reconstructed mutants colonized the air–liquid interface of static microcosms within 24 hr ( Figure 4—figure supplement 2 ) . Mutations found in the double and triple mutants ( Figure 5 ) were reconstructed individually and then combined . For the double and triple mutants , only combinations of the individual mutations with a PFLU4744 mutation were positive for calcofluor and Congo red binding ( Figure 5—figure supplement 1 ) . All individual and combined reconstructed mutants colonized the air–liquid interface of static microcosms within 24 hr with the exception of the genotype carrying solely the PFLU4443 E398* mutation ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 07074 . 011Figure 5 . Fitness effects of WS types arising from two and three mutational events . Individual mutations were recreated in the ancestral SM genotype plus combinations of mutations ( see text ) . ( A ) Fitness in a 1:100 invasion assay against the ancestral SM genotypes . ( B ) Fitness in a 1:1 competitive assay against the high fitness LSWS ( WspF S301R ) genotype . Error bars represent SD ( n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 01110 . 7554/eLife . 07074 . 012Figure 5—figure supplement 1 . Colony morphology of WS types arising from two and three mutational events on KB agar with and without Congo Red . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 01210 . 7554/eLife . 07074 . 013Figure 5—figure supplement 2 . Microcosms of WS types arising from two and three mutational events after 24 hr and 72 hr static growth . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 013 The preferential use of the Wsp , Aws , and Mws pathways to generate WS types when starting from the ancestral SM genotype could be a consequence of the WS types arising via the newly discovered pathways having lower fitness compared to those arising from mutation in the three known pathways . To investigate if reduced fitness explained the data , two types of competitive fitness assays were performed using the reconstructed mutants . The first assay focused on the capacity of the WS types to invade , from rare , a numerically dominant population of ancestral SM types marked with GFP . This mimics the situation where a random WS mutant occurs in a population and rises to high frequency without clonal interference from other WS mutants . The second assay involved a 1:1 competitive fitness assay in which each reconstructed WS type was directly pitted against one of the most fit and common of the WS mutants previously described—the so named LSWS type ( WspF S301R ) —also marked with GFP ( McDonald et al . , 2009 ) . This assay focused on competitive performance at the air–liquid interface—a realistic situation given that several WS mutants are likely to be present in the large populations ( ≈1010 cells ) used in the evolution experiments . Thus the two assays focus on different aspects of the adaptive radiation that are not necessarily directly correlated , for example , a mutant can be superior in initial attachment , but grow slowly at the air–liquid interface . In addition , fitness in static microcosms is frequency dependent ( Rainey and Travisano , 1998 ) . All WS types harbouring single mutations invaded the SM population from rare with selection coefficients ( s ) of 0 . 28–0 . 59 per generation compared to the control SM type ( s = 0 ) ( Figure 4A ) . Three WS mutants caused by mutations in the commonly followed mutation pathways ( wspF , awsX , and mwsR ) were included to see if lower invasion fitness was responsible for the rarity of the newly discovered mutational routes . All three produced high selection coefficients . Two newly discovered WS types containing mutations in PFLU5698 and PFLU5960 were significantly less fit than the LSWS ( p < 0 . 01 , two tailed t-test ) . However , taken together , the low ( in a few instances ) invasion rate for WS generated by mutations in the newly discovered pathways cannot account for the rarity of the alternative WS types found here . In competition with the high fitness common LSWS ( wspF S301R ) mutant ( s = 0 ) , seven of the 13 reconstructed WS types had significantly lower fitness ( p < 0 . 01 , two tailed t-test ) , including all four of the PFLU0085 mutants ( Figure 4B ) . One WS type harbouring a mutation in PFLU1349 had significantly higher fitness ( s = 0 . 039 , p < 0 . 01 , two tailed t-test ) . In the invasion assay all mutants , except the PFLU0458 E45* , PFLU4443 E398* , and PFLU4443 PFLU4414 , invaded the SM population successfully ( Figure 5A ) ( two tailed t-test p > 0 . 01 vs the control ) . In competition assay , the fitness of WS types harbouring two mutations were no different to WS types arising from a single mutation and were overall indistinguishable from the LSWS control strain ( Figure 5B ) . The individual PFLU0458 I835S mutant has a significantly higher fitness in the competition assay ( s = 0 . 039 , two tailed t-test p = 1 . 8 × 10−4 ) . The triple mutants and all combination of genotypes carrying constituent mutations had lower fitness than LSWS in the competition assay , with the highest fitness recorded for the single PFLU4744 I44T mutant ( Figure 5B ) . A similar finding came from the invasion assay where this mutant also showed high competitive ability . WS types with various combinations of single mutations ranked at the lower end of the fitness spectrum with the PFLU4443 , PFLU4443/PFLU4414 , and triple mutation WS types being no different from the SM control ( two tailed t-test p > 0 . 01 , Figure 5A ) . WS types carrying putative promoter mutations ( PFLU0956 , PFLU1349 , PFLU5698 , and PFLU0621 ) were expected to activate DGCs by increasing transcription . This is a possible mechanism of activation for the gene fusions ( PFLU0183 , PFLU4306 , and PFLU4308 ) where new promoters may have been captured by the deletion event or may have resulted in transcriptional terminators being lost . To examine the effect on transcription of representative promoter and gene fusion mutants , we performed quantitative PCR . The promoter mutations all increased transcription , up to 40-fold compared to SBW25 ( Figure 6 ) : the PFLU0621 mutant showed a more modest fivefold increase . The latter result is consistent with the fact that this mutation alone does not generate the WS morphology ( Figure 5—figure supplement 1 ) . WS types carrying gene fusion mutations ( PFLU4305-4306 and PFLU4313-4308 ) also showed large increases in transcription with 23- and 219-fold increases respectively , supporting promoter capture as the mechanism of activation . The PFLU0184-0183 fusion showed a 3 . 9-fold increase in transcription , which is not significantly different ( p = 0 . 09 , two tailed t-test ) to a control PFLU0085 mutation ( Figure 6 ) . 10 . 7554/eLife . 07074 . 014Figure 6 . Relative increase of mRNA levels in WS mutants with putative promoter or gene fusion mutations compared to SM SBW25 measured by quantitative PCR . PFLU0085 is the Δ361–414 deletion mutant not expected to result in increased mRNA levels used as a control . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 014 Understanding genetic evolution requires knowledge of the factors that affect the translation of mutation into phenotypic variation . While much is known about the nature of mutation ( Drake et al . , 1998 ) , knowledge of how change in DNA sequence is translated into phenotypic variation—the raw material for natural selection—is less well understood ( Gompel and Prud'homme , 2009 ) . Of particular relevance is the complex network of functional and regulatory connectivities that define the genotype-to-phenotype map . This network of interactions constrains ( channels ) evolution; restricts the pathways it takes and—by imposing limits to phenotype space—defines the rules by which it works ( Hansen , 2006; Gompel and Prud'homme , 2009; Stern and Orgogozo , 2009 ) . The notion that evolution follows a limited subset of pathways and conforms to rules is not new ( Geoffroy Saint-Hilaire , 1818; Vavilov , 1922; Smith et al . , 1985 ) , however , its relevance has often been challenged , largely through too frequent invocation of constraints to explain negative data and through lack of experimental insight ( Brakefield and Roskam , 2006 ) . Our unique experimental system has allowed unveiling of hitherto hidden evolutionary pathways by which the WS phenotype can be achieved . For the most part , WS types arising via the new pathways do not differ in fitness relative to known WS types and therefore their rarity cannot be attributed to a selective disadvantage . Genetic constraints , however , provide plausible explanations—certain pathways have a greater capacity than others to translate mutation into phenotypic variation ( Figure 7 ) . While bias is well known to arise as a consequence of localized mutational hotspots , the spectrum of mutations underpinning different routes to WS suggests a minor influence of any such bias ( Figure 3—source data 1 [McDonald et al . , 2011] ) . Conceivably , the presence of a mutational hot spot at the exact location of an intragenic activating mutational site or promoter could have a major impact on the rate of phenotypic production from that locus , but the diversity of mutants obtained for these mutational types suggest that such biases are not dominant in this system . 10 . 7554/eLife . 07074 . 015Figure 7 . Hierarchical principles of mutational gene activation . Certain pathways have a greater capacity than others to translate mutation into phenotypic variation depending on the functional and regulatory interactions involved . The estimated number of mutants is a first approximation of how mutations are expected to be distributed between different categories assuming a similar number of target genes in each category . This distribution can also be biased by mutational hot spots and differences in fitness effects of mutations in the gene involved . DOI: http://dx . doi . org/10 . 7554/eLife . 07074 . 015 All newly revealed mutational routes to the WS phenotype harbour proteins predicted to encode DGCs , all show the distinctive WS colony morphology and niche preference , and all over-produce cellulose . Despite underlying molecular similarity only three are routinely travelled: Wsp , Aws , and Mws . The newly discovered mutational spectra allow proposals regarding probable mechanisms for activation of the pathways specific to each DGC and with this a more complete understanding of why genetic evolution prefers certain pathways over others . From a purely genetic perspective there are no surprises: loss-of-function mutations are vastly more common that gain-of-function mutations ( Gompel and Prud'homme , 2009; Lee et al . , 2012; Herron and Doebeli , 2013 ) and thus those pathways containing DGCs subject to negative regulation will , by virtue of target size ( a product of length of DNA and function ) , translate mutation into WS variation more efficiently than those pathways containing DGCs subject to other forms of regulation . But our detailed analysis makes possible a finer scale of resolution . In common with the previously known Wsp , Aws , and Mws pathways , PFLU0085 also appears—on the basis of the spectrum of DGC-activating mutations—to be subject to negative regulation . That WS arising via the PLU0085 are not detected when the founding genotype is ancestral SM is readily understood based on the small mutational target ( ∼117 nucleotides ) and the fact that mutations within this region cannot disrupt the reading frame else DGC activity be lost . The second most common set of routes to WS , with 5–40 fold fewer mutations per gene compared to PFLU0085 , are those involving promoter mutations and gene fusions ( Figure 3 ) . It is not clear why promoter mutations were only found for a minority of the 39 DGC domain-containing proteins . Possibly this reflects the need for mutations to generate a high level of transcription , but it is also consistent with the fact that regulation of many DGC-containing proteins is post-translational ( Goymer et al . , 2006; Jenal and Malone , 2006 ) . The high frequency of gene fusions is surprising given the relatively few reports of this kind of mutational event in experimental evolution studies , with the notable exception of promoter capture that was central to the evolution of an Escherichia coli mutant that gained the ability to grow on citrate ( Blount et al . , 2012 ) . The loss of genetic material through beneficial gene fusions could contribute to deletional bias over evolutionary timescales as observed in bacteria ( Mira et al . , 2001 ) and presents a selectionist alternative to reductive evolution by genetic drift or loss of biosynthetically expensive genes ( Moran , 2002; Lee and Marx , 2012 ) . Moreover , beneficial promoter capture events can create new protein domain combinations and provide raw material for the evolution of genes with novel functions , even if initially selected only because of a difference in transcriptional regulation . The third most common mutation route to WS involves rare activating mutations that differ from intragenic negative regulators in that only specific base pair substitutions can lead to activation ( Figure 3 ) . Effects at the molecular level are unknown , but it appears that mutations modifying the active site are rare , given that only one mutation was found in proximity to the catalytic site . Assuming that any of the newly discovered pathways are used at a frequency that is at least 10-fold lower than for the common pathways to WS , it is possible to estimate that only one in one thousand activating mutations would directly modify the active site ( Figure 7 ) . This strongly supports the theory that the majority of advantageous mutations are likely to occur in regulatory regions , including promoters and regulatory proteins , or in the case of intragenic mutations , primarily in regions peripheral to the main catalytic domain ( McAdams et al . , 2004; Wray , 2007 ) . WS types arising as a consequence of two mutations arose at a similar frequency as the rare activating mutations . Such double ( and triple ) mutants are unexpected given the seeming improbability of such events based on estimates of the genome mutation rate ( Drake et al . , 1998 ) . Nonetheless their detection is not uncommon ( Drake , 2007 ) and they have been explained by the existence of transient phenotypic mutators in the population due to transcriptional or translational errors in the expression of key genes involved in replication or DNA repair ( Ninio , 1991; Drake , 2007 ) . This suggests that once the third most common pathways involving single mutations have been realized , the same evolutionary principles can be applied to double mutations starting with double mutations in negative regulators , followed by double mutations in one negative regulator and one promoter mutation ( Figure 3 , Table 1 ) . Taken together our findings provide the clearest evidence yet that the network of regulatory interactions and connectivities that define the genotype-to-phenotype map directly affects the translation of mutation into phenotypic variation and that this can profoundly bias the course of genetic evolution . This has a number of implications . Firstly , our findings contribute to the growing number of studies that show that parallel phenotypic evolution is often underpinned by parallel genetic changes ( see [Stern , 2013] Table 1 for a summary ) ( Jost et al . , 2008; Blount et al . , 2012; Gerstein et al . , 2012; Meyer et al . , 2012; Zhen et al . , 2012; Herron and Doebeli , 2013 ) . However , whereas it is common to attribute such genetic parallelism to selection , our work suggests the need for caution: genetic architecture can be a significant contributory factor . It is even conceivable , within the bounds of population parameters , such as population size and mutation rate , that certain high fitness phenotypes are never realized because of genetic bias . Secondly , the WS-based experimental system has allowed discovery of pathways that evolution would rarely ever follow and suggests a hierarchical set of rules . These rules are consistent with the concept of ‘target size’ ( Hansen , 2006; Gompel and Prud'homme , 2009 ) , which can be formulated more specifically in terms of size of the gene and the likelihood that changes generate viable phenotypes . The latter depends on the opportunity for loss-of-function mutations to generate adaptive phenotypes , which , in turn , depends on the function of the gene . But it is clear that while evolution will proceed most readily via loss-of-function mutations ( where possible ) , other kinds of genes that afford a much reduced target size cannot be overlooked . As we demonstrate , there exists opportunity for promoter mutations , gene fusions , and activating mutations to also contribute to new phenotypes , but with a greatly reduced likelihood . The existence of multiple genetic routes to a particular phenotypic end point is , given growing evidence of redundancy and evolvability in regulatory systems , possibly more common than currently appreciated ( Gompel and Prud'homme , 2009; Heineman et al . , 2009; Stern , 2013 ) . If so , then the kinds of mutational patterns unraveled here may be evident elsewhere . We cautiously suggest that in addition to relevance for interpreting patterns of mutation underlying numerous studies of parallel and convergent evolution in natural and laboratory populations ( Jost et al . , 2008; Blount et al . , 2012; Gerstein et al . , 2012; Meyer et al . , 2012; Zhen et al . , 2012; Herron and Doebeli , 2013; Stern , 2013 ) , our findings may also have value in understanding the spectrum of rare and common genetic variants underlying specific human diseases ( Gibson , 2011 ) . In the context of cancer , the mutational heterogeneity of different cancer types is well documented and it is possible that the common ( and less common ) mutational trajectories ( Yates and Campbell , 2012; Lawrence et al . , 2014 ) make sense in light of constraints due to genetic architecture . We have made little of our findings in terms of the regulatory networks underpinning c-di-GMP synthesis and cellulose expression in bacteria—including mechanisms of DGC activation—but the degree of redundancy ( under one set of laboratory conditions ) and capacity for evolutionary change by recruitment of different DGC-containing proteins and pathways is remarkable . The evident flexibility suggests that evolutionary rearrangement of interacting modules likely occurs in natural populations—it also underpins recent experimental findings on the evolution of multicellular life cycles ( Hammerschmidt et al . , 2014 ) . Indeed , such facility for re-wiring may partly explain the diverse modular arrangement of both DGC domain-encoding genes and DGC-containing operons found among even closely related strains of a single species ( Jenal and Malone , 2006 ) . The hierarchical rules for the evolution of new WS phenotypes revealed through the analysis of experimental Pseudomonas populations may be sufficiently general for them to be applied to other biological settings where an adaptive challenge can be solved by mutational gene activation . Such settings include the evolution of antibiotic resistance , the evolution of virulence in pathogens and the emergence of complex traits in eukaryotic populations . However , the relevance of these principles remains to be tested . This could occur via a priori predictions of mutational routes for specific populations in given selective contexts . If robust , then the claim to have moved closer to a more predictive theory of genetic evolution will have substance . All strains used are P . fluorescens SBW25 ( Silby et al . , 2009 ) or derivatives thereof except for E . coli strains used for strain construction and transposon mutagenesis ( E . coli DH5-α λpir , E . coli SM10 λpir IS-Ω-kan/hah , E . coli pRK2013 ) . P . fluorescens strains were grown in King's medium B ( KB ) ( King et al . , 1954 ) at 28°C and E . coli strains were grown in lysogeny broth ( LB ) ( Bertani , 1951 ) at 37°C . Solid media were KB or LB with 1 . 5% agar . Antibiotic concentrations for strain construction and plasmid maintenance were gentamycin ( 10 mg/l ) , kanamycin ( 100 mg/l ) , tetracycline ( 15 mg/l ) , nitrofurantoin ( 100 mg/l ) , and cycloserine ( 1000 mg/l ) . X-gal ( 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside ) was used at a concentration of 40 mg/l in agar plates . Calcofluor ( Fluorescent brightener 28 ) was added to agar plates at a concentration of 35 mg/l and Congo red at 10 mg/l . Individual colonies of the smooth ancestor strain PBR716 ( Δwsp Δaws Δmws ) ( McDonald et al . , 2009 ) were used to inoculate 200 glass microcosms ( 6 ml KB ) that were incubated statically for 72 hr at 28°C . The microcosms were then vortexed vigorously and diluted 1000 times into new KB microcosms that were incubated for another 72 hr under identical condition before suitable dilutions were spread onto agar plates . After incubation for 48 hr , the plates were screened ( 500–2000 colonies ) for colonies with different morphology than the SM ancestor . Transposon mutagenesis was used to find candidate genes for alternative WS mutations as previously described ( Giddens et al . , 2007 ) . Briefly , the plasmid pCM639 containing the IS-Ω-kan/hah transposon was conjugated from E . coli SM10 λpir into the recipient P . fluorescens WS strain using an E . coli pRK2013 helper strain . Suitable dilutions of successful transconjugants were selected on KB plates with kamamycin for selection of the transposon and nitrofurantoin for counterselection of E . coli . Fewer than 1000 transconjugants from each independent conjugation were screened for loss of the WS colony morphology and single colonies were isolated on agar plates . The insertion sites in the genome were found by an arbitrarily primed PCR approach and Sanger sequencing ( Macrogen , South Korea ) of the products ( Manoil , 2000 ) . Candidate genes from the transposon suppressor analysis were sequenced by Sanger sequencing ( Macrogen ) in all alternative WS in an iterative fashion , eliminating the common pathways before moving on to the next round of transposon mutagenesis and sequencing . For a few mutants that consistently failed to produce suppressors , were difficult to phenotypically distinguish from the ancestor or had low conjugation efficiency; we used genome sequencing to find the mutations . Genomic DNA was prepared using the Wizard Genomic DNA purification kit ( Promega ) , sequenced by the Australian Genome Research Facility using Illumina HiSeq2000 and assembled against the reference P . fluorescens SBW25 genome using Geneious 5 . 5 . 6 ( Biomatters ) . All oligonucleotide primers used in this study are available in Supplementary file 2 . Representative intragenic mutations ( PFLU0085 L408P , V447G , ΔR361-R414 , ΔR437-A458; PFLU3448 A200T; PFLU3571 W13R; PFLU5960 D160G ) , promoter mutations ( PFLU0956 T-54G; PFLU1349 ins TC -47/48 , PFLU5698 C-73T ) , and gene fusions ( PFLU0184 M1-T328 fused to PFLU0183 A29-G335; PFLU4305 M1-Y340 fused to PFLU4306 S21-G489; PFLU4313 M1-F115 fused to PFLU4308 A189-R820 ) were reconstructed in the SBW25 background to prove that they are the cause of the WS phenotype and to demonstrate that these mutational pathways are available in the ancestral strain and not dependent on deletion of the wsp , aws and mws loci . Mutations found in the double and triple mutants ( PFLU0458 I835S , PFLU0621 -51C > T; PFLU0458 E45* , PFLU4744 S39N; PFLU4414 Y652fs , PFLU4443 E398* , PFLU4744 I44T ) were reconstructed individually and then combined . We used a two-step allelic replacement method to transfer the mutations into the ancestral background as previously described ( Rainey , 1999; Bantinaki et al . , 2007 ) . In summary , PCR ( Phusion High-Fidelity DNA polymerase , Thermo Scientific ) was used to amplify an approximately 1000-bp region surrounding each mutation , and the product was subsequently cloned into the pCR8 plasmid and sequenced . The cloned fragment from pCR8 was then moved to the pUIC3 suicide vector and mobilized into P . fluorescens SBW25 where it integrates into the chromosome by homologous recombination . After non-selective growth in KB , 10 mg/l tetracycline was added to inhibit the growth of cells that had lost the pUIC3 insert . After 2 hr 1000 mg/ml cycloserine was added to kill growing cells and enrich for cells that had lost pUIC3 with the tetracycline resistance marker and incubated for 4 hr at 28°C . Suitable dilutions were then plated on agar plates with X-gal to allow screening for loss of the lacZ gene on pUIC3 . White colonies were confirmed to be tetracycline sensitive and single colonies were isolated . The region containing the desired mutation was then sequenced in a number of colonies , with both the ancestral and wrinkly phenotype to exclude picking bias and confirm that the wrinkly phenotype was linked to the mutation . We used the wild-type SBW25 and a previously described high fitness WS mutant LSWS ( wspF A901C , S301R ) ( Bantinaki et al . , 2007; McDonald et al . , 2009 ) to construct Green Fluorescent Protein ( GFP ) expressing strains for use in competition experiments to determine the relative fitness of the alternative WS . These strains were genetically tagged in the chromosome with a mini-Tn7 transposon expressing GFP and a gentamicin resistance marker ( miniTn7 ( Gm ) PrrnB P1 gfp-a ) ( Lambertsen et al . , 2004 ) that was transferred from E . coli by conjugation together with the pUX-BF13 plasmid carrying the transposase genes . Competitive fitness was determined relative to the LSWS strain ( Spiers et al . , 2002; Goymer et al . , 2006 ) ( SBW25 wspF A901C , S301R ) marked with GFP . Strains were grown for 16 hr , shaking , in KB at 28°C before they were mixed at equal volumes diluted 6 times and grown for 4 hr at the same conditions to ensure that the strains were in the same physiological state before the competition started . The initial ratios of alternative WS to LSWS GFP were determined by counting 100 , 000 cells using flow cytometry ( BD FACS Canto ) detecting GFP fluorescence on the 488 nm laser with 530/30 bandwidth filter . Suitable dilutions of the initial population were plated on KB agar plates to determine viable counts . The mix of alternative WS and LSWS was diluted 1000-fold in KB and incubated for 24 hr , static at 28°C . Final viable counts and ratios were determined in the same way . After incubation for approximately 40 hr at 28°C , viable counts were determined and the stability of the WS phenotypes and GFP marker was determined . This was possible as most WS types have slightly different colony morphologies , which allow us to determine if the GFP marker was lost and the emergence of smooth ancestral types is easily detected . Rarely , smooth colonies were found and if they made up more than 5% of the total population , the competition data were discarded for that microcosm . The number of generations was determined by ln ( final population/initial population ) /ln ( 2 ) . Selection coefficients were calculated using the regression model s = [ln ( R ( t ) /R ( 0 ) ) ]/[t] , as previously described ( Dykhuizen , 1990 ) where R is the ratio of alternative WS mutant to LSWS GFP and t is the number of generations . Control experiments with LSWS vs LSWS GFP were performed to compensate for the fitness cost ( s = 0 . 06 ± 0 . 01 ) of the miniTn7 with the GFP marker . For each strain , the competition assay was performed in quadruplicates at a minimum of two separate occasions . Invasion fitness was measured relative to the smooth ( SM ) ancestral SBW25 mini-Tn7 GFP . Invasion strains were grown in KB for 24 hr shaking and then mixed 1:100 with the SM GFP strain and a 1000-fold dilution of this mix was used to inoculate 6 ml static microcosms . After 48 hr at 28°C , the ratio of unmarked WS to GFP marked SM was determined by flow cytometry as described above . Viable counts on KB plates of initial and final populations were performed to calculate the number of generations during the invasion growth . The stabilities of the GFP marker and colony morphologies were confirmed and data from microcosms with >5% wrinkly GFP or smooth unmarked were discarded . Selection coefficients relative to the ancestral SBW25 strain were calculated as described above with compensation for the cost of the GFP marker determined by control invasion experiments with SBW25 vs SBW25 GFP . The averages presented are the result of at least two independent experiments with quadruplicates for each strain . Total RNA was isolated from reconstructed WS mutants with mutations in upstream promoter regions or fusions to other open reading frames using the SV Total RNA Isolation system ( Promega ) . Cells were harvested at OD600 = 0 . 6 and resuspended in Tris-1mM EDTA buffer with 0 . 4 mg/ml lysozyme to lyse the cells before using the supplied protocol . The RNA was reverse transcribed into cDNA using the High Capacity cDNA reverse transcription kit ( Applied Biosystems ) and diluted 40 times before use in quantitative real-time PCR ( DyNAmo Colorflash SYBR Green qPCR kit [Thermo Scientific] , PikoReal 96 Real-Time PCR System [Thermo Scientific] ) . Relative changes in mRNA levels between the ancestral SBW25 strain and the WS mutants were determined with the ΔΔCq method using recA as an internal control ( Livak and Schmittgen , 2001 ) . We used two biological replicates per strain and three technical replicates per run over at least two separate experiments .	Different living things often develop similar strategies to adapt to the environments in which they live . Sometimes two species that share a common ancestor independently evolve the same trait by changing the exact same genes . This is called ‘parallel evolution’ , and it has led some scientists to ask: are there certain traits that can only evolve in a limited number of ways ? Or are there other ways to evolve the same trait that , for some reason , are not explored ? Experimentally , investigating these questions is challenging , but parallel evolution occurs in the laboratory as well as in the wild . Many commonly studied organisms—such as fruit flies or bacteria—can be used in relevant studies , because they can be grown in large numbers and then exposed to identical environments . However , if this method fails to find a new way that a trait can evolve , it doesn't mean that alternative mechanisms do not exist . Lind et al . used a different approach that instead relies on removing all of the known pathways that can be mutated to produce a given trait and then seeing if that trait can still evolve via mutations elsewhere . The experiments involved a bacterium called Pseudomonas fluorescens that can evolve to grow flattened and wrinkled colonies ( instead of smooth , round ones ) when it has to compete for access to oxygen . Previous experiments had shown that the evolution of the so-called ‘wrinkly spreader’ form can be caused by mutations in one of three biological pathways . But P . fluorescens can survive unharmed without these pathways , which enabled Lind et al . to ask if there might be other ways that this trait could evolve . Bacteria without these three pathways were engineered and then grown under oxygen-deprived conditions . This experiment produced 91 new mutants that each had the wrinkly spreader phenotype . Further experiments revealed that together these mutants represented 13 previously unrecognized ways that the ‘wrinkly spreader’ phenotype can evolve . The new rare mutants had similar fitness as the previously known , common ones—so this cannot explain why they hadn't been seen before . Lind et al . instead suggest a set of principles to explain why these newly discovered pathways are rarely mutated and how genetic constraints can bias the outcome of evolution . Further work could investigate whether these principles can help us to predict the course of evolution in other biological contexts , such as in the evolution of antibiotic resistance .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	Experimental evolution reveals hidden diversity in evolutionary pathways
The proteostasis network has evolved to support protein folding under normal conditions and to expand this capacity in response to proteotoxic stresses . Nevertheless , many pathogenic states are associated with protein misfolding , revealing in vivo limitations on quality control mechanisms . One contributor to these limitations is the physical characteristics of misfolded proteins , as exemplified by amyloids , which are largely resistant to clearance . However , other limitations imposed by the cellular environment are poorly understood . To identify cell-based restrictions on proteostasis capacity , we determined the mechanism by which thermal stress cures the [PSI+]/Sup35 prion . Remarkably , Sup35 amyloid is disassembled at elevated temperatures by the molecular chaperone Hsp104 . This process requires Hsp104 engagement with heat-induced non-prion aggregates in late cell-cycle stage cells , which promotes its asymmetric retention and thereby effective activity . Thus , cell division imposes a potent limitation on proteostasis capacity that can be bypassed by the spatial engagement of a quality control factor . The proper folding of proteins is essential to cellular homeostasis , and an extensive collection of protein quality control ( PQC ) pathways , known as the proteostasis network , has evolved to protect nascent and metastable proteins from misfolding and to reactivate or remove proteins that have already misfolded ( Powers et al . , 2009; Wolff et al . , 2014 ) . The PQC network is tailored to buffer protein folding in a distinct homeostatic niche but can adapt when these buffering thresholds are exceeded by elevating the expression of PQC factors , including proteases and molecular chaperones , to clear accumulating misfolded proteins ( Morimoto , 2008; Powers et al . , 2009 ) . In cases such as thermal stress , these corrections are sufficient to restore balance , but in others such as aging , misfolded proteins assemble into ordered amyloid aggregates , which persist and dramatically alter cellular physiology by inducing disease ( Tuite and Serio , 2010; Voisine et al . , 2010; Taylor and Dillin , 2011; Kim et al . , 2013 ) . This proteostasis collapse has been linked to the unique ability of amyloids to incorporate and conformationally convert like protein to the misfolded state and to their high thermodynamic stability ( Chiti and Dobson , 2006; Jahn and Radford , 2008 ) . Together , these properties are thought to enhance the production and restrict the resolution of the misfolded protein to the point that the buffering capacity and adaptability of the proteostasis network is chronically exceeded . Despite this natural upper boundary on proteostasis capacity , the heterologous overexpression of molecular chaperones in Caenorhabditis elegans , mice , Drosophila , yeast , and tissue culture-cell models of amyloidoses reduces proteotoxicity ( Chernoff et al . , 1995; Morimoto , 2008; Broadley and Hartl , 2009; Holmes et al . , 2014 ) . While these observations are often interpreted as evidence of amyloid resolution , existing protein has not been demonstrated to transition from an amyloid to a non-amyloid form in any of the studies . Instead , two correlations have been observed where the reduced proteotoxicity has been linked to a change in amyloid state . Either amyloid accumulation is enhanced by chaperone overexpression ( Douglas et al . , 2009; Cushman-Nick et al . , 2013 ) , or amyloid accumulation is reduced . In the few cases where the mechanism has been determined , the reduction in amyloid accumulation results from an inhibition of amyloid assembly by the overexpressed chaperone ( Kobayashi et al . , 2000; Schaffar et al . , 2004; Tam et al . , 2006; Shorter and Lindquist , 2008; Winkler et al . , 2012 ) . Thus , even the specific overexpression of individual chaperones is unable to extend the proteostasis upper boundary in vivo to resolve protein amyloids . Although these targeted interventions have yet to succeed , studies conducted under conditions that reduce amyloid amplification indicate that amyloid clearance may not represent an insurmountable obstacle . For example , repressing expression of an amyloidogenic protein can reverse established toxicity and , at least in some cases , clear existing amyloid ( Yamamoto et al . , 2000; Mallucci et al . , 2003; Lim et al . , 2011 ) . In addition , expression of a dominant-negative mutant also promotes disassembly of wild-type amyloid in vivo ( DiSalvo et al . , 2011 ) . Together , these observations suggest that amyloid clearance mechanisms exist in vivo , and indeed amyloid resolution is biochemically feasible in vitro using purified chaperones such as yeast Hsp104 , alone or in combination with its co-chaperones Hsp40 , Hsp70 , and small heat shock proteins ( Inoue et al . , 2004; Shorter and Lindquist , 2004 , Lo Bianco et al . , 2008; Shorter and Lindquist , 2008 ) . What limitations , then , restrict the ability of cells to expand proteostasis capacity to effectively resolve continuously expressed wild-type protein amyloids in vivo ? To identify cell-based limitations on proteostasis capacity , we focused on the mechanisms controlling persistence of the yeast prion [PSI+] , the alternative , self-templating , amyloid form of the Sup35 protein ( Cox , 1965; Patino et al . , 1996; Paushkin et al . , 1996; Glover et al . , 1997; King et al . , 1997 ) . In this study , we report that a transient thermal stress surprisingly leads to the complete disassembly of existing Sup35 amyloid . This process requires the accumulation of heat-induced non-prion protein aggregates in cells primarily at the later stages of the cell cycle . The engagement of Hsp104 with these substrates , and its inability to resolve them before cell division , leads to asymmetric retention of the chaperone in cells that experienced the thermal stress . As a result , Hsp104 accumulates to a level that is sufficient to resolve amyloid aggregates . Thus , the kinetics of substrate engagement by a PQC factor and its partitioning during cell division impose cell-based limitations on proteostasis capacity . Under normal growth conditions , [PSI+] propagates faithfully ( Cox , 1965; Derkatch et al . , 1996 ) . However , at elevated temperatures where the PQC capacity is increased , [PSI+] becomes destabilized in a Sup35 conformation-specific manner . For example , the more thermodynamically stable but less efficiently propagated [PSI+]Weak variant is quantitatively ‘cured’ ( i . e . converted to the non-prion [psi−] state ) at elevated temperature in comparison with [PSI+]Strong , whose propagation is unaltered under the same conditions ( Cox et al . , 1988; Derkatch et al . , 1996; Jung et al . , 2000; Tanaka et al . , 2006; Newnam et al . , 2011 ) . This curing of [PSI+]Weak was linked to the inhibition of the molecular chaperone Hsp104 ( Newnam et al . , 2011 ) , an observation that is seemingly counter to the idea that proteostasis capacity increases in response to stress ( Morimoto , 2011 ) . However , in this study , stationary phase cultures were only briefly diluted into fresh medium to re-establish exponential growth before exposure to elevated temperature ( Newnam et al . , 2011 ) . Because stationary phase alters chaperone expression and blocks [PSI+] curing at elevated temperature ( Gasch et al . , 2000; Newnam et al . , 2011 ) , residual effects from the growth phase switch could alter the interaction between Sup35 aggregates and PQC factors . Therefore , we revisited the effects of elevated temperature on [PSI+] propagation , beginning with exponentially growing cultures . To monitor transitions from the prion [PSI+] to the non-prion [psi−] state , we used yeast strains encoding a premature termination codon ( PTC ) in the ADE1 gene . In [PSI+] strains , Sup35 is functionally compromised , leading to stop-codon read-through and the formation of white or pink colonies on rich medium , but in [psi−] strains , termination is faithful at the PTC , leading to the formation of red colonies on rich medium ( Chernoff et al . , 1995 ) . Transiently elevating the growth temperature from 30°C to 40°C had no effect on viability ( Figure 1—figure supplement 1A ) or on [PSI+]Strong propagation ( Figure 1A ) but induced [PSI+]Weak curing ( Figure 1A , B ) . Notably , both fully red and sectored colonies were observed , indicating that curing happened during both the thermal stress and subsequent recovery ( Figure 1B ) . Thus , [PSI+]Weak propagation is similarly sensitive to elevated temperature in exponentially growing cultures and in those that have recently exited stationary phase . 10 . 7554/eLife . 04288 . 003Figure 1 . Thermal stress induces curing through resolution of Sup35 amyloid . ( A ) [PSI+]Strong ( SLL2606 ) and [PSI+]Weak ( SLL2600 ) cultures were incubated for 30 min at the indicated temperatures before plating on rich medium at 30°C to analyze curing by colony color phenotype , as described in the text . ( B ) Quantification of [PSI+]Weak ( SLL2600 ) colony color phenotypes following treatment as described in ( A ) . Colonies were scored as completely [psi−] ( black ) , or sectored ( partially [psi−] , white ) . Data represent averages; error bars represent standard deviations; n = 3 . ( C ) Semi-native lysates of [psi−] ( SLL2119 ) , [PSI+]Weak ( SLL2600 ) , and [PSI+]Strong ( SLL2606 ) cultures were analyzed by semi-denaturing detergent agarose gel electrophoresis ( SDD-AGE ) and immunoblotting for Sup35 after treatment as described in ( A ) . ( D ) Sup35 released from amyoid aggregates in a [PSI+]Weak strain ( SLL2600 ) following treatment as described in ( A ) and recovery at 30°C in the presence of cycloheximide was determined by treating lysates with 2% SDS at 53°C , followed by SDS-PAGE and quantitative immunoblotting for Sup35 . Lines represent medians; boxes represents upper and lower quartiles , and whiskers represent maximum and minimum; n = 5; p = 0 . 02 , p = 0 . 01 by paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 00310 . 7554/eLife . 04288 . 004Figure 1—figure supplement 1 . Characterization of thermal stress effects . ( A ) . Exponentially growing [PSI+]Weak cultures ( SLL2600 ) were incubated at 30°C , 37°C , 40°C , or 37°C before 40°C for 30 min and plated to YPD at 30°C to quantify colony forming units . Data represent means; error bars represent standard deviations; n ≥ 3 . ( B ) [PSI+]Weak cultures ( SLL2600 ) grown exponentially for 2 hr after dilution from a saturated overnight culture ( left ) or for at least 24 hr ( right ) were incubated at 40°C for 30 min and allowed to recover for 2 hr at 30°C . Lysates isolated from these cultures were analyzed by SDD-AGE and immunoblotting for Sup35 . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 004 At the normal growth temperature , large Sup35 aggregates are fragmented into smaller complexes by Hsp104 ( Chernoff et al . , 1995; Eaglestone et al . , 2000; Ness et al . , 2002; Satpute-Krishnan et al . , 2007; Kawai-Noma et al . , 2009 ) . In a culture that recently exited stationary phase , the size of SDS-resistant Sup35 aggregates increased , as assessed by semi-denaturing detergent agarose gel electrophoresis ( SDD-AGE ) ( Kryndushkin et al . , 2003 ) , following incubation at 40°C and a 2 hr recovery at 30°C ( Figure 1—figure supplement 1B , left ) ( Newnam et al . , 2011 ) , consistent with an inhibition of fragmentation ( Newnam et al . , 2011 ) . In contrast , SDS-resistant Sup35 aggregates were immediately reduced in size ( Figure 1C , Figure 1—figure supplement 1B ) and completely lost after recovery ( Figure 1—figure supplement 1B , right ) following identical treatment of an exponentially growing [PSI+]Weak strain , a progression suggesting the resolution of existing Sup35 aggregates . To test this possibility , we incubated a [PSI+]Weak culture at 40°C , returned the culture to 30°C in the presence of cycloheximide to repress new protein synthesis , and monitored the conversion of existing Sup35 from the amyloid [PSI+] state ( i . e . SDS-resistant ) to the non-amyloid [psi−] state ( i . e . SDS-sensitive ) ( Serio et al . , 2000; Satpute-Krishnan and Serio , 2005 ) . In a control culture at 30°C , very little pre-existing Sup35 transitioned to an SDS-sensitive state despite the inhibition of new protein synthesis ( Figure 1D ) , as expected ( DiSalvo et al . , 2011 ) . However , following incubation at 40°C , over 70% of SDS-resistant Sup35 became detergent sensitive during recovery at 30°C ( Figure 1D ) , indicating disassembly of existing Sup35 amyloid . To determine if the prion curing resulting from this disassembly was mediated by Hsp104 , we chemically inhibited this factor with guanidine HCl ( GdnHCl ) treatment or reduced its dosage by creating a heterozygous disruption in a diploid strain ( Eaglestone et al . , 1999; Jung and Masison , 2001; Grimminger et al . , 2004; Kummer et al . , 2013; Tariq et al . , 2013; Zeymer et al . , 2013 ) , and in both cases , [PSI+]Weak curing was reduced by more than 50% relative to the wild-type untreated strain ( Figure 2A , B ) . Thus , Hsp104 promotes the disassembly of existing Sup35 amyloid in a [PSI+]Weak strain following thermal stress . 10 . 7554/eLife . 04288 . 005Figure 2 . Curing is mediated by Hsp104 and depends upon propagation efficiency . ( A ) [PSI+]Weak cultures ( SLL2600 ) were incubated at 40°C for 30 min in the absence ( untreated ) or presence of guanidine HCl ( GdnHCl ) and plated on YPD to quantify prion loss by colony color phenotype . Data represent means; error bars represent standard deviations; n = 3; p = 0 . 0004 by unpaired t-test . ( B ) A WT ( HSP104/+; SY945 ) and a heterozygous disruption ( HSP104/Δ; SY591 ) [PSI+]Weak diploid strain were incubated at 40°C for 90 min and plated on YPD to quantify prion loss by colony color phenotype . Data represent means; error bars represent standard deviations; n = 3; p < 0 . 0001 by unpaired t-test . ( C ) [PSI+]Strong strains expressing an extra copy of either WT ( SY1646 ) or G58D ( SY1648 ) Sup35 were incubated at 40°C for 90 min and plated on YPD to quantify prion loss by colony color phenotype . Data represent means; error bars represent standard deviations; n = 4; p < 0 . 0001 by unpaired t-test . ( D ) A WT ( SUP35/+; SLL3071 ) and a heterozygous disruption ( SUP35/Δ; SY957 ) diploid [PSI+]Strong strain were incubated at 40°C for 90 min and plated on YPD to quantify prion loss by colony color phenotype . Data represent means; error bars represent standard deviations; n = 3; p < 0 . 0001 by unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 005 At elevated temperature , we noted that the size of SDS-resistant Sup35 aggregates is reduced in a [PSI+]Strong strain ( Figure 1C ) , although curing does not occur ( Figure 1A ) . Because [PSI+]Strong propagates more efficiently than [PSI+]Weak , the former may be protected from curing at elevated temperature if the rate of Sup35 assembly continued to outpace the rate of its disassembly , a scenario that should be reversed by reducing the efficiency of [PSI+]Strong propagation . To test this idea , we subjected [PSI+]Strong diploid strains heterozygous for either a Sup35 mutant ( G58D ) or for a Sup35 disruption , which both reduce propagation efficiency ( Derdowski et al . , 2010; DiSalvo et al . , 2011 ) , to thermal stress . At 30°C , [PSI+] propagation is stable in both of these strains ( Figure 2C , D ) ; however at 40°C , both were now efficiently cured ( e . g . ∼100% for WT/G58D , ∼60% for SUP35/Δ ) ( Figure 2C , D ) . These observations not only provide additional support for Sup35 amyloid disassembly as the mechanism of prion curing in response to thermal stress but also reveal that the inability of chaperones to resolve amyloid in vivo results from both the physical characteristics of these aggregates and cell-based limitations , which are bypassed in the distinct proteostasis niche created at elevated temperature . Elevated temperature induces protein misfolding , and the cell responds to this stress by elevating the expression of PQC factors ( Morimoto , 2011 ) . To deconvolute the contributions of each of these events to [PSI+]Weak curing , we took advantage of the fact that we could modulate the efficiency of curing with variations in temperature . For example , while exposure to 40°C induced quantitative [PSI+]Weak curing , pretreatment at 37°C prior to exposure to 40°C slightly reduced curing ( Figure 1A , B , compare proportion of fully cured colonies ) , and incubation at 37°C did not induce curing at all ( Figure 1A , B ) . This failure to destabilize [PSI+]Weak at 37°C corresponded to an increase in aggregate size ( Figure 1C ) and a decrease in Sup35 solubilization ( Figure 1D ) relative to growth at 40°C alone , indicating a temperature-dependent modulation of amyloid resolution . To determine the molecular basis of these differences in curing efficiency , we first monitored the levels of Sup35 , Hsp104 , Ssa1/2 ( Hsp70 ) , and Sis1 ( Hsp40 ) proteins , which have all been implicated in Sup35 amyloid fragmentation ( Cox , 1965; Chernoff et al . , 1995; Song et al . , 2005; Higurashi et al . , 2008; Tipton et al . , 2008; Derdowski et al . , 2010 ) . By quantitative immunoblotting , neither Sup35 ( Figure 3—figure supplement 1A ) nor chaperone levels ( Figure 3A , Figure 3—figure supplement 1B ) correlated with curing efficiency ( Figure 1A ) , indicating that [PSI+]Weak curing could not be explained by simple changes in protein expression . Indeed , the specific overexpression of Hsp104 alone from a galactose-inducible promoter to levels that parallel those achieved during thermal stress ( Figure 3A and Figure 3—figure supplement 1C ) induces ∼40% [PSI+]Weak curing ( Figure 3—figure supplement 1D , 1 . 5 gen ) in comparison with the ∼95% [PSI+]Weak curing induced by thermal stress ( Figure 1B ) ( DiSalvo et al . , 2011 , Wegrzyn et al . , 2001 ) . Moreover , Hsp104 overexpression alone leads to an increase in the size of SDS-resistant Sup35 aggregates isolated from a [PSI+]Weak strain ( Kryndushkin et al . , 2003 ) , as previously reported for [PSI+]Strong ( Figure 3—figure supplement 1E ) ( Kryndushkin et al . , 2003 ) , but this outcome is in obvious contrast to the disassembly of Sup35 amyloid that we observe upon thermal stress ( Figure 1C , D ) . Thus , thermal stress and chaperone overexpression induce distinct changes in prion propagation . 10 . 7554/eLife . 04288 . 006Figure 3 . Heat-induced aggregate accumulation but not chaperone levels correlate with temperature . ( A ) A [PSI+]Weak strain ( SLL2600 ) was incubated at 30°C , 37°C , 40°C , or 37°C before 40°C for 30 min , and lysates were prepared and analyzed by SDS-PAGE and quantitative immunoblotting for Hsp104 ( black ) , Ssa1 ( gray ) , and Sis1 ( white ) . Data represent means; error bars represent standard deviations; n ≥ 3 . ( B ) Aggregates from lysates of a [PSI+]Weak strain ( SLL2600 ) following treatment as described in ( A ) were prepared and analyzed by differential centrifugation and Bradford assay . Data represent means; error bars represent standard error; n = 6; p = 0 . 0014 , *p = 0 . 0052 by paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 00610 . 7554/eLife . 04288 . 007Figure 3—figure supplement 1 . Effects of thermal stress and Hsp104 on protein accumulation . ( A ) Lysates were isolated from [PSI+]Weak strains ( SLL2600 ) incubated at 30°C , 37°C , 40°C , or 37°C before 40°C for 30 min and analyzed by SDS-PAGE and quantitative immunoblotting for Sup35 . Data represent means; error bars represent standard deviations; n = 4 . ( B ) Lysates were isolated from [PSI+]Weak strains ( SLL2600 ) incubated at 30°C , 37°C , 40°C , or 37°C before 40°C for 30 min and analyzed by SDS-PAGE and quantitative immunoblotting for Hsp104 , Ssa1 , Sis1 or phosphoglycerate kinase ( PGK ) as a loading control ( representative blot; see Figure 3A for quantification ) . ( C ) Lysates were isolated from a [PSI+]Weak strain containing a galactose-inducible HSP104 at the endogenous locus ( SY1749 ) after galactose treatment and Hsp104 protein was quantified by SDS-PAGE and immunoblotting . Data represent means; error bars represent standard deviations; n = 3 . ( D ) Galactose-inducible HSP104 [PSI+]Weak strains ( SY1749 ) were grown in the presence of galactose for various times and plated on YPD for analysis of [PSI+] phenotype . Data represent means; error bars represent standard deviation; n = 3 . ( E ) Lysates isolated from galactose-inducible HSP104 [PSI+]Strong ( SY1748 ) or [PSI+]Weak ( SY1749 ) cultures treated as described in ( D ) were analyzed by SDD-AGE and immunoblotting for Sup35 . ( F ) Lysates were isolated from [PSI+]Strong ( black ) ( SLL2606 ) and [psi−] ( white ) ( SLL2119 ) strains that were treated as described in ( B ) , and heat-induced protein aggregates were quantified following differential centrifugation and Bradford assay . Data represent means; error bars represent standard errors; n ≥ 5 . ( G ) Lysates were isolated from a [PSI+]Weak strain ( SLL2600 ) that was incubated at 30°C or 40°C for 30 min in the absence ( untreated ) or presence of GdnHCl , and heat-induced protein aggregates were quantified following differential centrifugation and Bradford assay . Data represent means; error bars represent standard error; n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 007 We next assessed the accumulation of misfolded proteins following shifts in temperature to determine if this event correlated with [PSI+]Weak curing efficiency . By differential centrifugation , protein aggregates accumulated independent of prion status at all elevated temperatures ( Figure 3B , Figure 3—figure supplement 1F ) , but in contrast to chaperone expression ( Figure 3A , Figure 3—figure supplement 1B ) , the severity of this accumulation was impacted by growth temperature . At 37°C , protein aggregation increased by less than 10% in comparison with a culture maintained at 30°C ( Figure 3B , column 1 ) , but in cultures treated at 37°C followed by 40°C or directly at 40°C , this level rose to ∼20% or ∼40% , respectively ( Figure 3B , columns 3 and 2 ) . Thus , the accumulation of protein aggregates ( Figure 3B ) correlates directly with curing efficiency at the various temperatures ( Figure 1A ) . We noted , however , that this correlation was not observed for a [PSI+]Weak culture treated with GdnHCl during a 40°C incubation , which strongly reduced curing efficiency ( Figure 2A ) but did not reduce the accumulation of protein aggregates ( Figure 3—figure supplement 1G ) . Nevertheless , numerous studies have reported the localization of chaperones to cytoplasmic quality control foci upon exposure to proteotoxic stresses ( Aguilaniu et al . , 2003; Erjavec et al . , 2007; Kaganovich et al . , 2008; Specht et al . , 2011; Wolfe et al . , 2013 ) , and GdnHCl blocks the association of Hsp104 with at least one substrate ( Winkler et al . , 2012 ) . To determine if Hsp104 localization to heat-induced aggregates rather than their accumulation per se determined prion-curing efficiency , we replaced endogenous HSP104 with an HSP104-GFP fusion , which supports [PSI+] propagation ( Figure 4—figure supplement 1A ) . At 40°C , this strain exhibited time-dependent [PSI+]Weak curing ( Figure 4—figure supplement 1B ) and accumulated protein aggregates ( Figure 4—figure supplement 1C ) and Hsp104-GFP to wild-type levels , albeit with slightly delayed kinetics ( Figure 4—figure supplement 1D ) . At elevated temperatures , we observed an increase Hsp104-interacting proteins as assessed by co-immunocapture ( Figure 4A ) and the localization of Hsp104-GFP to cytoplasmic foci ( Figure 4B , C ) , which also contain the model substrate firefly luciferase-mCherry ( Figure 4—figure supplement 1E ) . The amount of co-immunocaptured proteins ( Figure 4A [2 . 5-fold increase at 37°C and 4 . 2-fold increase at 40°C relative to 30°C] ) and the number and intensity of Hsp104-GFP fluorescent foci ( Figure 4C ) corresponded to both the accumulation of heat-induced protein aggregates ( Figure 3B ) and the efficiency of curing ( Figure 1A ) . Notably , the Hsp104-GFP fluorescence pattern was unaltered in a non-prion [psi−] strain ( Figure 4B ) , indicating that Hsp104-GFP was engaged with non-prion substrates . Treatment of a [PSI+]Weak culture with GdnHCl during an incubation at 40°C , which strongly reduces Hsp104-GFP association with heat-induced interacting proteins ( Figure 4A [1 . 7-fold decrease relative to 40°C in the absence of GdnHCl] ) and localization to cytoplasmic foci ( Figure 4—figure supplement 1F ) , also reduces the efficiency of curing ( Figure 2A , Figure 4—figure supplement 1G ) . Thus , the specific engagement of Hsp104 with heat-induced aggregates , rather than simply their presence , correlates with curing at elevated temperature . 10 . 7554/eLife . 04288 . 008Figure 4 . Hsp104 engages heat-induced substrates upon thermal stress . ( A ) A [PSI+]Weak strain with a GFP-tagged endogenous Hsp104 ( SY2126 ) was incubated at 30°C , 37°C , 40°C , or 40°C with GdnHCl for 30 min , and immunocapture in the presence ( + ) or absence ( − ) of anti-GFP antibodies ( Ab ) was performed on native lysates . Proteins were analyzed by SDS-PAGE and general protein staining ( Flamingo , top ) , or immunoblotting for GFP ( bottom ) . ( B ) [PSI+]Weak ( SY2126 ) or [psi−] ( SY2125 ) HSP104GFP strains were incubated at 30°C , 37°C , 40°C , or 37°C before 40°C for 90 min , and the pattern of Hsp104-GFP fluorescence was examined by microscopy . Scale bar = 1 μm . ( C ) Quantification of Hsp104-GFP fluorescence pattern in [PSI+]Weak ( SY2126 ) cells , treated as described in ( B ) : no localization ( white ) ; single dot ( light gray ) ; faint aggregate ( medium gray ) ; bright aggregate ( dark gray ) ; multiple bright aggregates ( black ) ; n > 25 . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 00810 . 7554/eLife . 04288 . 009Figure 4—figure supplement 1 . Characterization of HSP104GFP strain . ( A ) [PSI+]Weak ( SLL2600 ) and [PSI+]Weak HSP104GFP ( SY2126 ) strains were grown at 30°C , plated on YPD , and incubated at 30°C for analysis of [PSI+] colony color phenotype . ( B ) [PSI+]Weak ( SLL2600 , gray ) and [PSI+]Weak HSP104GFP ( SY2126 , white ) strains were incubated at 40°C for the indicated times and plated on YPD at 30°C for analysis of prion curing by colony color phenotype . Data represent means; error bars represent standard deviations; n = 3 . ( C ) Lysates were isolated from WT ( SLL2600 ) or HSP104GFP ( SY2126 ) [PSI+]Weak strains that were incubated at 30°C or 40°C for 30 min , and heat-induced protein aggregates were quantified by differential centrifugation and Bradford assay . Data represent means; error bars represent standard error; n = 3 . ( D ) Quantitative western blotting for Hsp104 was performed on lysates from [PSI+]Weak ( SLL2600 ) and [PSI+]Weak HSP104GFP ( SY2126 ) strains after incubation at 40°C for the indicated times . Data represent means; error bars represent standard deviations; n = 3 . ( E ) Hsp104GFP and an mCherry-tagged firefly-luciferase ( FFLmCh ) reporter were visualized in a [PSI+]Weak strain ( SY2802 ) by microscopy following incubation at 30°C or after a 30-min recovery from an incubation at 40°C for 90 min ( 30°C→40°C ) . Scale bar = 1 μm . ( F ) Hsp104GFP was visualized in a [PSI+]Weak strain ( SY2126 ) by microscopy after a 90-min recovery from incubation at 40°C for 90 min in the absence ( 40°C ) or presence of GdnHCl added before ( GdnHCl→40°C ) or after ( 40°C→GdnHCl ) heat treatment . Scale bar = 1 μm . ( G ) [PSI+]Weak HSP104GFP cultures ( SY2126 ) treated as described in ( F ) were plated on YPD and incubated at 30°C for analysis of [PSI+] colony color phenotype . Data represent means; error bars represent standard deviations; n = 3; p = 0 . 0001 , *p = 0 . 0089 . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 009 How does this chaperone engagement with heat-induced aggregates lead to the resolution of Sup35 amyloid ? One possibility is that the asymmetric localization of Hsp104 , resulting from its engagement with heat-induced protein aggregates ( Erjavec et al . , 2007 ) , increases its accumulation in a subpopulation of cells beyond that which can be achieved by its transcriptional up-regulation . To test this possibility , we first monitored the partitioning of Hsp104-GFP during cell division following incubation at various temperatures using microfluidics and fluorescence microscopy . Starting with budded cells , mother cells accumulated ∼60% of Hsp104-GFP following the completion of cell division at 30°C ( Figure 5A , gray ) , which is comparable to the accumulation of untagged GFP expressed from the same promoter ( Figure 5—figure supplement 1A ) and thus likely reflects the volume differences between mother and daughter cells . This baseline asymmetry progressively increased as the temperature was increased to 37°C ( ∼65% retention ) , 37°C followed by 40°C ( ∼73% retention ) , and finally 40°C ( ∼75% retention; Figure 5A , gray ) . Notably , both Ssa1-GFP and Sis1-GFP fusions also localized to cytoplasmic , and , in the case of Sis1 , nuclear foci ( Figure 5—figure supplement 1B , C ) , but neither was asymmetrically retained following incubation at 40°C ( Figure 5—figure supplement 1D , E ) , although their levels were elevated relative to 30°C ( Table 1 ) due to their enhanced expression ( Figure 3A ) . Thus , curing efficiency ( Figure 1A , B ) correlates directly with the asymmetric retention of Hsp104 in cells at elevated temperature . 10 . 7554/eLife . 04288 . 010Figure 5 . Curing results from the asymmetric localization of Hsp104 following thermal stress . ( A ) A [PSI+]Weak HSP104GFP culture ( SY2126 ) was imaged over time in a microfluidics chamber at 30°C after a 30 min incubation at 30°C , 37°C , 40°C , or 37°C before 40°C . Fluorescence intensity in daughter and mother cells was quantified at the first cell division in cells that were budded ( gray ) or unbudded ( orange ) after thermal stress . Lines represent medians; boxes represent upper and lower quartiles , and whiskers represent maximum and minimum . All pairwise comparisons are significantly distinct , with a p < 0 . 015 , except where indicated ( N . S . ) , by unpaired t-test; n ≥ 10 . ( B ) A [PSI+]Weak HSP104GFP WT ( SY2126 , gray ) or BNI1 deletion strain ( Δbni1 ) ( SY2486 , green ) was imaged over time in a microfluidics chamber at 30°C after a 30 min incubation at 40°C . Fluorescence intensity in daughter and mother cells was quantified at the first cell division . Lines represent medians; boxes represent upper and lower quartiles; and whiskers represent maximum and minimum; n ≥ 14; p = 0 . 0075 by unpaired t-test . ( C ) [PSI+]Weak WT ( SLL2600 ) or Δbni1 strains ( SY1888 ) , treated as described in ( B ) , were plated on YPD to analyze curing by colony color phenotype . Data represent means; error bars represent standard deviations; n = 3; p < 0 . 0001 by unpaired t-test . ( D ) A [PSI+]Weak HSP104GFP strain ( SY2126 ) was imaged over time in a microfluidics chamber at 30°C after a 30 min incubation at 40°C and with GdnHCl added before or after the 40°C incubation . Fluorescence intensity in daughter and mother cells was quantified at the first cell division . Lines represent medians; boxes represent upper and lower quartiles; and whiskers represent maximum and minimum; n > 11; p = 0 . 0003 , *p = 0 . 0026 by unpaired t-test . ( E ) A [PSI+]Weak strain ( SLL2600 ) was incubated at 40°C for 30 min and plated on rich medium . Mother and daughter pairs were separated by micromanipulation and allowed to form colonies , which were then dispersed to YPD for analysis of curing by colony color phenotype . n = 15 . ( F ) A [PSI+]Weak HSP104GFP culture ( SY2126 ) was incubated at 30°C ( dotted ) or at 40°C for 30 min and allowed to recover for 30 min at 30°C ( solid ) before analysis of GFP fluorescence intensity by flow cytometry . Based on these intensities , cells were sorted into four fractions ( orange , blue , purple , red ) by FACS . ( G ) Cells collected in ( F ) were plated on YPD to analyze curing by colony color phenotype . Data represent means; error bars represent standard deviations; n = 2; p = 0 . 02 by paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 01010 . 7554/eLife . 04288 . 011Figure 5—figure supplement 1 . Characterization of chaperone asymmetric retention following thermal stress . ( A ) A [psi−] strain expressing heat-inducible untagged GFP ( SY2091 ) was imaged over time in a microfluidics chamber at 30°C after 30 min incubation at 40°C ( red ) or 30°C ( gray ) . Fluorescence intensity in daughter and mother cells was quantified at the first cell division in budded cells . Lines represent medians , boxes represent upper and lower quartiles , and whiskers represent maximum and minimum; n ≥ 11 . ( B ) A [PSI+]Weak strain expressing a GFP-tagged endogenous Ssa1 and DsRedNLS ( SY2659 ) was imaged after a 90 min incubation at 30°C , 37°C , 40°C , or 37°C before 40°C . Scale bar = 2 μm . ( C ) A [PSI+]Weak strain expressing a GFP-tagged endogenous Sis1 and DsRedNLS ( SY2485 ) was imaged after a 90-min incubation at 30°C , 37°C , 40°C , or 37°C before 40°C . Scale bar = 2 μm . ( D ) A [PSI+]Weak SSA1GFP culture ( SY2658 ) was imaged over time in a microfluidics chamber at 30°C after a 30 min incubation at 40°C ( red ) or 30°C ( gray ) . Fluorescence intensity in daughter and mother cells was quantified at the first cell division in budded cells . Lines represent medians , boxes represent upper and lower quartiles , and whiskers represent maximum and minimum; n > 15 . ( E ) A [PSI+]Weak SIS1GFP culture ( SY2447 ) was imaged over time in a microfluidics chamber at 30°C after a 30 min incubation at 40°C ( red ) or 30°C ( gray ) . Fluorescence intensity in daughter and mother cells was quantified at the first cell division in budded cells . Lines represent medians , boxes represent upper and lower quartiles , and whiskers represent maximum and minimum; n ≥ 7 . ( F ) Quantitative immunoblotting for Hsp104 was performed on lysates from WT ( SLL2600 ) or Δbni1 ( SY1888 ) [PSI+]Weak cultures treated at 30°C ( black ) and 40°C ( white ) for 30 min following SDS-PAGE . Data represent means; error bars represent standard deviations; n = 3 . ( G ) Lysates were isolated from WT ( SLL2600 ) or Δbni1 ( SY1888 ) [PSI+]Weak strains that were incubated at 30°C or 40°C for 30 min , and heat-induced protein aggregates were analyzed by differential centrifugation and Bradford assay . Data represent means; error bars represent standard error; n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 01110 . 7554/eLife . 04288 . 012Table 1 . Relative fluorescence intensity in mother cellsDOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 012Treatment ( °C ) Hsp104 ( Relative to 30°C ) Ssa1 ( Relative to 30°C ) Sis1 ( Relative to 30°C ) 30°→30°1 ± 0 . 1 ( 24 ) 1 ± 0 . 2 ( 29 ) 1 ± 0 . 1 ( 18 ) 30°→37°1 . 6 ± 0 . 2 ( 11 ) 30°→40°3 . 5 ± 0 . 6 ( 52 ) 2 . 7 ± 0 . 5 ( 18 ) 1 . 5 ± 0 . 1 ( 7 ) 37°→40°3 . 4 ± 0 . 4 ( 46 ) [PSI+]Weak HSP104GFP ( SY2126 ) , SSA1GFP ( SY2658 ) , or SIS1GFP ( SY2447 ) cultures were treated at indicated temperatures and were imaged over time at 30°C using microfluidics and fluorescence microscopy . Average fluorescence intensity in mother cells with indicated standard deviations ( ± ) , which originated from budded cells at the time of thermal stress , was measured at the first cell division . Number of cells analyzed is indicated in parentheses . p values are <0 . 001 for all comparisons to 30°C treatment . To determine if this correlation was a requirement , we next disrupted the asymmetric retention of Hsp104 and determined its effects on curing . Disruption of the formin BNI1 ( Kohno et al . , 1996 ) did not alter Hsp104 expression levels or the accumulation of protein aggregates at 30°C and 40°C relative to a wild-type strain , ( Figure 5—figure supplement 1F , G ) but , Hsp104 asymmetric retention was reduced ( Figure 5B , green ) , as expected ( Liu et al . , 2010 ) . Strikingly , curing was dramatically suppressed from ∼80% for a wild-type strain to ∼10% in the Δbni1 strain ( Figure 5C ) . Likewise , GdnHCl treatment before thermal stress , which blocked both Hsp104 engagement with heat-induced aggregates ( Figure 4—figure supplement 1F ) and curing at elevated temperature ( Figure 2A , Figure 4—figure supplement 1G ) , also reduced Hsp104-GFP asymmetric retention following exposure to 40°C ( Figure 5D ) . Thus , the asymmetric retention of Hsp104 is required for curing . Our single-cell analyses of Hsp104-GFP partitioning indicated that a relatively minor change in chaperone retention from 65% to 75% , which corresponded to a 2 . 2-fold increase in accumulation based on fluorescence intensity ( compare 37°C–40°C , Table 1 , Figure 5A ) , correlated with a quantitative switch from prion stability to curing ( Figure 1A , B ) , suggesting the existence of a biological threshold in this range . To determine directly if cells accumulating Hsp104-GFP corresponded to those cured of [PSI+]Weak , we incubated a [PSI+]Weak culture at 40°C and then isolated single unbudded cells on rich solid medium at 30°C . Following budding and cell division , mother and daughter cells were separated by micromanipulation and grown into colonies , which were then dispersed on rich solid medium to quantify prion retention . Mother cells , which experienced the elevated temperature and accumulated Hsp104 ( Figure 5A ) , were more likely to be cured than their daughters ( Figure 5E , note most data points fall below the diagonal ) , as predicted by our hypothesis . To more quantitatively correlate Hsp104-GFP accumulation with curing efficiency , we analyzed the distribution of Hsp104-GFP in a population of cells by flow cytometry . At 30°C , Hsp104-GFP fluorescence was distributed normally in the population ( Figure 5F , dotted ) . Following incubation at 40°C , Hsp104-GFP fluorescence intensity in the population increased and its distribution was heterogeneous ( Figure 5F , solid ) . When these subpopulations were separated by FACS and analyzed for colony-based phenotype , the efficiency of curing correlated directly with the accumulation of Hsp104-GFP ( Figure 5F , G ) . Together , these observations indicate that cells exposed to elevated temperature accumulate heat-induced protein aggregates , asymmetrically retain Hsp104 in a manner that is proportional to these substrates , and ultimately cure [PSI+]Weak . But , is Hsp104 enzymatic activity required for this curing , or is its asymmetric localization alone sufficient ? As noted above , when cells are treated with GdnHCl before thermal stress , Hsp104 localization to cytoplasmic foci and asymmetric retention are both reduced ( Figure 5D , Figure 4—figure supplement 1F ) . However , we reasoned the Hsp104 association with its substrates would be dynamic and modulated by its ATPase cycle . Indeed , blocking the ATPase activity of Hsp104 after thermal stress with a 90-min treatment with GdnHCl failed to reduce Hsp104-GFP localization to cytoplasmic foci ( Figure 4—figure supplement 1F ) or its asymmetric retention ( Figure 5D ) , presumably because the chaperone bound to heat-induced substrates but was unable to release them once inhibited with GdnHCl . Despite the asymmetric localization of Hsp104-GFP under these conditions , [PSI+]Weak curing was reduced by nearly 50% ( Figure 4—figure supplement 1G ) . Thus , both Hsp104 asymmetric localization and activity are required to induce [PSI+]Weak curing following thermal stress . The distribution of Hsp104-GFP in a population of [PSI+]Weak cells that had been exposed to 40°C was very complex in contrast to the normal distribution of Hsp104-GFP at 30°C ( Figure 5F ) , suggesting that subpopulations of cells were differentially retaining the chaperone . One source of heterogeneity in the population was cell-cycle stage , as our experiments used asynchronous cultures ( Figure 6A ) . To determine if cell-cycle stage at the time of thermal stress impacted Hsp104 partitioning and explained this heterogeneity , we arrested cells in G1 with α-factor or at the G2/M transition with nocodazole ( Amon , 2002 ) , exposed these cultures to 40°C incubation , and analyzed them by flow cytometry . Treatment with α-factor ( Figure 6B ) and nocodazole ( Figure 6C ) efficiently synchronized cultures at the non-budded or large-budded stages , respectively , and did not alter Hsp104 protein levels or localization relative to the asynchronous culture at 30°C ( Figure 6—figure supplement 1A , B ) . At 40°C , Hsp104-GFP protein levels increased to similar extents in the asynchronous and arrested cultures ( Figure 6—figure supplement 1A ) , and its localization to cytoplasmic foci was similar in all cases ( Figure 6—figure supplement 1B ) . By flow cytometry , the distribution of Hsp104-GFP in the α-factor arrested culture remained normal ( Figure 6D ) , but in the nocodazole-arrested culture , this distribution became bimodal ( Figure 6E ) , indicating that Hsp104-GFP asymmetry is established immediately , even before cell division . 10 . 7554/eLife . 04288 . 013Figure 6 . Efficient curing occurs in late cell-cycle staged cells following thermal stress . ( A ) Single cells from an asynchronous WT [PSI+]Weak culture ( SLL2600 ) were scored for morphology following bright-field imaging by microscopy: unbudded ( black ) , tiny bud ( dark gray ) , small bud ( gray ) , medium bud ( light gray ) , large bud ( white ) . n = 153 . ( B ) α-factor-arrested cultures were analyzed as in ( A ) over time after release . n ≥ 250 . ( C ) Nocodazole-arrested cultures were analyzed as in ( A ) over time after release . n ≥ 175 . ( D ) A [PSI+]Weak HSP104GFP strain ( SY2126 ) released from α-factor arrest was incubated at 40°C ( solid black lines ) for 30 min before analysis by flow cytometry . 100 , 000 cells were analyzed per sample . ( E ) A [PSI+]Weak HSP104GFP strain ( SY2126 ) released from nocodazole arrest was incubated at at 40°C ( black lines ) for 30 min before analysis by flow cytometry . 100 , 000 cells were analyzed per sample . ( F ) α-factor-arrested cultures ( SLL2600 ) were incubated at 40°C for 30 min immediately or 30 min after release , and curing was quantified by colony color phenotype after plating on YPD at 30°C . Data represent means; error bars represent standard deviations; n = 3; p = 0 . 0255 by unpaired t-test . ( G ) Nocodazole-arrested cultures ( SLL2600 ) were incubated at 40°C for 30 min immediately or 30 min after release , and curing was quantified colony color phenotype after plating on YPD at 30°C . Data represent means; error bars represent standard deviations; n = 3; p = 0 . 0263 by unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 01310 . 7554/eLife . 04288 . 014Figure 6—figure supplement 1 . Characterization of chaperone accumulation and engagement in arrested cultures . ( A ) Lysates were isolated from asynchronous , α-factor-arrested , or nocodazole-arrested [PSI+]Weak cultures ( SLL2600 ) following incubation at 30°C ( black ) or 40°C ( white ) for 30 min , and the levels of Hsp104 were determined by quantitative immunoblotting following SDS-PAGE . Data represent means; error bars represent standard deviations; n = 3 . ( B ) Asynchronous , α-factor-arrested , and nocodazole-arrested [PSI+]Weak HSP104GFP ( SY2126 ) cultures were treated for 90 min at 30°C or 40°C and imaged by microscopy . Scale bar = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 014 We next assessed the impact of cell-cycle stage on [PSI+]Weak curing at elevated temperature . Arrest , without exposure to elevated temperature , did not induce curing ( Figure 6F , G ) . In α-factor arrested cells , exposure to 40°C at release inefficiently cured [PSI+]Weak ( ∼15%; Figure 6F ) , but in nocodazole-arrested cells , curing was nearly quantitative ( ∼90%; Figure 6G ) consistent with the asymmetric localization of Hsp104-GFP in the latter but not the former case ( Figure 6D , E ) . These observations suggest that cells at the end of the cell cycle are more sensitive to curing at elevated temperature than those at the beginning of the cell cycle . To test this idea , we released cultures from arrest and , after 30 min of growth at 30°C , exposed them to 40°C . For the culture originally arrested with α-factor , sensitivity to curing at elevated temperature increased ( Figure 6F ) as cells progressed into the late stages of the cell cycle ( Figure 6B ) , and for the culture originally arrested with nocodazole , this sensitivity declined ( Figure 6G ) with cell-cycle progression ( Figure 6C ) . Thus , curing occurs most efficiently when cells at a late stage of the cell cycle are exposed to elevated temperature . Our earlier experiments linked curing to the asymmetric retention of Hsp104 at elevated temperature ( Figure 5 ) . To determine if cell-cycle stage impacts this asymmetry , we analyzed Hsp104-GFP distribution in mother–daughter pairs resulting from the growth and division of unbudded cells isolated from asynchronous cultures that were exposed to elevated temperatures . In comparison with budded cells , Hsp104-GFP retention was significantly reduced at all temperatures when unbudded cells were exposed to elevated temperature , but the magnitude of the effect was most severe for conditions that induced curing ( 30°C→40°C and 37°C→40°C; Figure 5A , orange ) , indicating a cell-cycle stage dependence on Hsp104-GFP retention at elevated temperature . Because cell-cycle stage did not obviously alter the engagement of Hsp104-GFP with protein aggregates accumulating at elevated temperature ( Figure 6—figure supplement 1B ) , the more efficient partitioning of Hsp104-GFP and the reduced curing in unbudded cells could reflect the resolution of heat-induced protein aggregates and thereby the release of Hsp104-GFP during the extended time before cell division in comparison with budded cells . Indeed , nearly 100% of cells contained Hsp104-GFP foci immediately after thermal stress ( Figure 7A ) but only ∼80% still contained foci when cell division re-initiated ∼150 min after incubation at 40°C ( Figure 7A ) . Thus , the relative timing of substrate release and cell division could contribute to Hsp104 asymmetric retention and thereby curing . Consistent with this idea , 60% of unbudded cells , which are inefficiently cured ( Figure 6F ) , resolved Hsp104-GFP foci prior to cell division ( Figure 7B [165–210 min] , Figure 7C ) , allowing the partitioning of the chaperone ( Figures 7B and 5A , orange ) . In budded cells , which are efficiently cured ( Figure 6G ) , only ∼8% of cells had resolved heat-induced Hsp104-GFP foci by the time the cell divided ( Figure 7B [105 min] , Figure 7C ) , leading to the asymmetric retention of Hsp104-GFP ( Figures 7B and 5A , gray ) . Together , these observations indicate that Hsp104 is retained in cells exposed to elevated temperature if it is unable to resolve its heat-induced substrates prior to cell division . Because sensitivity to curing at elevated temperature correlated with cell-cycle stage ( Figure 6F , G ) and Hsp104-GFP localization to these cytoplasmic foci ( Figure 5 ) , substrate–chaperone dynamics must create a temporal limitation on proteostasis capacity . 10 . 7554/eLife . 04288 . 015Figure 7 . Substrate–chaperone engagement must exceed time to cell division to induce curing . ( A ) The number of [PSI+]Weak HSP104GFP ( SY2126 ) cells containing fluorescent foci was quantified in cultures recovering at 30°C over time following a 90 min incubation at 40°C ( white ) . Colony forming units in these cultures were quantified by plating ( black ) . Data represent means; error bars represent standard deviations; n = 3 . ( B ) [PSI+]Weak HSP104GFP cells ( SY2126 ) treated for 30 min at 40°C and imaged over time in a microfluidics chamber are shown . Cells that were budded at the time of thermal stress are outlined in white , while unbudded cells are outlined in orange . Solid lines mark mothers , and dotted lines mark daughters . Scale bar = 1 µm . ( C ) A [PSI+]Weak HSP104GFP strain ( SY2126 ) was imaged over time in a microfluidics at 30°C after a 30 min incubation at 40°C chamber . Budded or unbudded cells were scored at the first cell division for the presence or absence of fluorescent aggregates . Data represent means; error bars represent standard deviations; n = 3; p = 0 . 0005 by unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 015 In Saccharomyces cerevisiae , expression of the molecular chaperone Hsp104 , even at its low basal level , reduces organismal fitness at the normal growth temperature; however , survival at elevated temperatures is absolutely dependent on Hsp104 , whose expression is induced to high levels by heat shock ( Sanchez et al . , 1992; Escusa-Toret et al . , 2013 ) . Thus , cell-based limitations must finely tune proteostasis capacity not only to control protein misfolding induced by stress but also to allow normal protein folding in the absence of these challenges ( Morimoto , 2008 ) . Using the yeast prion [PSI+] as a model to understand the in vivo interactions between amyloid and PQC pathways , we have uncovered one such pathway . While [PSI+]Weak is mitotically stable at the normal growth temperature ( ∼3% loss ) ( Derkatch et al . , 1996 ) , a transient sub-lethal thermal stress induces quantitative curing ( Figure 1 and Figure 8 ) through the disassembly of existing Sup35 amyloid by Hsp104 ( Figures 1 , 2 and 8 ) . Our studies indicate that the increase in Hsp104 expression at elevated temperature alone is not sufficient to induce Sup35 amyloid resolution and [PSI+]Weak curing ( Figures 1 , 3 ) . Rather , Hsp104 must engage heat-induced protein aggregates for a period that exceeds the time to the next cell division ( Figures 7 , 8 ) . As a result , Hsp104 is asymmetrically localized to the cells that experienced the thermal stress ( Figures 5 , 8 ) , and this increase in chaperone accumulation , along with its activity , promotes curing in the same cells ( Figures 5 , 8 ) . Thus , chaperone spatial engagement , substrate processing dynamics , and partitioning during cell division represent cell based limitations on proteostasis capacity . 10 . 7554/eLife . 04288 . 016Figure 8 . Model for Sup35 amyloid resolubilization and curing upon thermal stress . Upon thermal stress , cellular proteins ( green ) misfold and aggregate , leading to the induction and recruitment of Hsp104 ( barrel ) . If thermal stress occurs in unbudded cells ( 1 ) , these aggregates are resolved prior to cell division , allowing the partitioning of Hsp104 to both mother ( black ) and daughter ( gray ) cells ( left ) . If thermal stress occurs in budded cells ( 2 ) , heat-induced aggregates persist upon cell division ( 3 ) , leading to the asymmetric retention of Hsp104 in mother cells . Both heat-induced aggregates ( green ) and Sup35 amyloid ( blue corkscrews ) are resolved in cells accumulating high levels of Hsp104 , leading to curing ( red , 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 016 Metazoans lack an Hsp104 homolog ( Torrente and Shorter , 2013 ) , but disaggregase activity has also recently been linked to a multi-component system in yeast comprised of Hsp110 , Hsp70 , and Hsp40 , and this activity is conserved in the C . elegans and human homologs of these chaperones ( Shorter , 2011; Rampelt et al . , 2012; Mattoo et al . , 2013 ) . This system is largely ineffective in the disaggregation of amyloid in vitro ( Shorter , 2011 ) but can promote the slow disassembly of amyloid from fiber ends in the presence of small heat shock proteins , such as Hsp26 and Hsp42 from yeast or HspB5 from humans ( Duennwald et al . , 2012 ) . Like Hsp104 in yeast , Hsp110 localizes to foci containing misfolded protein in human cells following thermal stress ( Rampelt et al . , 2012 ) and interacts with protein amyloids in vivo ( Ishihara et al . , 2003; Wang et al . , 2009; Olzscha et al . , 2011 ) , raising the possibility that Hsp110 engagement with stress-induced substrates could also promote its activity toward amyloidogenic substrates in vivo . The spatial engagement of PQC factors , including both chaperones and components of the ubiquitin–proteasome system , is a newly appreciated consequence of their function in vivo . Numerous cytoplasmic foci arise in response to stressors including heat , aging , oxidation , and/or proteasome inhibition . These foci include aggresomes , the insoluble protein deposit ( IPOD ) , the juxtanuclear quality control compartment ( JUNQ ) , StiF-inducible foci ( StiF ) , and Q-bodies , the latter of which form under the mild thermal stress conditions employed in our studies ( Johnston et al . , 1998; Erjavec et al . , 2007; Kaganovich et al . , 2008; Liu et al . , 2010; Specht et al . , 2011; Malinovska et al . , 2012; Weisberg et al . , 2012; Escusa-Toret et al . , 2013; Wolfe et al . , 2013 ) . While the relationship of each of these foci to one another is currently unclear , they are all defined by the co-localization of misfolded and/or aggregation-prone proteins with PQC factors , some of which can be found in more than one of type of focus . The PQC factors that localize to these foci , such as Hsp104 , clearly promote survival under stress ( Sanchez et al . , 1992; Escusa-Toret et al . , 2013 ) , but whether their localization into cytoplasmic foci specifically altered proteostasis capacity had not been previously established . Our studies indicate that the engagement of Hsp104 with heat-induced misfolded protein aggregates enhances proteostasis capacity by increasing the accumulation of this factor beyond the level attainable by changes in gene expression ( Figure 5 ) and thereby permitting the disassembly of existing Sup35 amyloid ( Figures 1 , 5 ) . While our studies indicate that chaperone partitioning imposes a limitation on proteostasis capacity , other aspects of this process may be more relevant to this upper boundary in post-mitotic cells , such as neurons . Indeed , our observations reveal other cell-based limitations beyond chaperone partitioning . For example , in contrast to the proteostasis enhancement we observe following thermal stress in yeast , previous studies have linked the accumulation of protein aggregates to reduced proteostasis capacity in vivo ( Broadley and Hartl , 2009 ) . In these cases , protein aggregates , including those resulting from oxidative damage with age or proteotoxic stresses , have been linked to reduced replicative lifespan , ( Aguilaniu et al . , 2003; Hernebring et al . , 2006; Rujano et al . , 2006; Erjavec et al . , 2007 , 2008; Tessarz et al . , 2009; Knorre et al . , 2010; Liu et al . , 2010; Unal et al . , 2011; Zhou et al . , 2011; Spokoini et al . , 2012 ) and , the presence of protein amyloids , such as polyglutamine-expanded proteins and other yeast prions , promote the misfolding of metastable proteins , interfere with proteolysis , reduce protein synthesis , inhibit endocytosis , and disrupt prion propagation through the sequestration of chaperones ( Gidalevitz et al . , 2006; Meriin et al . , 2003; Kirstein-Miles and Morimoto , 2013; Park et al . , 2013; Yang et al . , 2013; Yu et al . , 2014 ) . A comparison of these studies with our work suggests that the dynamics of chaperone engagement with distinct substrates , rather than simply their presence , correlates with the impact of these interactions on proteostasis capacity . In the studies resulting in chaperone sequestration , proteotoxicity correlates with imbalances in the system imposed by harsh conditions and/or the unnaturally high expression of amyloidogenic proteins . In contrast , we detect no differences in Hsp104-GFP localization in [PSI+] and [psi−] strains expressing Sup35 at its endogenous level ( Figure 4A , 30°C ) , and Sup35 amyloid can clearly be resolved at these native stoichiometries when the system is elevated to a distinct but accessible proteostatic niche ( Figure 1D ) . However , paralleling the studies in metazoans , [PSI+] can transition from a benign to a toxic state upon Sup35 overexpression ( Ter-Avanesyan et al . , 1993 ) , a condition that also induces Hsp70 co-localization ( Winkler et al . , 2012 ) . Thus , proteostasis capacity appears to be finely tuned to maintain a natively expressed proteome . Additional evidence of the importance of this balance can be gleaned by a comparison of chaperone overexpression in different cellular contexts . For example , overexpression of Hsp104 alone , to the same level achieved here through asymmetric retention of this factor ( Table 1 and Figure 3—figure supplement 1C ) , also induces curing ( Chernoff et al . , 1995; Wegrzyn et al . , 2001 ) . While it has been suggested that Hsp104 overexpression dissolves Sup35 amyloid in vivo , this interpretation is complicated by a lack of temporal resolution and the ability to monitor existing protein ( Park et al . , 2014 ) and is inconsistent with the increase in the size of SDS-resistant Sup35 aggregates under these conditions ( Figure 3—figure supplement 1D ) ( Kryndushkin et al . , 2003 ) . An alternative model , which is consistent with this biochemical evidence of Hsp104 inhibition , suggests that upon its overexpression , Hsp104 aberrantly and non-productively co-localizes with Sup35 amyloid ( Chernoff et al . , 1995; Kryndushkin et al . , 2003; Bagriantsev et al . , 2008; Winkler et al . , 2012 ) . In contrast , our studies indicate that Hsp104 overexpression within the context of a thermal stress transitions Sup35 amyloid from outside the buffering capacity of the proteostasis network to within its sphere of protection . Notably , Hsp104 co-localization with Sup35 amyloid varies based on its mode of overexpression ( i . e . individual vs network up-regulation ) , again implicating substrate–chaperone dynamics , rather than simply chaperone availability , in proteostasis capacity . Intriguingly , this interplay is distinct for individual PQC factors within the same cell , as thermal stress induces localization of Hsp104 , Ssa1 , and Sis1 to cytoplasmic foci , but only Hsp104 is asymmetrically retained upon cell division ( Figure 5—figure supplement 1D , E ) , suggesting an additional point of proteostasis regulation . Beyond these cell-based limitations on proteostasis capacity , our studies have deconvoluted the contributions of distinct physical characteristics of amyloid variants to their ability to exceed the PQC buffering capacity in vivo . Intriguingly , we find that [PSI+]Weak , the more thermodynamically stable but less efficiently amplified variant of Sup35 amyloid ( Tanaka et al . , 2006 ) , is susceptible to curing at elevated temperature , while the less thermodynamically stable and more efficiently amplified [PSI+]Strong variant is not ( Figure 1 ) , indicating that amyloid amplification rather than stability imposes the primary limitation on amyloid clearance . Consistent with this idea , reducing [PSI+]Strong amplification by either expressing a Sup35 mutant or decreasing the expression of wild-type Sup35 promotes curing at elevated temperature ( Figure 2 ) . Thus , manipulations that have minor effects on the dynamics of existing amyloid are also sufficient to move this alternative protein-folding pathway within the buffering capacity of the proteostasis network . Together , our observations suggest an alternative to the view that the physical characteristics of amyloid complexes alone preclude their accessibility to the cell's natural defenses against protein misfolding . Rather , the dynamics and balance of the system as a whole , including both protein-based and cell-based contributors , create not only a niche that allows amyloid to arise and persist but also another that promotes amyloid clearance . Our studies , therefore , raise the possibility that the proteostasis limitations that allow the accumulation of chronically misfolded proteins may be distinct in a native context and under conditions of their overexpression . Within this framework , our studies provide a proof-of-principle example to support the idea that proteostasis regulators , which are aimed at transitioning proteostasis landscapes to new thresholds , may be the most effective interventions into amyloidosis ( Lindquist and Kelly , 2011 ) . All plasmids used in this study were previously reported ( Table 2 ) except for SB1013 ( pRS306PGPD-FFL-mCherry ) , which contains firefly luciferase as an Xba1/BamHI fragment and mCherry as a BamHI/XhoI fragment , separated by a three-repeat glycine–serine linker . The ORFs were amplified by PCR using primers 5XbaI firefly/3BamHI firefly and 5BamHIGS3mCherry/3XhoImCherry , respectively ( Table 3 ) and confirmed by sequencing . All strains of Saccharomyces cerevisiae used in this study are derivatives of 74-D694 ( Table 4 ) ( Chernoff et al . , 1995 ) . A WT [PSI+]Weak diploid strain ( SY945 ) was generated by mating SY2600 with SLL3252 ( Table 4 ) . The diploid state was confirmed by sporulation . SY591 , a [PSI+]Weak strain containing a heterozygous deletion of HSP104 , was created by transformation of a PvuI-BamHI fragment of pYABL5 ( a kind gift of S . Lindquist ) into SY945 and selection on medium lacking leucine . Disruptions were verified by PCR and 2:2 marker segregation upon sporulation and dissection . SY957 , a [PSI+]Strong diploid strain containing a heterozygous disruption of SUP35 , was created by transforming a PCR-generated cassette using pFA6a-KanMX4 as a template with primers SD27 and SD28 ( Table 3 ) into SLL3071 ( Table 4 ) . Integration was confirmed by PCR using primers Psup352/PTEFCH and Sup35 3′chk/pFa6 test ( Table 3 ) . The galactose-inducible HSP104 strains were made by integrating BstBI-linearized SB630 ( Table 2 ) into SY197 ( Table 4 ) and selecting transformants on medium lacking uracil . Galactose-inducible expression of Hsp104 was confirmed by western blotting . [PSI+]Strong and [PSI+]Weak were then cytoduced into this strain from SY1698 and SY1699 , respectively , to create SY1748 and SY1749 ( Table 4 ) , respectively ( Conde and Fink , 1976 ) . Cytoductants were selected by growth on synthetic medium containing glycerol and lacking uracil and by colony color on YPD . The HSP104-GFP [psi−] strain ( SY2125 ) was created by transforming a PCR-generated cassette using pFA6a-GFP ( S65T ) -KanMX6 as a template with primers HSP104-GFP F-A and HSP104-GFP R-A ( Table 3 ) into WT [PSI+]Strong strains and selection on medium containing 300 μg/ml G418 . Integration was confirmed by PCR using primers Hsp104for/GFP-R and pFa6 test/Hsp104 3 flank R ( Table 3 ) , and expression was confirmed by fluorescence microscopy . These strains were cured of the prion by growth on YPD plates containing 3 mM GdnHCl ( Tuite et al . , 1981 ) . The [PSI+]Weak variant ( SY2126 ) was generated by mating SY2125 to a WT [PSI+]Weak strain ( SLL2600 ) and sporulation . Tetrads were dissected to recover haploids , and HSP104-GFP isolates were verified by G418 resistance , fluorescence microscopy , and quantitative immunoblotting for Hsp104 . The heat inducible GFP strain ( SY2091 ) was generated by transformation of a WT [psi−] strain ( SLL2119 ) with Bsu36I-digested SB849 ( Table 2 ) . Expression was confirmed by fluorescence microscopy . SSA1-GFP ( SY2658 ) and SIS1-GFP ( SY2447 ) [PSI+]Weak strains were created by transforming PCR-generated cassettes using pFA6a-GFP ( S65T ) -KanMX6 as a template with primers GFP-GS-Ssa1-F/GFP-Ssa1-R or Sis1-GFP-F GS/Sis1-GFP-R ( Table 3 ) , respectively , into WT [PSI+]Strong strains and selection on medium containing 300 μg/ml G418 . Expression was confirmed by fluorescence microscopy and quantitative immunoblotting for Ssa1/2 and Sis1 , respectively . These strains were cured of the prion by growth on YPD plates containing 3 mM GdnHCl , mitochondrial loss was induced by growth in 25 μg/ml ethidium bromide , and [PSI+]Weak was transferred to them by cytoduction ( Cox , 1965 ) , using SY1699 as a donor strain . Cytoductants were selected by growth on glycerol medium and 300 μg/ml G418 . SSA1-GFP and SIS1-GFP [PSI+]Weak strains containing a nuclear-localized fluorescent reporter protein ( DsRed-NLS , SY2659 , and SY485 , respectively ) were generated by transforming SY2658 or SY2447 with Bsu36I-digested SB503 ( Table 2 ) . Expression was confirmed by fluorescence microscopy . The Δbni1 [PSI+]Weak strain ( SY1888 ) was created by transforming a PCR-generated cassette using pFA6a-KanMX4 as a template with primers AD-BNI1-f and AD-BNI1-r ( Table 3 ) into a [PSI+]Weak diploid ( SY782 , a cross between SY2600 and SY86 , Table 4 ) . Transformants were selected on medium containing 300 μg/ml G418 and verified by PCR using primers AD-BNI1-fseq/PTEFCH and AD-BNI1-rseq/pFa6 test ( Table 3 ) . The haploid Δbni1 [PSI+]Weak strain was then generated by sporulation and tetrad dissection and verified by G418 resistance . The HSP104-GFP Δbni1 [PSI+]Weak strain ( SY2486 ) was created by transforming SY2126 with a PCR-generated cassette using pFA6a-hphMX4 as a template with primers AD-BNI1-f and AD-BNI1-r ( Table 3 ) . Transformants were confirmed by PCR using primers AD-BNI1-fseq/PTEFCH and AD-BNI1-rseq/pFa6 test ( Table 4 ) and growth on YPD plates containing 300 μg/ml hygromycin B . 10 . 7554/eLife . 04288 . 017Table 2 . PlasmidsDOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 017NameDescriptionSB20pRS306-PSup35N ( GS ) 3sGFP ( GS ) 3MCSB503pRS304-PGPDGST-DsRED-NLSSB630pRS306-PGALHsp104SB657pRS306-PtetO2Sup35SB658pRS306-PtetO2Sup35 ( G58D ) SB849pRS306-PHSEGFPSB1013pRS306-PGPDFFL-mCherry10 . 7554/eLife . 04288 . 018Table 3 . PrimersDOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 018NameSequence5XbaI firefly5′-TCTAGAATGGAAGATGCCAAAAACATTAAG-3′3BamHI firefly5′-GGATCCACCTTGAGACTGTGGTTGGAAAC-3′5BamHI GS3mCherry5′-GGATCCGGTAGTGGTAGTGGTAGTATGGTGAGCAAGGG CGAGGAG-3′3XhoI mCherry5′-CTCGAGTTACTTGTACAGCTCGTCCATGCCG-3′SD275′-ACTTGCTCGGAATAACATCTATATCTGCCCACTAGCAACA CAGCTGAAGCTTCGTACGC-3′SD285′-GGTATTATTGTGTTTGCATTTACTTATGTTTGCAAGAAATG CATAGGCCACTAGTGGATCTG-3′Psup3525′-GAGATGCTCATCAAGGG-3′PTEFCH5′-GCACGTCAAGACTGTCAAGG-3′Sup35 3′chk5′-TATTTACGAAGGAGACCCGGAG-3′pFa6 test5′-TGCCCAGATGCGAAGTTAAGTG-3′HSP104-GFP F-A5′-CGATAATGAGGACAGTATGGAAATTGATGATGACCTA GATCGGATCCCCGGGTTAATTAA-3′Hsp104-GFP R-A5′-TATTATATTACTGATTCTTGTTCGAAAGTTTTTAAAAATC GAATTCGAGCTCGTTTAAAC-3′Hsp104for5′-GGCACATCCTGATGTTTTGA-3′GFP-R5′-CCTTCACCCTCTCCACTGACAG-3′Hsp104 3 flank R5′-CCGTATTCTAATAATGGACCAATC-3′GFP-GS-Ssa1-F5′-AGCTCCAGAGGCTGAAGGTCCAACCGTTGAAGAAGTTG ATGGTTCTGGTTCTGGTTCTCGGATCCCCGGGTTAATTAA-3′GFP-Ssa1-R5′-ACCCAGATCATTAAAAGACATTTTCGTTATTATCAATTGC GAATTCGAGCTCGTTTAAAC-3′Sis1-GFP-F GS5′-ACTAAACGACGCTCAAAAACGTGCTATAGATGAAAATTT TGGTTCTGGTTCTGGTTCTCGGATCCCCGGGTTAATTAA-3′Sis1-GFP-R5′-ATTTATTTGAGTTTATAATTATATTTGCTTAGGATTACTAG AATTCGAGCTCGTTTAAAC-3′AD-BNI1-f5′-ATGTTGAAGAATTCAGGCTCCAAACATTCGAACTCAAAG GCAGCTGAAGCTTCGTACGC-3′AD-BNI1-r5′-TTATTTGAAACTTAGCCTGTTACCTGTCCTAGCCTCACCT GCATAGGCCACTAGTGGATCTG-3′AD-BNI1-fseq5′-GACATCGGTTAGAGGAAG-3′AD-BNI1-rseq5′-CACTGTGCTTGTCACTTA-3′10 . 7554/eLife . 04288 . 019Table 4 . Yeast strainsDOI: http://dx . doi . org/10 . 7554/eLife . 04288 . 019StrainGenotypePlasmids integratedReferenceFigureSLL2119MATa [psi−] ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112-Chernoff et al . , 19951c , 3SfSLL2600MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112-Derkatch et al . , 19961 , 2a , 3 , 5ce , 6abcfg , 1S , 3Sabg , 4Sabcd , 5Sfg , 6SaSLL2606MATa [PSI+]Strong ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112-Chernoff et al . , 19951ac , 3SfSLL3071MATa/α [PSI+]Strong ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52 leu2-3112/ leu2-3112-DiSalvo et al . , 20112dSLL3252MATa [psi−] ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112-Chernoff et al . , 1995‘Materials and methods’SY86MATα [psi−] ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112 sup35::N ( GS ) 3GFP ( GS ) 3MCSB20Derdowski et al . , 2010‘Materials and methods’SY197MATa [psi−] ade1-14 his3-11 , -15 trp1-1 ura3-1 leu2-3112 can1-100-J Weissman ( YJW513 ) ‘Materials and methods’SY591MATa/α [PSI+]Weak ade1-14/ade1-14 his3Δ200/his3Δ200 TRP/ trp1-289 ura3-52/ura3-52 leu2-3112/ leu2-3112 HSP104/hsp104::LEU2-This study2bSY782MATa/α [PSI+]Weak ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52 leu2-3112/ leu2-3112 SUP35/sup35::N ( GS ) 3GFP ( GS ) 3MC-This study‘Materials and methods’SY945MATa/α [PSI+]Weak ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52 leu2-3112/ leu2-3112-This study2bSY957MATa/α [PSI+]Strong ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52 leu2-3112/ leu2-3112 SUP35/sup35::KANMX4-This study2dSY1646MATa/α [PSI+]Strong ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52::URA3::PtetO2SUP35 leu2-3112/ leu2-3112 SUP35/sup35::KANMX4SB657DiSalvo et al . , 20112cSY1648MATa/α [PSI+]Strong ade1-14/ade1-14 his3Δ200/his3Δ200 trp1-289/ trp1-289 ura3-52/ura3-52::URA3::PtetO2SUP35 ( G58D ) leu2-3112/ leu2-3112 SUP35/sup35::KANMX4SB658DiSalvo et al . , 20112cSY1698MATα [PSI+]Strong ade1-14 his3Δ200 ura3-52 leu2-3 kar1-d15 ConR CyhR-This study‘Materials and methods’SY1699MATα [PSI+]Weak ade1-14 his3Δ200 ura3-52 leu2-3 kar1-d15 ConR CyhR-This study‘Materials and methods’SY1748MATa [PSI+]Strong ade1-14 his3-11 , -15 trp1-1 ura3-1::URA3::PGALHSP104 leu2-3112 can1-100SB630This study3SeSY1749MATa [PSI+]Weak ade1-14 his3-11 , -15 trp1-1 ura3-1::URA3::PGALHSP104 leu2-3112 can1-100SB630This study3ScdeSY1888MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112 Δbni1::KANMX4-This study5c , 5SfgSY2091MATa [psi−] ade1-14 his3Δ200 trp1-289 ura3-52::URA::PHSEGFP leu2-3 , 112SB849This study5SaSY2125MATα [psi−] ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3112 HSP104GFP::KANMX6-This study4bSY2126MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3112 HSP104GFP::KANMX6-This study4 , 5abdfg , 6de , 7 , 4Sabcdfg , 6SbSY2447MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3112 SIS1GFP::KANMX6-This study5SeSY2485MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289::TRP::PGPDGST-DsRed-NLS ura3-52 leu2-3112 SIS1GFP::KANMX6SB503This study5ScSY2486MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3 , 112 HSP104GFP::KANMX6 Δbni1::hphMX4-This study5bSY2658MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52 leu2-3112 SSA1GFP::KANMX6-This study5SdSY2659MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289::TRP::PGPDGST-DsRed-NLS ura3-52 leu2-3112 SSA1GFP::KANMX6SB503This study5SbSY2802MATa [PSI+]Weak ade1-14 his3Δ200 trp1-289 ura3-52::URA::PGPDFirefly-mCherry leu2-3112 HSP104GFP::KANMX6SB1013This study4Se Unless otherwise specified , yeast cultures were grown in rich YPD medium supplemented with 0 . 3 mM adenine . Cultures were maintained at an OD600 of less that 0 . 5 at 30°C for at least 10 doublings to ensure exponential growth . Where indicated , cultures were then transferred to 37°C or 40°C for the specified period . Pretreatment of cultures at 37°C prior to shift to 40°C was for 30 min . To analyze colony color phenotype , aliquots of cultures were diluted in H2O as needed to ensure well-separated single colonies upon plating to solid YPD medium . After growth at 30°C , each colony was scored based on colony color phenotype: fully cured ( completely red , [psi−] ) , sectored ( part red and part white ) , or [PSI+] ( completely white ) . Unless otherwise indicated , fully cured and sectored colonies were combined in the ‘cured’ category . For all colony counting assays , at least 150 colonies were counted for each experimental condition/timepoint . For the galactose-inducible Hsp104 experiments , cells were grown in rich YP medium containing 3% raffinose supplemented with 3% galactose during induction . α-factor and nocodazole arrests were performed in YPD liquid medium containing final concentrations of 5 μg/ml α-factor or 15 μg/ml nocodazole , respectively , for ∼2 hr . Following confirmation of arrest based on cell morphology by bright-field microscopy , cultures were washed three times with medium containing 1 mM DMSO followed by one wash in YPD before resuspension for indicated manipulation . GdnHCl treatment was performed at 3 mM final concentration in liquid YPD , and for experiments involving recovery , cultures were washed three times with medium before resuspension in YPD for indicated manipulation . SDS-PAGE and quantitative immunoblotting were performed as previously described ( Pezza et al . , 2009 ) . Anti-Ssa1/2 rabbit serum was provided by E . Craig ( U Wisconsin—Madison ) , and anti-Sis1 rabbit serum was provided by M . Tuite ( U Kent , Canterbury , UK ) . Semi-native agarose gel electrophoresis ( SDD-AGE ) was performed as previously described ( Kryndushkin et al . , 2003 ) . The cycloheximide SDS-sensitivity assay was performed as previously described ( DiSalvo et al . , 2011 ) with the following modifications: 1 ) cultures were treated at the various experimental temperatures for 30 min prior to the addition of cycloheximide to allow for the induction of chaperone proteins , and 2 ) after cycloheximide treatment , cultures were incubated with shaking at 30°C for 2 hr before lysis and analysis . For the aggregation analysis , native lysates were prepared as described previously ( Kryndushkin et al . , 2003 ) . Lysates were pre-cleared for 1 min at 500×g and total protein content was quantified using the BioRad Bradford assay in triplicate . Lysates were subjected to 15 , 000×g centrifugation for 15 min , and pellets were washed with 10 mM sodium phosphate buffer ( pH7 . 5 ) containing 2% NP-40 before being resuspended in 10 mM sodium phosphate buffer ( pH7 . 5 ) and quantified again in triplicate using the Bradford assay . For the Hsp104 immunocapture , native lysates were prepared at 4°C in IP buffer ( 50 mM HEPES–NaOH ( pH 7 . 5 ) , 150 mM NaCl , 10 mM MgCl2 , 1 mM EDTA , 1% NP-40 , 0 . 25% Na-deoxycholate , and protease inhibitors ( 2 mM PMSF , 5 µg/ml pepstatin , complete protease inhibitor tablets ( Pierce , Rockford , IL ) , protease inhibitor cocktail ( Sigma-Aldrich , St . Louis , MO ) ) . Lysates were pre-cleared for 1 min at 500×g , and then incubated for 1 hr with Protein G magnetic beads ( NEB , Ipswich , MA ) with nutation . Immunocapture was performed using Protein G magnetic beads and anti-GFP mouse monoclonal antibody ( Roche , Switzerland ) . Beads were washed 4× with IP buffer and 1× with 50 mM HEPES–NaOH ( pH 7 . 5 ) , and protein was eluted by boiling in SDS sample buffer . Co-captured proteins were resolved by SDS-PAGE and analyzed by gel staining with Flamingo ( Bio-Rad , Hercules , CA ) and fluorescent scanning on a Typhoon imager ( GE Lifesciences , Marlborough , MA ) according to the manufacturer's instructions or by western blot for GFP . Static images were obtained on a Zeiss Axio Imager M2 fluorescence light microscope equipped with a 100× objective . Confocal images were obtained on a Zeiss LSM 510 Meta confocal microscope using a 100× objective . Microfluidics experiments were performed on a Zeiss Axio Observer Z1 using a CellAsics microfluidics plate with temperature controls and media flow of 2 psi on a Y0C4 yeast perfusion plate ( channel size 3 . 5–5 μm ) . Imaging was performed in complete minimal medium supplemented with 2% glucose and 2 . 5 mM adenine . Fluorescence intensity was analyzed using the Zen software package ( Zeiss , Germany ) . Flow cytometry and cell sorting was performed on a BD FACSAria fluorescence-activated cell sorter using a 488 nm laser and a FITC-A filter to measure GFP fluorescence intensity in single cells . Data were obtained at least in triplicate with representative spectra shown . Data were analyzed using the FlowJo software package ( TreeStar Inc . , Ashland , OR ) .	Proteins must fold into specific shapes to work inside cells , and the misfolding of proteins is associated with a growing number of diseases . For example , prions are misfolded proteins that form insoluble aggregates called amyloids . These aggregates are not easily destroyed and can cause other nearby proteins to misfold and join the amyloid . This process of amyloid assembly leads to progressive diseases such as mad cow disease , Huntington's disease , Alzheimer's disease , and Parkinson's disease , which are collectively known as amyloidoses . A series of biological pathways called the proteostasis network control protein integrity in a cell . Under normal conditions or even mildly stressful conditions—such as at slightly increased temperatures—the proteostasis network is able to prevent proteins from misfolding . However , if a cell is placed under lots of stress this network may become overwhelmed and misfolded proteins can accumulate . To date , the proteostasis network has not been linked to the clearance of amyloids . A protein called Sup35 , which is found in budding yeast , can exist as two different prion forms . Previous studies have shown that briefly heating the yeast cells can ‘cure’ the so-called ‘weak’ form of the prion . The ‘strong’ prion form , however , was thought to be unaffected by elevated temperature . These previous studies had only tested yeast cells that had been dividing for a few generations; it was unknown if cells that had been dividing for longer might respond differently . Klaips et al . found that a protein called Hsp104—which helps to fold proteins properly—can break down the amyloid aggregates . This protein is normally only present in small amounts , but heating causes the levels of Hsp104 to rise . Klaips et al . found that the extra Hsp104 protein associated with the aggregates and led to their disassembly . When Hsp104 was prevented from associating with the prions , the aggregates were not cured even if high levels of Hsp104 were present in the cell . When budding yeast form new cells , a daughter cell ‘buds’ off from the mother cell . Klaips et al . found that mother cells exposed to heat retain most of the Hsp104 when the cell divides , and this retention allowed Hsp104 to accumulate to a level required for the breakdown of amyloid aggregates . Therefore , under normal conditions , amyloids persist because cell division keeps the amount of Hsp104 below this threshold . Previously it had been thought that the physical characteristics of amyloids accounted for their resilience in the face of the cell mechanisms designed to counteract protein misfolding . However , Klaips et al . show that the balance of the different mechanisms involved in proteostasis can be manipulated to create environments where amyloids are either created and maintained or destroyed . Targeting these mechanisms could therefore present new treatment options for amyloidosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Spatial quality control bypasses cell-based limitations on proteostasis to promote prion curing
Neurons rely on translation of synaptic mRNAs in order to generate activity-dependent changes in plasticity . Here , we develop a strategy combining compartment-specific crosslinking immunoprecipitation ( CLIP ) and translating ribosome affinity purification ( TRAP ) in conditionally tagged mice to precisely define the ribosome-bound dendritic transcriptome of CA1 pyramidal neurons . We identify CA1 dendritic transcripts with differentially localized mRNA isoforms generated by alternative polyadenylation and alternative splicing , including many that have altered protein-coding capacity . Among dendritic mRNAs , FMRP targets were found to be overrepresented . Cell-type-specific FMRP-CLIP and TRAP in microdissected CA1 neuropil revealed 383 dendritic FMRP targets and suggests that FMRP differentially regulates functionally distinct modules in CA1 dendrites and cell bodies . FMRP regulates ~15–20% of mRNAs encoding synaptic functions and 10% of chromatin modulators , in the dendrite and cell body , respectively . In the absence of FMRP , dendritic FMRP targets had increased ribosome association , consistent with a function for FMRP in synaptic translational repression . Conversely , downregulation of FMRP targets involved in chromatin regulation in cell bodies suggests a role for FMRP in stabilizing mRNAs containing stalled ribosomes in this compartment . Together , the data support a model in which FMRP regulates the translation and expression of synaptic and nuclear proteins within different compartments of a single neuronal cell type . A key feature in the molecular biology of learning and memory is protein synthesis-dependent synaptic plasticity , which involves translation of localized mRNAs in response to synaptic activity . Local translation has been demonstrated in neuronal dendrites and axons ( reviewed in Glock et al . , 2017; Lin and Holt , 2007; Rangaraju et al . , 2017 ) and allows for rapid and precise changes in the local proteome near active synapses . In dendrites , a brief burst of local translation has been shown to be necessary and sufficient for induction of the late phase of long-term potentiation ( L-LTP , occurring hours to days after potentiation ) ( Frey et al . , 1988; Kang and Schuman , 1996; Kang et al . , 1997 ) and long-term depression ( LTD ) ( Huber et al . , 2000 ) , and inhibiting protein synthesis blocks long-term memory formation ( Frey et al . , 1988; Sutton and Schuman , 2006 ) . Activity-dependent local translation depends on both the availability of specific mRNAs and the sensitivity with which their translation can be initiated upon local signaling events . Both rely on interactions between mRNAs , a host of RNA-binding proteins , and ribosomes . mRNAs are thought to be transported in a translationally repressed state into the neuronal processes via transport granules containing RNA-binding proteins such as the Fragile-X mental retardation protein ( FMRP ) , CPEB1 , ZBP-1 , and STAU1/2 ( Hüttelmaier et al . , 2005; Krichevsky and Kosik , 2001; Martin and Ephrussi , 2009 ) . Although dendritic targeting elements have been defined for a few mRNAs such as Camk2α , Actb , and Map2 ( Andreassi and Riccio , 2009 ) , and a few RNA-binding proteins have been found to regulate individual mRNAs , such as the interaction between ZBP-1 and the β-actin mRNA ( reviewed in Biswas et al . , 2019 ) , the functional relationship between the global dendritic transcriptome and individual RNA-binding proteins is still largely unknown . For at least some localized mRNA granules , signaling cascades initiated by synaptic activity lead to their dissolution and initiation of translation ( Dahm and Kiebler , 2005 ) , but the role of RNA regulatory factors in this process is incompletely understood . The integrated study of the dendritic transcriptome and the RNA-binding proteins responsible for regulation of local translation will provide critical insight into the mechanisms underlying protein synthesis-dependent synaptic plasticity . FMRP , the RNA-binding protein whose activity is lost in Fragile X syndrome , represses translation ( Bassell and Warren , 2008; Costa-Mattioli et al . , 2009; Darnell et al . , 2011; Laggerbauer et al . , 2001 ) and is thought to be a key regulator of activity-dependent local translation in neurons ( Banerjee et al . , 2018; Bear et al . , 2004; Huber et al . , 2002; Lee et al . , 2011 ) . Dendritic FMRP levels are increased upon neuronal activity , with evidence for local translation of the FMRP transcript itself ( Greenough et al . , 2001; Weiler et al . , 1997 ) and kinesin-mediated movement of FMRP-containing mRNA transport granules from the neuronal cell body ( Dictenberg et al . , 2008 ) . At the synapse , FMRP is proposed to be linked to local signal transduction , potentially through calcium-induced post-translational modification of the protein , which alters the FMRP granule and leads to translation of the mRNAs ( Lee et al . , 2011; Narayanan et al . , 2007 ) . FMRP knockout ( KO ) neurons show excess basal translation as well as an inability to produce activity-stimulated translation ( Ifrim et al . , 2015 ) . Direct FMRP targets have been identified in the whole mouse brain through CLIP studies ( Darnell et al . , 2011; Korb et al . , 2017 ) , indirectly through ribosome-binding studies ( translating ribosome affinity purification [TRAP]; Ceolin et al . , 2017; Kumari and Gazy , 2019 ) or through TRAP together with crosslinking immunoprecipitation ( CLIP ) ( Sawicka et al . , 2019 ) . Some recent studies have explored FMRP targets specifically in the excitatory CA1 neurons of the mouse hippocampus ( Ceolin et al . , 2017; Sawicka et al . , 2019 ) . FMRP target genes overlap significantly with autism susceptibility genes and include genes involved in both synaptic function and transcriptional control in the nucleus ( Darnell , 2020; Darnell et al . , 2011; Iossifov et al . , 2012; Sawicka et al . , 2019 ) , and loss of FMRP increases translation of chromatin modifiers such as BRD4 ( Korb et al . , 2017 ) and SETD2 ( Shah et al . , 2020 ) . These and other observations have suggested a model in which FMRP regulates the stoichiometry of its targets in two ways: globally , by translational control of transcription regulators in the cell body , and locally , by enabling activity-dependent local translation of synaptic proteins in dendrites ( Darnell , 2020 ) , but it is still unclear the extent to which such regulation occurs simultaneously in a single neuron . Here , we probe this model by exploring subcellular compartment-specific patterns of FMRP binding and regulation . We utilize compartment- and cell-type-specific profiling technologies to precisely define the transcriptome of mouse hippocampal CA1 pyramidal neurons . We use TRAP and conditionally tagged ( cTag ) mice that express tagged RNA-binding proteins in a single-cell type to study RNA regulation specifically in CA1 neurons combined with manual microdissection to isolate compartment-specific proteins and mRNAs . RNA profiling of subcellular CA1 compartments reveals that dendritic mRNAs are enriched for elongated 3′UTR isoforms and depleted for alternative splicing ( AS ) events driven by the neuronal splicing factor NOVA2 , indicating a nuclear role in the generation of the localized transcriptome in CA1 neurons . Integrating compartment-specific cTag-FMRP-CLIP and TRAP defined FMRP CLIP scores in the dendrites and cell bodies of CA1 neurons and identified 383 FMRP-bound dendritic targets . This allowed us to distinguish FMRP targets according to their site of regulation within neurons , revealing enrichment of FMRP-regulated mRNAs encoding nuclear proteins in the CA1 cell bodies and mRNAs encoding synaptic proteins in the CA1 dendrites . Moreover , although mRNA localization is unaffected in FMRP KO mice , mRNAs encoding these synaptic proteins show altered localized ribosome association . Together , these findings support a model in which distinct patterns of both mRNA and FMRP subcellular localization enable FMRP to regulate the expression of different proteins within different compartments in a single neuronal cell type . We developed a system that allows for parallel isolation of mRNAs and RNA-binding proteins that are enriched in the cell bodies or dendrites specifically in excitatory CA1 neurons in the hippocampus ( Figure 1A ) . We created three CA1-specific protein-tagged mouse lines by crossing cTag mice with mice in which Cre recombinase expression is driven from the Camk2a promoter ( Tsien , 1998 ) . In these mice , Cre is expressed only in pyramidal neurons of the hippocampus . The cTag-PABP ( Hwang et al . , 2017 ) and cTag-FMRP mice allow for Cre-dependent expression of GFP tagged polyA-binding protein c1 ( PABPC1 ) , , or FMRP ( Sawicka et al . , 2019; Van Driesche et al . , 2019 ) , respectively . The RiboTag ( Sanz et al . , 2009 ) mouse allows for Cre-driven expression of HA tagged RPL22 , a ribosomal subunit . Crossing these animals with Camk2a-Cre mice results in lines expressing tagged ribosomes , or in the case of cTag polyA-binding protein c1 ( PABPC1 or FMRP , ‘knock-in’ tagged proteins expressed from native genes . In the hippocampus , this expression is specific in the CA1 pyramidal neurons ( Figure 1—figure supplement 1A; Hwang et al . , 2017 ) . Microdissection of the CA1 neuropil compartments and immunoprecipitation ( IP ) allowed us to enrich dendritic tagged proteins that originated from the cell bodies ( CBs ) of the CA1 neurons . TRAP-seq ( Heiman et al . , 2014 ) , in which mRNAs bound to affinity-tagged ribosomes are immunoprecipitated and sequenced , enriched for mRNAs bound to dendritic ribosomes ( in Camk2a-Cre x RiboTag mice ) . cTag-PAPERCLIP ( conditionally tagged poly ( A ) -binding protein-mediated mRNA 3′ end retrieval by crosslinking immunoprecipitation ) ( Hwang et al . , 2017 ) allowed cell-type-specific CLIP of the polyA-binding protein , PABPC1 , and subsequent sequencing of 3′UTR-polyA tail junctions in order to determine the precise end of the 3′UTR of an expressed mRNA . Here , we use cTag-PAPERCLIP on microdissected CA1 cell bodies and neuropil to describe compartment-specific 3′UTR usage . cTagFMRP-CLIP ( in Camk2a-Cre × Fmr1-cTag mice ) allowed us to identify compartment-specific FMRP-binding events , extending previous studies ( Darnell et al . , 2011; Sawicka et al . , 2019 ) . Microdissected CA1 compartments from 8- to 10-week-old mice were subjected to bulk RNA-seq as a denominator for all transcripts in the neuropil , and TRAP as a denominator for all CA1 pyramidal neuron-specific , ribosome-bound dendritic transcripts . IP conditions were optimized to isolate relatively pure , intact , ribosome-bound mRNAs with minimal contamination by interneurons and glial cells found in the neuropil ( Figure 1—figure supplement 1B–E ) . As a negative control , RiboTag animals not expressing the Cre recombinase were microdissected and subject to affinity purification and sequencing , and only mRNAs enriched over these controls were considered for downstream analyses . We identified two groups of dendritic , ribosome-bound mRNAs: dendrite-present ( significantly enriched in CA1 neuropil TRAP-seq over CA1 neuropil bulk RNA-seq; 2058 mRNAs ) and dendrite-enriched ( dendrite-present and also significantly enriched in CA1 neuropil TRAP over cell bodies TRAP; 1211 mRNAs; Figure 1B , see Supplementary file 1A and B for a full list of mRNAs identified ) . 689 ( 34% ) of the dendrite-present mRNAs were previously identified in bulk RNA-seq of the microdissected rat CA1 neuropil ( Cajigas et al . , 2012 ) , and these RNAs were found to be significantly enriched in dendrites ( CA1 neuropil TRAP vs . bulk RNA-seq; Figure 1—figure supplement 2A and B ) . The identified dendrite-enriched mRNAs were significantly longer than the whole-cell transcriptome identified in CA1 pyramidal neurons ( Sawicka et al . , 2019 ) , whether considering full-length transcripts , 5′UTR , 3′UTR , or coding sequence ( CDS ) portions ( Figure 1—figure supplement 2C ) . Gene Ontology ( GO ) analysis of dendrite-enriched mRNAs showed strong enrichment for genes encoding proteins with important roles in the synapse such as synaptic signaling , anterograde synaptic signaling , and behavior ( Figure 1C ) , consistent with prior analyses ( Cajigas et al . , 2012 ) . We used RNAscope fluorescence in situ hybridization ( FISH ) to validate the presence in dendrites of several mRNAs that had not been identified in previous studies including Kmt2d ( a histone methyltransferase ) , Myo5a ( an actin motor protein involved in transporting cargo to dendrites ) , Ppp1r9b ( a scaffolding protein component of protein phosphatase 1a important for dendritic spine morphology ) , and Rbfox2 ( a neuronal splicing factor ) ( Figure 1D–G ) . Interestingly , approximating the distance from the cell body for each detected mRNA spot revealed variable mRNA distribution patterns for different transcripts , suggesting multiple potential paths for mRNA localization . For example , roughly 35% of the transcripts encoding Kmt2d and Rbfox2 were detected throughout in the neuropil , whereas ~74% of the transcripts encoding Ppp1r9b were abundant in the distal neuropil ( Figure 1F ) . By comparison , less than 15% of mRNAs encoding alpha-synuclein , a neuronal gene whose protein product is involved in presynaptic transmission ( Snca ) and an mRNA identified as enriched in the CA1 cell body compartment , were found in the CA1 neuropil . Subcellular localization of cytoplasmic mRNAs is thought to be at least partially mediated by 3′UTR elements ( Andreassi and Riccio , 2009; Blichenberg et al . , 2001; Mayford et al . , 1996; Tushev et al . , 2018 ) . However , analysis of 3′UTRs from RNA-seq data alone is complicated by mixed cell types , incomplete annotation , and difficulty in identifying internal polyA sites . To identify the expressed 3′UTRs in CA1 pyramidal neurons , we first used polyA sites determined by Camk2a-Cre-driven cTag-PAPERCLIP from whole hippocampus ( Hwang et al . , 2017 ) and microdissected CA1 compartments to define all 3′ ends . We next used splice junctions identified from TRAP to define 5′ end of each 3′UTR ( Figure 2—figure supplement 1 ) . This allowed us to identify the boundaries of potential 3′UTRs ( Figure 2A , Figure 2—figure supplement 2A and B ) , and revealed 15 , 322 3′UTRs expressed in Camk2a-expressing pyramidal neurons , including 3700 genes that give rise to mRNAs with more than one 3′UTR isoform . Analyzing expression of these 3′UTRs in the compartment-specific TRAP data revealed 219 3′UTR isoforms that were differentially localized to CA1 dendrites ( Figure 2B , Supplementary file 1D ) . Analysis of these differentially localized 3′UTRs revealed transcripts generated by two types of alternative polyadenylation ( APA ) , distinguished by their effect on the CDS of the resulting protein . APA events that do not affect the CDS of the resulting protein derive from transcripts with multiple polyadenylation sites in a single 3′UTR , resulting in isoforms with short ( proximal ) and long ( distal ) 3′UTRs ( 3′UTR-APA ) . APA events that truncate the CDS of the resulting protein utilize polyA sites in upstream regions , resulting in multiple ( short and long ) protein isoforms ( UR-APA ) ( Tian and Manley , 2017 ) . Of the 219 genes producing differentially localized 3′UTR isoforms in CA1 neurons , we found that 149 had no effect on the CDS , 48 resulted in altered CDS , and 22 generated both event types ( Figure 2B , left panel ) . Among isoforms with unchanged CDS , distal 3′UTRs were significantly enriched in dendrites , consistent with a previous study of CA1 neuropil RNAs analyzed by 3′ end sequencing ( Tushev et al . , 2018 ) . Conversely , proximal 3′UTRs were significantly enriched in the CA1 cell bodies ( Figure 2B , middle and right panels ) . We used FISH to validate these types of differential localization events , including Calmodulin 1 ( Calm1 ) ( Figure 2—figure supplement 2B–E ) , previously described to harbor differentially localized 3′UTR isoforms ( Tushev et al . , 2018 ) , F-box protein 31 ( Fbxo31 ) ( Figure 2—figure supplement 2F–H ) , an E3 ubiquitin ligase proposed to be involved in neuronal maintenance and dendritic outgrowth ( Vadhvani et al . , 2013 ) , and vesicle-associated membrane protein B ( Vapb ) ( Figure 2—figure supplement 2I–K ) , a membrane protein involved in vesicle trafficking . Approximately 20% of the differential isoform localization events ( 48 out of 219 ) involved a polyadenylation event that led to an extension or truncation of the CDS ( Figure 2B ) . For example , the gene for connector enhancer of kinase suppressor of Ras2 ( Cnksr2 or MAGUIN ) produces mRNAs with two 3′UTRs isoforms: a short isoform that is highly sequestered in the cell bodies ( less than 10% of transcripts were found in the CA1 neuropil by FISH ) and a longer isoform of which at least 40% of transcripts were localized in the CA1 neuropil ( Figure 2C–E ) . Analysis of the ankyrin repeat and sterile alpha motif domain containing 1B ( Anks1b ) gene revealed differential localization of an isoform generated from 5′ extension of the 3′UTR sequence , which was depleted in the CA1 cell bodies , and again validated by FISH ( Figure 2F–H ) . Finally , two mRNAs produced from the ank-repeat domain containing protein 11 ( Ankrd11 ) gene were identified , a full-length version that contains Ank repeats , as well as the C-terminal transcriptional repression and activation domain , and a previously uncharacterized isoform derived from a polyadenylation site found in intron 8 , which is able to produce a protein that contains only the Ank-repeat regions ( see PAPERCLIP profile in Figure 2I ) . The truncated isoform was predominantly detected in the cell bodies of CA1 neurons by both TRAP and FISH , while the full-length isoform was detected in both the cell bodies and dendrites ( Figure 2I–K ) . Together , these data demonstrate the utility of combining compartment- and cell-type transcriptomics and PAPERCLIP to define expressed 3′UTRs and reveal that dendritic transcripts with altered protein-coding capacity are generated by alternative processing of 3′UTRs . We next sought to identify alternative spliced RNA isoforms that were differentially abundant in the dendrites of CA1 pyramidal neurons . After analysis with rMATS ( Shen et al . , 2014 ) and filtering , we identified 165 AS events in 143 genes that were differentially expressed between the two compartments ( Figure 3A , Supplementary file 1E ) . Of these , 106 ( 64 . 2% ) were skipped exons , 32 ( 19 . 4% ) were alternative 3′ splice sites , 14 ( 8 . 5% ) were alternative 5′ splice sites , and 13 ( 7 . 9% ) were mutually exclusive exons ( Figure 3B ) . These alternatively spliced transcripts encode proteins involved in synaptic functions such as action potential , receptor localization , and synaptic signaling , as well as mRNA splicing ( Figure 3C ) . To determine splicing factors that may be responsible for these differentially localized AS events , we used existing datasets ( see Supplementary file 1F ) of splicing changes previously found to be mediated by neuronal AS factors . Of these , MBNL1/2 ( using data from Weyn-Vanhentenryck et al . , 2018 ) and NOVA2 ( using data from Saito et al . , 2016 ) were found to regulate the largest number of these events ( 37 for MBNL1/2 and 36 for NOVA2 , Figure 3D ) . Interestingly , we found that CA1 neuropil/cell body splicing changes were positively correlated with splicing changes in NOVA2 KO animals ( from analysis of Nova2-null vs . WT data , Pearson coefficient = 0 . 498 , p-value=9 . 87e-08 ) , which indicates that NOVA2 drives splicing changes that result in mRNAs that are preferentially sequestered in CA1 cell bodies ( Figure 3E , left panel ) . This effect was specific for NOVA2 as MBNL1/2-dependent splicing changes did not show such a correlation with the localized splicing changes ( Pearson coefficient = –0 . 00438 , p-value=0 . 9698 , Figure 3E , right panel ) . Among transcripts that exemplify differential exon usage in dendritic transcripts were Rapgef4/Epac2 and neuronatin ( Nnat ) . Rapgef4 ( Epac2 ) , the gene encoding a cAMP-activated guanine exchange factor for RAP1 and RAP2 involved in LTP in the hippocampus , expresses two isoforms in the brain , a full-length isoform ( Epac2A1 ) , and one that is lacking exon 7 ( Epac2A2 ) ( Hoivik et al . , 2013 ) . Of the Rapgef4 transcripts detected in the CA1 dendrites , only 25% included exon 7 , whereas in the cell bodies , 75% of the Rapgef4 contained exon 7 ( Figure 3F ) , indicating preferential localization of the transcripts without exon 7 to the CA1 dendrites . Nnat , a maternally imprinted gene whose protein is important for regulation of intracellular calcium levels , is expressed as either an α- and β-isoform in which exon 2 is included or skipped , respectively . We found that Nnat transcripts lacking exon 2 are predominantly sequestered in the cell bodies , with only ~12 . 5% of cell body transcripts containing exon 2 . Conversely , the majority of localized Nnat transcripts ( 50–75% ) contain exon 2 , indicating preferential localization of the exon 2 containing Nnat transcripts ( Figure 3G ) . These observations underscore the role of AS in generation of localized transcript isoforms . More generally , these data demonstrate that dendritic transcripts with altered protein-coding capacity are generated by both APA and AS . FMRP is thought to be a master regulator of local translation ( Ronesi and Huber , 2008 ) , leading us to examine the relationship between the FMRP targets previously defined in CA1 neurons ( Sawicka et al . , 2019 ) and those that we found to be present in the dendritic ribosome-bound transcriptome ( Figure 1 ) . We observed significant overrepresentation of FMRP targets in dendrite-present mRNAs , and even more so in dendrite-enriched mRNAs ( Figure 4A ) . Of 1211 dendrite-enriched mRNAs , about 35% ( 413 mRNAs ) were FMRP targets compared to 28 . 5% of dendrite-present mRNAs and 11 . 6% of all CA1-expressed mRNAs ( Figure 4B ) . We next compared the relative abundance ( as compared to the cell bodies ) of three groups of mRNAs: all CA1 FMRP targets , and dendrite-enriched mRNAs that either are or are not CA1 FMRP targets . Dendrite-enriched FMRP targets were significantly more abundant in dendrites than nontargets ( Figure 4C ) . Further characterization of these dendrite-enriched mRNAs revealed that they were generally longer than all CA1-expressed mRNAs ( Figure 1—figure supplement 2 ) , but that FMRP-bound dendritic mRNAs were significantly longer than the nontargets ( p-value=9 . 13e-46 , Figure 4D ) . These observations were consistent with prior observations that FMRP preferentially binds long mRNAs ( Darnell et al . , 2011; Sawicka et al . , 2019 ) , and taken together , suggest that FMRP binds the majority of long , dendritic mRNAs . Examination of the functional differences between dendrite-enriched FMRP targets and nontargets revealed an enrichment in dendritic FMRP targets for proteins involved in synaptic signaling , behavior , regulation of trans-synaptic signaling , and GTPase-mediated signal transduction ( Figure 4E ) . These data indicate that FMRP is a key regulator of local translation in the dendrite of mRNAs encoding proteins involved in important synaptic functions . Previous work on mRNA localization in FMRP KO cells in vitro has suggested a role for interactions between G-quadruplexes in the 3′UTRs of FMRP target mRNAs and the RGG-domain of the FMRP protein ( Goering et al . , 2020 ) . We examined dendrite-enriched FMRP targets for enrichment of potential G-quadruplexes . Importantly , we found that all dendrite-enriched mRNAs are highly G- and C-rich ( Figure 4—figure supplement 1B–D ) , so we analyzed differences in G-quadruplex containing transcripts between dendrite-enriched FMRP targets and dendrite-enriched non-FMRP targets ( Figure 4—figure supplement 1A ) . We searched for experimentally defined G-quadruplexes ( Guo and Bartel , 2016; Figure 4—figure supplement 1E ) and also predicted G-quadruplex motifs ( as defined in Goering et al . , 2020 ) in the 3′UTRs of dendrite-enriched FMRP targets and FMRP nontargets ( Figure 4—figure supplement 1F ) . We found no evidence for significant enrichment of G-quadruplexes in dendrite-enriched FMRP targets . FMRP ‘CLIP scores’ were previously developed as a metric to define FMRP-bound transcripts with greater amount of FMRP binding relative to other transcripts of similar abundance in CA1 neurons ( Sawicka et al . , 2019 ) . Dendrite-enriched mRNAs had significantly higher FMRP CLIP scores and hence greater FMRP binding than the dendrite-present group ( p-value=2 . 646e-05 , Figure 4F ) . Additionally , FMRP CLIP scores positively correlated with relative abundance in dendrites: when CA1 mRNAs were grouped according to the magnitude of their CA1 FMRP CLIP scores , those with increasingly higher scores were increasingly abundant in dendrites ( Figure 4G ) . Taken together , these results suggest that FMRP binds mRNAs that are more abundant in dendrites than in cell bodies . Moreover , the magnitude of CA1 FMRP CLIP scores are predictive of the relative dendritic abundance of its targets ( Figure 4G ) . We examined whether differential transcript isoforms were specifically bound by FMRP in hippocampal CA1 neurons . For example , the Ankrd11 transcript undergoes APA to express a short and long isoform , and only the long isoform is abundant on dendritic ribosomes ( Figure 2I–K ) . Interestingly , CA1 FMRP-CLIP tags were detected on the long , dendritic isoform , but only sparsely on the short isoform ( Figure 5A , gray dashed boxes ) . To look at this phenomenon on a transcriptome-wide scale , we isolated exon junction reads in whole hippocampus CA1 FMRP-cTag-CLIP data . While the length of CLIP tags ( 20–100 nts ) results in a low number of junction reads , we were able to confidently identify FMRP-CLIP tags covering 17 differentially abundant alternative splice events . For example , FMRP binding was largely absent on a shorter , CB-enriched isoform of the Cnksr2 transcript , while robust binding was evident on the longer , dendritic 3′UTR ( Figure 5B , gray dashed boxes ) . Of the 12 exon junction reads that originated from exon 20 of the Cnksr2 transcript , 10 were derived from the long isoform , suggesting that approximately 80% of the FMRP-bound Cnksr2 transcripts derived from the longer , dendritic isoform . This was especially striking since the shorter isoform was the predominant isoform in CA1 pyramidal neurons ( ~80% of exon junction reads in cell body TRAP belonged to the short isoform ) , indicating a high degree of selectivity of FMRP binding to this dendritic isoform ( Figure 5B ) . Globally , we compared the percent spliced in ( PSI ) values for the 17 detected alternative splice events detected in FMRP-CLIP with those in the CA1 cell body and neuropil TRAP data . This revealed that splicing events identified in FMRP-bound mRNAs show stronger correlation with PSI values determined in CA1 neuropil TRAP relative to cell body TRAP ( Figure 5C ) . Taken together , these results indicate that FMRP preferentially binds to specific processed transcripts that are fated for dendritic localization . In order to identify direct FMRP-bound mRNA targets in CA1 dendrites , we crossed FMRP cTag mice with Camk2a-Cre mice , tagging FMRP with GFP specifically in the CA1 pyramidal neurons ( Figure 1A ) . Hippocampal slices from cTag mice were crosslinked , microdissected into cell body and neuropil regions , and subjected to FMRP-CLIP using antibodies against GFP . This allowed purification of FMRP-bound RNA specifically in the CA1 cell bodies or dendrites . Across five biological replicates , we obtained 746 , 827 FMRP CA1-specific CLIP tags from the cell bodies and 80 , 749 tags from CA1 dendrites . Overall , we observed a similar distribution of FMRP CLIP tags across the CDS in these mRNAs and in the two compartments ( Figure 6—figure supplement 1 ) , consistent with prior CLIP analysis and the general observation that FMRP binds CDS to arrest ribosomal elongation ( Darnell et al . , 2011 ) . Combining compartment-specific TRAP and FMRP-CLIP experiments allowed us to determine compartment-specific FMRP CLIP scores for the CA1 cell bodies and dendrites ( Figure 6A , Figure 6—figure supplement 2 , Supplementary file 1G ) . From this , we identified 383 ‘dendritic FMRP targets , ’ defined as mRNAs that are reproducibly bound by FMRP in CA1 dendrites ( Supplementary file 1H ) . Of these dendritic FMRP targets , 60 . 8% ( 233 ) were mRNAs defined in Figure 1 as dendrite-enriched ( Figure 6B ) and 76 . 5% ( 293 ) were dendrite-present ( Figure 6—figure supplement 1 ) . Dendritic FMRP targets show greater relative abundance in ribosome-bound mRNAs ( TRAP ) when compared to all CA1 FMRP targets ( Figure 6C ) . Additionally , when comparing the FMRP-CLIP scores identified previously by whole hippocampus CA1 FMRP-CLIP , the FMRP-CLIP scores for the dendritic FMRP targets were significantly larger than the scores for the full set of dendrite-enriched mRNAs ( Figure 6D ) . These data suggest that dendritic FMRP targets are a subset of previously identified FMRP targets . Interestingly , we identified a number of experimentally defined dendritic FMRP targets that had low levels of whole-cell FMRP cell binding ( i . e . , had negative CA1 FMRP CLIP scores , Figure 6D ) , indicating that these mRNAs are significantly more FMRP-bound in dendrites than in cell bodies . Many directly bound FMRP target transcripts encode proteins that are implicated in autism spectrum disorders ( ASDs ) ( Darnell et al . , 2011; Iossifov et al . , 2012; Zhou et al . , 2019 ) . We hypothesized that FMRP may regulate functional subsets of its targets in a subcellular compartment-specific manner , a phenomenon that would be reflected by differences in compartment-specific FMRP binding . To test this , we segregated all whole-cell CA1 FMRP CLIP targets according to their function by module detection using the HumanBase software ( Krishnan et al . , 2016 ) . Eight functional modules were detected , three of which contained more than 100 genes ( Figure 7A , Supplementary file 1I ) . The FM1 cluster , which contains 393 genes , is highly enriched for genes involved in nuclear regulation of gene expression , with the top GO terms being chromatin organization and modification and histone modification . FM2 ( 292 genes ) is enriched for genes involved in ion transport and receptor signaling . The FM3 cluster ( 203 genes ) contains genes involved in the maintenance of cell polarity and autophagy ( Figure 7B ) . To determine if any of these functional modules might be differentially regulated by FMRP in the dendrites and cell bodies of CA1 neurons , we performed gene set enrichment analysis ( GSEA ) . We estimated enrichment of the FM1-3 transcripts among all FMRP-bound , CA1-expressed transcripts ranked by their dendritic or cell bodies-specific FMRP CLIP score . FM2 and FM3 clusters were highly enriched in FMRP-bound mRNAs in both the dendrites and the cell bodies , while the FM1 cluster was strongly enriched among cell body-bound FMRP targets , but only weakly enriched among the dendritic FMRP-bound transcripts ( Figure 7C ) . This suggests that FM2 and FM3 modules contain mRNAs that are directly bound and regulated by FMRP in dendrites , and the FM1 cluster contains highly bound FMRP targets in the cell bodies , indicating distinct , biologically coherent regulation . We further utilized compartment-specific FMRP-CLIP scores to identify functional modules of ASD candidate mRNAs subject to compartment-specific FMRP regulation ( Figure 7D , Figure 7—figure supplement 1A and B , Supplementary file 1J ) . One module , AM2 , contains transcripts enriched for glutamate signaling , learning , and memory , and is bound by FMRP in both the dendrites and cell bodies . The AM1 module consists of genes involved in chromatin modification and is highly enriched among mRNAs bound by FMRP in the cell bodies , but is not significantly enriched among dendritic FMRP-bound mRNAs . Taken together these observations suggest the possibility of compartmentalized roles for FMRP , in which mRNAs important for synaptic signaling are bound and regulated by FMRP near the synapses , while mRNAs bound by FMRP in the cell bodies are involved in the regulation of neuronal gene expression through chromatin regulation . To better understand FMRP-dependent regulation of dendritic mRNAs , we examined the dendritic ribosome-bound transcriptome in FMRP KO animals . We performed bulk RNA-seq and cell-type-specific TRAP on microdissected hippocampi from WT and FMRP KO littermates . Bulk RNA-seq of microdissected material in FMRP WT and KO mice showed no overall change in the localization of FMRP targets ( Figure 7E , left panel ) . In addition , the mRNAs found to be dendrite-present and dendrite-enriched in KO animals ( as in Figure 1 , Supplementary file 1K ) show large overlap with those in WT animals ( Figure 7—figure supplement 2A and B ) . We validated this finding by FISH in FMRP KO mouse brain slices and found no evidence for altered localization of FMRP targets into the neuropil ( Figure 7—figure supplement 2C–E ) . Global analysis of 3′UTR usage differences in TRAP between dendrites and cell bodies in FMRP KO and WT animals also showed no significant ( FDR < 0 . 05 ) instances of dysregulated localization of 3′UTR isoforms in FMRP KO animals ( Figure 7—figure supplement 3A ) . We validated this finding using FISH in FMRP KO mice , which revealed no differences in isoform localization for Cnksr2 or Anks1b mRNAs ( Figure 7—figure supplement 3B and C ) . Although the identities of the dendritic mRNAs found in FMRP WT and KO mice were similar , quantitative analysis of TRAP revealed that dendritic mRNA levels of ribosome-associated FMRP targets were increased in CA1 dendrites of KO mice ( Figure 7E , right panel ) . Interestingly , this was evident for FM2/3 , but not FM1 transcripts . While FMRP targets are generally downregulated in TRAP from hippocampal neurons ( Sawicka et al . , 2019 ) , a finding that we replicate in cell bodies ( Figure 7—figure supplement 3A ) , transcripts that encode synaptic regulatory proteins ( FM2/3 ) , which are bound by FMRP in the dendrites , show increased ribosome association in CA1 dendrites of KO animals ( Figure 7—figure supplement 3B ) . These results suggest a model in which FMRP differentially regulates translation of functionally distinct mRNAs in specific neuronal compartments ( see model in Figure 7F ) . Much effort has been put into molecular profiling of the localized transcriptome , translatome , and proteome using in vitro neuron or neuron-like cell models ( Zappulo et al . , 2017; Goering et al . , 2020; Middleton et al . , 2019; Taliaferro et al . , 2016 ) . In vivo systems , such as microdissection of hippocampal CA1 regions , offer the advantage of profiling neurons that have formed physiological levels of relevant connections with surrounding neurons . Although RNA-sequencing ( Cajigas et al . , 2012 ) , 3-Seq ( Tushev et al . , 2018 ) , and TRAP-seq ( Ainsley et al . , 2014 ) have been performed previously for microdissected CA1 neuropil , these studies were either not performed using cell-type-specific approaches or unable to capture full-length mRNAs in resting neurons . As the mRNAs presented here are intact and relatively free of contaminating cell types ( Figure 1—figure supplement 1 ) , this dataset can be used for definition of dendritic ribosome-bound mRNAs and for identification of differential usage of 3′UTRs ( Figure 2 ) and alternative splice isoforms ( Figure 3 ) in CA1 neuropil and cell bodies compartments , making it a valuable dataset for the community . Consistent with prior reports ( Tushev et al . , 2018 ) , we find that the majority of differentially localized 3′UTRs are longer than their sequestered counterparts , suggesting that APA events that lead to longer 3′UTR isoforms might allow inclusion of localization and regulatory elements , such as binding sites for RNA-binding proteins or AGO-miRNA complexes . Long 3′UTRs may also act to recruit binding partners for nascent proteins , which can affect the function and/or localization of the protein , as previously reported ( Berkovits and Mayr , 2015 ) . Future experiments analyzing compartment-specific cTag-CLIP of RNA-binding proteins that bind to 3′UTRs such as AGO , Staufen , NOVA1/2 , or ELAVL2/3/4 will provide further insight into the role of these 3′UTRs in mRNA localization and local translation . In addition to a role for 3′UTR-APA in RNA localization and regulation , we find that 20% of differentially localized 3′UTRs result from APA events that impact the CDS . This finding underscores the possibility that differential mRNA localization may be linked to expression of functionally distinct protein isoforms generated during nuclear processing . This is further supported by our observation of AS events that result in differentially localized mRNA alternatively spliced isoforms , which has not been reported previously . We found that NOVA2 , a neuron-specific splicing factor , is responsible for the generation of splicing isoforms that are sequestered to the neuronal cell bodies of CA1 neurons ( Figure 3E ) . NOVA1 and NOVA2 are examples of a relatively small number of mammalian splicing factors demonstrated to directly bind to pre-mRNA and thereby regulate AS ( Licatalosi et al . , 2008; Zhang and Darnell , 2011 ) and also bind 3′ UTRs of those same transcripts ( Eom et al . , 2013 ) . For example , in the case of GlyRa2 , NOVA proteins co-localize with the transcript in the nucleus to regulate exon 3A splicing and in neuronal dendrite ( Racca et al . , 2010 ) . These findings further underscore the many ways in which RNA-binding proteins contribute to neuronal complexity in specific subcellular compartments . The significant overlap between CA1 FMRP targets and dendrite-enriched mRNAs supports literature indicating that FMRP regulates a significant portion of the dendritic transcriptome ( Bagni and Zukin , 2019; Banerjee et al . , 2018; Liu-Yesucevitz et al . , 2011 ) . We do not find a role for FMRP in the localization of its targets in CA1 neurons , as demonstrated in our comparison of FMRP WT and KO brain . However , overall FMRP-binding affinity ( defined by FMRP-CLIP scores in hippocampal neurons ) correlates with relative dendritic abundance of a given mRNA ( i . e . , the enrichment of mRNAs in the neuropil over the cell body , Figure 4G ) , indicating a strong preference for FMRP binding on dendritic mRNA isoforms . This suggests the possibility that FMRP and localized mRNAs may be co-transported into the dendrites . It is also possible that FMRP may play a role in localization of its targets in other neuronal cell types , as has been suggested in radial glia from developing mouse brains ( Pilaz et al . , 2016 ) . Future studies dissecting compartment-specific regulation of FMRP targets in cell-type systems such as these will be very informative . Interestingly , through analysis of whole-cell cTag FMRP-CLIP data , we find multiple instances of FMRP selectively binding to specific dendritic isoforms ( Figure 5 ) . A striking example is the case of the Cnksr2 gene , which generates a short , sequestered mRNA and a longer , highly localized isoform . The protein encoded from the dendritic mRNA isoform contains an additional PDZ-binding domain that is not present in the shorter isoform . Cnksr2 has been identified in genome-wide association studies as an ASD candidate , and mutations in this gene have been shown to cause epilepsy and intellectual disability ( Aypar et al . , 2015 ) . In the cell body compartment , the shorter isoform is predominant , which can be seen by both PAPERCLIP and TRAP . However , FMRP-CLIP , which generally binds the CDS and at least the proximal 3′UTR regions of its targets , shows predominant binding on the 3′UTR of a minor isoform in the presence of a more highly expressed , shorter , sequestered mRNA isoform ( Figure 5B ) . Similar trends can be seen with a number of other mRNAs such as Ankrd11 ( Figure 5 ) and Anks1b ( data not shown ) . Taken together , these data indicate that FMRP can display binding preferences both on different transcripts and different isoforms generated from a single gene . This finding adds an additional layer to the already-complicated process of how FMRP recognizes and binds its targets and suggests that FMRP binding specificity may rely on localization-determining events in the nucleus , such as deposition of RNA-binding proteins on the 3′UTRs of alternatively spliced transcripts . FMRP binding to dendritic mRNA isoforms may also be a result of events that occur in the cytoplasm . For example , some mRNAs with longer 3′UTRs themselves may possess great propensity for entrance into FMRP-containing transport granules due simply to length . This would be consistent with observations that long mRNAs are preferentially found in stress granules due to lower translation efficiency and increased ability for RNA-RNA interactions to form , which are thought to stabilize RNA granules ( Khong et al . , 2017 ) . This suggestion is also supported by findings that FMRP is found in neuronal mRNA transport granules ( Dictenberg et al . , 2008 ) and is known to bind to RNA structural elements such as kissing complexes and G-quadruplexes ( Darnell et al . , 2005 ) , and support suggestions for a role for FMRP in maintaining the translationally repressed status of long mRNAs in transport granules . Although previous work has proposed a role for the interaction between G-quadruplexes and the RGG domain of the FMRP protein in mRNA localization in an in vitro system ( Goering et al . , 2020 ) , we did not find enrichment of G-quadruplexes in the 3′UTRs of dendritic FMRP targets when compared to dendritic non-FMRP targets . This is consistent with previous findings ( Sawicka et al . , 2019; Darnell et al . , 2011 ) that direct FMRP binding occurs primarily on CDS and proximal 3′UTR portions of its targets without observable sequence specificity . This discrepancy could be the result of cell-type-specific functions of FMRP or may indicate that FMRP-directed regulation of G-quadruplex-containing mRNAs is not the result of stable binding of FMRP to these sequences . We present here a list of mRNAs that are highly bound to FMRP in the dendrites of CA1 neurons . These 383 targets are significantly enriched in the dendrites and were found to have high FMRP CLIP scores in whole CA1 neurons ( Sawicka et al . , 2019 , Figure 6 ) , indicating higher than expected FMRP binding relative to other mRNAs of similar transcript abundance . Dendritic FMRP targets were determined by combining compartment-specific CLIP and TRAP experiments to determine compartment-specific FMRP CLIP scores . Importantly , these targets are the result of stringent filtering to include only high-confidence , experimentally defined dendritic FMRP targets . Moreover , we bioinformatically extended our findings using compartment-specific FMRP CLIP scores to identify functional clusters of previously identified FMRP targets that are differentially abundant in the dendrites in respect to CA1 cell bodies . We find a remarkable link between the function of the protein product of a given FMRP target mRNA and its subcellular localization . The FM1 cluster , which contains FMRP target transcripts encoding proteins with nuclear functions such as histone modification and chromosome organization , is enriched in CA1 cell bodies . Approximately 10% of mRNAs encoding chromatin modifiers in CA1 neurons are FMRP targets . In contrast , FM2 and FM3 FMRP target mRNAs , which encode for proteins with synaptic functions such as ion transport , receptor signaling , and cell polarity , are found in both cell bodies and dendrites . Approximately 15–20% of CA1 mRNA encoding synaptic genes are members of the FM2 and FM3 clusters of FMRP targets . Together , these results indicate highly specific FMRP targeting of these two biologically coherent subgroups of targets in two distinct neuronal compartments , suggesting differential translation of chromatin transcripts in the cell body and synaptic-related transcripts in dendrites , and potentially where axonal synapses make contacts in the cell soma . Interestingly , mRNAs from genes in the FM2 and FM3 clusters show increased ribosome association in the FMRP KO mouse in a pattern distinct from the FM1 genes ( Figure 7E ) . Bulk RNA-seq on the same compartments , as well as FISH on FMRP KO mouse brain , showed that overall FMRP targets levels were largely unchanged in abundance or localization in the neuropil . This suggests that in the absence of FMRP , while transcripts that are normally FMRP targets can still be localized to dendrites , they have an increased ribosome association . This supports the proposal ( Wang et al . , 2008 ) that FMRP in the neuronal processes may exist in a polyribosome-depleted granule , which is altered to become translationally competent upon neuronal activity . It is also consistent with the role of FMRP as a translational suppressor and detection of increased basal translation rates in mouse models of Fragile X syndrome ( Gross et al . , 2010; Liu et al . , 2012 ) . Taken together , we suggest a model in which FMRP specifically binds mRNAs that encode synaptic proteins and are fated for dendritic localization and maintains them in a translationally repressed , and potentially polyribosome-depleted state for transport into the processes . Further , within the dendrite , our findings in FMRP-null mice are consistent with a role for neuronal activity to induce polyribosome formation and local translation ( and concomitant increased polyribosome density ) of its specific targets in dendrites . This may be through activity-dependent removal of FMRP from its targets , for example , by dephosphorylation ( Narayanan et al . , 2007 , Figure 7F ) . Future experiments investigating how dendritic FMRP binding changes upon neuronal activity will help to elucidate the precise role of FMRP in regulation of activity-dependent local translation in dendrites . We have previously shown in bulk CA1 neurons that FMRP target mRNAs are destabilized in the absence of FMRP ( Sawicka et al . , 2019 ) . Earlier work also suggested that FMRP targets as a group are downregulated in the absence of FMRP ( Thomson et al . , 2017; Ceolin et al . , 2017 ) , and some evidence can be seen for translational activation of individual mRNAs . Our TRAP in CA1 cell bodies ( Figure 7—figure supplement 4A ) is consistent with this data , and with the proposal that the absence of FMRP leads to an overall decrease in steady-state mRNA abundance of its targets . We detect downregulation of both transcripts encoding chromatin regulators ( FM1 ) and synaptic regulators ( FM2/3 ) in the cell bodies in the absence of FMRP . We hypothesize that FMRP may act to protect mRNAs with stalled ribosomes from degradation , suggesting a role for FMRP in stabilization of translationally stalled mRNAs . A similar model has been proposed following the finding that the abundance of codon-optimized FMRP targets is decreased in FMRP KO ( Shu et al . , 2020; Richter and Zhao , 2021 ) . It is reasonable to suspect that loss of FMRP may lead to an increase in translation in the cell body , as seen in other systems ( Greenblatt and Spradling , 2018 ) ; however , TRAP does not allow for quantitation of ribosome occupancy , so we could not detect these changes using this method . Downregulation of steady-state mRNA levels in cell bodies in the absence of FMRP could also relate to homeostatic feedback on transcription ( Darnell , 2020 ) . However , in the dendrites we suggest that this pathway is either not present or is decreased in steady-state neurons , such that the absence of FMRP is seen as increased ribosome association of FMRP targets encoding synaptic regulators ( FM2/3 ) and thus translational regulation of these mRNAs predominates in dendrites ( Figure 7F ) . An emerging theme in the study of FMRP is that not all targets are regulated in the same manner . Ribosome profiling and RNA-seq in FMRP KO cells in vitro identified distinct groups of FMRP targets whose localization and translation is regulated by the RGG- and KH- domains of the FMRP protein , respectively ( Goering et al . , 2020 ) . Extensive ribosome profiling and RNA-seq in mouse brains showed functionally distinct groups of FMRP targets for which loss of FMRP leads to changes in either mRNA levels or translational efficiency ( Shah et al . , 2020 ) . Our work suggests that subcellular localization of FMRP targets may be a critical factor in these distinct modes of FMRP-mediated regulation . Further , we present three functionally distinct clusters of CA1 FMRP targets and suggest that the cluster that contains chromatin regulators ( FM1 ) are specifically regulated in the cell bodies , whereas synaptic regulators ( FM2/3 ) are regulated in both compartments . In summary , we demonstrate the ability to utilize compartment- and cell-type-specific RNA profiling technologies to precisely define the dendritic transcriptome . Our results underscore the role of FMRP as an important regulator of dendritic mRNAs , playing an important function in ribosome association of isoform-specific dendritic targets and local translational control . This finding , coupled with the identification of FM1 chromatin-associated transcripts regulated by FMRP exclusively in the cell bodies , supports the hypothesis ( Darnell , 2020 ) that FMRP acts as a sensor for neuronal activity through actions on both neuronal transcription and synaptic activity . Further studies into how these subsets of mRNAs are differentially FMRP-regulated in a subcellular compartment-specific manner will have important implications in the understanding of how dysregulation of FMRP and its targets leads to intellectual disability and ASD . All mouse procedures were conducted according to the Institutional Animal Care and Use Committee ( IACUC ) guidelines at the Rockefeller University . RiboTag ( B6N . 129-Rpl22tm1 . 1Psam/J , stock no . 011029 ) and Camk2a-Cre ( B6 . Cg-Tg ( Camk2a-cre ) T29-1Stl/J , stock no . 005359 ) were obtained from Jackson Laboratories . FMRP cTag ( Sawicka et al . , 2019 ) and PABPC1 cTag ( Hwang et al . , 2017 ) mice were previously described . B6 . 129P2-Fmr1tm1Cgr/J ( Fmr1 KO ) mice were a generous gift from W . T . Greenough maintained for multiple generations in our own facilities . Mice were housed up to five mice per cage in a 12 hr light/dark cycle . Breeding schemes for TRAP-seq ( producing RiboTag+/- , Fmr1+/+ , and RiboTag+/- , Fmr1Y/- male littermates ) and FMRP cTag-CLIP ( producing Cre+/-; Fmr1-cTag+/Y male offspring ) were described previously ( Sawicka et al . , 2019 ) . Immunofluorescence was performed as described previously ( Sawicka et al . , 2019 ) . Primary antibodies used were NeuN ( Millipore ABN90P , RRID:AB_2341095 , 1:2000 dilution ) and HA ( Cell Signaling , C29F4 , RRID:AB_1549585 , 1:4000 dilution ) . For each TRAP-seq replicate ( four replicates were performed ) , hippocampi from three adult mice ( 6–10 weeks ) were sectioned into 300 μm slices using a tissue chopper and microdissected in HBSS containing 0 . 1 mg/mL cycloheximide . For microdissection , the CA1 was excised from the hippocampal slices and separated into a cell body ( CB ) and neuropil layer . Microdissected tissue from each mouse was collected and resuspended in 0 . 5 mL ice-cold polysome buffer ( 20 mM HEPES , pH 7 . 4 , 150 mM NaCl , 5 mM MgCl2 , 0 . 5 mM DTT , 0 . 1 mg/mL cycloheximide ) supplemented with 40 U/ml RNasin Plus ( Promega ) and cOmplete Mini EDTA-free Protease Inhibitor ( Roche ) and homogenized by mechanical homogenization with 10 strokes at 900 rpm . NP-40 was added to 1% final concentration and incubated on ice for 10 min . Samples were pooled and centrifuged at 2000 × g for 10 min . Supernatant was subsequently centrifuged at 20 , 000 × g for 10 min . 10% of the resulting lysate was used for RNA-seq , and the remaining lysate was subject to pre-clearing with 1 . 5 mg ( 50 μL ) Protein G Dynabeads for 45 min . HA-tagged ribosomes were collected by indirect IP by adding 40 μg of anti-HA antibody ( Abcam ab9110 , RRID:AB_307019 ) to CB lysate pools and 5 μg to NP lysate pools . IP was performed overnight with rotation at 4°C . Antibody-ribosome complexes were collected by addition of 7 . 2 mg ( CB pools ) or 4 . 44 mg ( NP pools ) Protein G Dynabeads and further incubated with rotation at 4°C for 1 hr . Beads were washed with 1 mL polysome buffer containing 1% NP-40 once for 5 min and twice for 20 min , followed by 4 × 10 min washes in 50 mM Tris pH 7 . 5 , 500 mM KCl , 12 mM MgCl2 , 1% NP-40 , 1 mM DTT , 0 . 1 mg/mL cycloheximide . RNA was extracted from beads by incubating in 500 µL Trizol at room temperature for 5 min . RNA was collected by standard Trizol ( Invitrogen ) extraction via the manufacturer’s protocol and quantified with RiboGreen Quant-IT assays ( Invitrogen ) . Bulk RNA-seq samples were treated with RQ1 RNase-free DNase ( Promega ) prior to library preparation . RNA was further purified for polyadenylated RNA by using Dynabeads mRNA Purification Kit ( Ambion ) . The libraries were prepared by TruSeq RNA Sample Preparation Kit v2 ( Illumina ) following the manufacturer’s instructions . High-throughput sequencing was performed on HiSeq ( Illumina ) to obtain 100 nucleotide paired-end reads . Mice were anesthetized with isoflurane and transcardially perfused with PBS containing 10 U/mL heparin followed by perfusion with ice-cold PBS containing 4% paraformaldehyde . After perfusion , animals were decapitated and intact brains removed and postfixed overnight in 4% paraformaldehyde in PBS at 4°C . Brains were then transferred to PBS with 15% sucrose for 24 hr followed by PBS with 30% sucrose for a further 24 hr and then embedded and frozen in OCT medium . 12 μm coronal slices were prepared using a Leica CM3050 S cryostat and directly adhered to Fisherbrand 1 . 0 mm superfrost slides ( Cat # 12-550-15 ) and stored at –80°C until use . FISH was performed using the RNAscope Multiplex Fluorescent Kit v2 as recommended for fixed frozen tissue , with some exceptions . For pretreatment of samples prior to hybridization , slides were baked at 60°C for 45 min , followed by fixation in 4% paraformaldehyde in PBS at 4°C for 90 min . Samples were dehydrated in ethanol ( 50 , 70 , and 100% twice each ) and incubated at room temperature before hydrogen peroxide treatment for 10–20 min , followed by target retrieval as recommended . After probe hybridization , samples were washed three times for 15 min in wash buffer heated to 37°C . Probes used were conjugated with Alexa fluorescein ( 488 nm ) , Alexa Cyanine 3 ( 555 nm ) , and Alexa Cyanine5 ( 647 nm ) . RNAscope probes were designed to recognize unique 3′UTR sequences ( for UR-APA events ) or for common and distal 3′UTRs ( for UTR-APA events ) with at least 500–1000 nts between regions . See Supplementary file 1B . Each FISH experiment was performed on at least three slices from at least two different mice . Airyscan-Fast ( AS-F ) image capturing was performed using the Zen Black 2 . 3 SP1 FP3 acquisition software on an Inverted LSM 880 Airyscan NLO laser scanning confocal Microscope ( Zeiss ) outfitted with AS-F module ( 16 detectors ) and argon laser for 488 line . Objective: Zeiss Plan 63 × 1 . 4 NA Apochromat oil immersion; imaging at this objective was performed using Immersol 518F immersion media ( ne = 1 . 518 [23°C]; Carl Zeiss ) . Acquisition parameters include laser lines 405 nm , 488 nm , 561 nm , and 633 nm ( laser power adjusted until relative power for each line eliminates as much background as possible without diminishing signal ) . Emission filter for Airyscan detection: 405ch , BP 420–480+ BP 495–620; 488ch , BP 420–480 + 495-550; 561ch , BP 420–480 + 495-620; 633ch , BP 570–620+ LP645 . Settings: eight bit-depth and acquired with image size: 135 . 0 × 135 . 0 µm; pixel size: 0 . 14 µm ( step size is 0 . 159 using a piezo stage ) . All raw image data was sent directly to ZEN 2 . 3 software for reconstruction . Files underwent Airyscan processing ( parameters: auto strength at 6 for 3D images ) before being stitched at a normalized cross-correlation threshold set at 7 . Processed and stitched . czi files were converted to . ims files using Imaris File Converter x64 9 . 6 . 0 before being uploaded into Imaris x64 9 . 6 . Spots were quantified using the spot counting operation ( Imaris software ) with the default values and modifying the spot detection parameters ( ‘Model PSF-elongation along Z-axis’: estimated XY diameter: 0 . 8 μm; estimated Z diameter: 1 . 4 μm ) . Detection threshold was adjusted manually until all false/weak signals were eliminated . The mRNA coordinates ( X , Y , Z ) were downloaded for bioinformatic analysis . Max projections exported from Imaris were uploaded in Fiji . Images were adjusted to eight-bit , orientation is adjusted , and channels are separated . For detection of nuclei for bioinformatic analysis , threshold was adjusted until the majority of the DAPI stain was detected and applied . ‘Analyze particles’ operation was applied with the settings size 50-infinity ( pixel units ) ; circularity 0 . 0–1 . 0; show ‘masks . ’ Resulting text image files were used for downstream analysis . Microdissection of hippocampal slices from 5 to 8 adult Camk2a-FMRP-cTag mice was performed as described above , except that the slices were UV crosslinked in HBSS with 0 . 1 mg/mL cycloheximide three times using 400 mJ/cm2 after sectioning and before microdissection . After dissection , samples were collected and homogenized in lysis buffer ( 1× PBS , 0 . 1% SDS , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate supplemented , 1X cOmplete Mini EDTA-free Protease Inhibitor [Roche] and 0 . 1 mg/mL cycloheximide ) by passing through syringes with a 28 gauge needle . cTag FMRP-CLIP was performed as described previously ( Sawicka et al . , 2019 ) , with minor modifications . Cell body pools were lysed in 1 mL of lysis buffer and neuropil pools in 0 . 5 mL . Pre-clearing was performed with 6 and 1 . 5 mg of Protein G Dynabeads for CB and NP pools , respectively . IP was performed using mouse monoclonal anti-GFP antibodies conjugated to Protein G Dynabeads using 25 μg of each antibody for CB pools and 6 . 25 μg of each antibody for NP pools and rotated at 4°C for 1–2 hr . IPs washes were rotated 2–3 min at room temperature . RNA tags were cloned as described previously ( Sawicka et al . , 2019 ) , with cell bodies and neuropil samples being pooled after barcoding in order to increase yield for low-input samples . Collection and UV crosslinking of microdissected material was performed as described for compartment-specific cTag FMRP-CLIP . cTag-PAPERCLIP was performed as described previously ( Hwang et al . , 2017 ) with the following exceptions . Four replicates were performed , using 3–14 mice per replicate . CB pools were lysed in 1 mL of lysis buffer , NP pools in 0 . 5 mL . Additional IP washes were performed using stringent washes conditions ( described in Sawicka et al . , 2019 ) , and low-input samples were pooled after barcoding . Cell body pools were lysed in 1 mL of lysis buffer and neuropil pools in 0 . 5 mL . IP was performed using mouse monoclonal anti-GFP antibodies conjugated to Protein G Dynabeads using 25 μg of each antibody for CB pools and 6 . 25 μg of each antibody for NP pools and rotated at 4°C for 3–4 hr . RNA tags were cloned as described previously ( Hwang et al . , 2017 ) with cell bodies and neuropil samples being pooled after barcoding in order to increase yield for low-input samples .	The brain has over 100 billion neurons that together form vast networks to relay electrical signals . A neuron receives electrical signals from other neurons via branch-like structures known as dendrites . The signals then travel into the cell body of the neuron . If their sum reaches a threshold , they fire a new signal through a single outgoing projection known as the axon , which is connected to the dendrites of other neurons . A single neuron has thousands of dendrites that each receive inputs from different axons , and it is thought that the strengthening and weakening of these dendritic connections enables us to learn and store memories . Dendrites are filled with molecules known as messenger ribonucleic acids ( mRNAs ) that act as templates to make proteins . Axonal signals reaching the dendrites can trigger these mRNAs to make new proteins that strengthen or weaken the connections between the two neurons , which is believed to be necessary for generating long-term memories . A protein called FMRP is found in both the cell body and dendrites and is able to bind to and regulate the ability of mRNAs to make proteins . A loss of the gene encoding FMRP is the most common cause of inherited intellectual disability and autism in humans , but it remains unclear precisely what role this protein plays in learning and memory . Hale et al . used genetic and bioinformatics approaches to specifically study mRNAs in the dendrites and the cell body of a specific type of neuron involved in memory in mice . The experiments revealed that FMRP played different roles in the dendrites and cell body . In the dendrites , FMRP interacted with mRNAs encoding proteins that can change how the neuron responds to a signal from a neighboring neuron and may alter how strong the connections between the neurons are . On the other hand , FMRP in the cell body modulated the activities of mRNAs encoding proteins that in turn regulate the activities of genes . These findings change the way we think about how memory may work by suggesting that groups of mRNAs encoding proteins with certain activities are found in distinct parts of a single neuron . These observations offer new ways to approach intellectual disabilities and autism spectrum disorder .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience", "genetics", "and", "genomics" ]	2021	FMRP regulates mRNAs encoding distinct functions in the cell body and dendrites of CA1 pyramidal neurons
Due to the uniform cyto-architecture of the cerebellar cortex , its overall physiological characteristics have traditionally been considered to be homogeneous . In this study , we show in awake mice at rest that spiking activity of Purkinje cells , the sole output cells of the cerebellar cortex , differs between cerebellar modules and correlates with their expression of the glycolytic enzyme aldolase C or zebrin . Simple spike and complex spike frequencies were significantly higher in Purkinje cells located in zebrin-negative than zebrin-positive modules . The difference in simple spike frequency persisted when the synaptic input to , but not intrinsic activity of , Purkinje cells was manipulated . Blocking TRPC3 , the effector channel of a cascade of proteins that have zebrin-like distribution patterns , attenuated the simple spike frequency difference . Our results indicate that zebrin-discriminated cerebellar modules operate at different frequencies , which depend on activation of TRPC3 , and that this property is relevant for all cerebellar functions . Resolving structure–function relations remains one of the main challenges of modern neuroscience . The unique cyto-architecture of the cerebellum is characterized by the crystalline matrix of its sagittally oriented PC dendrites and climbing fibers and its orthogonally running parallel fibers ( Larsell , 1972; Voogd , 2011 ) . The ubiquitous nature of this relatively simple matrix throughout all lobules and modules of the cerebellar cortex made scientists predict in 1967 that this neuronal machine was probably the first to be elucidated ( Eccles , 1967 ) . Yet , about half a century later , we have collected a wealth of information about the molecular and physiological identity of the various cell types in the cerebellum ( Gao et al . , 2012 ) , but gross structure–function relations are still largely lacking . For example , the amount of evidence for physiological differences within the cerebellar cortex is limited and there is little comparative analysis of spiking activity throughout the cerebellar cortex in adult awake animals . In fact , even most slice physiology studies do not discriminate between lobules or modules , indicative of the fact that the cerebellum is still considered physiologically homogeneous . At the same time , several molecular markers have been identified that can subdivide the cerebellar cortex into distinct bands ( Apps and Hawkes , 2009 ) . The best-known of these molecules are the zebrins , which are highly expressed by specific bands of Purkinje cells ( PCs ) , that is the sole output of the cerebellar cortex . Immunostainings for both zebrin I and II give rise to symmetric stripes that are oriented perpendicular to the cerebellar folds ( Brochu et al . , 1990; Leclerc et al . , 1990 ) . The combined presence of zebrin-positive and zebrin-negative PCs can be found in all vertebrate classes , and zebra-like patterns are present in the cerebellum of birds and mammals , varying from pigeons and mice up to monkeys and humans ( Brochu et al . , 1990; Sillitoe et al . , 2003; Marzban and Hawkes , 2011; Graham and Wylie , 2012 ) . In most cases , adjacent PCs with zebrin II ( from hereon referred to as zebrin ) are located in the same bands , receive CF inputs from the same part of the inferior olive , and project their axons to the same part of the cerebellar nuclei ( Voogd and Ruigrok , 2004; Pijpers et al . , 2006; Sugihara and Shinoda , 2007; Apps and Hawkes , 2009; Sugihara et al . , 2009; Sugihara , 2011 ) . Moreover , although their various terminal rosettes may be located in different parts of the cerebellar cortex , individual mossy fibers often also adhere to the same zebrin signature ( Pijpers et al . , 2006 ) . As such zebrin may be regarded as a biomarker linking different cerebellar cortical zones , potentially binding activity of different olivo-cerebellar modules and mossy fiber systems ( Voogd and Ruigrok , 2004; Pijpers et al . , 2006; Ruigrok , 2011 ) . However , what the basic characteristics of this zebrin-related activity might be is unknown . Since zebrin has been identified as the glycolytic enzyme aldolase C , its presence might in principle be linked to the level of metabolic and/or electrophysiological activity . Indeed , the distribution of zebrin in the cerebellum is similar to that of the excitatory amino acid transporter 4 ( EAAT4 ) and complementary to splice variant b of the metabotropic glutamate receptor 1 ( mGluR1b ) ( Dehnes et al . , 1998; Mateos et al . , 2001; Wadiche and Jahr , 2005 ) . Intracellularly , several proteins in a molecular cascade linked to mGluR1 are also expressed in zebrin-like bands , including the IP3-receptor ( Furutama et al . , 2010 ) , PLCβ3/4 ( Sarna et al . , 2006 ) , PKCδ ( Barmack et al . , 2000 ) , and NCS-1 ( Jinno et al . , 2003 ) . This cascade controls the activity of the transient receptor potential cation channel type C3 ( TRPC3 ) ( Hartmann et al . , 2008 ) , which in turn can influence the firing activity of PCs ( Sekerkova et al . , 2013 ) . We hypothesized that differential activity of this cascade of proteins with zebrin-related expression might lead to differential activity of their effector channel , TRPC3 , and thereby to differences in simple spike ( SS ) firing frequency between modules ( Kim et al . , 2012a , 2012b ) . To test this hypothesis , we investigated the activity of PCs in awake mice at rest in relation to the zebrin-identity of their module . We demonstrate that there are zebrin-related differences in firing frequency of both SSs and complex spikes ( CSs ) , that these differences are intrinsically driven , and that they are consistently present throughout the cerebellar cortex contributing to all its functions . We performed extracellular recordings from PCs in the cerebellar cortex of awake , restrained C57Bl/6 mice at rest with the use of double-barrel electrodes marking the recording location with Alcian Blue ( Figure 1A , B ) . Purkinje cells were identified by the presence of SSs and CSs and the consistent presence of a pause in SS firing after a CS ( i . e . , climbing fiber pause ) ( Figure 1C–F ) . Recordings that were used for analysis had to meet several criteria including a minimum duration of 120 s , stable spike amplitude over the whole recording period and no detectable tissue damage in a 400-µm radius ( see also Figure 1—figure supplement 1 ) . Following perfusion of the animals and processing of their cerebella , the zebrin-negative ( Z− ) and zebrin-positive ( Z+ ) zones were identified by immunostaining . Of the 104 PCs included in the analysis ( 50 mice ) , 47 and 57 cells were located in Z− and Z+ zones , respectively ( Figure 1G , plot adapted from Sugihara and Quy , 2007 ) . The SS firing frequency was significantly higher in Z− zones than in Z+ zones ( Z−: 96 . 1 ± 15 . 4 Hz , Z+: 61 . 4 ± 19 . 3 Hz , t = 9 . 942 , p<0 . 001 ) ( Figure 1H ) . In line with this the climbing fiber pause was also significantly longer ( t = −7 . 482 , p<0 . 001 ) in Z+ zones ( Figure 1I ) ( see also Paukert et al . , 2010 ) . Both the SS firing frequency , SS regularity and CS firing frequency were stable over time ( Figure 1—figure supplement 2 ) . In contrast to the SS firing frequency and climbing fiber pause , the waveform and regularity of SSs did not consistently depend on zebrin identity in that average half-width and mean coefficient of variation for adjacent intervals ( CV2 ) were not significantly different between Z− and Z+ PCs ( half-width: t = −1 . 133 , p=0 . 260 , data not shown; CV2: t = 1 . 197 , p=0 . 234 ) ( Figure 1F–J ) . 10 . 7554/eLife . 02536 . 003Figure 1 . Simple spike firing activity differs between Purkinje cell populations . ( A ) Extracellular recordings were made from PCs in the cerebellar cortex of awake mice , using double barrel glass electrodes ( right ) . Dye injections were placed to histologically identify the recording location . ( B ) Photomicrographs of coronal sections with examples of zebrin-negative ( Z− , left ) and zebrin-positive ( Z+ , right ) identified Purkinje cells in lobule II and lobule IX , respectively . Cells are marked by dye injections ( blue , indicated by arrows ) , zebrin is stained brown , dotted lines demark zebrin borders . Note that Z+ stripes in lobules I–III are very narrow . ( C and D ) Example trace of Z− and Z+ Purkinje cell recordings identified by its hallmark feature , the occurrence of complex spikes ( asterisk ) and simple spikes . ( E ) Recordings were confirmed to be from a single neuron by the consistent pause in simple spike firing following each complex spike , in the overlay . ( F ) Overlay of simple spikes . ( G ) Distribution of recorded Z− and Z+ cells throughout the unfolded cerebellar cortex based on zebrin II compartments . ( H ) Simple spike firing frequency is significantly lower in identified Z+ PCs compared to Z− PCs ( Z−: n = 47 cells , 26 mice; Z+: n = 57 cells , 34 mice; t = 9 . 942 , p<0 . 001 ) . ( I ) In line with the lower simple spike firing frequency , the climbing fiber pause was longer in identified Z+ Purkinje cells ( CF pause; t = −7 . 482 , p<0 . 001 ) . ( J ) Simple spike regularity is not different between Z+ and Z− PCs ( CV2: t = 1 . 147 , p=0 . 234 ) . Error bars represent SD , p<0 . 05 , p<0 . 001 . Schematic drawing in A was adapted from Sugihara and Quy ( 2007 ) with permission . Scale bars in A and B indicate 100 and 200 µm , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 00310 . 7554/eLife . 02536 . 004Figure 1—figure supplement 1 . Experimental approach and histological verification . ( A ) To obtain recordings from all cerebellar areas in vivo , we placed craniotomies in different locations . We avoided recording close to the craniotomy to decrease the chance of having to exclude the cells due to tissue damage ( B ) . Five different approaches ( top row ) were used to cover all cerebellar areas , and recordings were done in a straight line running in either mediolateral or rostrocaudal direction , to assure reliable retrieval after histology . ( B ) To minimize the potential influence of tissue disruption on the recording results , we excluded all Purkinje cells that showed clear damage to the tissue , visible by light microscope , in a 400 µm radius circle around the dye identified recording location . Cells without detectable damage in surrounding tissue ( left ) were included , and those with clear damage ( right ) were excluded . Also , we did not observed obvious cell death in the Z-areas ( Welsh et al . , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 00410 . 7554/eLife . 02536 . 005Figure 1—figure supplement 2 . Stability of key parameters over the recording time . As a measure of recording stability , the key parameters simple spike firing frequency and mean CV2 , and complex spike firing frequency were analyzed in bins of 30 s . ( A ) Minimum duration was 120 s for all Purkinje cell recordings . For clarity , recordings were divided based on zebrin-identity . ( B ) Normalized values were obtained by dividing for each cell the absolute values binned per 30s in A by the average for the entire recording of that cell . Note that the deviation from average was typically <20% over 30 s bins . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 005 Due to the heterogeneous distribution of Z+ vs Z− Purkinje cells over the cerebellar cortex , the majority of the Z+ cells were recorded in the posterior half , whereas the Z− cells were predominantly from the anterior half . Hence , one could argue that the difference between Z+ and Z− is related to location , rather than directly linked with zebrin identity . Re-plotting the results , but now color-coded for simple spike frequency to facilitate individual comparisons , seems to largely contradict this possibility ( Figure 2—figure supplement 1 ) . To more thoroughly test our hypothesis that differences are indeed related to zebrin identity , we also attempted to record neighboring , online identified , Z+ and Z− PCs in a single experiment . To this end , we performed two-photon imaging in vivo in awake , head-fixed mice that express enhanced green fluorescent protein ( eGFP ) under the EAAT4 promoter in a pattern similar to that of zebrin ( Gincel et al . , 2007 ) . In the dorsal layer of lobule V , VI , and Crus I we identified Z+ and Z− bands and recorded PCs in adjacent zebrin bands ( Figure 2A , B ) . In line with our hypothesis , we observed higher simple spike activity in Z− than in Z+ Purkinje cells ( Z+: 36 . 0 ± 15 . 5 Hz , n = 8; Z−: 75 . 8 ± 19 . 5 Hz , n = 9; t = 4 . 618 , p<0 . 001 ) and concommitant longer climbing fiber pauses ( Figure 2C ) . In contrast to the immunohistochemically subdivided PC dataset ( Figure 1 ) , that covers the entire cerebellar cortex , this spatially restricted dataset did show a difference in simple spike regularity , suggesting that variations in regularity may occur more locally . 10 . 7554/eLife . 02536 . 006Figure 2 . Simple spike firing frequency correlates with the zebrin identity of Purkinje cells . To determine of the differences in simple spike activity are related to the location of the Purkinje cells , or to their zebrin identity , we compared PC activity of Z+ against Z– Purkinje cells in various smaller areas of the cerebellar cortex . ( A ) To more directly test the link with zebrin , we used EAAT4-eGFP mice that express eGFP in a pattern similar to that of zebrin . Two-photon images show an EAAT4+/Z+ band ( green ) in lobule V of an EAAT4-eGFP mouse: left , electrode ( blue ) positioned in the adjacent negative band; right , electrode ( blue ) in the positive band . ( B ) The activity of 17 zebrin-identity determined PCs ( Z+ , n = 8; Z– , n = 9 , 5 mice ) from lobule V , VI , and Crus I was recorded . ( C ) The difference in simple spike firing frequency was pertained in this subset of Purkinje cell recordings ( Z+: 36 . 0 ± 15 . 5 Hz; Z−: 75 . 8 ± 19 . 5 Hz; t = 4 . 618 , p<0 . 001 ) , indicating that this difference is linked to zebrin identity , rather than lobular location . In contrast to data obtained with immunostaining for zebrin , the regularity of simple spikes also differs in this subpopulation ( t = −2 . 715 , p<0 . 016 ) . ( D ) Cerebellar Purkinje cells can be subdivided based on the input they receive into four transverse zones: the anterior ( red ) , central ( orange ) , posterior ( yellow ) , and nodular ( green ) zone . ( E ) The difference in simple spike firing frequency between Z+ and Z– Purkinje cells is consistently present throughout all transverse zones . In each of the four transverse zones , the simple spike rate was significantly lower in Z+ compared to Z– Purkinje cells ( all p<0 . 05 , One-tailed Student's t test ) . Note that simple spike frequency within different Z+ subgroups was also variable , in that the frequency in the anterior zone was lower than that in the nodular zone ( p=0 . 018 , One-way ANOVA , followed by Bonferroni's posthoc test ) . Error bars represent SD , p<0 . 05 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 00610 . 7554/eLife . 02536 . 007Figure 2—figure supplement 1 . Overview with color-coded simple spike frequency for all identified Z+ and Z− Purkinje cells . Overview plot of the unfolded cerebellar cortex depicting all Purkinje cells of which the activity was recorded and the zebrin identity was determined using either immunostaining or two-photon imaging , with dot color indicating the simple spike firing frequency . Dot color ranges from the combined average simple spike firing frequency for Z+ Purkinje cells , 58 . 3 Hz , in green to that of Z– Purkinje cells , 92 . 9 Hz , in red . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 007 Finally , to extend this analysis over the entire cortex , we compared Z+ vs Z− PC activity per transverse zone . Along the rostro-caudal axis the cerebellum can be subdivided into four transverse zones: the anterior , central , posterior , and nodular zone ( Ozol et al . , 1999 ) . We consistently observed a similar difference in simple spike activity between Z+ and Z− PCs in each zone , independent of the location within the cerebellar cortex ( Figure 2D–E ) . This approach also revealed a difference within the population of Z+ PCs . Whereas the simple spike firing frequency of Z− PCs is comparable over different transverse zones , Z+ PCs firing rate is lower in the anterior zone when compared to the nodular zone ( p=0 . 018 , One-way ANOVA followed by Tukey's post-hoc test ) . If the SS activity of PCs depends on the presence of zebrin , one should also observe differences between lobules , as there is a gradual increase in zebrin-positive modules and , thus , average zebrin intensity from lobule I to lobule X in the vermis and the corresponding lobules in the hemispheres ( Sugihara and Shinoda , 2004 ) ( Figure 1G , Figure 3A ) . Indeed , when we extend the immunohistochemically analyzed dataset with recordings from all lobules in which the zebrin identity was not determined to generate one large , randomly sampled dataset ( combined n = 245 ) , our prediction is confirmed . Both the firing frequency and climbing fiber pause of SS activity , among the different lobules in the vermis and the hemispheres , show robust and consistent correlations with the averaged intensity of zebrin staining ( for firing frequency , vermis: r = 0 . 893 , p=0 . 007; hemisphere: r = 1 . 000 , p<0 . 001; for climbing fiber pause , vermis: r = −0 . 929 , p=0 . 003; hemisphere: r = −1 . 000 , p<0 . 001 ) ( Figure 3A–C , Figure 3—figure supplement 1 ) . In contrast , the CV2 of SSs could not be consistently related to the zebrin intensity in the vermis and hemispheres ( vermis: r = 0 . 929 , p=0 . 003; hemisphere: r = −0 . 300 , p=0 . 624 ) ( Figure 3D ) . 10 . 7554/eLife . 02536 . 008Figure 3 . Zebrin staining intensity and simple spike frequency are inversely correlated . To test the correlation to zebrin identity of modules throughout the cerebellar cortex , Purkinje cell activity was recorded from all parts of the cerebellar cortex , each followed by dye injection to identify the lobule . ( A1–2 ) The average zebrin staining intensities of Purkinje cell somata in vermis and hemispheres were obtained from the sagittal sections of three mice . Note that high intensity values equal weak staining , and vice versa . ( B1–2 and C1–2 ) The average simple spike firing frequency ( vermis: n = 192 cells , 70 mice , r = 0 . 893 , p=0 . 007; hemisphere: n = 53 cells , 30 mice , r = 1 . 000 , p<0 . 001 ) and CF pause ( vermis: r = −0 . 929 , p=0 . 003; hemisphere: r = −1 . 000 , p<0 . 001 ) show significant correlation with zebrin intensity over different parts of vermis and hemisphere . ( D1–2 ) The CV2 of SSs could not be consistently related with zebrin intensity ( vermis: r = 0 . 929 , p=0 . 003; hemisphere: r = −0 . 300 , p=0 . 624 ) . Error bars represent SD . HIV&V-Sim , hemispheral part of lobule IV&V and simple lobule; CrI-II , Crus I and II; PM , paramedian lobule; Cop-PF , copula of the pyramis and paraflocculus; Flocc , flocculus . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 00810 . 7554/eLife . 02536 . 009Figure 3—figure supplement 1 . Statistical analysis of PC spiking characteristics per lobule . Statistical comparisons of all recorded Purkinje cells , as presented in Figures 3 and 4C . These comparisons include all Purkinje cells from the vermis and hemispheres of which the location was determined based on dye injection ( one-way ANOVA , followed by Bonferroni's posthoc test ) . FF , firing frequency; CFP , climbing fiber pause . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 009 Reduced tonic SS activity of PCs at rest , as observed in Z+ modules , will lead to enhanced activity of the GABAergic neurons in the cerebellar nuclei that inhibit inferior olivary neurons ( Chen et al . , 2010; De Zeeuw et al . , 2011 ) . Therefore , one can expect the CS activity that results from activity in the climbing fibers originating in the inferior olive to be reduced as well . This prediction indeed holds ( Figure 4A ) . The CS activity of immunostaining identified Z+ PCs was significantly lower than that in Z− PCs ( same PCs as described in Figure 1G–J; Z−: 1 . 13 ± 0 . 25 Hz , Z+: 0 . 92 ± 0 . 28 Hz , t = 3 . 926 , p<0 . 001 ) . This difference persisted in the subset of two-photon imaging identified Z+ and Z− PCs recorded in lobule V–VI and Crus I , supporting the link to zebrin-identity , rather than cortical location ( Figure 4B ) . Moreover , the gradual trend that we observed for SS firing frequency , but not for CV2 , in the different lobules in both the vermis and hemispheres was also observed for CS activity ( Figure 4C , Figure 3—figure supplement 1 ) . Since climbing fibers evoked prolonged EPSCs in Z+ Purkinje cells ( Paukert et al . , 2010 ) , we also investigated the half-width of the first upward deflection in potential and the integrated deviation of the CS potential from zero . Both parameters were significantly higher in immunostaining identified Z+ PCs ( half-width: t = −3 . 269 , p=0 . 001 , spike area: t = −2 . 523 , p=0 . 013 ) ( Figure 4D ) . Given that SS activity and the wave of CS activity correlated with zebrin , the distribution of post-CS configurations of SS activity might in principle also be affected ( Simpson et al . , 1996 ) . Based on the peri-CS time histograms , we could distinguish four different types of SS responses following the climbing fiber pause . These included a neutral pattern ( i . e . , normal type ) , a pattern with increased SS activity ( i . e . , facilitation type ) , and two patterns with decreased SS activity , one without and one with a superimposed oscillatory effect ( i . e . , suppression and oscillation type , respectively ) ( Figure 4E ) . Thus , if there is a relation with zebrin expression , one could predict that the facilitation type of cells prevail in the Z− zones , whereas the suppression and oscillation type of cells occur predominantly in the Z+ zones . This prediction did hold . Even though the normal type dominated in both Z− and Z+ PCs , the suppression and oscillation types only occurred in Z+ PCs . The facilitation type occurred in both Z− and Z+ PCs , but significantly more in the Z− areas ( Z−: 17 /47 , Z+: 6/57; χ2 = 9 . 835 , p=0 . 002 , Pearson's Chi-squared test ) ( Figure 4F ) . Attempts to find other parameters correlating with the response type were largely unsuccessful , except for the oscillation type , which showed a combination of low SS frequency and low CV ( Figure 4G ) . 10 . 7554/eLife . 02536 . 010Figure 4 . Complex spike characteristics depend on zebrin identitiy . ( A ) Similar to simple spike frequency , complex spike frequency differs between immunostaining identified Z+ and Z− PCs ( data from zebrin-identified PCs shown in Figure 1; t = 3 . 926 , p<0 . 001 ) , ( B ) This difference is confirmed in the sample of Z+ and Z− PCs obtained by two-photon imaging in EAAT4-eGFP mice , in that Z+ Purkinje cells have a lower complex spike firing frequency here too ( t = 2 . 692 , p=0 . 017 ) . C , Moreover , complex spikes frequency shows significant correlation with zebrin intensity in vermis and hemisphere ( vermis: r = 0 . 929 , p=0 . 003; hemisphere: r = 1 . 000 , p<0 . 001 ) . Even though the regularity of CSs differs between immunostaining identified Z− and Z+ PCs ( A , bottom ) , this was not reproduced in the other two experimental data sets ( B–C , bottom ) . ( D ) Typical Z− and Z+ CS shapes ( −0 . 5 to +3 ms ) showing the characteristics analyzed: half-width and spike area ( left ) . Z+ PCs have a longer half-width and bigger spike area than Z− cells ( right ) . ( E ) Raster plots of simple spike activity around complex spikes ( event , −100 ms till +300 ms ) were converted in peri complex-spike time histograms . Based on these histograms , we could distinguish four different types of simple spike response types among the Purkinje cells recorded in all areas: normal , facilitation , suppression and oscillation . ( F ) The percentage of different types in Z− and Z+ PCs ( values indicate percentage ) . The facilitation type occurs predominantly in Z− PCs , whereas the suppression and oscillation type are restricted to the Z+ PCs . ( G ) Attempts to find other parameters correlating in all recorded cells ( n = 243 cells ) with the response type were largely unsuccessful . The exception is the oscillation type , which has a signature combination of simple spike frequency and CV ( 11 out of 13 , SS freq . range 35–60 Hz and CV <0 . 32 ) . Two-photon imaging data are only included in panel B; panels D–F are based on immunostaining identified Z+ and Z− PCs only and panels C and G on all recorded PCs . Error bars represent SD , p<0 . 05 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 010 In general , SS activity of PCs results from the integration of their excitatory input , inhibitory input , and intrinsic pace-making activity ( De Zeeuw et al . , 2011 ) . This raises the question as to what extent the difference in SS firing frequency between Z+ and Z− PCs results from differences in input or intrinsic activity . We used two approaches to tackle this question . First , we completely removed the impact of external inputs onto the PCs using blockers for AMPA , NMDA , and GABAA receptors during cell-attached recordings from sagittal slices ( Figure 5A ) . The average SS firing frequency was , on average , 22 ± 7% lower over all lobules in vitro than that in vivo indicating that the larger part of SS activity is internally driven by PCs . The dominant impact of intrinsic PC activity was also reflected by the finding that the differential firing frequency pattern over all lobules in vitro correlated with that in vivo ( r = 0 . 916 , p=0 . 010 , Pearson's correlation ) ( Figure 5B ) . For example , the in vitro SS firing frequency of PCs in lobules III , which are predominantly Z− , was alike the in vivo recordings significantly higher than that in lobule X , where PCs are Z+ ( t = 2 . 844 , p=0 . 007 ) . This higher firing frequency in lobule III was associated with a higher intrinsic excitability in lobule III compared to lobule X , reinforcing the interpretation that the difference is predominantly intrinsic to Purkinje cells ( Figure 5—figure supplement 1 ) ( see also Kim et al . , 2012a ) . To confirm that under these in vitro conditions , without excitatory or inhibitory input , the difference in simple spike firing frequency still correlates with zebrin identity , we also recorded the activity of fluorescence-identified Z+ and Z− PCs in adjacent bands in slices from the EAAT4-eGFP mice . Comparison of sets from lobules II–V and lobule VIII–IX confirmed this presumption ( II–V: t = 2 . 910 , p=0 . 017; VIII–IX: t = 2 . 352 , p=0 . 043 ) ( Figure 5C ) . 10 . 7554/eLife . 02536 . 011Figure 5 . Zebrin-related differences are present in the intrinsic activity of Purkinje cells . To test if intrinsic or input-related differences underlie the difference in simple spike frequency , we recorded PC activity in conditions of limited or no synaptic input . ( A ) PC activity was recorded in vitro ( n = 107 cells , 15 mice ) under complete block of synaptic inputs . ( B ) Spiking frequency in vitro ( red ) was lower than that in vivo ( black ) over the range of lobules , but the shape of the curve was similar ( r = 0 . 916 , p=0 . 010 , Pearson's correlation ) . ( C1–3 ) To verify the correlation with zebrin , we recorded activity of EAAT4/zebrin-positive and negative PCs in slices of EAAT4-eGFP mice . Both in lobules II–V ( Z+: n = 7 cells , Z−: n = 4; 3 mice; t = 2 . 910 , p=0 . 017 ) and lobules VIII-IX ( Z+: n = 6 , Z−: n = 5; 2 mice; t = 2 . 352 , p=0 . 043 ) the difference in simple spike firing frequency was present , further confirming the link with zebrin . ( D ) Next , extracellular recordings were made in vivo in a6-Cacna1a and PC-Δγ2 mutant mice that have minimized excitatory and no synaptic inhibitory inputs to their PCs , respectively . ( E ) PC activity in Z+ lobule X of both mutants was lower than that in the predominantly Z− lobules I–III ( wild types , lobules I–III: n = 43 cells , 18 mice , lobule X: n = 32 cells , 25 mice , t = 6 . 808 , p<0 . 001; a6-Cacna1a , I–III: n = 16 cells , 2 mice; X: n = 11 cells , 2 mice; t = 3 . 979 , p<0 . 001; PC-Δγ2 , I-III: n = 11 cells , 3 mice; X: n = 17 cells , 3 mice; t = 4 . 876 , p<0 . 001 ) . Inset compares the absolute differences in firing frequency between lobules I–III and X . ( F ) CV2 values of Z− and Z+ SS activity from in vitro recordings ( lobules I–III and X: both p<0 . 001 ) and in vivo recordings of both a6-Cacna1a mutants ( lobule I–III: t = 5 . 613 , p<0 . 001; lobule X: t = 2 . 062 , p=0 . 046 ) and PC-Δγ2 mutants ( lobules I–III and X: both p<0 . 005 ) were significantly lower than the wild type recordings . Abbreviations: cf , climbing fiber; GC , granule cell; IO , inferior olive; MLI , molecular layer interneuron; PC , Purkinje cell; pf , parallel fiber . Error bars represent SD , p<0 . 05 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 01110 . 7554/eLife . 02536 . 012Figure 5—figure supplement 1 . Purkinje cell intrinsic excitability is higher in lobule III than in X . To test intrinsic excitability Purkinje cells were current clamped at −65 mV , 1 s current steps were injected increasing from 0 . 1 to 1 . 0 nA and the response frequencies were determined . ( A ) Example traces with insets ( green ) showing a magnification of the last 100 ms . ( B ) Response frequency was determined for the entire range , and the slope was calculated . The slope of the input–output curves was significantly higher in Purkinje cells from lobule III compared to lobule X ( III: mean ± SEM 226 . 0 ± 14 . 5 spk/nA , n = 7; X: 124 . 1 ± 20 . 6 spk/nA , n = 9; t = 4 . 311 , p<0 . 001 ) . Error bars denote SEM , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 012 To further assess the impact of excitatory and inhibitory inputs , we investigated the in vivo SS activity in mouse mutants , in which either the glutamatergic input ( a6-Cacna1a mutants ) or the GABAergic input ( PC-Δγ2 mutants ) to PCs was affected . The a6-Cacna1a mutants are characterized by a silenced parallel fiber output in the vast majority of their granule cells due to a lack of voltage-gated calcium channels required for neurotransmission ( Galliano et al . , 2013 ) , while the PC-Δγ2 mutants are characterized by the absence of synaptic inhibition from the molecular layer interneurons through ablation of the γ2 subunit of the GABAA-receptor in PCs ( Wulff et al . , 2009; Figure 5D ) . In both a6-Cacna1a and PC-Δγ2 mutant mice the differences in vivo in SS firing frequency between lobules I–III and lobule X were still significant , analogous to that in normal mice ( a6-Cacna1a , I-III: 75 . 1 ± 19 . 0 Hz , X: 50 . 2 ± 10 . 2 Hz , t = 3 . 979 , p<0 . 001; PC-Δγ2 , I-III: 89 . 8 ± 14 . 9 Hz , X: 60 . 9 ± 15 . 6 Hz , t = 4 . 876 , p<0 . 001 ) ( Figure 5E ) . In contrast , the CV2 values of SS activity were significantly reduced not only in vitro , but also in vivo in both a6-Cacna1a and PC-Δγ2 mutants as compared to wild-types ( Figure 5F ) . These differences held true for lobules I–III ( in vitro: t = 32 . 647; a6-Cacna1a: t = 5 . 613 , p<0 . 001; PC-Δγ2: t = 3 . 068 , p=0 . 003 vs in vivo wild-types ) , as well as for lobule X ( in vitro: t = 14 . 593 , p<0 . 001; a6-Cacna1a: t = 2 . 062 , p=0 . 046; PC-Δγ2: t = 3 . 292 , p=0 . 002 ) . Together , these data suggest that the SS firing frequency is largely determined by intrinsic properties of PCs , whereas the level of regularity appears to be predominantly determined by external inputs . The finding that the difference in firing frequency between Z+ and Z− PCs must predominantly reflect their different intrinsic properties raises the question whether PC proteins other than zebrin also play a mechanistic role . This may be particularly relevant as zebrin , or aldolase C , is a glycolytic enzyme and probably plays a secondary role via energy consumption without a direct impact on the electrophysiological properties of PCs . In fact , when the products of aldolase C , that is glyceraldehyde-3-phosphate ( GAP ) and dihydroxyacetone phosphate ( DHAP ) , were introduced to the ACSF in our in vitro recordings , SS firing increased in both the largely zebrin-negative lobule III and zebrin-positive lobule X ( Figure 6—figure supplement 1 ) , arguing against the possibility that aldolase C's enzymatic function directly contributes to a lower SS firing frequency in Z+ PCs . Hence , we shifted our focus to TRPC3 , which can be associated with zebrin-negative PCs ( Mateos et al . , 2001; Hartmann and et al , 2008; Kim et al . , 2012b ) , and underlies the mGluR1-mediated slow EPSCs ( Hartmann and et al , 2008 ) and mGluR1-agonist ( DHPG ) -induced currents ( Nelson and Glitsch , 2012 ) , that have been shown to affect SS activity even in the absence of synaptic input ( Yamakawa and Hirano , 1999; Coesmans et al . , 2003; Chanda and Xu-Friedman , 2011 ) . We first tested the effect of blocking TRPC3 on the activity of PCs in vitro in lobules III and X , in the absence of synaptic input , using two blockers , genistein and Pyr3 ( Kim et al . , 2012b; Kiyonaka et al . , 2009 ) . Both TRPC3 blockers had a significant impact on PC activity reducing the firing frequency in lobule III ( genistein , p<0 . 001; Pyr3 , p<0 . 001 vs vehicle control , one-way followed by Tukey's post-hoc test ) without a significant effect in lobule X ( p=0 . 271 and p=1 . 000 vs vehicle respectively , one-way ANOVA followed by Tukey's post-hoc test ) ( Figure 6A , B ) , an effect that is in line with that of blocking mGluR1 ( Yamakawa and Hirano , 1999; Coesmans et al . , 2003; Chanda and Xu-Friedman , 2011 ) . To more directly compare the effect of blocking TRPC3 between lobule III and X , we recorded PC activity during wash-in of the blockers ( Figure 6C ) . Wash-in of Pyr3 had a robust effect on SS firing frequency of PCs in lobule III ( pre: 49 . 9 ± 7 . 9 Hz; post: 25 . 5 ± 9 . 9 Hz; t = 5 . 412 , p=0 . 002 , paired Student's t test ) and this effect was significantly larger than that in lobule X ( reduction , lobule III: 48 . 2 ± 19 . 7%; lobule X: 8 . 5 ± 16 . 6%; t = 4 . 069 , p=0 . 002 ) ( Figure 6D , E ) . 10 . 7554/eLife . 02536 . 013Figure 6 . Blocking TRPC3 attenuates the simple spike frequency difference . In search for the underlying mechanism , we tested the contribution of TRPC3 , which can be indirectly linked to zebrin-like expression . ( A and B ) The presence of TRPC3 blocker genistein ( 10 µM ) or the more selective Pyr3 ( 100 µM ) reduced Purkinje cell firing frequency in lobule III ( vehicle: n = 47 cells , 6 mice; genistein: n = 25 cells , 7 mice; Pyr3: n = 33 cells , 7 mice; both p<0 . 001 vs vehicle , One-Way ANOVA followed by Tukey's post-hoc test ) , but not in lobule X ( vehicle: n = 48 cells , 6 mice; genistein: n = 44 cells , 7 mice; Pyr3 , n = 24 cells , 7 mice; p=0 . 271 and p=1 . 000 vs vehicle , respectively , One-Way ANOVA followed by Tukey's post-hoc test ) , virtually eliminating the difference between averages for lobule III–X ( inset ) . To more directly quantify the effect of blocking TRPC3 , we washed-in Pyr3 during the recording of Purkinje cells in lobule III and X . ( C–E ) Pyr3 wash-in significantly decreased the simple spike firing frequency in lobule III ( n = 7 cells , 7 mice; t = 5 . 412 , p=0 . 002 , paired Student's t test ) , and this decrease was larger in lobule III than in lobule X ( t = 4 . 069; p=0 . 002 ) . ( F–G ) In line with the in vitro data , in vivo blocking of TRPC3 by application of genistein ( 240 mg/kg , i . p . ) or Pyr3 ( 200 µg , i . c . v . ) decreased simple spike firing in lobule I–III ( vehicle: n = 27 cells , 3 mice; genistein: n = 37 cells , 3 mice and Pyr3: n = 30 cells , 2 mice; both p<0 . 001 vs vehicle , one-way ANOVA followed by Tukey's post-hoc test ) , but had no effect in lobule X ( vehicle: n = 32 cells , 3 mice; genistein: n = 23 cells , 4 mice and Pyr3: n = 31 cells , 4 mice; p=0 . 546 and p=0 . 887 vs vehicle , respectively one-way ANOVA followed by Tukey's post-hoc test ) , resulting in a pronounced reduction of the difference ( inset ) . Error bars represent s . d . , p<0 . 05 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 01310 . 7554/eLife . 02536 . 014Figure 6—figure supplement 1 . Aldolase C enzymatic reaction products GAP and DHAP increase the activity in lobules III and X . To test the possibility that zebrin , or aldolase C , is responsible for the difference in simple spike activity due to its enzymatic activity , we bath-applied its reaction products . If this difference in enzymatic activity is responsible for the difference in simple spike activity between Z+ and Z– Purkinje cells , addition of the reaction products of zebrin should decrease the activity of Zebrin-negative Purkinje cells in lobule III . Simultaneous application of 250 µM of glyceraldehyde-3-phosphate ( GAP ) and dihydroxyacetone phosphate ( DHAP ) caused a subtle but significant increase in simple spike activity in both lobule X and lobule III ( X: p<0 . 001; III: p=0 . 019 vs vehicle , One-Way ANOVA followed by Bonferroni's post-hoc tests ) . This increase was even more pronounced when we increased the concentration to 1 mM of each ( X: p<0 . 001; III: p<0 . 001 vs vehicle , one-way ANOVA followed by Bonferroni's post-hoc tests ) . The increase of activity , particularly in lobule III , and the persistence of the difference between lobule III and X argue against a role for aldolase C in generating the difference in simple spike activity between Z+ and Z– Purkinje cells . Minimum recording duration was 60 s . Error bars represent SD , p<0 . 05 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 01410 . 7554/eLife . 02536 . 015Figure 6—figure supplement 2 . Effects of blocking EAAT4 on Purkinje cell activity in lobule III and X in vitro . In search for the underlying mechanism , we tested the contribution of EAAT4 , which is expressed in a pattern similar to that of zebrin , to Purkinje cell activity in vitro . ( A and B ) EAAT4 blocker TBOA ( 25 µM ) did not affect the firing frequency of Purkinje cells in lobule III ( vehicle: n = 47 cells , 6 mice; DL-TBOA: n = 30 cells , 5 mice; t = 1 . 219 , p=0 . 227 ) , or those in lobule X ( vehicle: n = 48 cells , 6 mice; DL-TBOA: n = 30 cells , 5 mice; t = −0 . 597 , p=0 . 553 ) , largely maintaining the difference between averages for lobule III–X ( inset ) . ( C–E ) Wash-in of DL-TBOA had no significant effect ( III: n = 7 cells , 6 mice; t = 0 . 457 , p=0 . 664 , X: n = 4 cells , 4 mice; t = −2 . 202 , p=0 . 115 , paired Student’s t-test ) . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 01510 . 7554/eLife . 02536 . 016Figure 6—figure supplement 3 . Effects of TRPC3 blockers on other PC activity parameters . Application of TRPC3 blockers genistein and Pyr3 affected , apart from simple spike firing frequency , several other characteristics of Purkinje cell activity . ( A ) Whereas effects of TRPC3 blockers on simple and complex spike regularity were subtle or absent , Pyr3 increased the climbing fiber pause and decreased the complex spikes firing frequency in lobules I–III towards levels comparable to those found in lobule X , suggesting that the differences in these parameters between the predominantly zebrin-negative lobules I–III and zebrin-positive lobule X are also , at least in part , dependent on the activity of TRPC3 . The effects of Pyr3 on complex spike half width appears to be comparable , in that it increases half-width towards levels observed in lobule X , while genistein predominantly has an attenuating effect on half-width and spike area , possibly due to its less selective nature . ( B ) Interestingly , blocking TRPC3 with Pyr3 induces the occurrence of the suppression and oscillation response types in the post complex spike simple spikes activity . These types are under normal and vehicle conditions restricted to PCs in lobule X , suggesting a role for TRPC3 in the post complex spike activity . Error bars represent SD p<0 . 05 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02536 . 016 An alternative candidate protein that has a zebrin-related expression in adult animals and could potentially influence spiking activity is EAAT4 , a glutamate transporter that is expressed in a zebrin-like pattern and carries a depolarizing current ( Wadiche and Jahr , 2005 ) . Based on the higher expression of EAAT4 in Z+ Purkinje cells , blocking EAAT4 would arguably affect lobule X more than lobule III , but the general EAAT blocker DL-TBOA had no significant effect on the activity of Purkinje cells in either lobule ( Figure 6—figure supplement 2A , B ) ( lobule III: t = 1 . 219 , p=0 . 227; lobule X: t = −0 . 597 , p=0 . 533 ) , maintaining the difference between lobule III and X ( inset , t = 3 . 641 , p<0 . 001 ) . Wash-in of DL-TBOA did also not affect activity in lobule III or X ( lobule III: 2 . 0 ± 10 . 8%; lobule X: −3 . 6 ± 3 . 4%; t = 0 . 982 , p=0 . 352 ) ( Figure 6—figure supplement 2C–E ) . Next , we studied whether the effects of TRPC3 blockers were sufficiently robust to also induce measurable effects in vivo . In line with the in vitro data , both genistein and Pyr3 ( i . p . and i . c . v . , 240 mg/kg and 200 µg , respectively ) caused a decrease in SS activity in lobules I–III , lasting for several hours ( vehicle: 90 . 6 ± 13 . 3 Hz; genistein: 74 . 6 ± 14 . 9 Hz; Pyr3: 61 . 6 ± 15 . 5 Hz; both p<0 . 001 vs vehicle , One-way ANOVA followed by Tukey's post-hoc test ) , while no significant effect was recorded in lobule X ( vehicle: 52 . 9 ± 13 . 3 Hz; genistein: 57 . 5 ± 20 . 8 Hz , ; Pyr3: 51 . 1 ± 14 . 3 Hz , p=0 . 546 and p=0 . 887 vs vehicle , One-way ANOVA followed by Tukey's post-hoc test ) ( Figure 6F , G ) . Together with changes in simple spike frequency , several other parameters , related to the complex spike of Purkinje cells in lobule I-III shifted , upon Pyr3 application , towards the values for lobule X with or without drugs . These included including climbing fiber pause , complex spike frequency and width and type of simple spike response following complex spikes ( Figure 6—figure supplement 3 ) . Effects of genistein were less consistent , probably due to its less selective nature . Together , these data suggest that TRPC3 contributes to the elevated SS activity in zebrin-negative PC zones . The difference in SS firing frequency between zebrin-positive and zebrin-negative PCs in vivo was robust ( i . e . , approximately 60 Hz vs 90 Hz ) and highly significant , present throughout the cerebellar cortex and could be reproduced by directly comparing the activity of Purkinje cells in adjacent modules . Although we cannot exclude the possibility that lobule-specific effects contribute to the observed differences in simple and complex spike firing frequency , the results obtained in EAAT4-eGFP mice in vivo and in vitro argue against a contribution of their rostro-caudal or lobular location . The difference in SS firing frequency is probably largely determined by the intrinsic properties of PCs , as this difference was maintained in the reduced slice preparation , in which the inputs are blocked , as well as in mouse mutants , in which the excitatory ( a6-Cacna1a mice ) or inhibitory ( PC-Δγ2 mice ) inputs are attenuated . Comparing the PC activity in zebrin-positive and zebrin-negative PCs per transverse zone confirmed its link to zebrin-identity , but also revealed a more subtle difference within the population of zebrin-positive PCs . These differences could be lobule-specific and/or originate from differences in input , in more subtle variations in zebrin or its related proteins or even in the expression pattern of other proteins . It should be noted , however , that blocking TRPC3 shifts the activity of Z− PCs towards that of Z+ PCs , indicating that the potential differences within the group of Z+ PCs do not affect our conclusions . In contrast , the level of regularity of firing ( i . e . , CV2 ) was not consistently dependent on zebrin identity , but significantly altered by impairing the excitatory and/or inhibitory inputs . In line with the notion that reduced SS activity of PCs , as observed in the zebrin-positive modules , should lead to enhanced firing of the GABAergic neurons in the cerebellar nuclei and thereby to reduced activity in the inferior olivary neurons ( Chen et al . , 2010; De Zeeuw et al . , 2011 ) , we found that CS activity induced by activity of olivary climbing fibers was reduced in zebrin-positive PCs . In fact , reduction of simple spike frequency in lobule III in vivo for several minutes to hours by Pyr3 application also reduced complex spike frequency , supporting the indirect control of simple spikes on complex spike activity . Interestingly , temporary increases in climbing fiber-evoked CS activity suppress SS frequency providing a homeostatic control mechanism within an olivocerebellar module ( Mathews et al . , 2012; Coddington et al . , 2013 ) . Thus , whereas the external inputs to PCs may control the precise temporal coding of SS activity at rest as well as the firing frequency and dynamic range during natural sensory stimulation ( Badura et al . , 2013; Galliano et al . , 2013 ) , their intrinsic properties appear to determine the baseline frequencies of SSs as well as CSs at rest , around which they can operate . These findings raise the question which proteins in the zebrin-positive and zebrin-negative zones may actually determine the difference in intrinsic firing frequencies of their PCs . Since zebrin's enzymatic reaction products did not underlay the differences in SS firing frequency , we shifted our attention to other proteins that are expressed in pattern similar or complementary to that of zebrin . We targeted EAAT4 that is expressed in a pattern similar to that of zebrin , and TRPC3 , the effector channel of a cascade of proteins that has zebrin-like expression patterns ( Dehnes et al . , 1998; Mateos et al . , 2001; Wadiche and Jahr , 2005 ) . Although blocking EAAT4 in vitro had no detectable effect on SS firing frequency , blocking TRPC3 reduced SS activity in lobule III ( largely zebrin-negative ) , but not in lobule X ( zebrin-positive ) , both in vitro and in vivo . It should be noted that , although TRPC3 gene expression appears to vary from anterior to posterior with higher levels in the anterior , largely zebrin-negative , lobules ( Allen Brain Atlas , www . brain-map . org ) , there is no immunohistochemical evidence for differences in protein levels ( Hartmann and et al , 2008; Becker et al . , 2009; Sekerkova et al . , 2013 ) . If the expression of TRPC3 is indeed homogeneous throughout the cerebellum , the differential effect of blocking TRPC3 suggests that its activity might be higher in Z− PCs . Two mutually non-exclusive mechanisms could contribute to this difference in activity . First , several proteins in the molecular cascade related to TRPC3 are expressed in zebrin-like bands , including the IP3-receptor ( Furutama et al . , 2010 ) ( TRPC3 modulator [Kim et al . , 2012b] ) , PLCβ3/4 [Sarna et al . , 2006] ( TRPC3 activator [Kim et al . , 2012b] ) , PKCδ [Barmack et al . , 2000] , and NCS-1 [Jinno et al . , 2003] . In fact , zebrin II or aldolase C , which is not likely to be involved via its enzymatic function , bears the capacity to bind IP3 ( Baron et al . , 1999 ) , and thus could potentially reduce the activation of TRPC3 in Z+ PCs , through the IP3-receptor . At the same time , mGluR1 subtype b is expressed in a pattern complementary to that of zebrin ( Mateos et al . , 2001 ) , and it has been shown that mGluR1 can be tonically activated , that mGlurR1 blockers can reduce SS firing frequency ( Yamakawa and Hirano , 1999; Coesmans et al . , 2003; Chanda and Xu-Friedman , 2011 ) , and that mGluR1-evoked depolarizing currents can be blocked with TRPC3-selective blockers ( Kiyonaka et al . , 2009; Kim et al . , 2012b ) . However , the possibility that an alternative pathway , independent of mGluR1 , leads to TRPC3 activition cannot be excluded . Knowledge of this pathway of proteins and their exact interactions is at current presumably incomplete and beyond the scope of this study . The findings that the expression patterns of mGluR1b , PLCβ , PKCδ , and IP3R1 , all of which are key proteins in calcium release from intracellular calcium stores , are linked to cerebellar modules ( Mateos et al . , 2001 ) and intimately connected with TRPC3 , provokes the speculation that this entire pathway contributes to the difference in SS activity between zebrin-positive and zebrin-negative PCs ( Hartmann and et al , 2008; Becker et al . , 2009 ) . Our finding that SS activity and indirectly also CS activity at rest are determined by the intrinsic properties of PCs implies that they operate around these baseline frequencies during natural stimulation and behaviour . Interestingly , the low and high baseline frequencies of zebrin-positive and zebrin-negative PCs also appear to be in line with their propensities for induction of long-term potentiation ( LTP ) and long-term depression ( LTD ) , respectively ( Wadiche and Jahr , 2005; Wang et al . , 2011 ) . Thus , PCs operating at lower frequencies may be preferentially potentiated , whereas PCs with higher SS firing frequency may have less ‘space’ for increasing the firing rate and may be more prone to express LTD . Likewise , entrainment of cerebellar nuclei neurons by synchronized SS input from PCs , which results in phase-locking of connected neurons , may occur at 50–80 Hz , but is impaired at 100 Hz ( Person and Raman , 2012 ) . If correct , this mechanism predicts that the phase-locking mechanism is engaged in contacts between zebrin-positive PCs and cerebellar nuclei neurons , whereas those involved in zebrin-negative zones may be more prone for rebound excitation , which follows strong forms of inhibition ( De Zeeuw et al . , 2011; Person and Raman , 2012 ) . The cerebellar nuclei can also be divided based on the zebrin expression pattern , with a rostral half that receives , predominantly , input from zebrin-negative PC's and a caudal part that mostly receives zebrin-positive inputs ( Sugihara and Shinoda , 2007; Sugihara et al . , 2009; Sugihara , 2011 ) . This implies that PC input to cerebellar nuclei neurons may be segregated on the basis of frequency , and that as a consequence the output of cerebellar nuclei neurons located within zebrin-positive and zebrin-negative territories may be distinctly different . Although this remains speculative at this stage , similar phenomena have been described for highly active neurons in cerebral cortex ( Yassin et al . , 2010 ) and hyperpolarization-activated currents in affiliated olfactory bulb mitral neurons ( Angelo et al . , 2012 ) . Combined with the zebrin- or lobule-related prevalence of plasticity mechanisms ( Wadiche and Jahr , 2005; Wang et al . , 2011 ) , our results suggest that the biochemically identified bands in the structurally homogenous cerebellar cortex are physiologically different with distinct biophysical signatures that probably have significant implications downstream in the cerebellar nuclei and thereby on motor behaviour and cognition . We recorded in vivo single-unit Purkinje cell activity in adult male C57Bl/6 mice ( C57Bl/6J , Charles River ) , aged 10–35 weeks . Mice were prepared for recordings by placing an immobilizing construct ( pedestal ) and a craniotomy on their skulls ( Badura et al . , 2013 ) . In short , the skin over the skull was shaven , and opened along the rostro-caudal midline . Using Optibond ( Kerr , Salerno , Italy ) and Charisma ( Heraeus Kulzer , Hesse , Germany ) , a U-shaped holder ( 6 × 4 mm ) with a magnet inside ( 4 × 4 mm , MTG , Weilbach , Germany ) was fixed on the skull , overlying the frontal and parietal bones . Next , the medial neck muscles overlying the occipital bone were removed , a craniotomy was made over the interparietal or occipital bone and a recording chamber was placed around it , allowing in vivo electrophysiological recordings throughout different areas in the cerebellum of awake mice . The exact location of the craniotomy depended on the target area , see Figure 1—figure supplement 1 for details . After recovery of >24 hr , mice were head-fixed to a bar , their bodies restrained in a custom-made plastic tube and the dura was opened to facilitate the recording of extracellular Purkinje cell activity , as previously described ( Hoebeek et al . , 2005 ) . Electrophysiological activity in the cerebellar cortex was recorded using double barrel borosilicate glass pipettes ( theta septum , 1 . 5 OD , 1 . 02 ID , WPI , FL , USA ) . To do so , one of the barrels was opened laterally , approximately 10 mm from the taper , to allow entrance of the electrode wire and sealed with glass glue at the back . The other barrel was filled with a blue dye ( Alcian Blue , 0 . 1–0 . 2% solution in saline; Sigma–Aldrich , St . Louis , MO , USA ) . The recording half of the double barrel pipettes were filled with 2 M NaCl-solution , and had a tip size of 3–6 μm , respectively . Pipettes were advanced into the cerebellum with an oil micro-drive ( Narishige , Tokyo , Japan ) and signals were pre-amplified ( custom-made preamplifier , 1000x DC ) , filtered ( CyberAmp 320 , Axon , Molecular Devices , Sunnyvale , CA , USA ) , digitized ( Power1401 , CED , Cambridge , UK ) , and stored for offline analysis . After successful recordings , brief pressure pulses were delivered through the other barrel of the electrode , using a custom-built device , to mark the recording site . To obtain a6-Cacna1a and PC-Δγ2 mice , we used the Cre/loxP system to delete exon 4 of the gene coding for the P/Q-type voltage-gated calcium channel ( Cacna1a ) selectively from granule cells and exon 4 of the GABAA receptor γ2 subunit gene ( Gabrg2 ) selectively from PCs , respectively , as described previously ( Wulff et al . , 2009; Galliano et al . , 2013 ) . In short , we crossed Cacna1alox/lox mice with Gabra6::Cre ( or Δa6::cre ) mice ( Aller et al . , 2003 ) and Gabrg2lox/lox mice with Pcp2::Cre ( or L7::Cre ) mice ( Oberdick et al . , 1990 ) , respectively . From the offspring , that was heterozygous for the floxed genes ( i . e . , Cacna1alox/+ and Gabrg2lox/+ ) , Cre-negative males were crossed with Cre-positive females to generate , amongst others , Δa6::cre;Cacna1alox/lox ( or Cacna1aΔ/Δ , here named a6-Cacna1a ) and Pcp2::cre;gabrg2lox/lox ( or Gabrg2Δ/Δ , here named PC-Δγ2 ) mice , respectively . Both lines were maintained in a C57Bl6 background . In the experiments with mutant mice and blocker injections , double and single barrel ( 2 . 0 mm OD , 1 . 16 mm ID , Harvard Apparatus , MA , USA ) borosilicate glass pipettes were used , and alcian blue was injected to confirm that the recordings were from lobules I–III or X . EAAT4-eGFP mice express enhanced green fluorescent protein ( eGFP ) under control of the EAAT4 promoter , and were generated using the bacterial artificial chromosome ( BAC ) ( Gincel et al . , 2007 ) . Targeted recordings of eGFP-positive and eGFP-negative Purkinje cells were made after visualizing the eGFP-positive bands using in vivo two-photon imaging of 5 awake EAAT4-eGFP mice ( 3 females , 2 males , 10–26 weeks old ) . Images were acquired using a TriM Scope II ( LaVision BioTec , Bielefeld , Germany ) attached to an upright microscope with a 40x/0 . 8 NA water-immersion objective ( Olympus , Tokyo , Japan ) . Laser illumination was provided by a Chameleon Ultra titanium sapphire laser ( Coherent , Santa Clara , CA ) . We aimed to image Purkinje cells in the superficial layer of a restricted part of the cortex ( lobules V-VI and Crus I ) at a depth of ∼250 µm using an excitation wavelength of 920 nm , and their location in relation to zebrin bands was determined online . The recording pipette was filled with Alexa-594 ( 10 µM in 2 M NaCl; Life Technologies , Carlsbad , CA ) and visualized with an excitation wavelength of 800 nm , the minimum recording duration was 30 s . Images from eGFP and Alexa-594 were filtered using a Gaussian kernel , contrast-optimized and subsequently merged in Photoshop ( Adobe , San Jose , CA ) . Purkinje cells recorded in vivo from EAAT4-eGFP mice are included in Figure 2 , Figure 2—figure supplement 1 , Figure 4B . Purkinje cells were identified by the occurrence of simple and complex spikes and were confirmed to be from a single unit by the presence of a pause in simple spikes after each complex spike . To assure the quality and reliability of the recording the following criteria were imposed: ( 1 ) a minimum recording duration of 120 s , ( 2 ) stable simple spike amplitude , ( 3 ) no clear signs of tissue damage in a circle with 400 µm radius , around the recording site ( Figure 1—figure supplement 1B ) . All in vivo data were analyzed using SpikeTrain ( Neurasmus BV , Rotterdam , The Netherlands , www . neurasmus . com ) , running under Matlab ( Mathworks , MA , USA ) . SpikeTrain uses wave clustering to identify simple and complex spikes , and in case of doubt manual checking ( and correcting ) would be performed . For each cell the firing rate , CV and mean CV2 were determined for simple and complex spikes , as well as the climbing fiber pause . CV is the standard deviation of inter-spike intervals ( ISI ) divided by the mean , the mean CV2 is calculated as the mean of 2·\| ( ISI ) n+1−ISIn\|/ ( ISIn+1 + ISIn ) . Both are measures for the regularity of the firing , with CV reflecting that of the entire recording and mean CV2 that of adjacent intervals , making the latter a measure of regularity on small timescales . The climbing fiber pause is determined as the minimum duration between a complex spike and the following simple spike . To extend this analysis , we also plotted histograms of simple spike activity time locked on the complex spike , and labelled the shape of this time histogram as normal , facilitation , suppression , and oscillation ( see Figure 3 for examples ) . The spike characteristics half maximum width ( HMW ) and spike area were determined from the normalized average signal of simple and complex spikes of individual recordings . Half-width was calculated as the width of the first peak at half of its maximum amplitude . The spike area was defined as the integral of the rectified complex spike wave form in a time window of 0 . 5 ms pre and 3 ms post spike onset . Acute sagittal slices ( 250 μm thick ) were prepared from the cerebellar vermis of 3–5 month old male C57BL/6J mice ( Charles River ) in ice-cold slicing medium that contains the following ( in mM ) : 240 sucrose , 2 . 5 KCl , 1 . 25 Na2HPO4 , 2 MgSO4 , 1 CaCl2 , 26 NaHCO3 , and 10 D-glucose , bubbled with 95% O2 and 5% CO2 . Subsequently , slices were incubated in ACSF containing ( in mM ) : 124 NaCl , 2 . 5 KCl , 1 . 25 Na2HPO4 , 2 MgSO4 , 2 CaCl2 , 26 NaHCO3 , and 10 D-glucose equilibrated with 95% O2 and 5% CO2 at 34 . 0°C for 30 min , and then at room temperature . Slices were typically used within 5 hr ex vivo . NBQX ( 10 μM ) , DL-AP5 ( 50 μM ) , and picrotoxin ( 100 μM ) were bath-applied to block AMPA- , NMDA- , and GABA subtype A ( GABAA ) -receptors , respectively . Borosilicate glass pipettes ( WPI ) were filled with ACSF and had an open pipette resistance of 2–4 MΩ . Purkinje cells were identified using visual guidance by DIC video microscopy and water-immersion 40X objective ( Axioskop 2 FS plus; Carl Zeiss , Jena , Germany ) . Slices were transferred to the recording chamber and incubated for at least 10 min before starting the recordings . We recorded the Purkinje cell activity in cell-attached mode ( 0 pA injection ) at 33 . 0 ± 1 . 0°C , with a distance of 0 . 5 cm between temperature probe and slice . Current clamp recordings were performed with the same setting as cell-attached recording , except that pipettes were filled with intracellular solution contains the following ( in mM ) : 120 K-Gluconate , 9 KCl , 10 KOH , 3 . 48 MgCl2 , 4 NaCl , 10 HEPES , 4 Na2ATP , 0 . 4 Na3GTP , and 17 . 5 sucrose , pH 7 . 25 and Osm 295 . For experiments in EAAT4-eGFP mice , all experimental conditions were the same as cell-attached experiments above , except that coronal slices ( 300 μm thick ) were used to record from identified EAAT4-positive and EAAT4-negative Purkinje cells within the same lobules . We first identified the lobules by their locations and band patterns with a 10X objective , and then zoomed in with a 40X objective to proceed to recording . Electrophysiological data were acquired using an EPC9 amplifier ( HEKA , Lambrecht , Germany ) , filtered at 10 kHz and digitized at 25 kHz . Acquisition was controlled using PULSE software ( HEKA ) and the data were exported and analyzed using Minianalysis ( v6 . 0 . 3 ) software ( Synaptosoft , Fort Lee , NJ , USA ) or Matlab . The typical signal-to-noise ratio was larger than 5:1 , and minimum recording duration was 120 s , unless stated otherwise . Cells were included based on the following criteria: ( 1 ) the CV over the whole period of recording was <0 . 2; ( 2 ) the average frequency changed less than 20% between the first and the last ( 30 s ) , except for those in the wash-in experiment . To minimize the day-to-day and slice-to-slice variations , recordings were targeted at different lobules for every slice . For the Pyr3 wash-in experiment , firing frequencies were normalized to the average frequency over the 2-min period before adding the drug to the ACSF ( pre ) . The wash-in effect was determined by calculating the firing frequency in the period of 5–7 min after the drug was in the recording chamber ( post ) . In the whole-cell patch recording , the membrane potential of Purkinje was held at −65 mV using current injection to avoid spontaneous spiking activity ( average: −454 ± 38 pA ) . We recorded the intrinsic excitability by injecting depolarizing currents ranging from 100 to 1000 pA ( 100 pA steps ) relative to the holding current . Data were exported and analysed using threshold search with Clampfit ( v10 . 4 , Molecular Devices , Sunnyvale , CA , USA ) . DL-TBOA ( EAAT blocker , Tocris , Ellisville , MO , USA ) , genistein ( TRPC3 blocker , Sigma–Aldrich ) , and Pyr3 ( TRPC3 blocker , Tocris ) were dissolved in dimethyl sulfoxide ( DMSO , Carl Roth GmbH , Karlsruhe , Germany ) and ACSF . For in vivo recordings , mice were injected with 240 mg/kg genistein ( i . p . ) and 200 µg Pyr3 ( i . c . v ) dissolved in saline or DMSO . Vehicle control mice were injected with 100 µl DMSO ( i . p . ) . After injection , extracellular Purkinje cell activity was recorded as described above , and alcian blue was injected to verify that the recordings were from lobules I-III or X . The minimum recording duration was 60 s and recordings were made for up to 4 hr after injection for these experiments . For in vitro experiments all blockers were prepared in 1:1000 stock solutions in DMSO , stored at −20°C and used within 4 weeks after preparation . Blockers were bath applied where indicated in the following concentrations: DL-TBOA ( 25 μM ) , genistein ( 10 μM ) , and Pyr3 ( 100 μM ) . Except for the wash-in experiments , all recordings started after the slice was incubated in the drug-containing ASCF for at least 15 min . In the cases where we observed an effect in the wash-in experiments , the firing frequency of Purkinje cells typically dropped the moment the drug reached the bath , an effect maximizing within a few minutes and sometimes followed by a smaller recovery . The experimenter was blind to the presence and type of drug applied until analysis was completed . The tubing was changed after every blocker experiment . After recordings , mice were deeply anesthetized with Nembutal and perfused with 75 ml of 4% paraformaldehyde ( PFA ) . The brains were removed from the skull and post-fixed for 1–2 hr in 4% PFA , and stored in 0 . 1M PB containing 10% sucrose . After embedding in 10% gelatin and 10% sucrose , blocks were hardened in a solution containing 10% formaldehyde , 30% sucrose for 1–2 hr at room temperature and then stored overnight in 0 . 1M PB with 30% sucrose at 4°C . To identify if recordings were made in a Z+ or Z− band , coronal sections with a thickness of 40 µm were processed based on a standard immunostaining procedure; next , they were thoroughly rinsed with 0 . 1MPB . The goat-derived zebrin II antibody ( Santa cruz , TX , USA ) was diluted at 1:1000 in PBS , pH 7 . 6 , containing 2% normal horse serum and 0 . 4% Triton . Rabbit anti-goat secondary antibody HRP conjugate diluted at 1:200 was used as a secondary antibody ( Dako , Glostrup , Denmark ) . The sections were thoroughly rinsed three times with PBS and PB , followed by diaminobenzidine ( DAB ) incubation ( 0 . 66% DAB , and 0 . 033% H2O2 for 10–20 min ) . The sections were put on glass and then dehydrated by different grades of ethanol ( 70% , 80% , 90% , 96% , 96% , 100% , 100% , 100% , 2 min per grade ) , xylene was applied to clean the ethanol , and subsequently the sections were covered with Permount . To only determine the recording sites for recordings throughout the cerebellum , sagittal sections were cut at 80 µm followed by neutral red staining . The injection sites were located by the light microscope . If the same injection could be found in several slices , the injection was allocated to the slice with the highest density of Alcian Blue . If the injection site was at the border of two cerebellar lobules , the cell was allocated to the most rostral lobule . Particularly in the hemipsheres , a distinction between positive and lightly positive areas can be made . In this study , they were taken together as positive , and compared to the—clearly identifiable—negative bands . Recordings were excluded from further analysis if there was clear tissue damage within a circle with a radius of 400 µm around the injection site . The recording sites of 7 out 8 cells in the flocculus ( FL ) were confirmed by the response to visual stimulation instead of histology ( Galliano et al . , 2013 ) . Example sections and sections used to determine the staining intensity of zebrin II were photographed using a Leica DMRB microscope equipped with Leica DC300 camera . To compare zebrin intensity between lobules the average pixel intensity of the PC somas in each lobule was determined using ImageJ software , based on the average of a total of 3 sections for vermis and 2 sections for the hemisphere per mouse , from 3 mice . Sections compared in the same panel , were processed in parallel . To correct for minor differences in overall staining intensity between vermis and hemispheres , soma intensities were normalized based on background ( granule cell layer ) intensity . All values are shown as mean ± SD , unless stated otherwise . Unpaired Student's t test were used for comparisons and Spearman's r test for correlations , unless stated otherwise , and p<0 . 05 was considered to be significant . Comparisons in which at least one of the groups had n ≤ 4 were re-tested using a Mann–Whitney U-test , and in all cases the outcome was confirmed . All experiments were performed under the GGO license no . IG 04-197 , and approved by the Dutch animal ethical committee ( DEC , EMC 2168/2545/2999/3002/3057 ) .	The cerebellum , located at the back of the brain underneath the cerebral hemispheres , is best known for its role in the control of movement . Despite its small size , the cerebellum contains more than half of the brain's neurons . These are organized in a repeating pattern in which cells called Purkinje cells receive inputs from two types of fibers: climbing fibers , which ascend into the cerebellum from the brainstem; and parallel fibers , which run perpendicular to the climbing fibers . This gives rise to a characteristic ‘crystalline’ structure . As a result of this uniform circuitry , it was widely believed was that all Purkinje cells throughout the cerebellum would function the same way . However , the presence of distinct patterns of gene expression in different regions suggests that this is not the case . Molecules called zebrins , for example , are found in some Purkinje cells but not others , and this gives rise to a pattern of zebrin-positive and zebrin-negative stripes . A number of other molecules have similar distributions , suggesting that these differences in molecular machinery could underlie differences in cellular physiology . Zhou , Lin et al . have now provided one of the first direct demonstrations of such physiological differences by showing that zebrin-positive cells generate action potentials at lower frequencies than zebrin-negative cells . This pattern is seen throughout the cerebellum , and is evident even when the positive and negative cells are neighbors , which indicates that these differences do not simply reflect differences in the locations of the cells or differences in the inputs they receive from parallel fibers . Additional experiments revealed that the distinct firing rates are likely not generated by zebrin itself , but rather by proteins that are expressed alongside zebrin , most notably those that work through an ion channel called TRPC3 . By showing that cells arranged in the same type of circuit can nevertheless have distinct firing rates , the work of Zhou , Lin et al . has revealed an additional level of complexity in the physiology of the cerebellum . In addition to improving our understanding of how the brain controls movement , these findings might also be of interest to researchers studying the increasing number of neurological and psychiatric disorders in which cerebellar dysfunction has been implicated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Cerebellar modules operate at different frequencies
A large group of bacterial virulence autotransporters including AIDA-I from diffusely adhering E . coli ( DAEC ) and TibA from enterotoxigenic E . coli ( ETEC ) require hyperglycosylation for functioning . Here we demonstrate that TibC from ETEC harbors a heptosyltransferase activity on TibA and AIDA-I , defining a large family of bacterial autotransporter heptosyltransferases ( BAHTs ) . The crystal structure of TibC reveals a characteristic ring-shape dodecamer . The protomer features an N-terminal β-barrel , a catalytic domain , a β-hairpin thumb , and a unique iron-finger motif . The iron-finger motif contributes to back-to-back dimerization; six dimers form the ring through β-hairpin thumb-mediated hand-in-hand contact . The structure of ADP-D-glycero-β-D-manno-heptose ( ADP-D , D-heptose ) -bound TibC reveals a sugar transfer mechanism and also the ligand stereoselectivity determinant . Electron-cryomicroscopy analyses uncover a TibC–TibA dodecamer/hexamer assembly with two enzyme molecules binding to one TibA substrate . The complex structure also highlights a high efficient hyperglycosylation of six autotransporter substrates simultaneously by the dodecamer enzyme complex . Protein glycosylation is one of the most abundant post-translational modifications in all domains of life ( Spiro , 2002 ) . Recent studies have appreciated protein glycosylation in bacteria , which is often associated with pathogen virulence and immune modulation ( Szymanski and Wren , 2005; Abu-Qarn et al . , 2008; Nothaft and Szymanski , 2010 ) . For instance , flagella glycosylation is found in many bacteria species including Campylobacter jejuni , Helicobacter pylori , Clostridium spp . and Pseudomonas aeruginosa , contributing to bacterial locomotion or virulence ( Schirm et al . , 2003; Nothaft and Szymanski , 2010 ) . Recently , we and others have shown that secreted effectors from enteropathogenic Escherichia coli ( EPEC ) and related enteric pathogens harbor an arginine N-acetylglucosamine transferase activity that modifies host death-domain proteins and is essential for bacterial colonization in infected mice ( Li et al . , 2013; Pearson et al . , 2013 ) . The autotransporter secretion pathway ( also known as the type V secretion system ) in bacteria delivers autotransporters onto the bacterial surface . Autotransporters , representing the largest family of bacterial virulence factors , share a similar structural organization containing a signal peptide followed by a functional passenger domain and a C-terminal β-barrel translocation domain . Autotransporters play a critical role in diverse aspects of bacterial physiology including proteolytic digestion of host proteins , biofilm formation , adhesion and invasion of host cells , and intracellular motility ( Henderson et al . , 2004; Lazar Adler et al . , 2011; Wells et al . , 2007 ) . A subfamily of autotransporters , including AIDA-I ( adhesin involved in diffuse adherence ) from diffusely adhering E . coli ( DAEC ) 2787 ( Benz and Schmidt , 1989 ) and TibA from enterotoxigenic E . coli ( ETEC ) H10407 ( Elsinghorst and Weitz , 1994 ) , are glycosylated in their passenger domains ( Lindenthal and Elsinghorst , 1999; Benz and Schmidt , 2001; Sherlock et al . , 2006 ) , which functions in bacterial autoaggregation and adhesion to host cells ( Benz and Schmidt , 1989; Sherlock et al . , 2004; Charbonneau and Mourez , 2007 ) . AIDA-I and TibA-like autotransporters are present in diverse bacterial species . Previous studies carried out in the ectopic system ( Benz and Schmidt , 2001; Moormann et al . , 2002 ) indicate a role for AAH ( autotransporter adhesin heptosyltransferase ) in DAEC and its ETEC homologue TibC in AIDA-I and TibA glycosylation . However , AAH and TibC harbor no sequence homology to known glycosyltransferases and there has been no biochemical evidence demonstrating their glycosyltransferase activity . In a separate parallel study ( Lu et al . , 2014 ) we showed that AAH is a bona fide heptosyltransferase belonging to a large bacterial autotransporter heptosyltransferase ( BAHT ) family . Here we determine the crystal structures of TibC heptosyltransferase , both alone and in complex with ADP-D-glycero-β-D-manno-heptose ( ADP-D , D-heptose ) . The structure shows a symmetric ring-shape dodecamer . The protomer features a β-hairpin thumb and an iron-finger motif , both required for the dodecamer assembly . The ligand-bound structure reveals a sugar transfer mechanism and determinants for the ligand stereoselectivity for TibC and AAH . Furthermore , electron cryomicroscopy ( cryo-EM ) analyses of a TibC–TibA dodecamer/hexamer enzyme–substrate complex reveal the structural basis for high efficient autotransporter hyperheptosylation by the TibC dodecamer . We first confirmed the autotransporter heptosyltransferase activity of TibC and its function in mediating bacterial adhesion to host cells ( Lu et al . , 2014 ) . Co-expression of TibC or AAH together with AIDA-I in E . coli BL21 cells resulted in a tight adhesion of the bacteria to HeLa cells ( Figure 1A ) and AIDA-I glycosylation ( Figure 1B ) . Co-expression of TibC also induced TibA glycosylation in the heterologous system ( Figure 1C ) . We further identified a minimal passenger domain fragment of TibA ( TibA305-350 ) that could be efficiently glycosylated by TibC ( Figure 1D , E ) . TibA305-350 tagged with Flag epitopes at both termini was then fused C-terminal to the Small Ubiquitin-like Modifier ( SUMO ) . The SUMO-Flag-TibA305-350-Flag fusion protein showed a slower migration on the SDS-PAGE gel when co-expressed with TibC ( Figure 1F ) . Electrospray ionization ( ESI ) mass spectrometry analysis gave a mass of 20 , 758 Da that matched the theoretical mass ( 20 , 757 Da ) of SUMO-Flag-TibA305-350-Flag ( Figure 1G ) . Co-expression of TibC resulted in mass increase of 576 , 768 , or 960 Da , corresponding to the addition of 3 , 4 , and 5 heptoses , respectively ( Figure 1G ) . The modified fusion protein was further digested with the endoproteinase Asp-N for liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) analysis . Asp-N digestion is expected to yield two major peptide fragments , DK-305SASKVIQNSGGAVITNTSAAVSGTN329 and 330DNGSFSIAGGSAVNMLLENGG350 . We were able to detect abundant signals for both unmodified and modified DK-305SASKVIQNSGGAVITNTSAAVSGTN329 in full scan MS analyses and only weak signals of the modified 330DNGSFSIAGGSAVNMLLENGG350 peptide , probably due to the hydrophobic nature . Mass measurements indicated that both peptides had multiple heptose modifications; a mixed population of modified peptides , containing different numbers of heptose conjugations , was observed . Similarly to that performed with AAH modification of AIDA-I ( Lu et al . , 2014 ) , serine residues ( Ser-313 and Ser-322 within DK-305SASKVIQNSGGAVITNTSAAVSGTN329 ) , but not threonine residues , were identified to be the modification sites by electron transfer dissociation ( ETD ) mass spectrometry ( Figure 1H ) . Supporting the heptosyltransferase activity of TibC , recombinant TibC was found to directly modify synthetic peptides derived from the passenger domains of either AIDA-I or TibA in which ADP-D , D-heptose , but not the anomer ADP-L , D-heptose , could serve as the sugar ligand ( Figure 1I ) . 10 . 7554/eLife . 03714 . 003Figure 1 . TibC catalyzes TibA/AIDA-I heptosylation and confers AIDA-I-mediated bacterial adhesion to host cells . ( A ) E . coli BL21 expressing EGFP was transformed with a plasmid harboring Flag-AIDA-I ( pAIDA-I ) together with a vector or Myc-tagged autotransporter adhesin heptosyltransferase ( AAH ) /TibC . TibCWT , wild-type TibC; TibCmut , the TibC mutant used for crystallization . Infected cells and the bacteria were visualized by differential interference contrast microscopy and the green fluorescence , respectively . Scale bar 20 μm . ( B and C ) Assays of TibC-catalyzed AIDA-I and TibA glycosylation . Lysates of E . coli BL21 ( DE3 ) cells transformed with indicated constructs were subjected to anti-Flag and -Myc immunoblotting or glycosylation analysis . The Flag tag in AIDA-I/TibA was inserted between the signal peptide and the passenger domain . ( D–F ) Indicated TibA truncations fused C-terminal to the SUMO-Flag were co-expressed with a vector or Myc-TibC in E . coli BL21 . Shown in ( F ) is Coomassie brilliant blue staining of purified SUMO-Flag-TibA305-350-Flag . ( G ) ESI mass spectrometry analyses of TibC modification of SUMO-Flag-TibA305-350-Flag in the E . coli system . The red peaks mark the glycosylated species . ( H ) ETD tandem mass spectrum of a triply charged Asp-N digested peptide from TibC-modified SUMO-Flag-TibA305-350-Flag . The fragmentation patterns generating the observed c and z ions are illustrated along the peptide sequence on top of the spectrum . Asterisk marks a modification on the serine by a heptose . ( I ) In vitro heptosyltransferase activity of TibC . Two peptides derived from the passenger domains of AIDA-I ( left ) and TibA ( right ) , respectively , were left untreated ( control ) or reacted with TibC in the presence of indicated sugar ligands . Shown are MALDI-TOF mass spectra of the reacted peptides ( * , peptides modified by one heptose ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 00310 . 7554/eLife . 03714 . 004Figure 1—figure supplement 1 . Multiple sequence alignment of the BAHT Family . Alignment was generated in the GeneDoc program . The genus names are indicated on the left . Secondary structures determined from the TibC crystal are on top of the sequence . The β-hairpin thumb and the iron-finger motif are highlighted as red arrows and a blue line , respectively . Identical residues are in black and other conserved residues are in grey . The ligand binding residues are in yellow and the four iron-finger cysteine residues are in blue . Green color highlights the residue that determines the sugar substrate stereoselectivity , and the catalytic base is in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 00410 . 7554/eLife . 03714 . 005Figure 1—figure supplement 2 . Amino acid sequence of AIDA-I passenger domain ( residues 58–588 ) arranged by β-helix repeat units . Each row lists the sequence of one repeat unit , which is predicted to form a β-helix composed of three small β-strands ( green arrows ) . The predicated secondary structure is marked on top of the sequence . The solid line represents putative loops . The numbers in each row mark the positions of the amino acids in each repeat unit . Putative glycosylated serine residues are yellow shaded . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 005 AAH and TibC homologues , defined as the BAHT family ( Lu et al . , 2014 ) , are widely present in pathogenic bacteria species including Citrobacter rodentium , Salmonella enterica serovar Urbana , Shigella sp . D9 , Laribacter hongkongensis , Cronobacter sakazakii , and several Burkholderia species ( Figure 1—figure supplement 1 ) . These proteins generally bear more than 50% sequence homology to TibC/AAH and contain residues critical for the heptosyltransferase activity ( see below ) . In the genomic locus , genes encoding BAHT are often followed by an autotransporter or a putative autotransporter , consistent with the notion that BAHT modifies its autotransporter partner for functioning ( Lu et al . , 2014 ) . Consistent with the modification of AIDA-I by TibC , the passenger domain of AIDA-I shows sequence similarities to that of TibA and contains a number of heptosylation motifs identified in TibA previously ( Lu et al . , 2014 ) ( Figure 1—figure supplement 2 ) . To reveal the mechanism of the BAHT family , we attempted to solve the crystal structure . Among several BAHTs analyzed , TibC behaved the best in recombinant expression and homogeneity . Purified His6-TibC was crystallized , but the good-looking crystals did not diffract due to internal disorders . TibC obtained from glutathione-S-transferase ( GST ) -fusion expression was also crystallized , which only diffracted to 8 Å despite extensive optimizations . A number of point mutations designed to reduce the surface entropy were screened and a TibC variant with three mutated epitopes ( E83A/E84A , K400A/K401A , and Q215A/E216A ) produced the highest quality crystal . This mutant exhibited identical biochemical and biological functions to wild-type TibC ( Figure 1A–C ) . A 2 . 9 Å structure was solved by single wavelength dispersion phasing using anomalous signals from selenomethionine ( SeMet ) and the naturally bound ferric ion ( Table 1 ) . 10 . 7554/eLife . 03714 . 006Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 006CrystalsTibC-SeMetTibC D110A in complex with D , D-heptoseData collection Space groupP21P21 Wavelength ( Å ) 0 . 97890 . 9792 a , b , c ( Å ) 87 . 8 , 314 . 4 , 164 . 583 . 3 , 313 . 2 , 164 . 7 α , β , γ ( º ) 90 , 101 . 4 , 9090 , 101 . 3 , 90 Resolution range ( Å ) 20 . 0–2 . 90 ( 2 . 95–2 . 90 ) 20 . 0–3 . 87 ( 3 . 94–3 . 87 ) No . of unique reflections194 , 303 ( 9760 ) 79 , 888 ( 3968 ) Completeness ( % ) 99 . 9 ( 99 . 9 ) 99 . 9 ( 100 ) Redundancy7 . 3 ( 6 . 1 ) 4 . 6 ( 4 . 6 ) I/σI19 . 5 ( 2 . 6 ) 9 . 18 ( 2 . 0 ) Rmerge ( % ) 12 . 7 ( 98 . 6 ) 23 . 7 ( 92 . 8 ) Refinement statistics Rwork/Rfree ( % ) †20 . 8/24 . 426 . 4/27 . 6 No . of protein atoms37 , 20137 , 136 No . of ligands atoms32504 No . of waters50 RMSD bond lengths ( Å ) 0 . 0130 . 006 RMSD bond angles ( º ) 1 . 401 . 11 Average overall B-factor69 . 10150 Iron atoms B-factor58 . 90150 ADP-heptose B-factor150Ramachandran plot statistics Most favored regions ( % ) 94 . 5194 . 73 Additional allowed regions ( % ) 5 . 364 . 45 Outlier regions ( % ) 0 . 130 . 83The data for the highest resolution shell are shown in parentheses . †Rfree is calculated by omitting 5% of the total number of reflections in model refinement . RMSD , root-mean-square deviation; SeMet , selenomethionine; Strikingly , one asymmetric unit in the solved structure contains 12 TibC molecules , which are assembled into a large ring structure ( Figure 2A ) . The overall shape of the dodecamer ring resembles a circular garland with an external diameter of 145 Å and a height of 72 Å . Parts of the 12 protomers are lined side by side at the central plate to form the stem of the garland with an inner diameter of ∼110 Å ( Figure 2A ) . The inner surface of the stem is decorated with uniformly distributed 12 ferric ions , forming a characteristic iron belt along the ring ( Figure 2A ) . The presence of ferric ions is expected from the brownish color of recombinant TibC protein in solution . Six symmetric TibC protomers project upwards from the central plate while the other six project downwards , generating a gear-like shape at both sides . Two adjacent TibC protomers projecting oppositely contact each other in a back-to-back manner around a dyad axis; six back-to-back dimers are further linked together in a hand-in-hand fashion around a sixfold axis to form the garland-like ring ( Figure 2B ) . Thus , the dodecamer assembly can be viewed as a hexameric oligomerization of dimers . The back-to-back interface is extensive and tilts across the central plate by 45° ( Figure 2A , B ) . The hand-in-hand connection creates 12 tilted grooves extending to the interior of the garland from both up and down sides . 10 . 7554/eLife . 03714 . 007Figure 2 . The overall structure and assembly of the TibC dodecamer . ( A ) The surface view of the TibC dodecamer . The two adjacent symmetric protomers are colored blue and magenta , respectively . The ferric ions are shown as red spheres . Shown on the left is a top view along the sixfold axis and on the right is a titled view . ( B ) Assembly of the TibC dodecamer . Two adjacent protomers form a back-to-back dimer with a twofold symmetry . Six symmetric dimers are further assembled through hand-in-hand contacts into a large ring . ( C ) Cryo-EM reconstruction of TibC at 11 . 5 Å resolution . The crystal structure was fit into the cryo-EM envelope . Shown are top and side views of the EM structure . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 00710 . 7554/eLife . 03714 . 008Figure 2—figure supplement 1 . Cryo-EM images and 3D reconstruction of the TibC dodecamer . ( A ) The cryo-EM image of TibC dodecamer acquired at a nominal magnification of 75 , 000× . The white and green boxes indicate typical top and side view particles , respectively . Scale bar , 50 nm . ( B ) Initial model reconstruction from class averages generated from reference-free classification . C1 symmetry was applied in the reconstruction process . New density was obtained by projection-matching refinement of the starting density using the Fourier reconstruction approach against the phase-flipped particles iteratively with gradually reduced step angle . D6 symmetry was enforced during the refinement . A , average; P , projection . ( C ) Resolution assessment by Fourier shell correlation ( FSC ) . The FSC coefficient from the final round of the refinement is plotted as a function of spatial frequency ( 1/Å ) . The resolution of the final reconstruction is 11 . 5 Å according to the FSC = 0 . 5 criterion . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 00810 . 7554/eLife . 03714 . 009Figure 2—figure supplement 2 . Analytical ultracentrifugation sedimentation velocity analysis of TibC dodecamer and TibC–TibA dodecamer/hexamer complex . In the left upper panels , fringes were collected by the interference optics and raw sedimentation velocity scans were overlaid with the best-fit curves obtained from sedimentation coefficient distribution analysis . Two or three scans are included and the calculated systematic signal offsets are removed . The left lower panels show the residual fringes after fitting the raw data with modeled curves . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 009 To investigate whether the dodecamer observed in the crystal structure results from crystal packing , single particle cryo-EM analyses of highly purified TibC protein were performed ( Figure 2—figure supplement 1 ) . A 3D reconstruction of 11 . 5 Å was obtained from the cryo-EM images , which showed a remarkably similar dodecamer architecture to that determined by X-ray crystallography ( Figure 2C ) . This suggests that the dodecamer ring architecture is the predominant form of TibC protein in solution and likely represents a functional physiological state . Consistently , purified TibC eluted as an oligomer of ∼500 kDa on the gel filtration column ( see below ) and the molecular weight was determined to be 578 kDa by analytic ultracentrifugation ( Figure 2—figure supplement 2 ) . All the 12 TibC protomers adopt an identical structure containing an N-terminal β-barrel domain , a core catalytic domain , a β-hairpin thumb , and an iron-finger motif which are arranged into a palm-like shape ( Figure 3A ) . The β-hairpin thumb and the iron-finger motif are insertions into the catalytic domain . The flattened β-barrel ( residues 1–97 ) is structured from two four-stranded sheets that pack against each other . The barrel is stacked against the catalytic domain from the side and makes extensive contacts with β9 , β12 , and the loop linking α3 and β12; several hydrophobic residues from both domains are in close proximity at the interface ( Figure 3—figure supplement 1A ) . The β-barrel is well separated from the catalytic domain and is unlikely to play a direct role in sugar transfer . Notably , deletion of the N-terminal 80 or 95 residues abolished TibC heptosylation of TibA ( Figure 3—figure supplement 1B ) , indicating a possible function of the β-barrel domain in either substrate recognition or maintaining structural integrity of the enzyme . 10 . 7554/eLife . 03714 . 010Figure 3 . The TibC protomer structure and the glycotransferase domain . ( A ) Ribbon diagram of the TibC protomer structure . The upper right shows tetrahedral coordination of the ferric ion by four cysteine residues . Shown on the lower right is a lateral view with the β-hairpin thumb highlighted as magenta ribbons . The domain organization along the primary structure is on the lower left . ( B ) Structural comparison of TibC with GT-B glycosyltransferases . Left , the catalytic domain of ADP-D , D-heptose-bound TibC structure determined in this study; Center , WaaC bound with ADP-2-deoxy-2-fluoro-heptose ( PDB ID: 2H1H ) ; Right , MurG bound with UDP-GlcNAc ( PDB ID: 1NLM ) . The overall structures and sugar ligands are in ribbon diagram and orange sticks , respectively . The N- and C-lobe are colored blue and yellow , respectively . The β-hairpin thumb insertion and the iron-finger motif in TibC are omitted for clarity . The linker between the two lobes and an extra C-terminal helix in MurG are in green and magenta , respectively . The dashed circle highlights the absence of α-helices at the N terminus of the N-lobe in TibC . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01010 . 7554/eLife . 03714 . 011Figure 3—figure supplement 1 . Interaction between the β-barrel and the N-lobe of the catalytic domain in TibC protomer . ( A ) Structural interactions between the β-barrel and N-lobe . Shown in the upper panel is the overall structure using the same color scheme as that in Figure 2A . The interface between the two domains is circled with dashed lines and the detailed interactions are in the lower panel with interacting residues in sticks . Polar interactions are represented by black dashed lines with a number denoting the distance in angstrom . R97 and Y162 form a pair-wise cation-π interaction; Y70 and P179 have a π-stacking interaction . Interface residues including F28 , F71 , L195 , I101 , V99 , L94 , and Y165 are clustered together to form a buried hydrophobic core . ( B ) Effects of β-barrel deletion on TibA glycosylation . TibC or indicated TibC truncations were co-expressed with Flag-TibA in E . coli BL21 ( DE3 ) . Expression of TibA and TibC was detected by anti-Flag and anti-TibC immunoblotting , respectively . TibA glycosylation was detected with the ECL glycoprotein detection kit . ΔN80 and ΔN95 refer to deletion of the N-terminal 80 and 95 residues in TibC , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 011 The catalytic domain contains two lobes ( residues 98–195 and 218–406 , respectively ) linked by a long loop ( residues 196–217 ) . The N-lobe , containing β9–12 , α1–4 and surrounding loops , adopts a compact Rossmann-like globular fold . Different from the classical Rossmann fold that has α-helices flanking the central β-sheet , the β-sheet in the N-lobe is flanked by α-helices ( α1–4 ) at one side but the β-barrel at the other side ( Figure 3A ) . The C-lobe ( α5–11 and β13–18 ) features a central β-sheet ( β13 and β16–18 ) sandwiched by α7/α8 at one side and α6/α11 at the other side , adopting a typical Rossmann fold . The catalytic cores of nearly all known glycosyltransferases , despite the extraordinary sequence divergence , converge onto two general folds—namely , GT-A and GT-B ( Unligil and Rini , 2000; Breton et al . , 2006; Lairson et al . , 2008; Breton et al . , 2012 ) . The typical GT-A fold consists of two α/β/α domains with a continuous central β-sheet . The GT-A glycosyltransferase features a D×D motif that binds to the nucleotide phosphate in the sugar donor via a metal ion ( Bourne and Henrissat , 2001 ) . The GT-B fold has two Rossmann-like domains separated by a cleft , in which a sugar donor is positioned for nucleophilic attack by the substrate . The GT-B glycosyltransferase does not require metal ion for catalyzing sugar transfer ( Wrabl and Grishin , 2001; Hu and Walker , 2002 ) . The catalytic domain of TibC structurally more resembles the GT-B fold such as WaaC and MurG , two bacterial glycosyltransferases involved in cell wall synthesis ( Hu et al . , 2003; Grizot et al . , 2006 ) , with its own distinctions ( Figure 3B ) . One major difference is that the N-lobe of the TibC catalytic domain is not a typical Rossmann fold; its central β-sheet is four-stranded rather than six-stranded for a Rossmann fold and also misses the flanking α-helices at one side ( Figure 3B ) . The TibC catalytic core is accessorized by two unique structural modules , the β-hairpin thumb and the iron-finger motif , both of which are important for the dodecamer assembly . The β-hairpin thumb consists of β15 , β16 , and a four-residue hairpin loop projecting away from the globular C-lobe ( Figure 3A ) . The β-hairpin thumb mediates the hand-in-hand contact between adjacent dimers ( Figure 2B ) . Phe-265 and Val-266 at the tip of the thumb , together with a nearby Phe-368 from the catalytic domain , contact the same set of residues from the adjacent protomer to form a hydrophobic cluster ( Figure 4A ) . To validate the structural observations , the three residues were each subjected to mutational analyses . TibC F368D mutant was largely insoluble . Both F265D and V266D mutants eluted as a dimer from the gel filtration column ( Figure 4B ) and failed to catalyze TibA heptosylation in E . coli ( Figure 4C ) . These results highlight the importance of the β-hairpin thumb in mediating the hand-in-hand association for assembly of an active TibC dodecamer . The data also explain the unique presence of two successive hydrophobic residues on TibC β-hairpin loop , which differs from other β-hairpin loops that are usually dominated by hydrophilic residues ( Sibanda and Thornton , 1985 ) . The β-hairpin thumb is highly conserved in the BAHT family ( Figure 1—figure supplement 1 ) , supporting the view that an intact dodecamer is critical for this family of heptosyltransferase to modify its partner autotransporter . 10 . 7554/eLife . 03714 . 012Figure 4 . The β-hairpin thumb-mediated hand-in-hand contact required for TibC dodecamer assembly and catalytic activity . ( A ) β-hairpin thumb-mediated hand-in-hand contact . Upper , surface presentation of two adjacent TibC protomers with the β-hairpin thumb in ribbons . Lower , interface details with key residues in sticks . Black lens , the twofold axis . ( B and C ) Effects of hand-in-hand contact mutations on TibC dodecamer formation and catalyzing TibA glycosylation . TibC ( WT or indicated mutant ) proteins were loaded onto a gel filtration column in ( B ) . Black arrows mark the molecular weight calibration . Vo , void volume . The experiments in ( C ) were performed and data are presented similarly to those in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 012 The other insertion module ( residues -368 ) features two knuckles with long exposed loops capping and embellishing the C-lobe ( Figure 3A ) . Within the module , a ferric ion is tetrahedrally coordinated by four conserved cysteine residues ( Cys-339 , 342 , 358 , and 370 in TibC ) ( Figure 3A and Figure 1—figure supplement 1 ) , analogous to the zinc-finger motif . This iron-finger motif lies ∼20 Å from the catalytic cleft ( see below ) and its architecture belongs to the 1Fe-0S cluster . Differing from the most common zinc-finger motif , the 1Fe-0S iron-finger motif is only observed once in rubredoxin protein found in sulfur-metabolizing bacteria and archaea ( Lovenberg and Sobel , 1965; Adman et al . , 1975 ) . In rubredoxin , the 1Fe-0S cluster acts as an electron transfer carrier ( Sieker et al . , 1994 ) . The iron-finger motif in TibC , however , does not seem to function in electron transfer according to our structural and functional analyses . Notably , mutation of any of the four iron-coordinating cysteine residues resulted in a colorless TibC protein ( Figure 5A ) due to the loss of iron binding . The four cysteine mutants eluted from the gel filtration column as heterogeneous interconverting oligomers with their sizes smaller than a dodecamer ( Figure 5B ) . Thus , the iron finger is required for maintaining structural integrity of the TibC dodecamer . Consistently , the four iron binding-deficient mutants all failed to catalyze TibA heptosylation in the E . coli system ( Figure 5C ) . 10 . 7554/eLife . 03714 . 013Figure 5 . The iron-finger motif critical for TibC dodecamer assembly and heptosyltransferase activity . ( A ) Purified TibC ( WT and the four iron-finger cysteine mutants , ∼8 mg/ml ) were loaded onto a SDS-PAGE gel followed by Coomassie blue staining . ( B ) Gel filtration chromatography of the iron-finger cysteine mutants of TibC . Black arrows mark the molecular weight calibration and Vo denotes the void volume . ( C ) Analyses of TibA glycosylation by the iron-finger cysteine mutants of TibC . Indicated proteins were co-expressed in E . coli BL21 ( DE3 ) and the lysates were analyzed by anti-Flag ( TibA ) and anti-TibC immunoblotting and also TibA heptosylation assay . ( D ) The back-to-back dimer formation . Shown on the left is the surface presentation colored as indicated . Two symmetric dimerization interfaces are marked by dashed circles . Structural details of one interface are shown on the right with interacting residues in sticks . Polar interactions are represented by black dashed lines with a number denoting the distance in angstrom . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 013 Structurally , the iron-finger motif contributes to the back-to-back dimerization . A portion of the iron-finger motif , together with several surface residues from the catalytic domain of one protomer , makes extensive contacts with the β-barrel as well as the catalytic domain of the other protomer in the back-to-back dimer ( Figure 5D ) . A symmetric interface is formed via the same kind of interactions between the two protomers . Extensive polar and hydrophobic contacts occur at the interface ( Figure 5D ) , burying ∼1467 Å2 solvent-accessible surface area of TibC ( ∼9% of the total surface areas ) . Dimerization via this large interface provides building blocks for TibC dodecamer assembly . We also crystallized the TibC dodecamer in complex with its natural ligand ADP-D-glycero-β-D-manno-heptose ( ADP-D , D-heptose ) . This was accomplished by soaking the sugar ligand into crystals of the catalytically inactive TibC D110A mutant . Structural determination was achieved by molecular replacement , which revealed extra omit density in the TibC catalytic domain . Placing the sugar into the omit density led to a concomitant loss of Fo–Fc density after the refinement . ADP-D , D-heptose was convincingly modeled into each protomer and a final model of 3 . 88 Å was obtained ( Table 1 ) . The 12 sugar ligands form an array of two parallel circles along the inner surface of the dodecamer ring and face towards the hollow center ( Figure 6A ) . ADP-D , D-heptose is located at the cleft between the two lobes of the catalytic domain ( Figure 6B and Figure 3B ) . The adenine heterocycle fits snugly into a hydrophilic pocket and its N1 atom bears a hydrogen bond contact with Arg-286 of TibC ( Figure 6B ) . The hydrogen bond contact , though weak , is incompatible with a guanine due to its saturated N1 , providing a structural basis for the specificity of TibC in using ADP-activated sugar donor . Moreover , the heptose hydroxyl O4 forms a hydrogen bond with Trp-305 indolic N1 ( Figure 6B ) . The β-phosphate has strong interactions with Thr-226 and Lys-230 in the ligand-binding pocket of TibC . Lys-230 is also close to the heptose hydroxyl O5; its positive charge could potentially alleviate the negative charge of β-phosphate , thereby facilitating nucleophilic attack of the heptose C1 by the substrate acceptor . 10 . 7554/eLife . 03714 . 014Figure 6 . Crystal structure of the TibC-ADP-D , D-heptose complex and sugar ligand stereoselectivity . ( A ) Structure of the TibC-ADP-D , D-heptose complex . Left , a top view of the dodecamer . Right , a half-cut view of the back-to-back dimer . ADP-D , D-heptose and ferric ions are shown as ball-and-stick models and red spheres , respectively . ( B ) Binding of ADP-D , D-heptose to TibC protomer . Key sugar-binding residues are in sticks . Polar interactions are indicated by black dashed lines with a number denoting the distance in angstrom . ADP-D , D-heptose is shown as orange sticks meshed with σA-weighted 2mFo-DFc electron density map contoured at 1 . 0 σ . The mFo–DFc omit density of the ligand at a contour level of 3 . 0 σ is shown in blue on the right . ( C ) Sugar ligand stereoselectivity and effects of swapping Pro-300 in TibC with the corresponding Ser-294 in autotransporter adhesin heptosyltransferase ( AAH ) . The raw experimental data are in Figure 6—figure supplement 2A . ( D ) Comparison of ADP-D , D-heptose binding residues in TibC ( green ) and AAH ( pink ) . The AAH structure was modeled from that of TibC . Conserved residues are shown as lines; divergent residues are in sticks . ADP-D , D-heptose is in orange sticks with the chiral C6 as a sphere . ( E ) A proposed model for the ligand stereoselectivity . The stereoselectivity determinants ( Pro-300 in TibC and Ser-294 in AAH ) are in sticks . The distance between the sugar ligand and the Pro-300/Ser-294 are denoted as a number in angstrom . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01410 . 7554/eLife . 03714 . 015Figure 6—figure supplement 1 . Requirement of TibC heptosyltransferase activity for bacterial adhesion to host cells . ( A ) Assays of TibA and AIDA-I glycosylation by TibC mutants in E . coli . BL21 ( DE3 ) cells were transformed with a Flag-TibA ( upper ) or AIDA-I ( lower ) expression plasmid alone ( vector ) or together with an indicated TibC mutant construct ( pTibC ) . TibA and AIDA-I glycosylation was detected with the Amersham ECL glycoprotein detection kit . ( B ) Assays of TibC mutants in supporting bacterial adhesion to HeLa cells . BL21 ( DE3 ) cells were transformed with an AIDA-I expression plasmid together with an indicated TibC mutant . Adhesion assay was performed and data were presented similarly to those in Figure 1A . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01510 . 7554/eLife . 03714 . 016Figure 6—figure supplement 2 . Assays of the ligand stereoselectivity of TibC and autotransporter adhesion heptosyltransferase ( AAH ) . ( A ) A TibA-derived synthetic peptide ( GGVQHVSSGGSATSSTINSGG ) ( theoretical molecular weight 1846 . 5 Da ) was reacted with wild-type ( WT ) or indicated TibC or AAH mutants in the presence of indicated sugar donors . The peptides were then subjected to MALDI-TOF mass spectrometry analysis and shown are the spectra obtained . Asterisks mark the peptides modified by one ( * ) or two ( ** ) heptose molecules . ( B ) In vitro heptosylation assays of recombination AAH using GST-AIDA-I531-611-Flag substrate . 1 μM GST-AIDA-I531-611-Flag , 100 μM indicated sugar ligand and recombinant AAH with a concentration gradient ( from 0 . 01 μM to 1 μM ) were used . The reaction mixtures were subjected to immunoblotting and glycoprotein detection analyses . Purified GST-AIDA-I531-611-Flag and GST-AIDA-I531-611-Flag coexpressed with AAH were used as negative ( NC ) and positive control ( PC ) . Short and long exposures of the glycosylated AIDA-I are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 016 Superimposition of the TibC sugar binding site with those of the GT-B glycosyltransferases WaaC and MurG identified Asp-110 as the candidate catalytic base that activates the substrate acceptor . Asp-110 is located close to the anomeric hydroxyl of heptose in the TibC–ADP-D , D-heptose complex structure ( Figure 6B ) . All four sugar-binding residues ( Arg-286 , Trp-305 , Thr-226 , and Lys-230 ) as well as the catalytic base ( Asp-110 ) are highly conserved in the BAHT family ( Figure 1—figure supplement 1 ) . Alanine substitution of Asp-110 , Arg-286 , Trp-305 , or Lys-230 in TibC all abolished or largely diminished TibA/AIDA-I heptosylation in E . coli ( Figure 6—figure supplement 1A ) ; these point mutants also failed to support AIDA-I-mediated bacterial adhesion to HeLa cells ( Figure 6—figure supplement 1B ) . The results also confirm the functional importance of the heptosyltransferase activity of TibC and the BAHT family . AAH accepts both ADP-D , D-heptose and ADP-L , D-heptose as the sugar donor despite a possible slight preference for the latter ( Figure 6C and Figure 6—figure supplement 2 ) . In contrast , TibC exhibited a high stereoselectivity for ADP-D , D-heptose . To reveal the underlying structural basis , we modeled the structure of sugar ligand-bound AAH using the TibC complex as the template and examined TibC/AAH residues within 5 Å from the sugar ligand ( Figure 6D ) . This led to the identification of residue 300 in TibC ( 294 in AAH ) as a possible structural determinant . In TibC , residue 300 is a proline and its main-chain carbonyl oxygen forms two hydrogen bonds with D , D-heptose hydroxyl O4 and O6 , respectively ( Figure 6E ) . Notably , the latter interaction is incompatible with L , D-heptose due to the chiral inversion of C6 and the resulting spatial distance . For AAH , the side-chain hydroxyl of Ser-294 can have a polar interaction with hydroxyl O6 in D , D-heptose or hydroxyl O7 in L , D-heptose ( Figure 6E ) , thereby capable of tolerating both sugar ligands . Consistent with these structural indications , TibC P300S mutant could use both ADP-D , D-heptose and ADP-L , D-heptose while AAH S294P mutant instead became selective for ADP-D , D-heptose ( Figure 6C and Figure 6—figure supplement 2A ) . Thus , Pro-300-mediated polar interaction is a critical structural determinant for TibC sugar ligand stereoselectivity . We further succeeded in purifying a TibC–TibA enzyme/substrate complex by co-expression of the sugar binding-deficient TibC K230A mutant with TibA55-350 in E . coli . The complex eluted from the gel filtration column as a homogenous fraction with a size larger than the TibC dodecamer . The molecular weight of the giant enzyme/substrate assembly was determined to be 750 kDa by analytical ultracentrifugation ( Figure 2—figure supplement 2 ) , matching well the composition of a TibC dodecamer and six TibA molecules . Cryo-EM micrographs revealed a well-defined shape of the TibC–TibA dodecamer/hexamer complex at an intermediate resolution ( Figure 7—figure supplement 1A ) , allowing for single particle structural analysis ( Figure 7—figure supplements 1 , 2 ) . Averaging and refinement without classification from the whole dataset of 53 , 303 particles produced an electron density map of 9 . 7 Å in which the TibC12–TibA6 stoichiometry became immediately evident ( Figure 7A ) . In this low-resolution model , density agreeing well with the crystal structure of the TibC dodecamer could be clearly identified , but the TibA moiety is slightly blurred and segmented , indicating a conformational heterogeneity . The reconstructed maps could be further classified into two classes based on different features of the TibA density ( Figure 7—figure supplement 2 ) . Two maps of higher resolution at 8 . 2 Å and 8 . 9 Å were then reconstructed by averaging 35 , 300 and 18 , 003 sorted particles , respectively ( Figure 7—figure supplement 2 ) . As a result , a higher contrast in the TibA structure was observed between the two models despite a similar overall architecture for the entire complex ( Figure 7A ) . These analyses indicate two major conformational states of the TibC–TibA enzyme/substrate complex in the absence of the sugar donor . 10 . 7554/eLife . 03714 . 017Figure 7 . Cryo-EM models of the TibC–TibA55-350 dodecamer/hexamer complex . ( A ) Cryo-EM maps of the TibC–TibA55-350 complex . The left is the overall average map obtained from classification-free reconstruction . A two-class classification was employed to distinguish different features observed with the TibA density ( Figure 7—figure supplement 2 ) , and the resulting two maps are shown in the middle and right panels . ( B ) Model fitting of TibA55-350 crystal structure into the two classes of experimental maps and comparison with the post-molecular dynamics ( MD ) model . The cross-correlation coefficient ( CCC ) between TibA and the experimental map is listed underneath the models . The protrusion loop is highlighted in orange and the to-be-glycosylated Ser-176 is in sticks . ( C ) Cryo-EM reconstruction of the TibC–TibA dodecamer/hexamer complex . The atomic model after MD simulation was fit into the cryo-EM map ( the upper panel ) . Ribbon diagram of TibC dodecamer is colored as that in Figure 2A . The lower panel shows the TibA–TibC2 trimer in the active and resting states . ADP-heptose and Ser-176 are in sticks . The conformationally changed TibA loop is highlighted in orange . ( D ) The CCC for TibA during the steered MD simulation . The distance from TibA Ser-176 OG to heptose C1 ( blue line ) and the local CCC value for TibA ( red line ) are plotted as a function of time . ( E ) Protrusion of the TibA loop sampled by the steered MD simulation . The snapshots at 100 , 200 , 300 , and 400 ps were fitted into the active state cryo-EM map . The protrusion loop is in orange and Ser-176 is in sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01710 . 7554/eLife . 03714 . 018Figure 7—figure supplement 1 . Cryo-EM images of the TibC-TibA55-350 complex . ( A ) The cryo-EM image of TibC–TibA55-350 complex acquired at a nominal magnification of 75 , 000× . The white and green boxes indicate typical top and side view particles , respectively . Scale bar , 50 nm . ( B ) Fourier shell correlation ( FSC ) of cryo-EM reconstruction of the overall , resting , and active conformation states . Using the gold standard FSC = 0 . 143 criterion , the resolutions for the overall average , resting , and active states are 9 . 7 Å , 8 . 7 Å , and 8 . 2 Å , respectively . ( C ) Euler angle distribution plot of the resting and active conformation states . The size of the black dot is proportional to the number of particles that belong to the specific Euler angle . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01810 . 7554/eLife . 03714 . 019Figure 7—figure supplement 2 . Cryo-EM reconstruction of the TibC12-TibA6 complex . ( A ) Low-pass filtered TibC dodecamer was used to initiate the reconstruction . ( B ) After 3D classification , three conformational states appeared . Two similar ones ( middle and right ) were merged into one super class ( Class II ) due to their similar TibA density ( red circles ) . The two classes were then subjected to independent auto-refinement . ( C ) Following the auto-refinement , the two reconstructed density maps showed more evident discrepancy . Particles in each class were sorted into two independent stacks and subjected to further projection-matching refinement . ( D ) The resulting final maps had higher resolution with more structural details compared with the overall average map ( gray ) obtained by auto-refinement and reconstruction without classification . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 01910 . 7554/eLife . 03714 . 020Figure 7—figure supplement 3 . Molecular dynamics ( MD ) -assisted cryo-EM analysis reveals two conformational states of the TibC–TibA complex . ( A ) Catalytic center of the resting state TibC–TibA55-350 complex . Left , structure of one TibC protomer ( TibC-A ) in a back-to-back dimer . Right , structure of the other TibC protomer ( TibC-B ) aligned with TibC-A using the Cα trace as the reference . The dashed line with a number in angstrom marks the candidate substrate serine to Asp-110 or heptose C1 . ( B ) Comparison of the rigid body crystal structure model fitting and the post-MD model fitting into the active state EM map . Red arrows mark the major structural differences . ( C ) Catalytic center of the active state TibC–TibA55-350 complex . Upper , the active state TibC protomer ( TibC-A ) in a back-to-back dimer . Lower , structure of the other TibC protomer ( TibC-B ) aligned with the resting state TibC-A using the Cα trace as the reference . The dashed line with a number in angstrom marks the candidate substrate serine to Asp-110 or heptose C1 . ( D ) Overall structure of one TibC back-to-back dimer ( blue and magenta ) bound with one TibA ( green ) in the active state . DOI: http://dx . doi . org/10 . 7554/eLife . 03714 . 020 The 8 . 9 Å EM map fits remarkably well with crystal structures of TibA55-350 ( PDB ID code: 4Q1Q ) ( Lu et al . , 2014 ) and the TibC dodecamer ( Figure 7B ) , allowing for reliable atomic model docking . The cross-correlation coefficient ( CCC ) for TibA in the final model is 0 . 76 ( only a small N-terminal portion of TibA55-350 is out of the density ) . The reconstruction revealed that six TibA55-350 molecules are distributed along the inner surface of the TibC dodecamer ring and share its sixfold axis ( Figure 7C ) . Importantly , each TibA55-350 binds to two TibC protomers of a back-to-back dimer , suggesting that two enzyme molecules are responsible for catalyzing full heptosylation of one TibA substrate . Ser-176 is the closest sugar acceptor situated at the entrance of the catalytic cleft of one TibC protomer ( TibC-A ) , but the distance to the catalytic base exceeds 20 Å ( Figure 7C and Figure 7—figure supplement 3A ) . A similar distance was observed between the other TibC protomer ( TibC-B ) and its possible substrate acceptor ( Figure 7—figure supplement 3A ) . Thus , the 8 . 9 Å TibC–TibA dodecamer/hexamer complex model represents a sugar transfer incompetent state , designated hereafter as the resting state . Moreover , the spiral shaft of TibA β-helix adopts a trend roughly parallel to the sixfold axis of the dodecamer which , together with the patterned solenoid-like surface distribution of heptosylation sites on TibA ( Lu et al . , 2014 ) , indicates a possible screw propelling-like mechanism for processive heptosylation of TibA by the dodecamer ring . The static X-ray structures of TibC dodecamer and TibA55-350 were also docked into the 8 . 2 Å EM map , but the TibA moiety had a few degrees of bending with a protrusion into the catalytic center of one TibC protomer ( Figure 7A , B and Figure 7—figure supplement 3B ) . The extra protruding density could not be appropriately fit despite extensive attempts at possible solutions . The heterogeneity observed in the cryo-EM image likely reflects the dynamic catalytic motions; the protruding density was tentatively interpreted as the TibA loop bearing the sugar acceptor Ser-176 . This interpretation was supported by molecular dynamics ( MD ) simulation of a resting state TibC2–TibA55-350 trimer . In the 400 ps steered MD simulation , pulling Ser-176 in TibA55-350 towards the heptose led to a proportional increase in the CCC ( Figure 7D , E ) . A subsequent 20 ns unbiased MD simulation produced an ensemble of TibA models that matched the EM density with a best overall CCC value of 0 . 87 ( Figure 7B ) . The final model of the entire complex was well accommodated by the experimental map ( Figure 7B and Figure 7—figure supplement 3B ) . The protruding loop was assigned to residues 170–180 of TibA , resulting from disruption of a local β-strand within these residues . The loop placed Ser-176 into a position ready for attacking the sugar donor in one TibC protomer ( TibC-A ) ( Figure 7C and Figure 7—figure supplement 3C ) . This ‘protrusion’ state of TibA was designated as the catalytically competent active state . Notably , the other TibC protomer ( TibC-B ) in the back-to-back dimer and its possible substrate acceptor appeared to be in the resting state ( Figure 7—figure supplement 3C , D ) , suggesting that only one of the two enzyme molecules in the dimer can be productive in transferring the sugar onto TibA at a time . It is worth mentioning here that the EM dataset presumably covers a continuous series of catalytic stages; the two discrete classifications employed in our reconstruction may not give a complete picture of all catalytic motions of the TibC–TibA enzyme/substrate complex . The active state structure , which was reconstructed from two-thirds of the total particles , probably contains the majority of representative conformational stages in the catalytic process and therefore is used to represent the overall structure of the TibC–TibA dodecamer/hexamer complex ( Figure 7C ) . TibC is distinct from known glycosyltransferases in that it adopts an unusual giant dodecamer ring architecture in catalyzing simultaneous glycosylation of six molecules of TibA autotransporter . The selection of a double-layered ring with a large hollow channel is advantageous and provides a favorable environment for cooperative catalysis in a structurally economical manner . Assembly of the dodecamer into a ring can reduce the solvent-accessible surface area and renders a stable enzymatic machinery complex beneficial to efficient catalysis . Moreover , the spatial arrangement of the back-to-back TibC dimer , binding to the different glycosylation sites in one TibA , suggests a divalent catalysis is also probably contributing to the high catalytic efficiency required for hyperglycosylation . Thus , the dodecamer/hexamer enzyme/substrate complex with six symmetric units may represent a best compromise between structural economy and catalytic cooperativity/efficiency . The TibC complex , compared with other macromolecular assemblies with large interior volumes such as ferritin and multimeric chaperones/proteases , is more permeable with 12 titled grooves extending from outside to the interior chamber . These grooves are optimally imbedded and close to the catalytic center , allowing for easy access of ADP-heptose from outside into the TibC particle . The catalytic center , on the other hand , is structurally best accessible from the interior , where TibA is located and can readily deliver the to-be-glycosylated serine by loop protrusion . We further speculate that the energy used to drive the rotation and shuttle motion of the TibA substrate is partially derived from breakdown of the ADP-heptose glycosidic bond occurring during catalysis . Further experiments such as obtaining a higher resolution cryo-EM structure of the TibC12–TibA6 complex , which may capture more catalytic stages/events , are necessary to validate the proposed model for TibC-catalyzed processive heptosylation of TibA . Another most unusual feature of the TibC protomer structure is the presence of the 1Fe-0S iron-finger motif . The iron-finger motif does not fulfill the previously defined function of serving as an electron carrier in rubredoxin protein . Biochemical and structural studies have underscored the singular importance of this motif in mediating the back-to-back dimer formation and TibA substrate recognition . Interestingly , the iron-finger motif , adopting a topological fold similar to the zinc-finger motif , follows a right-handed helical path in binding to TibA β-helix , which is analogous to DNA binding by zinc-finger motifs . In addition , the recognition specificity and the interaction strength of this motif will have been precisely tuned during macromolecular evolution so that specific substrate recognition , sliding of the TibA β-helix along the reaction channel , and efficient protrusion of the TibA loop can all occur in a catalytically productive manner . It will be interesting to elucidate the detailed mechanism by which iron-finger motif-mediated substrate recognition is effectively coupled with its promoting of the TibA loop protrusion . The cDNAs encoding TibC and TibA were PCR amplified from genomic DNA of ETEC strain H10407 . The full-length tibC was cloned into pGEX-6p-2 vector ( GE Healthcare , UK ) for recombinant expression in E . coli . To co-express TibA55-350 and TibC K230A for capturing the TibC–TibA55-350 enzyme–substrate complex , the tibA fragment was cloned into the pGEX-6p-2 vector and a catalytically inactive mutant of TibC was inserted into the pACYCDuet vector . AAH and AIDA-I constructs have been described previously ( Lu et al . , 2014 ) . All constructs were generated by standard molecular cloning . Point mutations were generated by the QuickChang Site-Directed Mutagenesis Kit ( Stratagene , La Jolla , CA ) . All the plasmids were verified by DNA sequencing . All recombinant proteins and protein complexes were expressed in E . coli BL21 ( DE3 ) Gold strain ( Agilent Technologies , Santa Clara , CA ) with 0 . 5 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) for overnight induction at 22°C in Luria–Bertani ( LB ) medium . TibC was purified by glutathione affinity chromatography and the GST tag was removed by PreScission protease digestion . TibC was further purified by anion exchange and gel filtration chromatography using HiTrap Q HP and Superdex 200 Hi-load columns ( GE Healthcare ) , respectively . The TibC–TibA55-350 complex was initially purified using one-step glutathione affinity chromatography . The GST moiety on TibA was removed by PreScission protease cleavage . The complex was further separated from free TibA55-350 by gel filtration chromatography using the Superdex 200 Hi-Load column . SeMet-labeled TibC and TibA were expressed in the methionine-auxotrophic E . coli B834 ( DE3 ) strain ( Novagen , Germany ) using SeMet-supplemented M9 medium . SeMet protein was purified following the same scheme as that used for the native protein . All the crystallization experiments were carried out at 20°C using the hanging drop vapor diffusion method . The purified TibC was concentrated to 20 mg/ml in a buffer containing 10 mM Tris–HCl ( pH 7 . 6 ) , 100 mM NaCl , and 2 mM DTT . Crystals of both native and SeMet-TibC protein were grown in 8% ( wt/vol ) PEG 8000 , 120 mM magnesium acetate , and 100 mM MES buffer ( pH 5 . 5 ) for two days . Crystals were cryoprotected with the mother liquid containing 19% ( vol/vol ) ethylene glycol and 1% ( vol/vol ) DMSO and flash-frozen with liquid nitrogen for data collection . The TibC D110A mutant crystals were grown at the same condition . Before freezing , the crystals were soaked with 5 mM ADP-D , D-heptose in the mother liquid for 1 hr . All the diffraction data were collected at BL-17U of Shanghai Synchrotron Radiation Facility ( SSRF ) at 100 K and processed with the HKL 2000 suite by the routine procedure ( Otwinowski and Minor , 1997 ) . The structure of TibC dodecamer was solved by the single wavelength anomalous dispersion ( SAD ) method . In the TibC data , a total of 36 heavy atoms were determined using the program SHELXD ( Schneider and Sheldrick , 2002 ) . The identified heavy atom sites were then refined and the initial phases were calculated using the program autoSHARP ( Global Phasing Ltd ) . The phase was then refined and extended to 2 . 9 Å by NCS averaging , solvent flattening and histogram matching using the DM module in CCP4 ( Dodson et al . , 1997 ) . After density modification , the map was sufficiently interpretable for model building . The model was automatically built using the Buccaneer software ( Cowtan , 2006 ) , which generated ∼70% complete model . The remaining parts were manually built and adjusted in Coot ( Emsley et al . , 2010 ) . The model was refined in Refmac ( Murshudov et al . , 1997 ) . The TibC–ADP-D , D-heptose complex structure was solved by molecular replacement in Phaser MR ( McCoy et al . , 2007 ) using the apo-TibC structure as the template . The final model was adjusted in Coot and refined in Refmac . The structural pictures were generated in Pymol ( www . pymol . org ) . Purified recombinant TibC ( WT or indicated mutants ) were loaded onto a HiLoad 16/600 Superdex 200 column ( GE Healthcare ) in a buffer containing 20 mM Tris–HCl ( pH 7 . 6 ) and 20 mM NaCl on an ÄKTA Purifier ( GE Healthcare ) . Fractions corresponding to the specified elution volumes were collected and loaded onto SDS-PAGE gels followed by Coomassie blue staining . In vivo heptosyltransferase activity of TibC was monitored by co-expressing pACYCDuet-Flag-TibA ( the Flag tag was inserted at the junction between the signal peptide and the passenger domain ) and pET21a-TibC ( WT and indicated mutants ) . 1/10 of 100 μl of total lysates from 1 mL of E . coli culture was used for immunoblotting and glycoprotein detection analyses . Expression levels of TibA and TibC were detected by anti-Flag M2 mouse monoclonal antibody ( Sigma-Aldrich , St . Louis , MO ) and in-house generated rabbit anti-TibC serum , respectively . Glycosylated proteins were detected with the ECL glycoprotein detection module ( GE Healthcare , product code RPN2190 ) according to the manufacturer’s protocol . For in vitro analysis of TibC and AAH heptosyltransferase activities , 10 μg of the TibA-derived synthetic peptide ( GGVQHVSSGGSATSSTINSGG , synthesized by Scilight Biotechnology LLC , China ) were incubated with 2 μg of purified TibC or AAH protein ( WT or indicated mutants ) for 4 hr at 30°C in a 20 μl reaction containing 10 mM Tris–HCl ( pH 7 . 5 ) , 50 mM NaCl , 10 mM MgCl2 , 1 mM DTT , and 0 . 5 mM ADP-L , D-heptose or ADP-D , D-heptose ligands ( Zamyatina et al . , 2003 ) . The reaction mixtures were subjected to mass spectrometry analysis . Analytic ultracentrifugation was carried out with an XL-I analytical ultracentrifuge ( Beckman-Coulter , Indianapolis , IN ) with An-60 Tirotor at 4°C . The TibC and TibC–TibA complex proteins were prepared at 0 . 8 mg/ml in a buffer containing 20 mM Tris–HCl ( pH 8 . 0 ) , 150 mM NaCl , and 2 mM DTT . All data were collected at a speed of 25 , 000/32 , 000 rpm . 3 mM path length charcoal-filled Epon centerpieces were used for the TibC–TibA complex due to the sample quality limitation and a 12 mM path length aluminum centerpiece was used for the TibC sample . Samples were monitored real time using interference optics at intervals of 4 min . Buffer alone was used as the reference . The molecular weights were calculated by the SEDFIT program ( Schuck , 2000 ) . 10 μg/ml bacitracin ( Sigma ) was added to the purified protein to obtain mono-dispersed particles and make the orientation distribution more anisotropic . An aliquot of 3 . 5 μl of TibC or TibC–TibA55-350 complex ( ∼1 mg/ml ) was absorbed onto a glow-discharged Quantifoil holey carbon grid ( R2 . 1 , 300 mesh; Jena Biosciences , Germany ) . The grid was blotted in an FEI Vitrobot Mark IV ( FEI Company , Hilsboro , OR ) using 4 s blotting time and blotting force 2 with 100% humidity at 283 K , and then plunged into the chilled liquid ethane . The grids were immediately transferred into a Titan Krios electron microscope ( FEI Company ) operated at 300 kV . For the TibC dodecamer , approximately 3000 images were semi-automatically collected following the Leginon procedure ( Carragher et al . , 2000 ) using a 4 K × 4 K Gatan UltraScan4000 CCD ( model 895 , Gatan Inc , Pleasanton , CA ) at a nominal magnification of 75 , 000× . For the TibC–TibA55-350 complex , ∼1800 cryo-EM images were recorded using a 2K × 2K Gatan Tridiem CCD camera ( Gatan Inc ) at a nominal magnification of 81 , 000× with the energy filter turned off . The corresponding pixel size on the specimen level is 1 . 196 and 1 . 778 Å/pixel for TibC and TibC–TibA55-350 complex datasets , respectively . The range of nominal defocus is 2 . 3–3 . 6 μm for TibC and 2 . 7–4 . 1 μm for TibC–TibA55-350 data . The electron dose is 20 e−/Å·s for each CCD frame . For both TibC and TibC–TibA55-350 micrographs , the particles were semi-automatically boxed using e2boxer . py in EMAN2 package ( Tang et al . , 2007 ) followed by interactively screening . The particles with the wrong shape and size were discarded . Approximately 10 , 200 and 56 , 000 particle images were cropped out and normalized ( mean = 0 , standard deviation = 1 ) for TibC and TibC–TibA55-350 complex , respectively . After determining the defocus and B-factor , the particles were phase-flipped for contrast transfer function ( CTF ) correction using e2ctf . py in EMAN2 ( Tang et al . , 2007 ) . For the TibC dataset , two-dimensional multivariate statistical analysis-based reference-free classification was performed in EMAN2 to further eliminate particles belonging to the class averages with blurry appearance . Particles contained in the averages , which were redundant in the ring-like view , were also removed to avoid introducing orientation isotropy . A total of 8546 particles from the TibC data were used for subsequent data processing . The initial model was built using e2initialmodel . py with C1 symmetry applied and was iteratively refined against the phase-flipped particles until no further improvement could be obtained . During the projection-matching refinement process , a Fourier ring correlation ( FRC ) was chosen to compare the similarity between the class averages and re-projections of the model , and C6 symmetry was enforced as indicated by the reference-free class averages . 20% of the worst particles in each class were discarded in each round of iteration . The information over a 20–100 Å resolution range and SSNR weighting were considered in the comparison in several early iterations . In the last four rounds of iterations , the information was expanded to the full resolution range and the step angle was assigned to 2 . 0° . The map obtained in the last iteration was then subjected to B-factor correction with a damping factor of −210 Å2 to enhance the high resolution information using the program embfactor ( Fernandez et al . , 2008 ) . For the TibC–TibA55-350 complex data , the processing procedure before reconstruction was similar to that described for the TibC data above . A total of 53 , 303 particles were finally employed for 3D reconstruction . The starting model used was a 60 Å low-pass filtered TibC dodecamer reconstruction ( C6 symmetry applied ) . After several rounds of refinement running , the phase-flipped particle images were imported into RELION ( Scheres , 2012 ) for 3D classification . After 30 rounds of iterations with three-class classification , two classes that contained 35 , 300 members in total and showed almost identical reconstruction in the TibA density were merged together for subsequent 3D refinement . Finally , two clusters that corresponded to the ‘active’ and ‘resting’ conformations of the complex were acquired . After the 3D refinement converged , the particles corresponding to the two clusters were sorted for subsequent single-model projection-matching refinement in EMAN2 . Final maps for the two conformations were independently reconstructed through projection-matching refinement with the step angle of 1 . 68° for the active conformation and 3 . 6° for the resting conformation . Resolution was estimated by the gold standard FSC using the 0 . 143 cutoff value ( Rosenthal and Henderson , 2003; Scheres and Chen , 2012 ) . An additional projection-matching reconstruction test for potential model bias was performed with the resting volume as the initial reference against the active state particle set and vice versa . The structural pictures of cryo-EM reconstruction were drawn in UCSF chimera ( Pettersen et al . , 2004 ) . Homology modeling of AAH was performed in the program MODELLER ( version 9v7 ) ( Marti-Renom et al . , 2000; Bredenberg and Nilsson , 2001 ) using the TibC structure as the template . The quality of the modeled structure was ensured by the high sequence identity between AAH and TibC ( 68% ) ( Figure 1—figure supplement 1 ) . The top DPOE scored model was chosen and validated with PROCHECK ( Laskowski et al . , 1993 ) . Structural models of AAH S294P and TibC P300S mutants were generated directly from the corresponding wild-type structures . Residues within a radius of 5 Å from the mutated residue were refined using the Protein Local Optimization Program ( PLOP ) ( Jacobson et al . , 2002; Zhu et al . , 2007 ) . Molecular dynamics flexible fitting ( MDFF ) ( Trabuco et al . , 2009 ) was used to refine the complex structure by incorporating the EM density map as an external potential into MD sampling . All MD simulations were performed in the program NAMD 2 . 9 ( Phillips et al . , 2005 ) using the CHARMM27 force field ( MacKerell et al . , 1998 ) including the CMAP correction ( Mackerell et al . , 2004 ) . The MDFF simulations were carried out at T = 300 K with a scaling factor ζ = 1 kcal for 10 ns . The snapshot with the top CCC was selected for further unbiased MD simulation . The value of CCC was calculated using the program VMD ( version 1 . 9 . 1 ) ( Humphrey et al . , 1996 ) . The CHARMM force field files ( topology and parameter ) for ADP-D , D-heptose were automatically generated using the server SwissParam ( Zoete et al . , 2011 ) , except that the partial atomic charges of ADP-D , D-heptose were manually assigned using the analogs in the existing CHARMM force field ( Best et al . , 2012 ) . The parameters of ferric ions with four coordinating cysteine residues were modified according to the available parameters developed for the zinc-finger motif ( Eriksson et al . , 1995; Bredenberg and Nilsson , 2001 ) . To sample the intermediate state of loop protrusion in the TibC–TibA complex , steered molecular dynamics ( SMD ) simulation ( Isralewitz et al . , 2001 ) was carried out with one unit of TibC12–TibA6 complex . By fixing the position of the C1 atom of ADP-D , D-heptose , a force was applied to the OG atom of Ser-176 along a vector connecting the two atoms . The constant pulling velocity was set to 0 . 06 Å/ps with a spring constant of 10 kcal/mol/Å2 . The trajectories of TibA were extracted from a total of 400 ps SMD simulation and then fitted into the cryo-EM density of TibA , which was extracted from the total map using ‘Volume Eraser’ in CHIMERA ( Pettersen et al . , 2004 ) . The correlation between the CCC values and the distances of loop protrusion was analyzed . The last snapshot of the TibC–TibA subunit from SMD with minimum distance between C1 and OG was chosen as the initial model of the active state conformation , and a 20 ns unbiased MD simulation was then carried out for further refinement . The MD snapshot with the top CCC score was used to generate the final active state TibC–TibA dodecamer/hexamer complex structure . The MDFF-refined TibC12–TibA6 complex structure was solvated using TIP3P water molecules and then neutralized . The system was minimized for 10 , 000 integration steps and equilibrated for 100 ps with a time step of 1 fs , and the temperature was gradually increased from 25 K to 300 K . Following this , an unbiased MD simulation was performed under a constant temperature of 300 K and a constant pressure of 1 atm using the Nosé–Hoover Langevin piston method ( Feller et al . , 1995 ) with the integration time step of 2 fs . The cutoff distances for electrostatic and van der Waals calculations were set at 10 Å . The long-range electrostatic forces were computed using the particle mesh Ewald method ( Darden et al . , 1993 ) with a grid spacing of 1 Å . All covalent bonds involving hydrogen atoms were constrained with the SHAKE algorithm . Crystal structural data for apo-TibC and TibC/ADP-heptose complex are deposited in the Protein Data Bank ( PDB ) under the accession numbers 4RAP and 4RB4 , respectively . Cryo-EM structure data are deposited in Electron Microscopy Data Bank ( EMD ) with accession codes EMD-2755 ( apo-TibC ) , EMD-2756 ( averaged TibC12–TibA6 complex ) , EMD-2757 ( active state TibC12–TibA6 complex ) , and EMD-2758 ( resting state TibC12–TibA6 complex ) .	Bacteria release proteins known as virulence factors to help them infect host cells . Many bacteria are surrounded by two membranes , so virulence factors must be able to pass through both of these membranes . Autotransporters are a group of virulence factors that pass through the inner membrane and anchor themselves in the outer membrane; this allows part of the autotransporter to project from the surface of the bacterial cell and stick to the surface of the cell that the bacterium is about to infect . Many autotransporters are coated with sugar molecules and this increases their ability to adhere to cells . Enzymes called glycosyltransferases ensure that this sugar coating process takes place . Autotransporters contain two sections: the passenger domain and the beta domain . The passenger domain is important for virulence , while the beta domain forms a pore in the outer membrane that the passenger domain passes through to reach the outer surface of the bacterium . TibA and AIDA-1 are autotransporters associated with two types of bacteria that infect the intestines and cause diarrhea . Similar autotransporters are found in a wide range of bacteria , but the precise details of how these autotransporters are coated with sugar molecules are not fully understood . Yao et al . now show that a glycosyltransferase called TibC , which is found in many bacteria , adds large numbers of sugar molecules to the passenger domains of both TibA and AIDA-1 . To learn more about this process Yao et al . used X-ray diffraction to work out the structure of TibC . Strikingly , this revealed that TibC proteins come together to form a large circular structure that contains two rings , each made of six TibC proteins . The integrity of this structure is maintained by the presence of iron atoms , which also gives TibC a characteristic brown colour . Yao et al . also studied what happens when TibC binds to TibA; a technique called electron cryo-microscopy revealed that six TibA molecules are distributed along the inner surface of the circular TibC structure , with each TibA protein binding to two TibC proteins . This arrangement allows for the efficient transfer of sugar molecules from the glycosyltransferase to the autotransporter .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2014	A structural mechanism for bacterial autotransporter glycosylation by a dodecameric heptosyltransferase family
The synaptonemal complex ( SC ) is an ultrastructurally conserved proteinaceous structure that holds homologous chromosomes together and is required for the stabilization of pairing interactions and the completion of crossover ( CO ) formation between homologs during meiosis I . Here , we identify a novel role for a central region component of the SC , SYP-4 , in negatively regulating formation of recombination-initiating double-strand breaks ( DSBs ) via a feedback loop triggered by crossover designation in C . elegans . We found that SYP-4 is phosphorylated dependent on Polo-like kinases PLK-1/2 . SYP-4 phosphorylation depends on DSB formation and crossover designation , is required for stabilizing the SC in pachytene by switching the central region of the SC from a more dynamic to a less dynamic state , and negatively regulates DSB formation . We propose a model in which Polo-like kinases recognize crossover designation and phosphorylate SYP-4 thereby stabilizing the SC and making chromosomes less permissive for further DSB formation . Meiosis is a specialized cell division program during which the diploid germ cell genome is halved to generate haploid gametes , and therefore it is critical for sexual reproduction . This halving is achieved by following one round of DNA replication with two consecutive rounds of cell division where homologs segregate away from each other at meiosis I , and sister chromatids segregate at meiosis II . Accurate chromosome segregation at meiosis I depends on several earlier key steps including pairing and the assembly of the synaptonemal complex ( SC ) between homologous chromosomes ( homologs ) , formation of programmed DNA double-strand breaks ( DSBs ) and the repair of a subset of these DSBs as interhomolog crossover ( CO ) recombination events ( Page and Hawley , 2003 ) . The SC plays a conserved and central role during meiosis . The assembly of the SC is required for the stabilization of pairing interactions between the homologs and for interhomolog CO formation ( Nag et al . , 1995; Storlazzi et al . , 1996; Page and Hawley , 2001; MacQueen et al . , 2005; Colaiácovo et al . , 2003; de Vries et al . , 2005; Smolikov et al . , 2009 , Smolikov et al . , 2007b ) . While the components of the SC do not share a high degree of sequence conservation between species , this tripartite proteinaceous structure is conserved at the ultrastructural level ( Colaiácovo , 2006 ) . The SC is comprised of lateral element components that assemble along chromosome axes and central region components that bridge the gap between each pair of axes , much like the steps on a ladder . In the nematode C . elegans , there are four central region components of the SC , SYP-1 , SYP-2 , SYP-3 and SYP-4 , which begin to localize between the homologs in the leptotene/zygotene stage of meiotic prophase and are fully assembled throughout the interface between homologs by pachytene ( MacQueen et al . , 2002; Colaiácovo et al . , 2003; Smolikov et al . , 2009 , Smolikov et al . , 2007b , Smolikov et al . , 2007a ) . However , the SC is not static and has been shown to be a dynamic structure that undergoes continuous turnover . In budding yeast , Zip1 , the central region component of the SC , has been shown to be continuously incorporated into a fully assembled SC in a non-uniform spatial pattern influenced by recombination ( Voelkel-Meiman et al . , 2012 ) . In C . elegans , recent studies have shown that the central region of the SC contains mobile subunits ( Rog et al . , 2017 ) and that the SC persists in a more dynamic state in the absence of DSBs , leading to the suggestion that CO-committed intermediates may stabilize the SC ( Machovina et al . , 2016 ) . However , how these changes in SC dynamics are regulated is not well understood . Interhomolog CO formation is extremely important not only for producing genetic diversity but also for generating physical linkages ( chiasmata ) between homologs , which are essential for the proper bi-orientation and separation of homologs at meiosis I ( Page and Hawley , 2003 ) . Meiotic recombination initiates with the formation of programmed DSBs induced by the conserved topoisomerase-like protein Spo11/SPO-11 ( Keeney et al . , 1997; Bergerat et al . , 1997; Dernburg et al . , 1998 ) . DSBs are made in excess to ensure that at least one CO ( obligate CO ) is established for each pair of homologs ( Yokoo et al . , 2012; Jones , 1984 ) . However , the engagement of homologous chromosomes during recombination and/or SC assembly has been proposed to turn off further programmed DSB formation ( reviewed in Keeney et al . , 2014 ) . One possibility is that once a DSB is designated to become a CO a feedback mechanism turns off further programmed DSBs from forming . An alternative , albeit non-mutually exclusive possibility , is that SC formation may result in structural changes along the chromosomes which suppress further DSB formation . The molecular basis for transmission of this feedback regulation remains an open question . Here , we provide a mechanistic basis for how the switch in SC dynamics and the transmission of feedback regulation on further DSB formation work during meiosis in C . elegans . We show that the Polo-like kinases , PLK-1 and PLK-2 , are critical components of this negative feedback loop that couple CO designation with DSB formation . We show that a central region component of the SC , SYP-4 , is phosphorylated at Serine 269 in a PLK-1/2-dependent manner . Phosphorylation of SYP-4 coincides with the appearance of CO-promoting factor CNTD1/COSA-1 and is dependent on the formation of DSBs and CO designation/precursor formation , which in turn inhibits additional DSBs from being formed . Further , phosphorylation of SYP-4 by PLK-1/2 switches the central region of the SC from a more dynamic to a less dynamic state . Thus , we propose a model in which PLK-1/2 mediate the phosphorylation of SYP-4 in response to CO designation/precursor formation , which in turn stabilizes SC dynamics and prevents further DSB formation . plk-2 encodes one of the three Polo-like kinases in C . elegans and is the ortholog of mammalian PLK1 . PLK-2 localizes to chromosome-associated patches at the nuclear periphery of leptotene/zygotene nuclei , which correspond to the regions of the chromosomes referred to as pairing centers that are tethered to the nuclear envelope at this stage , and relocalizes to synapsed chromosomes in pachytene nuclei ( Labella et al . , 2011; Harper et al . , 2011 ) . To test whether PLK-2 localization on chromosomes is dependent on central region components of the SC , we analyzed the localization of PLK-2 in the syp-1 ( me17 ) and syp-4 ( tm2713 ) null mutants , which fail to synapse . We found that PLK-2 localized to the chromosome-associated patches at the nuclear periphery in leptotene/zygotene stage nuclei , but we did not detect PLK-2 signal on chromosomes in pachytene nuclei in syp-1 and syp-4 mutants ( Figure 1 ) . In C . elegans , PLK-1 functions redundantly with PLK-2 and it can partially substitute for the function of PLK-2 during pairing and synapsis of homologous chromosomes ( Nishi et al . , 2008; Harper et al . , 2011; Labella et al . , 2011 ) . To determine whether PLK-1 also localized to synapsed chromosomes , we examined the immunolocalization of PLK-1 in wild type germlines . While PLK-1 is observed on chromosome-associated aggregates at the nuclear periphery during leptotene/zygotene as expected , we did not detect any PLK-1 signal in pachytene stage nuclei even in the absence of PLK-2 , although we cannot rule out the possibility that PLK-1 signal could be below threshold levels of detection by immunofluorescence ( Figure 1—figure supplement 1 ) . Given that all of the four SYP proteins are interdependent on each other for their localization and for SC formation ( Colaiácovo et al . , 2003; Smolikov et al . , 2009 , Smolikov et al . , 2007a ) , we infer from our analysis of the syp-1 and syp-4 mutants that PLK-2 localization to chromosomes during pachytene is dependent on all four SYP proteins . 10 . 7554/eLife . 23437 . 003Figure 1 . PLK-2 localization on synapsed chromosomes is dependent on the SYP proteins . High-magnification images of leptotene/zygotene ( A ) and pachytene stage ( B ) nuclei from wild type , syp-4 , and syp-1 mutant gonads stained with HTP-3 ( green ) , PLK-2 ( red ) , and DAPI ( blue ) . ( A ) PLK-2 is observed localizing to aggregates at the nuclear periphery during leptotene/zygotene in synapsis-defective mutants . These aggregates , which have been previously shown to correspond to the pairing centers ( Labella et al . , 2011; Harper et al . , 2011 ) , are larger in wild type than in the syp-1 and syp-4 mutants given that the pairing centers of homologous chromosomes are not held in close juxtaposition in these mutants as frequently as in wild type at this stage in meiosis . ( B ) The more extensive localization of PLK-2 observed along chromosomes during pachytene in wild type is lost in syp-1 and syp-4 mutants . The thinner continuous tracks of HTP-3 staining observed in syp null mutants are due to HTP-3 localizing to unsynapsed axes . 27 , 17 , and 15 gonad arms were analyzed for wild type , syp-4 , and syp-1 , respectively . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 00310 . 7554/eLife . 23437 . 004Figure 1—figure supplement 1 . PLK-1 localization in the C . elegans germline . High-magnification images of leptotene/zygotene ( A ) and pachytene ( B ) stage nuclei from wild type and plk-2 mutant gonads stained with HTP-3 ( green ) , PLK-1 ( red ) , and DAPI ( blue ) . While PLK-1 is observed localizing to aggregates at the nuclear periphery during leptotene/zygotene , PLK-1 signal is not detected in pachytene nuclei in either wild type or plk-2 mutants . 14 and 15 gonad arms were analyzed for wild type and plk-2 mutant gonads , respectively . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 004 To determine whether the SYP proteins are potential targets for phosphorylation by the Polo-like kinases , we examined the SYP-1/2/3/4 proteins for the presence of potential PLK phosphorylation sites using the phosphorylation site prediction programs GSP-polo and PHOSIDA ( Liu et al . , 2013; Gnad et al . , 2011 , 2007 ) . We identified S269 in SYP-4 as a potential Polo-like kinase phosphorylation site ( Figure 2A ) . To test whether the S269 site is phosphorylated in vivo , we generated a phosphopeptide antibody against the SYP-4 S269 site . Phospho-specificity of the pSYP-4 antibody was confirmed by lack of pSYP-4 signal in the syp-4 ( S269A ) phosphodead mutant ( Figure 2D ) . Affinity purified phospho-specific SYP-4 antibody signal ( pSYP-4 ) was detected on chromosomes starting at early pachytene , even though SYP-4 localizes at the interface between homologous chromosomes earlier , starting at the leptotene/zygotene stage ( Smolikov et al . , 2009 ) . pSYP-4 was first observed as foci and short stretches on chromosomes in early pachytene nuclei and then fully colocalized with SYP-1 throughout the full length of the chromosomes by mid-pachytene ( Figure 2B and C ) . In late pachytene , as the SC starts to disassemble , pSYP-4 signal was also lost from most of the interface between homologs and was restricted to a single portion of each bivalent referred to as the short arm of the bivalent , which also retains residual SC proteins until later in prophase I and which we have previously shown to correspond to the shortest distance from the position of the single CO event to a chromosome end ( Figure 2B and C; ( Nabeshima et al . , 2005 ) . To test whether the S269 site on SYP-4 is phosphorylated in a PLK-2-dependent manner in vivo , we immunostained plk-2 ( ok1936 ) gonad arms with the pSYP-4 antibody . Although we observed a strong reduction in the pSYP-4 signal compared to wild type , there was residual phosphorylation of SYP-4 in the plk-2 mutant . Since PLK-1 can substitute for the function of PLK-2 during the pairing of homologous chromosomes , we examined whether SYP-4 was phosphorylated in the absence of both PLK-1 and PLK-2 . We found that pSYP-4 signal , but not SYP-4 , was completely lost upon depletion of plk-1 by RNAi in the plk-2 mutant ( Figure 2D and Figure 2—figure supplement 1 ) . Taken together , our data suggest that SYP-4 is phosphorylated at the S269 site and that its phosphorylation is dependent on Polo-like kinases PLK-1 and PLK-2 . 10 . 7554/eLife . 23437 . 005Figure 2 . Phosphorylation of SYP-4 at the S269 site is dependent on Polo-like kinase . ( A ) Schematic representation of the SYP-4 protein with the predicted Polo-like kinase phosphorylation site ( S269 ) indicated . CC indicates the coiled-coil domains and aa indicates amino acid . ( B ) Low-magnification images of whole-mounted gonads depicting SYP-1 ( green ) and phosphorylated SYP-4 ( pSYP-4; red ) localization in wild type . pSYP-4 signal is observed starting at early-pachytene although SC assembly , as observed here by SYP-1 immunostaining , starts earlier at the leptotene/zygotene stage . PMT ( premeiotic tip ) and LZ ( leptotene/zygotene ) . ( C ) High-magnification images of wild type germline nuclei at the indicated stages stained with DAPI ( blue ) , anti-SYP-1 ( green ) and anti-pSYP-4 ( red ) . 16 gonad arms were analyzed for wild type ( B–C ) . Phosphorylated SYP-4 signal is absent at leptotene/zygotene , observed colocalizing with SYP-1 in mid-pachytene and acquiring a similar restricted localization as SYP-1 during the disassembly of the SC starting at late pachytene . ( D ) High-magnification images of mid-pachytene nuclei from wild type , plk-2 ( ok1936 ) , plk-2 ( ok1936 ) ; plk-1 ( RNAi ) , and syp-4 ( S269A ) mutants stained with DAPI ( blue ) , anti-SYP-1 ( green ) and anti-pSYP-4 ( red ) . Phosphorylated SYP-4 signal is absent in the syp-4 ( S269A ) mutant indicating specificity of the phospho-specific antibody . Phosphorylated SYP-4 signal is reduced in mid-pachytene nuclei in plk-2 mutants and absent at that stage in plk-2; plk-1 ( RNAi ) germlines . Note that an uneven SYP-1 signal intensity is observed along chromosomes in plk-2; plk-1 ( RNAi ) germlines , but pSYP-4 signal is not detected even on chromosomes with strong SYP-1 signal . 22 , 17 , 16 , and 18 gonad arms were analyzed for wild type , plk-2 ( ok1936 ) , plk-2 ( ok1936 ) ; plk-1 ( RNAi ) , and syp-4 ( S269A ) mutants , respectively . Scale bar , 20 μm for ( B ) and 3 μm for ( C ) and ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 00510 . 7554/eLife . 23437 . 006Figure 2—figure supplement 1 . RT-PCR analysis of plk-1 knockdown by RNAi . ( A ) High-magnification images of pachytene stage nuclei from wild type and plk-2; plk-1 ( RNAi ) mutant gonads stained with SYP-4 ( green ) and DAPI ( blue ) . Note that an uneven SYP-4 signal intensity is observed along chromosomes in plk-2; plk-1 ( RNAi ) germlines . Scale bar , 3 μm . 15 gonad arms were analyzed for each genotype . ( B ) RT-PCR of plk-1 ( RNAi ) compared to control RNAi ( empty vector ) . Both depletions were performed in the plk-2 mutant . gpdh-1 expression was used as a loading control . Each lane corresponds to a pooled worm lysate of ~30 worms and indicates the effective depletion of plk-1 by RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 006 Although the SYP proteins load on chromosomes starting in leptotene/zygotene , SYP-4 is phosphorylated starting at early pachytene , which coincides with the time when CO precursor markers are observed ( Yokoo et al . , 2012 ) . We next tested the hypothesis that SYP-4 is phosphorylated in response to CO precursor formation . As we predicted , phosphorylation of SYP-4 coincided with the appearance of pro-crossover factor GFP::COSA-1 foci ( Figure 3A ) . In early pachytene , GFP::COSA-1 foci were first observed only in some , but not on all , of the six pairs of homologous chromosomes . Strikingly , at that same stage , phosphorylated SYP-4 was first observed on 76% of chromosomes ( n = 78 ) with GFP::COSA-1 foci , whereas a low percentage of chromosomes showed either only pSYP-4 signal ( 5 . 8%; n = 6 ) or COSA-1 foci ( 18 . 4%; n = 19 ) ( Figure 3A and Figure 3—figure supplement 1 ) . This result suggests that SYP-4 may be phosphorylated in response to CO precursor formation . 10 . 7554/eLife . 23437 . 007Figure 3 . SYP-4 is phosphorylated in response to CO precursor formation . ( A ) Low-magnification image of a whole-mounted gonad from an animal expressing GFP::COSA-1 stained with anti-GFP ( green ) , anti-pSYP-4 ( red ) , and DAPI ( blue ) . GFP::COSA-1 foci are detected from early to late pachytene , coinciding with the temporal window in which pSYP-4 signal is observed on chromosomes . Insets on the right are high-magnification images showing that stretches of pSYP-4 signal are first observed at early pachytene mainly on chromosomes that have a GFP::COSA-1 focus ( top row of insets ) . pSYP-4 signal is then observed continuously along the length of the chromosomes at mid-pachytene . Finally , pSYP-4 signal starts to be lost from some chromosome subdomains and is retained from the off-centered site of the CO event marked by GFP::COSA-1 through the shortest distance to one end of the chromosomes ( this will later become the short arm of the bivalent; bottom row of insets ) . PMT ( premeiotic tip ) and LZ ( leptotene/zygotene ) . Scale bar , 20 μm . 21 gonads were analyzed . ( B ) High-magnification images of pachytene stage nuclei stained with anti-SYP-1 ( green ) , anti-pSYP-4 ( red ) , and DAPI ( blue ) for the indicated genotypes . pSYP-4 signal is mostly lost in spo-11 , dsb-1 , dsb-2 , mre-11 , rad-51 and rad-54 mutants . Arrow points to the chromosome that still shows pSYP-4 localization in these mutants . The extensive localization of pSYP-4 along synapsed chromosomes can be rescued by exogenous induction of DSBs via γ-IR in spo-11 mutants . pSYP-4 signal is not detected in pro-crossover mutants cosa-1 and zhp-3 . Scale bar , 20 μm for ( A ) and 3 μm for ( B ) . At least 15 animals were examined for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 00710 . 7554/eLife . 23437 . 008Figure 3—figure supplement 1 . Quantitation of colocalization between COSA-1::GFP foci and pSYP-4 signal on chromosomes in early pachytene . Histogram showing the percentage of chromosomes exhibiting only pSYP-4 signal ( 5 . 8% ) , only COSA-1::GFP foci ( 18 . 4% ) , or COSA-1::GFP foci colocalizing with pSYP-4 signal ( 75 . 7% ) . 84 nuclei from 11 gonads were analyzed . Only nuclei where COSA-1::GFP foci and pSYP-4 tracks first begin to appear ( first 2 to 3 rows of nuclei in early pachtyene ) were scored . Therefore , nuclei with six COSA-1::GFP foci colocalizing with pSYP-4 tracks were excluded from this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 00810 . 7554/eLife . 23437 . 009Figure 3—figure supplement 2 . SYP-4 is phosphorylated in response to CO precursor formation . Low-magnification images show portions of the gonads starting at leptotene/zygotene ( LZ ) and extending through pachytene ( boundary between these two stages is indicated by vertical white dotted lines ) . Gonads were stained with anti-SYP-1 ( green ) , anti-pSYP-4 ( red ) , and DAPI ( blue ) for the indicated genotype . Images show that pSYP-4 signal starts to be detected on chromosomes by early pachytene , later then when central region components of the SC , such as SYP-1 , are observed associating with chromosomes . pSYP-4 signal is mostly , albeit not completely , lost in spo-11 mutants , and can be fully rescued by induction of exogenous DSBs by exposure to γ-irradiation . A similar reduction in pSYP-4 signal is observed in mre-11 , rad-51 and rad-54 mutants . pSYP-4 signal is completely lost in pro-crossover mutants , cosa-1 and zhp-3 . Scale bars , 20 μm . At least 15 animals were examined for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 00910 . 7554/eLife . 23437 . 010Figure 3—figure supplement 3 . Assessing the residual SYP-4 phosphorylation observed in spo-11 mutants . ( A ) High-magnification image of pachytene stage nuclei from spo-11; COSA-1::GFP mutants stained with anti-pSYP-4 ( red ) , anti-GFP ( green ) , and DAPI ( blue ) . Image shows the colocalization of the single pSYP-4 track observed in <20% of nuclei on chromosomes with a COSA-1 focus . 21 gonad arms were analyzed . Scale bar , 3 μm . ( B ) Image of pachytene stage nuclei from a spo-11 mutant stained with anti-pSYP-4 ( red ) , anti-HIM-8 ( red ) , and DAPI ( blue ) . Nuclei exhibiting a pSYP-4 track colocalizing with HIM-8 , which marks the pairing center end of the X chromosome , are indicated with yellow circles and a nucleus where the pSYP-4 track does not colocalize with HIM-8 is indicated with a green circle to facilitate visualization . 12 gonad arms were analyzed . Scale bar , 3 μm . ( C ) Histogram showing the quantitation of the percentage of pachytene nuclei in spo-11 mutants in which the pSYP-4 signal was observed on the X chromosome ( 46 . 7% ) , chromosome III ( 9 . 5% ) , and chromosome V ( 11 . 6% ) , where the X chromosome was identified by HIM-8 staining , and chromosomes III and V were identified with chromosome-specific probes by FISH . 117 , 42 and 43 nuclei with a pSYP-4 track from at least 10 gonad arms were scored for chromosomes X , III , and V , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 010 To further examine the link between SYP-4 phosphorylation and recombination , we examined phosphorylation of SYP-4 in the spo-11 ( ok79 ) null mutant . We found that pSYP-4 was largely abrogated in spo-11 mutants except on a single chromosome track in 19 . 7% ( n = 126 ) of mid to late pachytene nuclei ( Figure 3B and Figure 3—figure supplement 2 ) . Moreover , a COSA-1::GFP focus was observed on each of these chromosomes exhibiting pSYP-4 signal ( Figure 3—figure supplement 3A ) . These residual COSA-1 foci and SYP-4 phosphorylation tracks may reflect recombination initiated from spontaneous DSBs or other DNA lesions , and is similar to the observation of SPO-11-independent COSA-1 foci in late pachytene nuclei made by Pattabiraman et al . , 2017 . To determine which chromosome exhibited the pSYP-4 track in spo-11 mutants , we identified the X chromosome by HIM-8 staining and chromosomes III and V by fluorescence in situ hybridization . We found that 46% of pSYP-4 tracks corresponded to the X chromosome , 11 . 6% to chromosome V and 9 . 5% to chromosome III ( Figure 3—figure supplement 3B and C ) . The higher frequency of pSYP-4 tracks along the X chromosome could be due to its distinct chromatin structure and/or a result of it undergoing delayed replication compared to autosomes ( Bender et al . , 2004; Kelly et al . , 2002; Jaramillo-Lambert et al . , 2007 ) , which might lead to a higher number of DNA lesions that could be processed for CO precursor formation during meiosis . Introduction of exogenous DSBs by γ-irradiation restored SYP-4 phosphorylation , suggesting that the absence of SYP-4 phosphorylation in spo-11 mutants was due to lack of DSBs and not due to the lack of the SPO-11 protein ( Figure 3B and Figure 3—figure supplement 1 ) . We observed a similar loss of phosphorylated SYP-4 , except for a single chromosome track in <20% of nuclei , in mutants for genes involved in DSB formation ( spo-11 , dsb-1 , and dsb-2 ) , resection ( mre-11 ) and strand invasion/exchange ( rad-51 and rad-54 ) during homologous recombination ( Figure 3 and Figure 3—figure supplement 2 ) . However , the SYP-4 phosphorylation signal was completely absent in pachytene stage nuclei in the zhp-3 and cosa-1 pro-crossover defective mutants ( Figure 3 and Figure 3—figure supplement 2 ) . Taken together , our data suggest that SYP-4 phosphorylation is dependent on CO precursor formation . Previous work has shown that Polo-like kinases regulate pairing in C . elegans ( Labella et al . , 2011; Harper et al . , 2011 ) . To test whether phosphorylation of SYP-4 regulates pairing , we analyzed pairing in syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants . Analysis of homologous pairing for chromosome V by fluorescence in situ hybridization ( FISH ) and for the X chromosome using a HIM-8 antibody that recognizes the X-chromosome pairing center revealed no significant difference in pairing levels throughout the germline for these chromosomes in either the phosphodead or phosphomimetic mutants compared to wild type ( Figure 4A ) . To test whether Polo-like kinases regulate meiotic progression through phosphorylation of SYP-4 at the S269 site , we analyzed the localization of phosphorylated SUN-1 with a SUN-1 S8 phospho-specific antibody in syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants . SUN-1 encodes for a conserved inner nuclear envelope protein , which clusters at chromosome ends associated with the nuclear envelope , and its phosphorylation is dependent on CHK-2 and PLK-2 ( Woglar et al . , 2013 ) . In wild type , SUN-1 S8 phosphorylation is observed upon entry into meiosis in leptotene/zygotene stage nuclei and its signal is no longer detected around mid-pachytene , except in a few nuclei ( Woglar et al . , 2013 ) . We did not observe any significant difference in SUN-1 S8P phospho-specific antibody localization in either syp-4 ( S269A ) phosphodead or syp-4 ( S269D ) phosphomimetic mutants compared to wild type ( Figure 4B ) . Altogether , our data indicate that meiotic pairing and early meiotic progression are not dependent on the phosphorylation of SYP-4 . 10 . 7554/eLife . 23437 . 011Figure 4 . SYP-4 phosphorylation at the S269 site does not affect pairing or SUN-1 S8pi localization . Homologous pairing for chromosomes V and X was analyzed using FISH and immunostaining against the X chromosome pairing center protein HIM-8 , respectively . ( A ) Schematic representation of the C . elegans germline indicating the different zones scored for homologous chromosome pairing . Graphs depict no statistically significant difference in the percentage of nuclei with paired FISH signals for chromosome V and paired HIM-8 signals for the X-chromosome in the germline of the indicated mutants compared to wild type ( signals were scored as paired when separated by ≤0 . 75 µm ) . X-axes indicate the position along the germline . PMT- premeiotic tip , L/Z- leptotene/zygotene , EP- early pachytene , MP- mid pachytene , and LP- late pachytene . Six gonad arms were analyzed for each genotype . ( B ) Low-magnification images of whole-mounted gonads indicate no difference in the length of the germline region stained with anti-SUN-1 S8pi ( green ) for wild type ( top ) , syp-4 ( S269A ) phosphodead ( middle ) , and syp-4 ( S269D ) phosphomimetic mutants ( bottom ) . Left white vertical bar indicates entrance into meiosis and right vertical bar indicates end of zone where SUN-1 S8pi signal is detected . Scale bar , 15 μm . >18 gonad arms were analyzed for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01110 . 7554/eLife . 23437 . 012Figure 4—source data 1 . Numerical data for the percentage of nuclei with paired FISH signals for chromosome V and paired HIM-8 signals for the X-chromosome in the germline of the indicated mutants compared to wild type shown in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 012 C . elegans chromosomes exhibit strict CO interference and undergo a single CO per homologous chromosome pair ( Yokoo et al . , 2012; Libuda et al . , 2013 ) . Partial depletion of central region components of the SC has been shown to attenuate CO interference and elevate the number of COs observed between homologs ( Libuda et al . , 2013 ) . To test the hypothesis that phosphorylation of SYP-4 regulates CO interference , we crossed GFP::COSA-1 into the syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants and analyzed CO levels by counting the number of COSA-1 foci observed per nucleus . If phosphorylation of SYP-4 functions as a signal to promote CO interference and prevent more than one CO per chromosome , we would expect to see an increase in the number of COs per nucleus in the phosphodead mutants as that signal is no longer present . However , we did not observe a significant increase in the number of COs , as marked by COSA-1 , per nucleus in syp-4 ( S269A ) phosphodead mutants ( Figure 5A ) . Similarly , we did not observe a significant decrease in the number of COs in syp-4 ( S269D ) phosphomimetic mutants ( Figure 5A ) . We observed a significant reduction in brood size in both syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants compared to wild type ( Supplementary file 1 ) . However , we only saw 6–8% embryonic lethality accompanied by 0 . 3–1% males among the progeny from these mutants , suggesting only a mild increase of autosomal and X chromosome nondisjunction during meiosis in these mutants , which could stem in part from the <4% of oocytes observed having a reduced number of COSA-1 foci ( Supplementary file 1 ) . These results suggest that PLK-dependent phosphorylation of SYP-4 does not regulate CO interference . 10 . 7554/eLife . 23437 . 013Figure 5 . SYP-4 phosphorylation at the S269 site by Polo-like kinase regulates DSB formation . ( A ) Graph depicts no statistically significant difference in the percentage of GFP::COSA-1 foci scored per nucleus for either the syp-4 ( S269A ) or syp-4 ( S269D ) mutants compared to wild type . Color code at the bottom indicates the number of GFP::COSA-1 foci observed per nucleus . Number on the top of the histogram bars represent the total number of nuclei scored for each genotype . ( B ) Graph depicts the increase in the mean number of germ cell corpses detected for both syp-4 ( S269A ) and syp-4 ( S269D ) mutants compared to wild type ( *p<0 . 0001 and p<0 . 0008 , respectively , by the two-tailed Mann-Whitney test , 95% C . I . ) . The syp-1 mutant was used as a positive control given its elevated levels of germ cell apoptosis . 56 , 72 , 61 , and 40 animals were analyzed for wild type , syp-4 ( S269A ) , syp-4 ( S269D ) and syp-1 ( me17 ) mutants , respectively . ( C ) Schematic representation of the C . elegans germline indicating the different zones scored for the number of RAD-51 foci/nucleus . ( D ) Histogram depicts the increase in the mean number of RAD-51 foci observed per nucleus in the germlines of syp-4 ( S269A ) mutants compared to wild type and the decrease in the mean number of RAD-51 foci observed in syp-4 ( S269D ) mutants compared to wild type . ( E ) Similar analysis as in ( D ) except it is performed in a rad-54 ( RNAi ) background allowing for a quantification of the total number of DSBs observed in each indicated genotype . This analysis reveals that the altered numbers of RAD-51 foci are due to increased levels of DSBs in syp-4 ( S269A ) mutants and decreased levels of DSBs in syp-4 ( S269D ) mutants . ( D–E ) X-axes indicate the position along the germline . PMT- premeiotic tip , L/Z- leptotene/zygotene , EP- early pachytene , MP- mid-pachytene , and LP- late pachytene . Six gonad arms were analyzed for each genotype . ** Indicates p<0 . 0001 by the two-tailed Mann-Whitney test , 95% C . I . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01310 . 7554/eLife . 23437 . 014Figure 5—source data 1 . Numerical data for the percentage of GFP::COSA-1 foci scored per nucleus for either the syp-4 ( S269A ) or syp-4 ( S269D ) mutants compared to wild type shown in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01410 . 7554/eLife . 23437 . 015Figure 5—source data 2 . Numerical data used to calculate the mean number of germ cell corpses for both syp-4 ( S269A ) and syp-4 ( S269D ) mutants compared to wild type shown in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01510 . 7554/eLife . 23437 . 016Figure 5—source data 3 . Numerical data used to calculate the mean number of RAD-51 foci observed per nucleus throughout zones 1–6 from whole mounted gonads of the indicated genotypes shown in Figure 5D and E . Sheet one shows all raw data while sheet two shows combined summaries . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01610 . 7554/eLife . 23437 . 017Figure 5—figure supplement 1 . RT-PCR showing depletion of rad-54 by RNAi in wild type , syp-4 ( S269A ) rad-54 ( RNAi ) and syp-4 ( S269D ) rad-54 ( RNAi ) animals . ( A ) High-magnification images of mid-pachytene stage nuclei ( zone five indicated in Figure 5C and D ) in wild type and syp-4 ( S269A ) mutants stained with RAD-51 ( red ) and DAPI ( blue ) . Scale bar , 3 μm . ( B ) RT-PCR of rad-54 ( RNAi ) compared to control RNAi ( empty vector ) in syp-4 ( S269A ) and syp-4 ( S269D ) animals scored for levels of RAD-51 foci per germline nucleus in Figure 5E . gpdh-1 expression was used as a loading control . Each lane corresponds to a pooled worm lysate of ~30 worms and indicates the effective depletion of rad-54 by RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 017 Unprocessed meiotic DSBs can trigger germ cell apoptosis in late pachytene due to activation of a DNA damage checkpoint in the C . elegans germline ( Schumacher et al . , 2001 ) . Our analysis revealed a significant increase in the levels of germ cell apoptosis in both phosphodead and phosphomimetic mutants ( Figure 5B ) . To test whether this is due to a role for PLK-dependent phosphorylation of SYP-4 in regulating the progression of meiotic recombination , we examined the levels of RAD-51 foci , a protein involved in DNA strand invasion/exchange during DSB repair , in nuclei throughout the germline of the syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants . We observed a significant increase in the number of RAD-51 foci per nucleus starting at mid pachytene in syp-4 ( S269A ) phosphodead mutants compared to wild type , and the levels of foci remained higher than wild type through late pachytene ( Figure 5D and Figure 5—figure supplement 1 ) . In contrast , the number of RAD-51 foci observed in early and mid-pachytene nuclei was significantly lower than in wild type in syp-4 ( S269D ) phosphomimetic mutants ( Figure 5D ) . These results suggest that PLK-dependent phosphorylation of SYP-4 regulates progression of meiotic recombination . To distinguish whether this phosphorylation regulates DSB formation or repair we quantitated the levels of RAD-51 foci observed per nucleus throughout the germline in the absence of RAD-54 . Depletion of rad-54 traps RAD-51 association at DSB sites , which allows for the scoring of the total number of DSBs repaired via a RAD-51 intermediate ( Mets and Meyer , 2009 ) . We observed significantly elevated levels of RAD-51 foci in syp-4 ( S269A ) phosphodead mutant germlines upon depletion of rad-54 by RNAi compared to rad-54 ( RNAi ) in an otherwise wild type background ( Figure 5E ) . In contrast , syp-4 ( S269D ) phosphomimetic mutant germlines displayed significantly reduced RAD-51 foci levels upon rad-54 ( RNAi ) compared to control ( Figure 5E ) . These results indicate that phosphorylation of SYP-4 at S269 regulates DSB formation . Altogether , these data support a role for PLK-dependent phosphorylation of SYP-4 in negatively regulating DSB formation in the C . elegans germline . In C . elegans , the assembly of the SC is independent of DSB formation ( Dernburg et al . , 1998 ) . To understand the role of PLK-dependent phosphorylation on the SC structure , we analyzed the immunolocalization of the lateral element protein HTP-3 , and all four central region components of the SC , SYP-1/2/3/4 , in syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants . We did not observe obvious defects in either SC assembly or maintenance in either mutant background ( Figure 6—figure supplement 1 ) . Moreover , western blot analysis revealed no change in SYP-1 or SYP-3 protein levels in whole worm lysates from either syp-4 ( S269A ) phosphodead or syp-4 ( S269D ) phosphomimetic mutants compared to wild type ( we were precluded from examining SYP-2 and SYP-4 in this way since the available antibodies do not work on westerns ) ( Figure 6—figure supplement 1 ) . To test whether phosphorylation of SYP-4 affects the dynamics of the SC central region we performed Fluorescence Recovery After Photobleaching ( FRAP ) in live animals expressing SYP-3::GFP in an otherwise wild-type background as well as in the syp-4 ( S269A ) phosphodead and syp-4 ( S269D ) phosphomimetic mutants . We measured the extent of recovery of the fluorescence signal post-photobleaching for nuclei in the leptotene/zygotene stage , where SYP-4 is not phosphorylated , and at mid-pachytene , where SYP-4 is phosphorylated on all chromosomes in wild type . First , we observed that the majority of the SYP-3::GFP signal was recovered within 20 min after photobleaching in wild type nuclei . However , the extent of recovery of the fluorescence signal was higher in leptotene/zygotene nuclei compared to mid-pachytene nuclei in wild type ( p<0 . 0001 by the Dunn’s multiple comparisons test ) , suggesting that the central region of the SC transitions from a more dynamic state into a less dynamic or more stable state as meiotic prophase progresses ( Figure 6 , Supplementary file 2 , Video 1 and Video 2 . In contrast , in syp-4 ( S269A ) phosphodead mutants , the extent of recovery for the SYP-3::GFP signal was high in both leptotene/zygotene and mid-pachytene stage nuclei , and similar at mid-pachytene to that measured for leptotene/zygotene nuclei in wild type ( Figure 6 and Supplementary file 2 ) . This result suggests that phosphorylation of SYP-4 can alter the dynamics of the central region of the SC by stabilizing it . This is further supported by our observation of a lower extent of recovery of the SYP-3::GFP signal after photobleaching in leptotene/zygotene nuclei in syp-4 ( S269D ) phosphomimetic mutants compared to that observed for wild-type nuclei at that stage ( p<0 . 0001 ) ( Figure 6 ) . Specifically , the lower rate of recovery observed after photobleaching for the phosphomimetic mutant was similar for leptotene/zygotene and mid-pachytene nuclei and comparable to that observed in mid-pachytene stage nuclei for wild type ( Figure 6A and Supplementary file 2 ) . Altogether , these data suggest that PLK-dependent phosphorylation of SYP-4 in pachytene changes the central region of the SC from a more dynamic to a less dynamic , and therefore more stable , state , which in turn impinges on DSB formation . 10 . 7554/eLife . 23437 . 018Figure 6 . PLK-dependent phosphorylation of SYP-4 changes the SC central region from a more dynamic to a less dynamic state . ( A ) Representative images showing the GFP::SYP-3 fluorescence detected in leptotene/zygotene and mid-pachytene nuclei for the indicated genotypes during FRAP experiments . Arrows indicate the small region that was bleached and measured for fluorescence recovery on each nucleus . GFP::SYP-3 signal recovery after photobleaching during leptotene/zygotene is faster in both wild type and syp-4 ( S269A ) mutants compared to syp-4 ( S269D ) mutants . At mid-pachytene , while the rate of fluorescence signal recovery is slower compared to earlier prophase for wild type , the syp-4 ( 269A ) mutants continue to exhibit a rapid recovery rate comparable to that observed earlier in leptotene/zygote , while the syp-4 ( S269D ) mutants continue to exhibit a slower recovery rate . ( B ) Graph showing quantitation of GFP::SYP-3 fluorescence recovery in leptotene/zygotene and mid-pachytene stage nuclei in wild type compared to syp-4 ( S269A ) , and syp-4 ( S269D ) mutants . The total of number of nuclei measured in each group were: Mid-pachytene_WT: n = 20; Mid-pachytene _syp-4 ( S269D ) : n = 18; Mid-pachytene _ syp-4 ( S269A ) : n = 20; Leptotene/zygotene _WT: n = 28; Leptotene/zygotene_ syp-4 ( S269D ) : n = 20; Leptotene/zygotene _ syp-4 ( S269A ) : n = 21 . Error bars represent standard error of the mean . Bar code in the bottom indicates the stage of the nuclei and genotype . LZ – leptotene/zygotene; MP – mid-pachytene . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01810 . 7554/eLife . 23437 . 019Figure 6—source data 1 . Final data , after corrections , used to make graph for FRAP analysis shown in Figure 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 01910 . 7554/eLife . 23437 . 020Figure 6—source data 2 . Raw numerical data used for graph of FRAP analysis shown in Figure 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 02010 . 7554/eLife . 23437 . 021Figure 6—figure supplement 1 . Phosphorylation of SYP-4 at S269 does not affect SC assembly and maintenance . ( A ) High-magnification images of pachytene nuclei in wild type , syp-4 ( S269A ) and syp-4 ( S269D ) mutants co-stained with SYP-1 ( green ) , SYP-3 ( red ) , and DAPI ( blue ) . ( B ) High-magnification images of pachytene nuclei in wild type , syp-4 ( S269A ) and syp-4 ( S269D ) mutants co-stained with HTP-3 ( green ) , SYP-2 ( red ) , and DAPI ( blue ) . ( C ) High-magnification images of pachytene nuclei in wild type , syp-4 ( S269A ) and syp-4 ( S269D ) mutants co-stained with HTP-3 ( green ) , SYP-4 ( red ) , and DAPI ( blue ) . ( D ) Western blot of whole worm lysates showing no changes in the levels of the SYP-1 and SYP-3 proteins in wild type , syp-4 ( S269A ) and syp-4 ( S269D ) animals . α-tubulin and α-histone H3 were used as loading controls . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 02110 . 7554/eLife . 23437 . 022Video 1 . Example of a nucleus included in the FRAP analysis for Figure 6 . Mid-pachytene stage nuclei from wild type hermaphrodite animals expressing SYP-3::GFP included for analysis for FRAP experiment in Figure 6 . Playback time is 447x in real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 02210 . 7554/eLife . 23437 . 023Video 2 . Example of a nucleus excluded from FRAP analysis in Figure 6 . Mid-pachytene stage nuclei from wild type hermaphrodite animals expressing SYP-3::GFP excluded from the FRAP analysis since this nucleus is rotating on its own axis . Playback time is 447x in real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 23437 . 023 Polo-like kinase is a highly conserved kinase present from yeast to humans and has been shown to play multiple roles during both mitotic and meiotic cell divisions ( Archambault and Glover , 2009 ) . The yeast Polo-like kinase Cdc5 has been implicated in the resolution of double Holliday junctions in homologous recombination , promotion of CO formation , breakdown of the SC , co-orientation of sister chromatids , and loss of cohesion from chromosome arms at anaphase I ( Clyne et al . , 2003; Lee and Amon , 2003; Brar et al . , 2006; Attner et al . , 2013; Sourirajan and Lichten , 2008 ) . In mouse spermatocytes , PLK1 has been shown to phosphorylate the central element proteins SYCP1 and TEX12 to promote SC disassembly and is required for exit from meiotic prophase ( Jordan et al . , 2012 ) . In C . elegans , PLK-2 functions redundantly with PLK-1 to promote pairing and synapsis of homologous chromosomes in the leptotene/zygotene stage of meiosis ( Harper et al . , 2011; Labella et al . , 2011 ) . Our study reveals a novel role for PLK-1/2 in recognizing CO designation and generating an inhibitory signal that prevents further DSB formation . The mechanism by which PLK-1/2 recognize CO designation remains to be determined . Even though DNA damage can be very deleterious , programmed DSBs are induced and repaired in a highly regulated manner during meiosis to ensure interhomolog CO formation , which is critical for achieving accurate chromosome segregation at meiosis I ( Martinez-Perez and Colaiácovo , 2009 ) . Regulatory mechanisms need to be set in place to ensure that sufficient DSBs are formed to generate at least one CO per pair of homologous chromosomes and then turn off DSB formation once interhomolog engagement is achieved . Studies in different model organisms have described the existence of feedback mechanisms to regulate meiotic DSB number and distribution . Work done in both mice and Drosophila has shown that ATM kinase , which is activated in response to DNA damage , is also activated by SPO11-mediated DSBs and is implicated in controlling meiotic DSB levels by inhibiting further DSB formation via a negative feedback mechanism ( Lange et al . , 2011; Joyce et al . , 2011 ) . Studies in yeast , have shown both cis- and trans-acting regulation of DSB formation acting respectively either along the same or between different homologous chromosomes or sister chromatids . During regulation in cis , presence of a strong DSB hotspot inhibits DSB formation nearby ( Wu and Lichten , 1995; Xu and Kleckner , 1995; Fukuda et al . , 2008 ) whereas during regulation in trans , DSB formation at one site inhibits DSB formation at either the same site or a nearby site on the homologous partner . Communication between interhomolog interaction and DSB formation is important to establish an even spacing of total recombination events , and the Mec1 ( ATR ) DNA damage response kinase is proposed to mediate this by trans inhibition ( Zhang et al . , 2011 ) . Meanwhile , the Tel1 ( ATM ) kinase exerts cis inhibition at the local scale , on the same chromatid and between sister chromatids ( Garcia et al . , 2015 ) . Following meiotic DSB formation , Tel1 and Mec1 may phosphorylate Rec114 , a Spo11-accessory protein required for DSB formation , which has been proposed to inhibit further DSB formation by reducing the interaction of Rec114 with DSB hotspots ( Carballo et al . , 2013 ) . In C . elegans , our data suggests that PLK-1/2-dependent phosphorylation of SYP-4 functions as a signal to constrain further DSB formation when one of the DSBs undergoes CO designation/precursor formation . However , we cannot rule out a role for ATM/ATR kinases in PLK-1/2-mediated DSB inhibition . Studies in yeast and mice suggest that defects in interhomolog engagement result in increased DSB formation . In yeast , the ZMM proteins ( Zip1-4 , Msh4-5 , Mer3 , and others ) are required for both SC and CO formation and studies in zmm mutants implicate interhomolog engagement as directly responsible for inhibiting DSB formation ( Keeney et al . , 2014 ) . In mice , unsynapsed chromosomes exhibit elevated DSB levels ( Kauppi et al . , 2013; Thacker et al . , 2014; Hayashi et al . , 2010; Lam and Keeney , 2015 ) and the illegitimate ( non-homologous ) synapsis observed in Spo11-/- mutants is sufficient to evict the HORMA-domain proteins HORMAD1/2 from chromosomes ( Wojtasz et al . , 2009 ) suggesting a role for SC formation in the feedback mechanism that inhibits further DSB formation once interhomolog engagement is achieved . However , the molecular basis for transmission of the interhomolog engagement-mediated feedback loop on DSB formation is not understood . Axial element proteins Hop1 and Red1 in yeast , HTP-1 and HTP-3 in C . elegans , and HORMAD1 , which is a component of unsynapsed axes in mice , have been shown to be required for normal levels of programmed DSBs ( Mao-Draayer et al . , 1996; Woltering et al . , 2000; Blat et al . , 2002; Niu et al . , 2005; Carballo et al . , 2008; Lam and Keeney , 2015; Goodyer et al . , 2008; Couteau and Zetka , 2005; Martinez-Perez and Villeneuve , 2005; Sanchez-Moran et al . , 2007; Latypov et al . , 2010 ) . Our work implicates a central region component of the SC in functioning as a signaling unit to inhibit further DSB formation along synapsed chromosomes once a CO is designated or after CO precursor formation in C . elegans . This SC-mediated feedback mechanism may also be present in humans since a central element protein of the SC , SYCE1 , has PLK phosphorylation sites at S29 and S191 predicted by the phosphorylation site predictor GSP polo and both of these sites are identified as phosphorylated in vivo in humans by Phosphosite Plus ( Hornbeck et al . , 2015 ) . C . elegans strains were cultured at 20°C under standard conditions and the N2 Bristol strain was used as the wild-type background ( Sulston and Brenner , 1974 ) . The following mutations and chromosome rearrangements were used: LG I: syp-4 ( rj48 ( S269A ) I , syp-4 ( rj49 ( S269D ) I , syp-4 ( rj48 ( S269A ) I; mels9 ( unc-119 ( + ) pie-1promoter::gfp::syp-3 , syp-4 ( rj49 ( S269D ) I; mels9 ( unc-119 ( + ) pie-1promoter::gfp::syp-3 , syp-4 ( rj48 ( S269A ) I; meIs8 [pie-1p::GFP::cosa-1 + unc-119 ( + ) ] II , syp-4 ( rj49 ( S269D ) I; meIs8 [pie-1p::GFP::cosa-1 + unc-119 ( + ) ] II , syp-4 ( tm2713 ) I/hT2 [bli-4 ( e937 ) let- ? ( q782 ) qIs48] ( I;III ) , plk-2 ( ok1936 ) I , zhp-3 ( jf61 ) /hT2[bli-4 ( e937 ) let- ? ( q782 ) qIs48 ( I:III ) ]I . LG II: rol-1 ( e91 ) dsb-2 ( me96 ) /mnC1 ( dyp-1 ( e128 ) unc-52 ( e444 ) ) II . LG III: cosa-1 ( me13 ) /qC1 ( qls26 ) III . LG IV: dsb-1 ( tm5034 ) IV/nT1 ( unc- ? ( n754 ) let- ? ) ( IV;V ) , syp-1 ( me17 ) V/nT1 [unc- ? ( n754 ) let- ? qIs50] ( IV;V ) , spo-11 ( ok79 ) IV/nT1 [unc- ? ( n754 ) let- ? ] ( IV;V ) , mre-11 ( ok179 ) IV/nT1 [unc- ? ( n754 ) let- ? ] ( IV;V ) , rad-51 ( lg8701 ) IV/nT1 [let- ? ( m435 ) ] ( IV;V ) , rad-54 ( tm1268 ) / hT2 [bli-4 ( e937 ) let- ? ( q782 ) qIs48] ( I;III ) . A rabbit phospho-specific polyclonal antibody was generated by Abmart ( China ) using the phospho-peptide C-QFDR ( pS ) FILAS encompassing the S269 site in SYP-4 . Two peptides were synthesized: An antigen peptide with phosphoserine C-QFDR ( pS ) FILAS and a control peptide without phosphorylation ( C-QFDRSFILAS ) . Serum harvested from the rabbits immunized with the antigen peptides went through two rounds of affinity purification . First , serum was passed through a column to which the antigenic peptide was coupled in order to isolate phospho-specific SYP-4 antibodies . The eluate from the first column was then passed through a second column to which the control peptide was coupled and flow through was collected . This second step was done to remove any non-phosphorylated SYP-4 antibodies . An ELISA titer of ≥1:50 , 000 against the modified peptide and a modified/unmodified titer ratio of ≥8 were used as validation criteria . A rabbit polyclonal antibody was generated against a SYP-2 peptide ( RRVSFASPVSSSQ ) by Abmart and used at a 1:200 dilution for immunofluorescence . A polyclonal antibody against a SYP-1 peptide ( VDAPTEALIETPVDDQSSGFLC ) was generated by the GenScript Antibody Group ( Piscataway , NJ ) and used at a 1:3000 dilution for immunofluorescence and 1:5000 dilution for western blot analysis . All other primary antibodies were used at the following dilutions for immunofluorescence: chicken α-GFP ( 1:400; Abcam , Cambridge , MA ) , rabbit α-SYP-1 ( 1:200; MacQueen et al . , 2002 ) , rabbit α-SYP-3 ( 1:200; Smolikov et al . , 2007b ) , rabbit α-SYP-4 ( 1:200; Smolikov et al . , 2009 ) , guinea pig α-HTP-3 ( 1:400; Goodyer et al . , 2008 ) , rabbit α-RAD-51 ( 1:10 , 000; Novus Biological ( SDI ) , Littleton , CO ) , rabbit α-HIM-8 ( 1:500; Novus Biological ( SDI ) ) , rabbit α-PLK-1 ( 1:50; Labella et al . , 2011 ) , rabbit α-PLK-2 ( 1:200; Nishi et al . , 2008 ) , guinea pig α-SUN-1 Ser8-pi ( 1:700; Woglar et al . , 2013 ) . The following secondary antibodies from Jackson ImmunoResearch ( West Grove , PA ) were used at a 1:200 dilution: α-chicken FITC , α-rabbit Cy3 , α-goat alexa 488 , and α-guinea pig alexa 488 . Vectashield containing 1 μg/μl of DAPI from Vector Laboratories ( Burlingame , CA ) was used as a mounting media and anti-fading agent . Primary antibodies were used at the following dilutions for western blot analysis: rabbit α-SYP-3 ( 1:200; Smolikov et al . , 2007b ) , mouse α-tubulin ( 1:2000; Sigma , St . Louis , MO ) , and histone H3 ( 1:2000; Abcam ) . HRP-conjugated secondary antibodies , donkey anti-goat , rabbit anti-mouse , and mouse anti-rabbit from Jackson ImmunoResearch were used at a 1:10 , 000 dilution . Whole mount preparation of dissected gonads and immunostaining procedures were performed as in ( Colaiácovo et al . , 2003 ) . Immunofluorescence images were captured with an IX-70 microscope ( Olympus , Waltham , MA ) fitted with a cooled CCD camera ( CH350; Roper Scientific ) driven by the Delta Vision system ( Applied Precision , Pittsburgh , PA ) . Images were subjected to deconvolution by using the SoftWoRx 3 . 3 . 6 software ( Applied Precision ) . CRISPR-Cas9 genome editing technology was used to engineer syp-4 phosphodead and phosphomimetic mutations at the endogenous locus ( Tzur et al . , 2013 ) . To generate a phosphodead syp-4 mutant , serine 269 ( S269A ) was mutated to alanine . To generate a phosphomimetic mutant , serine 269 ( S269D ) was mutated to aspartic acid . We used the sgRNA recognition site ( AATTTGTGGAAGTCTCAGTCTGG ) 2414 base pairs downstream of the start codon . We cloned the sgRNA in to the pU6::unc-119_sgRNA plasmid ( Friedland et al . , 2013 ) . The donor sequence containing the genomic sequence of syp-4 extending from 943 bp upstream to 2128 bp downstream of the start codon with the following changes: TC to GA change at positions 1021 and 1022 , to generate the phosphomimetic mutant , and a T to G change at position 1021 , to generate the phosphodead mutant ( resulting in a silent mutation that is expected to prevent re-cutting by Cas9 ) , was cloned into the BglII site of the pCFJ104 vector expressing Pmyo-3::mCherry::unc-54 ( Frøkjaer-Jensen et al . , 2008 ) . A cocktail consisting of a plasmid expressing the sgRNA ( 200 ng/μl ) , a plasmid expressing the donor sequence ( 97 . 5 ng/μl ) , Cas9 ( 200 ng/μl ) and the co-injection marker pCFJ 90 ( Pmyo-2::mCherry::unc-54utr; 2 . 5 ng/μl ) was microinjected into the gonad arms of the worms ( P0s ) . F1 animals expressing the co-injection marker were sequenced to identify mutants and homozygous animals were picked from among the F2 generation and confirmed by sequencing . Young ( ~18 hr post-L4 stage ) wild type and spo-11 mutant adult animals were irradiated with approximately 10 Gy from a Cs137 source . Irradiated and untreated control worms were dissected 6–8 hr post-irradiation and immunostained with the phospho-specific SYP-4 antibody . Feeding RNAi experiments were performed as in ( Govindan et al . , 2006 ) with the following modifications: three L4-stage animals were placed on each RNAi plate and 24 hr post-L4 animals from the next generation were screened for a phenotype . Control RNAi was performed by feeding HT115 bacteria expressing the empty pL4440 vector . For complete depletion of rad-54 , L1-stage animals were placed on the RNAi plates and F2 animals were scored for a phenotype . Animals were grown on the rad-54 ( RNAi ) plates the entire time . FISH probes were generated as in ( Smolikov et al . , 2007b ) . The 5S rDNA probe was generated by PCR with the primers 5'-TACTTGGATCGGAGACGGCC-3' and 5'-CTAACTGGACTCAACGTTGC-3' . FISH probes were labeled using Terminal Transferase ( NEB M0315S , Ipswich , MA ) with Fluorescein-12-dCTP ( Perkin Elmer NEL-424 , Waltham , MA ) . The average numbers of nuclei scored per zone ( n ) for a given genotype to analyze pairing for chromosome V using FISH and pairing for the X chromosome using HIM-8 staining were as follows: zone 1 ( n = 263 ) , zone 2 ( n = 145 ) , zone 3 ( n = 157 ) , zone 4 ( n = 153 ) , zone 5 ( n = 148 ) , and zone 6 ( n = 134 ) . Statistical comparisons between genotypes were conducted using the two-tailed Mann-Whitney test , 95% confidence interval ( C . I . ) . Quantitative analysis of RAD-51 foci/nucleus was performed as in ( Colaiácovo et al . , 2003 ) . The average number of nuclei scored per zone ( n ) for a given genotype was as follows: zone 1 ( n = 235 ) , zone 2 ( n = 122 ) , zone 3 ( n = 119 ) , zone 4 ( n = 108 ) , zone 5 ( n = 99 ) , and zone 6 ( n = 95 ) . Statistical comparisons between genotypes were conducted using the two-tailed Mann-Whitney test , 95% C . I . Germ cell corpses were scored in adult hermaphrodites ( ~18 hr post-L4 ) as in ( Kelly et al . , 2000 ) . Statistical comparisons between different genotypes were conducted using the two-tailed Mann-Whitney test , 95% C . I . Day-1 ( ~18 hr post-L4 ) animals expressing SYP-3::GFP were anaesthetized with 0 . 1% levamisole and loaded with levamisole onto 4% agarose pads on slides ( VWR; cat#16004–368 , Radnor , PA ) and then covered with No . 1 . 5 coverslips . Images were acquired with a Nikon 60X/1 . 4 Plan Apo VC objective on an inverted Nikon Ti Microscope with Perfect Focus System and a Spectral Borealis-modified Yokogawa CSU-X1 spinning disc confocal head . EGFP fluorescence was excited with a 488 nm solid-state laser ( ~220µW at the sample ) and emission was collected through an ET525/50m emission filter ( Chroma , Bellows Falls , VT ) onto an ORCA-ER CCD camera ( Hamamatsu , Boston , MA ) with binning of 2 × 2 for an effective pixel size of 217 nm . Z-stacks of five planes with a 0 . 75 µm step-size were acquired every 30 s for a total of 80 volumes ( 40 min ) . After two baseline volumes were acquired , a diffraction-limited volume in roughly 10 nuclei across the field of view was photobleached throughout the Z-stack encompassing the entire depth of the nuclei using a MicroPoint ( Photonic Instruments , Saint Charles , IL ) equipped with a Coumarin 440 dye cell; MicroPoint laser power was adjusted in pilot experiments so that bleaching resulted in a ~90% reduction in local fluorescence intensity . Hardware and image acquisition were controlled with MetaMorph 7 . 8 . 4 ( Molecular Devices , Sunnyvale , CA ) . Z-stack sum-intensity projections were used for FRAP analysis . Images were qualitatively evaluated for animal health , and data sets in which animals ceased movement during the time-series were rejected . Gross lateral animal movement and drift were corrected for translation and rotation using the StackReg plugin ( Thévenaz et al . , 1998 ) in Fiji ( Schindelin et al . , 2012 ) , rectangular regions encompassing individual nuclei were extracted for further rigid body registration in StackReg to minimize the impact of chromosomal movement on intensity measurements . Only nuclei for which the photobleached region ( region of interest , ROI ) was in focus for the entire time period of fluorescence recovery were examined , whereas nuclei demonstrating obvious rotation around the lateral ( XY ) plane were excluded from analysis . After registration , the average intensity of the bleached region in each nucleus was measured over the duration of the time-lapse by manually tracing the photobleached region to correct for chromosomal movement . Background measured in a non-fluorescent region of the image was subtracted from each time point , and intensity measured in a region of interest encompassing all nuclei was used for double-normalization to correct for photobleaching during acquisition as well as lost signal due to the bleach pulse as described in ( Phair et al . , 2004 ) . Double-normalized traces were fit to a one-phase association curve: Y=A* ( 1-e-Kt ) , where A is the mobile fraction ( fit plateau ) , and K is the association rate . Traces that could not be fit to a one-phase association were rejected . The significance of the differences between the mobile fractions of different groups was assessed with a Kruskal-Wallis test . Statistical analysis was performed in Prism six ( GraphPad , La Jolla , CA ) . The difference across groups was p<0 . 0001 .	The majority of DNA in animal cells is stored in structures called chromosomes . Most cells contain two sets of chromosomes , one inherited from the mother and one from the father . Sperm and egg cells , however , contain only a single set of chromosomes . A specialized type of cell division called meiosis generates these cells . During meiosis , the chromosomes in a cell replicate to produce a cell that contains four copies of each chromosome . The equivalent chromosomes from the mother and the father are initially kept close together by a zipper-like structure called the synaptonemal complex . This allows the chromosomes to exchange segments of DNA , before the cell divides twice in successive rounds to produce four cells , each containing one set of chromosomes . Severing both of the DNA strands that make up a DNA molecule forms what is known as a double-strand break . To exchange DNA segments with another chromosome , double-strand breaks that form in the DNA of one chromosome are repaired in a process known as crossover formation . Only a subset of the double-strand breaks are designated to be repaired by crossover formation , but at least one crossover needs to form between each chromosome pair . This generates diversity and ensures that the chromosomes separate correctly at the first cell division . Since the synaptonemal complex holds equivalent chromosomes close together it ensures that at least some of the breaks are repaired by crossover formation . Nadarajan et al . have now investigated how a chemical modification called phosphorylation affects how the synaptonemal complex behaves in the roundworm Caenorhabditis elegans . A combination of genetic and cell-based approaches revealed that enzymes called polo-like kinases phosphorylate one of the proteins – called SYP-4 – that makes up the synaptonemal complex . This phosphorylation occurs after double-strand break sites have been designated to become crossovers . The synaptonemal complex is normally a dynamic structure , with the proteins that it consists of being continuously replaced . However , Nadarajan et al . found that phosphorylating SYP-4 made the synaptonemal complex less dynamic than it had previously been , which prevented further double-strand breaks from forming . Polo-like kinases are found in many organisms , from yeast to humans . Further work is now needed to investigate whether polo-like kinases phosphorylate the synaptonemal complex – and hence prevent the continuous formation of double-strand breaks – in animals such as mice and humans . This is important because failing to shut down the formation of double-strand breaks can result in cancer , infertility , miscarriages and birth defects in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2017	Polo-like kinase-dependent phosphorylation of the synaptonemal complex protein SYP-4 regulates double-strand break formation through a negative feedback loop.
The TAM receptor tyrosine kinases Tyro3 , Axl , and Mer regulate key features of cellular physiology , yet the differential activities of the TAM ligands Gas6 and Protein S are poorly understood . We have used biochemical and genetic analyses to delineate the rules for TAM receptor–ligand engagement and find that the TAMs segregate into two groups based on ligand specificity , regulation by phosphatidylserine , and function . Tyro3 and Mer are activated by both ligands but only Gas6 activates Axl . Optimal TAM signaling requires coincident TAM ligand engagement of both its receptor and the phospholipid phosphatidylserine ( PtdSer ) : Gas6 lacking its PtdSer-binding ‘Gla domain’ is significantly weakened as a Tyro3/Mer agonist and is inert as an Axl agonist , even though it binds to Axl with wild-type affinity . In two settings of TAM-dependent homeostatic phagocytosis , Mer plays a predominant role while Axl is dispensable , and activation of Mer by Protein S is sufficient to drive phagocytosis . The three receptor tyrosine kinases of the TAM family–Tyro3 , Axl , and Mer—are expressed in sentinel cells of the immune system , endothelial cells of the vasculature , neurons and glia of the nervous system , and professional phagocytes of the immune , nervous , and reproductive systems ( Lemke , 2013 ) . In these settings , TAM signaling regulates multiple functions but two are especially prominent . The first is the ‘homeostatic’ or silent phagocytosis of the billions of apoptotic cells ( ACs ) that are generated on a daily basis throughout life ( Scott et al . , 2001; Lemke and Burstyn-Cohen , 2010 ) . The second is the inhibition of the innate immune inflammatory response in dendritic cells ( DCs ) , macrophages , and other sentinel cells of the immune system ( Rothlin et al . , 2007; Lemke and Rothlin , 2008 ) . Deficiencies in TAM expression or activity lead to autoimmunity , blindness , and infertility ( Lu et al . , 1999; D'Cruz et al . , 2000; Lu and Lemke , 2001; Duncan et al . , 2003; Lemke and Lu , 2003; Rothlin and Lemke , 2010; Lemke , 2013 ) . Conversely , over-expression or aberrant activation of Tyro3 , Axl , or Mer ( gene name Mertk ) is associated with the development , progression , and metastasis of cancers ( Avilla et al . , 2011; Cummings et al . , 2013; Lemke , 2013; Meyer et al . , 2013; Paccez et al . , 2014 ) . Axl in particular has been shown to mediate the resistance of solid tumors and leukemias to epidermal growth factor receptor ( EGFR ) -directed chemotherapeutic agents ( Hong et al . , 2008; Zhang et al . , 2012; Meyer et al . , 2013 ) . Elevated TAM expression has also recently been implicated in increased susceptibility to infection by enveloped viruses ( Shimojima et al . , 2006; Morizono et al . , 2011; Meertens et al . , 2012; Bhattacharyya et al . , 2013 ) . The importance of TAM activity notwithstanding , the differential signaling capacity of the two soluble ligands that activate TAM receptors—Gas6 and Protein S ( gene and protein names Pros1 ) ( Stitt et al . , 1995; Varnum et al . , 1995 ) —is incompletely understood . Gas6 is thought to function as a ligand for all three receptors ( Stitt et al . , 1995; Nagata et al . , 1996; Chen et al . , 1997 ) , but the role of Pros1 as a ligand for one or more TAM receptors has until recently been controversial ( Godowski et al . , 1995; Anderson et al . , 2003; Hafizi and Dahlback , 2006 ) . At the same time , the unique structural features of Gas6 and Pros1—in which a C-terminal ‘SHBG domain’ binds to the Ig-like domains of TAM receptors while an N-terminal γ-carboxylated ‘Gla domain’ binds , in a Ca2+-dependent reaction , to the phospholipid phosphatidylserine ( PtdSer ) ( Lemke and Rothlin , 2008; Lemke , 2013 ) —have led to conflicting conclusions as to the relative importance of these two domains in receptor binding vs activation ( Mark et al . , 1996; Nakano et al . , 1997; Tanabe et al . , 1997 ) . Although Gas6 and Pros1 binding to the PtdSer-containing membranes of enveloped viruses potentiates TAM activation ( Meertens et al . , 2012; Bhattacharyya et al . , 2013 ) , and the Gla domain is thought to be required to ‘bridge’ a TAM receptor expressed on the surface of a phagocyte to the PtdSer expressed on the surface of its AC target ( Lemke and Rothlin , 2008; Lemke and Burstyn-Cohen , 2010; Nagata et al . , 2010 ) , a preparation of recombinant mouse Gas6 that lacks the Gla domain entirely is sold commercially as a TAM activator . Potentially distinct roles for Pros1 and Gas6 in vivo have only recently begun to be appreciated ( Burstyn-Cohen et al . , 2012; Carrera Silva et al . , 2013 ) , and to date only a single genetic study of the importance of Gas6 vs Pros1 in a TAM-dependent process has been reported ( Burstyn-Cohen et al . , 2012 ) . This study addressed TAM signaling during the phagocytosis of photoreceptor ( PR ) outer segments by retinal pigment epithelial ( RPE ) cells ( Strauss , 2005 ) , which express Mer and Tyro3 but no Axl ( Prasad et al . , 2006 ) . RPE phagocytosis of the PtdSer-displaying tips of PR outer segments ( Ruggiero et al . , 2012 ) requires Mer—mice , rats , and humans that carry loss-of-function mutations in the Mertk gene are blind , due to the cell-non-autonomous degeneration of nearly all PRs ( D'Cruz et al . , 2000; Gal et al . , 2000; Duncan et al . , 2003; Prasad et al . , 2006; Mackay et al . , 2010; Nandrot and Dufour , 2010 ) . This cell death results from the accumulation of toxic oxidated proteins that are generated during the course of phototransduction and are removed by phagocytosis . Retinal loss of either the mouse Gas6 or Pros1 gene alone yields a retina with a normal number of PRs , but the combined loss of Pros1 and Gas6 results in a PR degeneration phenotype that fully phenocopies the cell death seen in Mertk−/− mice ( Burstyn-Cohen et al . , 2012 ) . While this analysis demonstrated that Pros1 is sufficient to drive Mer-dependent phagocytosis in RPE cells , the presence of Tyro3 in these cells raised the possibility that Pros1 activation of the Mer kinase might be dependent on Pros1 binding to Tyro3 . In the current study , we have used biochemistry , receptor activation profiling , and genetic analyses of single and compound mouse mutants to establish the basic rules for TAM ligand–receptor interaction , signaling , and function . We find that Gas6 activates all three TAM receptors , but is an especially potent ligand for Axl , with which it has a unique association . In contrast , Pros1 activates Tyro3 and Mer but is inactive as an Axl agonist . Notably , we find that the Gla domains of TAM ligands are dispensable for receptor binding but are critical for optimal receptor activation , and that for Gas6 activation of Axl , this requirement is absolute . We conclude that a complete TAM signaling module is composed of a receptor , a γ-carboxylated protein ligand , and the phospholipid PtdSer , a tripartite arrangement that is unique to the TAM family . Finally , we show that Mer is the functionally predominant TAM receptor in two different settings of ‘homeostatic’ PtdSer-dependent phagocytosis in vivo–in the retina and the testes–and that Pros1 binding to and activation of Mer alone is sufficient to ensure wild-type levels of phagocytosis in these settings . We first generated highly pure preparations of recombinant mouse Gas6 and Pros1 . We transfected pCEP4-based expression plasmids containing cDNAs for both full-length and Gla domain-deleted ( ‘Gla-less’ ) versions of these TAM ligands into HEK293 EBNA cells ( Sasaki et al . , 2002 ) and selected stable transformants . Lines expressing the highest levels of each ligand were grown in serum-free production medium supplemented with vitamin K2 , required for γ-carboxylation of Gla domain glutamic acid residues ( Bandyopadhyay , 2008 ) . Ligands were purified to apparent homogeneity from conditioned medium , using affinity purification on a nickel-NTA resin followed by size exclusion and anion exchange chromatography ( see ‘Materials and methods’ ) . All recombinant mouse ligands ran as single bands under reducing conditions on SDS polyacrylamide gels and eluted from an anion exchange column as a single peak suggesting that the ligands were pure ( Figure 1A and data not shown ) . Commercially available Protein S purified from human plasma ( hPros1 ) ran as a single predominant band under reducing conditions , together with a secondary band of lower molecular weight ( Figure 1A ) . Size exclusion chromatography suggested that full-length Gas6 in solution is a mixture of monomers , dimers , and/or higher order multimers , whereas the Gla-less form appeared as a monomer ( Figure 1—figure supplement 1 ) . These elution profiles are consistent with earlier work on hPros1 , indicating that it forms disulfide-linked multimers , and that multimerization is enhanced by apoptotic cells ( Uehara and Shacter , 2008 ) . 10 . 7554/eLife . 03385 . 003Figure 1 . Recombinant TAM ligands and surface expression of TAM receptors . ( A ) Purified recombinant full-length and Gla-less mouse Gas6 ( rmGas6 and Gla-less rmGas6 , respectively ) and full-length and Gla-less Pros1 ( rmPros1 and Gla-less rmPros1 , respectively ) were run under both reducing ( r ) and non-reducing ( nr ) conditions in SDS-PAGE . In parallel , purified human Pros1 ( hPros1 ) was also run under reducing and non-reducing conditions . Gel was stained with Gel Code Blue ( Pierce ) . ( B ) Live cell labeling of Tyro3 ( top panels ) and Axl ( bottom panels ) on the surface of clonal populations of TAM TKO MEF lines expressing HA-tagged recombinant mouse Tyro3 ( left ) , recombinant mouse Axl ( middle ) , or recombinant mouse Mer ( right ) . Bar: 100 μm . ( C ) Induction of HA-tagged Mer expression in a Mer_TAM TKO MEF line in the presence of increasing concentrations of TetExpress transactivator protein ( Clontech ) . In this and all the subsequent blots in which the HA tag was used for both immunoprecipitation ( IP ) and immunoblotting ( IB ) , two different anti-HA antibodies were used: an anti-HA high affinity for IP and an anti-HA . 11 for IB . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 00310 . 7554/eLife . 03385 . 004Figure 1—figure supplement 1 . Size exclusion chromatography of full-length and Gla-less Gas6 . Purified full-length recombinant mouse Gas6 ( rmGas6 , black trace ) and Gla-less rmGas6 ( red trace ) were loaded onto and eluted from a Superdex 200 HR 10/30 gel filtration column . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 00410 . 7554/eLife . 03385 . 005Figure 1—figure supplement 2 . TAM receptor expression in immortalized cell lines . Lysates from the indicated cell lines were subjected to SDS-PAGE and subsequent immunoblotting with antibodies against Tyro3 ( top ) , Axl ( middle ) , and Mer ( bottom ) . Anti-Gapdh serves as a loading control . All established cell lines we have analyzed express at least one TAM receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 005 In order to have cellular assay targets in which individual TAM receptors could be expressed and their expression level normalized between cell lines , we first prepared immortalized mouse embryo fibroblast ( MEF ) lines from embryonic day ( E ) 13 . 5 Tyro3−/−Axl−/−Mertk−/− triple mutants ( ‘TAM TKO’ ) and all three possible double mutants ( e . g . , Tyro3−/−Mertk−/− , or ‘TM DKO’ ) ( Lu et al . , 1999; see ‘Materials and methods’ ) . The TAM TKO MEF lines provide a baseline cell population that lacks expression of any TAM receptor and the double mutant lines isolate a single receptor . This was particularly important , since in surveys of established cell lines we were unable to identify any human , mouse , rat , hamster , or monkey immortalized line that did not express at least one endogenous TAM receptor . [Selected examples of TAM expression in cultured cells are shown in Figure 1—figure supplement 2] This suggests that all prior analyses of TAM receptor activation have been encumbered by the expression of endogenous Tyro3 and/or Axl and/or Mer . We generated multiple clonal cell lines in the background of the TAM TKO MEFs that express HA-tagged versions of mouse Tyro3 , Axl , and Mer , respectively . We then selected one clonal population from the Tyro3 and Axl sets that expressed equivalent levels of these receptors , based on detection of the HA tag ( Figures 1B and 2D; see ‘Materials and methods’ ) . Expression in these Tyro3- and Axl-expressing MEFs was not maximal among the clones we isolated but was physiological in that we did not observe receptor activation ( autophosphorylation ) in the absence of added ligand ( Figure 2 ) , a phenomenon frequently observed when RTKs are over-expressed . Although we were also able to isolate MEF lines expressing Mer , these lines always ( for unknown reasons ) expressed much lower levels of this receptor ( Figure 1C and data not shown ) . For the Mer-expressing MEF line analyzed in this paper , we used a stable clonal line generated by expressing HA-tagged recombinant mouse Mer under the control of a TetExpress-inducible promoter ( Clontech; Figure 1C ) . For the Tyro3- and Axl-expressing MEF lines , we used Tyro3 and Axl antibodies directed against the extracellular domains of the receptors to live label non-permeabilized cells in culture and demonstrate surface expression of the individual receptors ( Figure 1B ) . Mer expression in clonal , Tet-inducible Mer_TAM TKO MEF lines , while detectable by western blot ( Figure 1C ) , was too low to be detected by immunocytochemistry . 10 . 7554/eLife . 03385 . 006Figure 2 . Gas6 and Pros1 exhibit TAM receptor selectivity . ( A ) Tyro3-expressing MEFs were stimulated with either human Pros1 ( hPros1 ) or recombinant mouse Pros1 ( rmPros1 ) . ( B and C ) Tyro3-expressing MEFs were stimulated with increasing concentrations of full-length recombinant mouse Gas6 ( rmGas6 ) , recombinant mouse Pros1 ( rmPros1 , B ) , or purified human Pros1 ( hPros1 , C ) . ( D ) Tyro3- and Axl-expressing MEFs were stimulated with increasing concentrations of rmGas6 . ( E ) Axl-expressing MEFs were stimulated with increasing concentrations of rmGas6 or rmPros1 . ( F ) Tyro3- and Axl-expressing MEFs were stimulated with increasing concentrations of hPros1 . ( The lower molecular weight doublet in lanes 8–14 of the anti-pY blot of this panel , in lanes 1 and 8–11 of the anti-pY blot of panel E , and in lane 8 of the anti-pY blot of panel D is seen only in Axl-expressing MEFs in which Axl is not activated by exogenous ligand and its intensity does not increase with increasing concentration of hPros1 or rmPros1 . It may represent basal Axl phosphorylation caused by low levels of MEF-produced ( endogenous ) Gas6 . See also lanes 1 and 7–12 of the anti-pY blot of Figure 3B ) ( G ) Mer-expressing MEFs were stimulated with increasing concentrations of rmGas6 or hPros1 . ( H ) Dexamethasone-treated bone marrow-derived macrophages ( Dex_Mϕ ) were stimulated with increasing concentrations of rmGas6 or hPros1 . Following stimulation with ligand for 10 min at 37°C , HA-tagged Tyro3 ( A–D and F ) , Axl ( D–F ) , Mer ( G ) , or endogenous Mer ( H , figure supplement 1 ) were immunoprecipitated from total cell lysates and subjected to SDS-PAGE and quantitative Licor western blotting ( panels G , H , ECL western blotting system ) with the indicated antibodies . In this and subsequent figures , receptor activation was assessed by blotting immunoprecipitates with an anti-phosphotyrosine antibody ( pY ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 00610 . 7554/eLife . 03385 . 007Figure 2—figure supplement 1 . Mer activation by rmGas6 and rmPros1 . Dex_Mφ were stimulated with increasing concentrations of rmGas6 or rmPros1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 007 We first assayed full-length recombinant mouse Gas6 and Pros1 , together with purified human Pros1 , for their ability to activate mouse Tyro3 , Axl , and Mer . We measured activation of single receptors expressed in MEFs by monitoring their tyrosine phosphorylation 10 min after ligand addition to cells in serum-free medium , using immunoprecipitation with antibodies to the individual receptors ( or their C-terminal HA tag ) followed by immunoblotting with an anti-phosphotyrosine antibody ( see ‘Materials and methods’ ) . We found that both purified human and recombinant mouse Pros1 ( hPros1 and rmPros1 , respectively ) were capable of rapidly activating Tyro3 autophosphorylation ( Figure 2A–C ) . In our hands , purified hPros1 was more stable biochemically and often more potent than rmPros1 ( Figure 2A ) , and we therefore used the human protein for the majority of the Pros1 experiments described below . However , both the mouse and human ligands were active ( Figure 2B , C ) . Consistent with these observations , both mouse and human Pros1 were also recently found to be active as ligands for a chimeric receptor comprised of the mouse Tyro3 ectodomain linked to the cytoplasmic domain of the R1 chain of the interferon ( IFN ) -gamma ( γ ) receptor ( Tsou et al . , 2014 ) . ( Activation of this chimera , expressed in CHO cells , was monitored by tyrosine phosphorylation of STAT1 . ) Mouse Tyro3 was also strongly activated by rmGas6 ( Figure 2B–D ) . In contrast , we found that mouse Axl could only be activated by rmGas6 ( Figure 2D–F ) . While we detected 10 min activation of Axl at rmGas6 concentrations as low as 1 nM ( Figure 2D ) , we could not detect Axl activation by either mouse Pros1 ( Figure 2E ) or human Pros1 ( Figure 2F ) at any concentration tested . Correspondingly , hGas6 but not mPros1 has been found to induce STAT1 phosphorylation downstream of a chimeric mAxl/γR1 receptor expressed in CHO cells ( Tsou et al . , 2014 ) . We used similar dose–response titrations to assess the ability of Gas6 and Pros1 to activate Mer . We found that both ligands were capable of activating this receptor in Mer-expressing MEFs , but that rmGas6 was more potent than hPros1 in this assay ( Figure 2G ) . Finally , as our Mer-expressing MEFs have much lower receptor levels than Tyro3 or Axl-expressing MEFs , we also performed the same analysis using dexamethasone-treated mouse bone marrow-derived macrophages ( Dex_Mϕ ) . These cells have levels of Mer comparable to those of Tyro3 or Axl in the MEF cell lines and do not express any other TAM receptor ( McColl et al . , 2009; Zagórska et al . , 2014; and data not shown ) . We found that both rmGas6 and hPros1/rmPros1 also function as a ligand for Mer in these cells ( Figure 2H and Figure 2—figure supplement 1 ) . Together with several components of the blood coagulation cascade , including prothrombin , Protein C , and factors VII , IX , and X , both Gas6 and Pros1 carry ‘Gla domains’—polypeptide segments of ∼45 amino acids that contain clusters of 10–12 glutamic acid residues whose gamma carbons are post-translationally carboxylated in a vitamin K-dependent reaction ( Stitt et al . , 1995; Nelsestuen et al . , 2000; Huang et al . , 2003; Bandyopadhyay , 2008 ) . This γ-carboxylation of glutamic acid is required for the Ca2+-dependent binding of Gla domains to PtdSer , which is displayed on the surface of apoptotic cells ( ACs ) and enveloped viruses ( Bhattacharyya et al . , 2013 ) . It is indispensible for the bioactivity of the Gla domain-containing proteins of the coagulation cascade and is a principal reason that vitamin K is an essential vitamin ( Freedman et al . , 1995; Ishimoto et al . , 2000; Bandyopadhyay , 2008; Rajotte et al . , 2008; Lemke , 2013 ) . The anticoagulant warfarin and related blood thinners antagonize vitamin K-dependent γ-carboxylation of Gla domains ( Stafford , 2005 ) . We first compared the ability of full-length vs Gla-less mouse Gas6 to activate Tyro3 . Gla-less Gas6 is deleted for the first 115 residues of the protein , including the Gla domain , but retains all four EGF-like domains and the SHBG-like domain . We found that although both full-length and Gla-less rmGas6 were capable of activating Tyro3 , the full-length protein was markedly ( ∼20-fold ) more potent ( Figure 3A ) . When we performed this same comparison for Axl-expressing MEFs , we observed an even stronger dependence . Full-length Gas6 was a very potent Axl ligand ( Figure 3B , left 6 lanes ) , but the Gla-less version was incapable of activating Axl at any concentration tested ( up to 750 nM; Figure 3B , right 6 lanes ) . We found that this stark disparity was unique to Axl . When we compared the ability of full-length and Gla-less rmGas6 to activate Mer in Mer-expressing Dex-Mϕ , we observed a pattern similar to that seen for Tyro3 . Although both forms of the ligand were capable of activating Mer , full-length Gas6 was ∼20-fold more potent than its Gla-less counterpart ( Figure 3C ) . Thus , the TAM receptors fall into two groups based on features of their ligand and PtdSer activation profiles: Tyro3 and Mer are activated by both Gas6 and Pros1 , whereas Axl is activated only by Gas6; and while Gla-less Gas6 is a substantially weaker ligand for Tyro3 and Mer , it is effectively dead as a ligand for Axl . 10 . 7554/eLife . 03385 . 008Figure 3 . The role of the Gla domain in TAM receptor activation . ( A , B , C ) Tyro3-expressing MEFs , Axl-expressing MEFs , or dexamethasone-treated BM-derived macrophages expressing Mer , respectively , were stimulated with the indicated increasing concentrations of full-length rmGas6 or Gla-less rmGas6 , respectively . Total cell lysates were immunoprecipitated with either HA antibodies ( A and B ) or Mer-specific antibodies ( C ) and subsequently subjected to SDS-PAGE and quantitative Licor western blotting ( panel C , ECL western blotting ) with the indicated antibodies . ( D ) MEFs expressing Axl or a chimeric receptor composed of the Tyro3 extracellular domain linked to the Axl tyrosine kinase domain ( ECDTTKA ) were stimulated with either full-length rmGas6 or Gla-less Gas6 . Receptors were immunoprecipitated from cell lysates using an HA antibody , and immunoprecipitates were subjected to SDS-PAGE and Licor western blotting with anti-phosphotyrosine and HA antibodies . The higher basal activation of the ECDTTKA construct may reflect its higher level of expression . ( E ) Binding assays were performed on Axl-expressing MEFs using a single concentration of 125I-labeled full-length rmGas6 in the presence of increasing concentrations of either unlabeled full-length rmGas6 ( black ) or Gla-less rmGas6 ( red ) . Measured concentrations of unlabeled ligand required for 50% inhibition of displaceable binding ( IC50 ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 008 The difference in the ability of Gla-less Gas6 to activate Axl vs Tyro3/Mer is unrelated to any difference in the intrinsic tyrosine kinase activity of the three receptors . We found that a Tyro3–Axl chimeric receptor composed of the Tyro3 extracellular domain linked to the Axl tyrosine kinase , when expressed in TAM TKO MEFs , displayed diminished activation by Gla-less Gas6 similar to that seen with Tyro3 ( Figure 3A , D ) . This suggests that the interaction of Gla-less Gas6 to Tyro3 and Mer is distinct structurally from its interaction with Axl . Together , these results indicate that the Gas6 Gla domain is required for optimal activation of Tyro3 and Mer , and that this requirement is absolute for Axl . We asked whether the marked difference in Axl activation displayed by full-length vs Gla-less rmGas6 was reflected in a difference in Axl binding . Since there have been conflicting reports as to the relative binding affinity of full-length vs Gla-less Gas6 to Axl-Fc fusion proteins , as measured by using BIAcore sensor chips ( Nakano et al . , 1997; Tanabe et al . , 1997; Demarest et al . , 2013 ) , we measured their Axl binding affinities in a more biologically relevant context; that is , to intact Axl receptor expressed on the surface of TM DKO MEFs ( prepared from Tyro3−/−Mertk−/− embryos ) in culture . We radio-iodinated full-length rmGas6 and then performed conventional competitive binding assays using displacement with increasing concentrations of either unlabeled full-length Gas6 or unlabeled Gla-less Gas6 ( see ‘Materials and methods’ ) . We found that full-length and Gla-less Gas6 displayed essentially the same IC50 for Axl binding—approximately 0 . 7 nM in this assay ( Figure 3E ) . Thus , the inability of Gla-less Gas6 to activate Axl is unrelated to its ability to bind the receptor; and conversely , the ability of a Gas6 variant to bind Axl is unrelated to its ability to activate the receptor . These properties suggest that a crystal structure of the Axl ectodomain bound to Gla-less Gas6 ( Sasaki et al . , 2006 ) represents a ligand-occupied but inactive Axl configuration . In related work , we have found that these properties are also exceptionally important to Gas6–Axl binding and signaling in tissues in vivo , where Gas6 appears to be specifically and constitutively bound to Axl without significant activation of the receptor ( Zagórska et al . , 2014 ) . Consistent with the failure of Pros1 to activate Axl in culture ( Figure 2E , F ) , we detected no binding of Pros1 to Axl-expressing MEFs ( data not shown ) . The importance of PtdSer for TAM ligand activity suggests that this phospholipid must be available to the purified TAM ligands that are added to MEF cultures ( as in Figures 2 and 3 ) . This is in spite of the fact that PtdSer is largely confined to the inner leaflet of the plasma membrane bilayer of non-apoptotic cells through the action of a set of P4-ATPases—so-called flippases ( van Meer et al . , 2008 ) . Consistent with this hypothesis , we found that the PtdSer-binding protein Annexin A5 antagonized Gas6 activation of Tyro3 in Tyro3_TAM TKO MEFs ( Figure 4A ) . We therefore probed for the presence of PtdSer in MEF cultures directly . We co-incubated MEFs ( grown under the same conditions we used for activation studies ) with propidium iodide ( PI ) and the fluorescent Annexin B12 derivative pSIVA , a polarity-sensitive biosensor that fluoresces only when bound to PtdSer ( Kim et al . , 2010; Ruggiero et al . , 2012 ) , and then analyzed the cells by flow cytometry . We detected significant number of pSIVA+PI− apoptotic cells and pSIVA+PI+ late apoptotic/necrotic cells in the culture medium and even a measurable number of pSIVA+PI− and pSIVA+PI+ cells in the adherent MEFs on the plate ( Figure 4B ) . Together , these results demonstrate that there are PtdSer-displaying membranes present in the MEF cultures . It is important to note that the molecular weight of PtdSer is ∼385 g/mol , while Gas6 and Pros1 have molecular weights of ∼80 , 000 g/mol , and that the crystal structure of the prothrombin Gla domain bound to lyso-PtdSer contains only a single molecule of this phospholipid ( Huang et al . , 2003 ) . Thus , a relatively low level of PtdSer can easily be in molar excess over the Gla domain-containing proteins with which it interacts . In this regard , the addition of either PtdSer-displaying enveloped viruses ( Bhattacharyya et al . , 2013 ) or PtdSer-displaying apoptotic cells ( Zagórska et al . , 2014 ) has been found to markedly shift the dose–response curves for Pros1 activation of Mer and Gas6 activation of Axl/Mer to lower ligand concentrations . 10 . 7554/eLife . 03385 . 009Figure 4 . Phosphatidylserine and TAM activation . ( A ) Tyro3-expressing MEFs were treated with the indicated concentrations of Annexin A5 for 10 min and then stimulated with 5 nM full-length rmGas6 for 10 min . Total cell lysates were immunoprecipitated with HA antibodies and subsequently subjected to SDS-PAGE and western blotting with the indicated antibodies . ( B ) FACS analysis . Adherent cells ( left ) and cells from culture medium ( ‘supernatant’ , right ) from Axl_TAM TKO MEF cultures were analyzed by staining with propidium iodide , which is only taken up by dead cells , and pSIVA , which only fluoresces when bound to PtdSer ( right panels ) . Numbers in the four quadrants of the panels indicate percent of signal in that quadrant . 21 . 2% of the gated material in the MEF culture supernatant ( upper right quadrant in the right panel ) represents PtdSer-expressing apoptotic cells . ( C and D ) Axl_TAM TKO MEFs ( C ) and control TAM TKO MEFs ( D ) were incubated +/− rmGas6 or Gla-less rmGas6 ( as indicated ) and +/− 10 mM EDTA for 90 min at 4°C and then live-stained with an anti-Gas6 antibody ( green ) and Hoechst to visualize nuclei . ( E ) Axl-expressing Axl_TAM TKO MEFs were stimulated +/− 10 nM rmGas6 in the presence of the indicated concentrations of EDTA for 10 min . Total cell lysates were either: ( top two blots ) immunoprecipitated with HA antibodies and subsequently subjected to SDS-PAGE and western blotting with the indicated antibodies; or ( bottom three loading control blots ) western blotted with the indicated antibodies . Scale bar ( for C and D ) 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 009 The binding of Gla domains to PtdSer is also Ca2+-dependent ( Huang et al . , 2003 ) and can be rapidly disrupted by Ca2+-chelating agents such as EDTA . We therefore used immunocytochemistry to ask if rmGas6 binding to Axl in Axl-expressing MEFs is Ca2+-dependent . We found that it is not: binding of rmGas6 ( at 25 nM ) to the surface of Axl-expressing MEFs was readily observed with an anti-Gas6 antibody at 90 min after application ( at 4°C ) , and this binding was unaffected by the inclusion of 10 mM EDTA ( Figure 4C ) . We also detected equivalent binding of Gla-less Gas6 ( Figure 4C ) . In control experiments , no Gas6 binding was detected in TAM TKO MEFs ( Figure 4D ) . These results argue that the observed Gas6 binding is via SHBG domain binding to Axl and not via Gla domain binding to any PtdSer expressed by the MEFs . We then asked whether Gas6 bound to Axl in the presence of EDTA ( Figure 4C , third panel ) is capable of activating Axl . We found that it is not: although 10 nM rmGas6 resulted in strong Axl activation in the absence of EDTA , the inclusion of 2 , 10 , or 50 mM EDTA all blocked this activation ( Figure 4E ) . These results indicate that the binding of full-length Gas6 alone , in the absence of simultaneous Ca2+-dependent binding of its Gla domain to PtdSer , is insufficient to trigger Axl activation . We next extended these cell culture observations to an in vivo setting . We injected either full-length or Gla-less rmGas6 intravenously ( IV ) into Gas6−/− mice and then used immunohistochemistry to monitor the binding of these injected proteins to splenic red pulp macrophages . F4/80+ red pulp macrophages express both Axl and Mer and are normally also Gas6+ ( Zagórska et al . , 2014 ) . By 30 min after IV injection , we observed equally strong binding of both full-length and Gla-less rmGas6 to these cells ( Figure 5A ) . However , this equivalent in vivo binding was not reflected in equivalent Axl activation . When we immunoprecipitated Axl from the spleen 30 min after IV injection of full-length Gas6 , we observed that Axl was strongly activated ( Figure 5B ) . In marked contrast , Axl immunoprecipitated after injection of Gla-less rmGas6 showed zero activation ( Figure 5B ) . This was in spite of the fact that equivalent amounts of full-length and Gla-less Gas6 could be recovered from the spleen ( Figure 5B ) . The PtdSer required for activation of Axl by full-length rmGas6 injected IV is presumably provided by both apoptotic cells and circulating PtdSer-exposing microparticles , derived from platelets , erythrocytes , leukocytes , and endothelial cells , which are abundant in the blood ( Lacroix and Dignat-George , 2012 ) . 10 . 7554/eLife . 03385 . 010Figure 5 . Gla-less Gas6 binds but does not activate Axl in vivo . ( A ) Sections of spleens from Gas6−/− mice 30 min after IV injection of saline ( top row ) , 30 μg full-length rmGas6 ( middle row ) , or 30 μg Gla-less rmGas6 ( bottom row ) , and stained with an anti-Gas6 antibody ( first column , green ) and an anti-F4/80 antibody to identify splenic red pulp macrophages ( second column , red ) . Merged images from the first and second columns are displayed in the third column . ( B ) Splenic lysates from Gas6−/− mice injected IV as in ( A ) were immunoprecipitated with the indicated antibodies 30 min after injection , and the immunoprecipitates then immunoblotted for pY and Axl ( top two panels ) , pY and Tyro3 ( middle two panels ) , or Gas6 ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 010 We found that the distinct receptor–ligand activation profiles described above reflect differences that are intrinsic to the Ig-like domains of the TAM ectodomains . We compared the ability of rmGas6 and hPros1 to activate TAM receptor autophosphorylation in MEF lines expressing either full-length Tyro3 or a chimeric receptor composed of the Tyro3 extracellular domain linked to the tyrosine kinase domain of Axl ( Figure 6A ) . This chimeric receptor was also expressed well in clonal TAM TKO MEF lines ( Figure 6B and Figure 6—figure supplement 1 ) . The activation profiles of both rmGas6 ( Figure 6C ) and hPros1 ( Figure 6D ) were similar for MEF lines expressing equivalent levels of wild-type Tyro3 or the chimeric Tyro3–Axl hybrid . We found that specificity resides within the two N-terminal Ig-like domains of the receptors , and that both of these domains are required: a hybrid receptor composed of the two Ig-like domains of Tyro3 linked to the remaining domains of Axl ( Figure 6A ) was , like Tyro3 , potently activated by Pros1 ( Figure 6D , lanes 5–8 ) , whereas a chimera that contained only the first Ig-like domain of Tyro3 ( Figure 6A ) was refractory to activation by Pros1 ( Figure 6D , lanes 1–4 ) . These results indicate that the intrinsic catalytic activity of Axl and Tyro3 are stimulated to comparable levels by ligand binding—of either Gas6 or Pros1–to the Tyro3 extracellular domain . 10 . 7554/eLife . 03385 . 011Figure 6 . Differences in TAM receptor activation are not due to differences in TAM kinase activity . ( A ) Schematic of wild-type Tyro3 and Axl , a Tyro3/Axl chimeric receptor carrying the complete Tyro3 ectodomain linked to the Axl transmembrane ( TM ) and tyrosine kinase ( TK ) domains ( ECDTTKA ) , and chimeric receptors carrying both Tyro3 Ig domains ( ECDTIgI/IITKA ) or only the first Tyro3 Ig domain ( ECDTIgITKA ) . FnIII: fibronectin type III repeat . ( B ) Immunostaining of Tyro3- , Axl- , and Tyro3/Axl chimera expressing TAM TKO MEFs , using an anti-HA antibody ( green ) . Bar: 50 μm . ( C and D ) MEFs expressing Tyro3 or the indicated Tyro3/Axl chimeric receptors were stimulated with increasing concentrations of either rmGas6 or hPros1 , respectively . Cell lysates were immunoprecipitated with anti-HA antibodies and subjected to SDS-PAGE electrophoresis followed by quantitative Licor western blotting with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 01110 . 7554/eLife . 03385 . 012Figure 6—figure supplement 1 . Surface expression of Tyro3/Axl chimeric receptors . Live cell labeling of surface expression of Tyro3 or Tyro3/Axl chimeric receptors , respectively . Cells were incubated with Tyro3 and Axl antibodies in combination on ice for 1 . 5 hr . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 012 We extended our in vitro observations of differential TAM interactions to two in vivo settings in which genetic analyses have shown that TAM signaling plays an essential role . The first of these is the phagocytosis of the distal tips of photoreceptor's ( PR ) outer segments by the retinal pigment epithelial ( RPE ) cells of the eye ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) . As noted above , Mer is absolutely required for this phagocytosis ( D'Cruz et al . , 2000; Duncan et al . , 2003; Prasad et al . , 2006; Nandrot and Dufour , 2010; Ostergaard et al . , 2011 ) . In Mertk−/− mice , a phagocytic deficiency results in the nearly complete death of PRs by 12 weeks of age ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) . We have previously shown that Tyro3 mouse mutants , Gas6 mouse mutants , and retina-specific Pros1 mouse mutants all have a normal number of PRs at this time ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) ; but that Gas6/Pros1 double mutants display PR death and retinal degeneration that fully phenocopies the degeneration of the Mertk mutants ( Burstyn-Cohen et al . , 2012 ) . These analyses demonstrate that Pros1 acts as an effective Mer activator during RPE cell phagocytosis but not that this occurs through direct binding of Pros1 to Mer . To address this question , we generated Tyro3−/−Gas6−/− double mutants . In these mice , the only TAM ligand remaining is Pros1 , and the only TAM receptor through which it can bind and signal in RPE cells is Mer . As noted above , RPE cells do not express Axl ( Burstyn-Cohen et al . , 2012 ) , which , in addition , is only activated by Gas6 ( Figure 2E , F ) . As described previously , PR death can be assessed quantitatively in retinal sections by measuring the radial thickness of the outer nuclear layer ( ONL ) , which is composed exclusively of PR nuclei ( Burstyn-Cohen et al . , 2012 ) . In 10–12-week old wild-type ( Figure 7A , first panel ) , Gas6−/− ( Figure 7A , third panel ) , and Tyro3−/− retinae ( Figure 7A , fourth panel ) , the ONL has a normal radial thickness of ∼50 μm and consists of 12–15 compact PR nuclei ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) . In marked contrast , the Mertk−/− ONL is reduced to a thickness of only 1–4 PR nuclei ( Figure 7A , second panel ) ( Burstyn-Cohen et al . , 2012 ) . ( The Figure 7A panel illustrates a region of severe PR degeneration in which the ONL is only one nucleus thick . ) We found that the Tyro3−/−Gas6−/− ONL is also of wild-type thickness ( Figure 7A , fifth panel ) . Thus , Pros1–Mer signaling alone is sufficient to mediate apparently normal phagocytosis in the retina . 10 . 7554/eLife . 03385 . 013Figure 7 . The Pros1–Mer signaling axis is sufficient for RPE phagocytosis in the retina . ( A ) Dorsal–ventral H&E stained sections from 12–14 week mouse retinae . The outer nuclear layer ( ONL ) , composed exclusively of PR nuclei , is delimited by the vertical white line . The ONL in the Mertk−/− mutant ( second panel ) is reduced to a thickness of a single nucleus ( arrow ) and the outer segment ( OS ) layer above is absent . In contrast , the ONL in the Gas6−/− , Tyro3−/− , and the Tyro3−/−Gas6−/− retinae is of a thickness that is indistinguishable from wild-type control . Bar: 50 μm . ( B ) Pros1-mediated activation of Mer in the RPE cell layer is independent of Tyro3 expression . Eye cups were acutely isolated from wild-type and Tyro3−/− mice . Cornea , iris , and lens were removed , leaving the RPE cell layer exposed . The eyecup was cultured under starvation conditions for 3 hr , and subsequently stimulated with increasing concentrations of hPros1 . Cell lysates were immunoprecipitated with anti-Tyro3 or anti-Mer antibodies and receptor activation monitored by immunoblotting with anti-phosphotyrosine antibody ( pY ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 013 We also assessed the ability of hPros1 to activate Mer in a dissected RPE layer preparation in vitro ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) . As before ( Prasad et al . , 2006 ) , we isolated the eyecup from either wild-type or Tyro3−/− adult mice and removed the cornea , lens , and retina to leave the apical surface of the RPE cell layer exposed . We then treated this surface , where Mer and Tyro3 are localized ( Prasad et al . , 2006 ) , with hPros1 . We have shown previously that purified Pros1 can enhance Mer autophosphorylation in this preparation ( Prasad et al . , 2006; Burstyn-Cohen et al . , 2012 ) . As shown in Figure 7B ( upper two panels ) , we demonstrate that purified Pros1 can also activate Tyro3 in this preparation and can activate Mer in both wild-type and Tyro3−/− RPE cells ( Figure 7B , lower two panels ) . Thus , Pros1 induces Mer autophosphorylation in the RPE directly , in the absence of any other TAM receptor , and this Pros1-to-Mer activation alone is sufficient to maintain a normal number of PRs in the retina in vivo . A second setting in which PtdSer-dependent TAM signaling plays an essential role is the homeostatic phagocytosis of apoptotic cells in the seminiferous tubules of the testes ( Lu et al . , 1999 ) . This setting is distinct from retina , where RPE cells phagocytize only a portion of a living cell . In the testes , Sertoli cells must engulf and clear millions of apoptotic germ cells that are generated normally during every cycle of mammalian spermatogenesis ( >108/day in a human male ) . Sertoli cells express all three TAM receptors and both ligands ( Lu et al . , 1999; Chen et al . , 2009 ) , and Tyro3/Axl/Mer triple mouse mutants are infertile due to the toxic accumulation of ACs ( Lu et al . , 1999 ) . We used immunostaining with antibodies to activated ( cleaved ) caspase 3 ( cCasp3; see ‘Materials and methods’ ) to quantify the number of ACs present in the 10–12-week seminiferous tubules across an informative set of TAM receptor and ligand mutants . We counted all activated cCasp3+ cells throughout an entire testis section and then normalized these counts to the number of tubule cross-sections present per testis section , which averaged 262 . 1 ± 20 . 6 ( 1 S . D . ) and did not vary significantly for any of the genotypes we analyzed ( data not shown ) . We found that the major receptor required for Sertoli cell phagocytosis of apoptotic germ cells in the mouse testis is Mer ( Figure 8A , B and Figure 8—figure supplement 1 ) . While Tyro3−/− , Axl−/− , and Gas6−/− single mutant mice all displayed wild-type number of cCasp3+ ACs—∼0 . 02 cells per tubule cross-section ( Figure 8A , and Figure 8—figure supplement 1 ) –Mertk−/− mice displayed nearly 20 times more ( Figure 8B ) . In contrast to other tissues we have analyzed ( e . g . , the spleen ) , the number of ACs was not increased in Mertk−/−Axl−/− double knock-outs compared to Mertk−/− single knock-outs ( Figure 8—figure supplement 1A , C ) . More tellingly , Tyro3−/−Axl−/− double knock-outs displayed wild-type number of ACs in the testes ( Figure 8—figure supplement 1A , C ) , and this was also the case for Tyro3−/−Gas6−/− double knock-outs , in which the only possible TAM signaling axis is Pros1 to Mer ( Figure 8A , B ) . 10 . 7554/eLife . 03385 . 014Figure 8 . Pros1–Mer signaling is sufficient for TAM-dependent homeostatic clearance of apoptotic cells in the testis . ( A ) Representative H&E images ( top row ) or immunostaining for cleaved caspase 3 ( cCasp3 ) ( bottom row ) in testis sections from 12–14 week old wild-type and TAM ligand and/or TAM receptor mutants . For the Mertk−/− sections ( second column ) , apoptotic cells are highlighted with white asterisks in the H&E-stained section , and cCasp3+ cells are highlighted with black asterisks . Bar: 100 μm . ( B ) Quantitative analysis of the number of apoptotic cells in wild-type and various TAM receptor/ligand mutants . Cleaved caspase 3+ cells and tubule cross-sections were counted in four testis cross-sections per mouse . Data are expressed as average number of ACs per number of total tubule cross-sections in each testis section . Error bars represent standard error of the mean for three independent animals . ( C ) TAM receptor expression in TAM receptor and TAM ligand mouse mutants . Testes were collected from 12- to 14-week old mice of the indicated genotypes and lysed . Lysates were subjected to SDS-PAGE and immunoblotting with anti-Tyro3 , Axl , and Mer antibodies , with anti-Akt serving as a loading control . ( D ) Comparative analysis of TAM expression in the testis . Lysates from either wild-type testis or TAM TKO MEFs expressing HA-tagged Tyro3 , Axl , or Mer were subjected to SDS-PAGE and subsequent quantitative Licor immunoblotting with the indicated antibodies . ( E ) Band intensity of the HA-tagged TAM receptors relative to each other and to wild-type testis was calculated using Licor Odyssey software and utilized to calculate the TAM protein expression relative to Tyro3 in the wild-type testis . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 01410 . 7554/eLife . 03385 . 015Figure 8—figure supplement 1 . Tyro3 and Mer are key regulators of homeostatic apoptotic cell clearance in the testis . ( A ) Representative H&E images ( top rows ) and immunostaining for cCasp3 ( bottom rows ) in 10- to 12-week old wild-type and TAM-deficient animals of the indicated genotypes . All mice except for TAM TKOs were on a pure C57Bl/6 background . Bar is 100 μm . ( B ) Comparison of the number of apoptotic cells identified by cCasp3 immunostaining in wild-type ( WT ) and Tyro3−/−Mer−/− double mutant testis at 10–12 weeks ( young ) vs the same genotypes at 9–12 months ( old ) . Bar is 100 μm . ( C ) Quantitative analysis of apoptotic ( cleaved Casp3+ ) cells in wild-type and various TAM ligand and/or TAM receptor mutants . Apoptotic cells as determined by cCasp3 staining and tubule cross-sections were counted in four testis sections per animal and averaged per number of total tubule cross-sections . All mice except for TAM TKOs were on a pure C57Bl/6 background . Error bars represent standard error of the mean for three independent animals ( Pros1f/−Gas6−/− , n = 2 ) . ( D ) Quantitative analysis as in ( C ) for the young/old testis comparison illustrated in ( B ) . These mice were on a hybrid C57Bl/6 × 129sv background . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 015 In contrast to the retina , we observed a significant effect of Tyro3 mutation in the presence of an existing Mertk mutation: Tyro3−/−Mertk−/− double mutants displayed substantially more ACs in the testis ( ∼2 cCasp3+ cells per each tubule cross-section ) than did Mertk−/− single mutants ( Figure 8—figure supplement 1A , C ) . Combining all three receptor knock-outs in Tyro3−/−Axl−/−Mertk−/− triple mutants led to a massive accumulation of ACs and cCasp3+ material that made accurate counting of ACs impossible ( Figure 8—figure supplement 1A ) . This suggests that in the absence of Mer and Tyro3 , Axl can also mediate AC phagocytosis . Finally , we examined the effect of removing all Gas6 and half the normal Pros1 throughout the mouse and found that the Prosfl/−Gas6−/− testes also contained a wild-type number of ACs ( Figure 8—figure supplement 1C ) . Thus in both the testes and the retina ( Burstyn-Cohen et al . , 2012 ) , only half the normal level of only a single TAM ligand ( Pros1 ) is sufficient to maintain an essential level of TAM-dependent phagocytosis . All three TAM receptors are detectable by western blot in the testis ( Figure 8C ) , and there was no significant compensatory change in Tyro3 , Axl , or Mer expression in Mertk−/− , Tyro3−/− , Gas6−/− , or Tyro3−/−Gas6−/− mice ( Figure 8C ) . We used comparative signal intensities on Licor immunoblots of lysates from testis and TAM TKO MEFs expressing HA-tagged Tyro3 , Axl , and Mer ( Figure 8D ) to determine that the relative protein expression level of TAM receptors in the testis is Mer > Tyro3 > Axl ( Figure 8E ) . Finally , we confirmed that the phenotypes we analyzed at 10–12 weeks were fully developed by this time . We measured the number of cCasp3+ cells in wild-type vs Tyro3−/−Mertk−/− mice at 10–12 weeks vs 9–12 months and saw that this number was the same in both the young and aged testis ( Figure 8—figure supplement 1D ) . Together , these results indicate that Mer is the major facilitator of steady-state homeostatic phagocytosis in both the retina and the testis , and that binding and activation of this receptor by Pros1 alone at only half its normal gene dose is sufficient to maintain normal levels of phagocytosis . From the above analyses , we draw seven salient conclusions with respect to TAM receptor–ligand–phospholipid interactions , which are summarized in Figure 9 . First , Gas6 alone is capable of binding to and activating Tyro3 , Axl , and Mer when these receptors are expressed in isolation and is especially potent as a ligand for Axl . Second , Pros1 alone is capable of binding to and activating Tyro3 and Mer but does not function as a ligand for Axl . Third , the PtdSer-binding Gla domain of Gas6 , PtdSer itself , and calcium are all required for optimal receptor activation but none is required for receptor binding . Fourth , ligand binding does not translate into receptor activation: Gla-less Gas6 binds as well to Axl as does its full-length counterpart , but is dead as an activator , and full-length Gas6 binds to Axl in the presence of EDTA but does not activate . Fifth , receptor heterodimerization is not an essential feature of TAM activation , since the TAM receptors can be activated by purified single ligands when expressed as single receptors in MEFs in the absence of any other TAM receptor , and in the retina , Mer activation in wild-type and Tyro3-deficient RPE cells is equivalent . Sixth , in the two settings of PtdSer-dependent ‘homeostatic’ phagocytosis—by RPE cells in the adult retina and Sertoli cells in the adult testes—Mer is the functionally predominant TAM receptor . And seventh , in these settings , Pros1 binding to and activation of Mer alone is sufficient to ensure wild-type levels of phagocytosis . 10 . 7554/eLife . 03385 . 016Figure 9 . Rules of engagement for TAM receptor ligand interaction and signaling . Gas6 activates all three TAM receptors independently , but Axl is uniquely dependent on Gas6: Pros1 activates Tyro3 and Mer but not Axl . Optimal activation of any receptor by any ligand requires the simultaneous presence of PtdSer , which binds to the Gla domain of the ligands , and calcium ions ( Ca2+ ) . Gla-less Gas6 is dead as a ligand when assayed against Axl in isolation . Its Axl-bound orientation is schematized differently from Tyro3- and Mer-bound Gla-less Gas6 , since the latter two result in partial kinase activation , although there are no structural data for these complexes . The right-most signaling configuration , in which full-length Gas6 is also inactive as an Axl ligand in the absence of PtdSer and Ca2+ , is speculation based on the results of this paper and data in ( Zagórska et al . , 2014 ) . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03385 . 016 Prior to this study , the relative binding and selectivity of Gas6 and Pros1 to the individual TAM receptors were incompletely understood . While Gas6 was widely regarded as a universal TAM ligand , the role of Pros1 in TAM-mediated signaling has , until recently , been controversial ( Godowski et al . , 1995; Stitt et al . , 1995; Burstyn-Cohen et al . , 2012 ) , and older data regarding the relative potency of Gas6 and Pros1 in the binding and activation of the specific TAM receptors in vitro were conflicting ( Ohashi et al . , 1995; Nagata et al . , 1996; Chen et al . , 1997 ) . Perhaps most importantly , all prior analyses of receptor activation following the addition of TAM ligands to cultured cells were confounded by the presence of endogenous TAM receptors in the cell lines used for these analyses . Together with earlier work from our group in the retina ( Burstyn-Cohen et al . , 2012 ) , the studies of Figures 7 and 8 represent the only use of molecular genetics to parse differential TAM receptor–ligand signaling interactions in vivo . Our experiments suggest a model in which the interaction between distinct TAM receptors and ligands defines a signaling signature that is specific for each ligand–receptor pair ( Figure 9 ) . This is best illustrated in the case of Gas6 activation of Axl , which is uniquely dependent on Gas6 for activation . Both full-length Gas6 and a Gla-less Gas6 variant that lacks its PtdSer-binding Gla domain bind to Axl with the same sub-nanomolar equilibrium dissociation constant , yet the former is an extremely potent Axl agonist while the latter is inactive , whether assayed in vitro or in vivo . Furthermore , full-length Gas6 cannot activate Axl in the presence of EDTA , which disrupts the binding of the Gas6 Gla domain to PtdSer . In a related study ( Zagórska et al . , 2014 ) , we have found that just as Axl is uniquely dependent on Gas6 for activation , so is Gas6 uniquely dependent on Axl for its stable localization in several different tissues in vivo . We find that Axl is constitutively pre-bound to Gas6 in these tissues but that this binding does not result in significant Axl activation ( Zagórska et al . , 2014 ) . Together , all of these observations suggest that the induced exposure of PtdSer is actually the regulated trigger for TAM activation and that the TAM receptor–ligand pair is in reality a PtdSer detector . They also suggest that a crystal structure of the two Ig-like domains of the Axl ectodomain bound to two molecules of Gla-less Gas6 ( Sasaki et al . , 2006 ) reflects the structure of an inactive complex . We propose that an effective TAM signaling complex in vivo is always tripartite , in that it is always composed of a TAM receptor , a Gla domain-containing TAM ligand , and the phospholipid PtdSer ( Figure 9 ) . This configuration is unique to the TAMs . In this regard , it is important to note that all settings of TAM-dependent ‘homeostatic phagocytosis’ are strictly co-dependent upon the exposure of PtdSer on the surface of the phagocytic target ( Shiratsuchi et al . , 1997; Ruggiero et al . , 2012; Lemke , 2013 ) . In the case of RPE cell phagocytosis of juxtaposed PR outer segment tips , PtdSer is locally exposed only on these outer segment tips ( i . e . , not on the remainder of the photoreceptor ) and only during the narrow window after subjective dawn in which phagocytosis occurs each day ( Ruggiero et al . , 2012 ) . The activity of Gla domains is also Ca2+-dependent , as these ions both stabilize the folded structure of the domain and also interact with PtdSer ( Bandyopadhyay , 2008; Huang et al . , 2009 ) . Thus , the TAM receptor tyrosine kinases represent a novel addition to the set of enzymes , such as protein kinase C , whose activities are regulated by both calcium and lipid binding ( Lemmon , 2008; Leonard and Hurley , 2011 ) . The key difference for the TAMs is that neither the lipid nor the calcium binds the kinase directly . Instead , they bind the TAM ligands , which are outside the cell . The dependence of Axl on Gas6 may be relevant to the fact that all of the very low level of Gas6 that is present in blood appears to be complexed with soluble Axl ( sAxl ) extracellular domain ( Ekman et al . , 2010c ) . Upon Axl activation , this ‘ectodomain’ is proteolytically cleaved from the rest of the receptor ( Costa et al . , 1996; O'Bryan et al . , 1995; Wilhelm et al . , 2008 ) , which results in the generation of an sAxl–Gas6 complex . In related work , we have shown that this complex is also generated when Axl is activated in tissues subsequent to the injection of activating ( cross-linking ) α-Axl antibodies ( Zagórska et al . , 2014 ) . Although there are conflicting reports as to the presence of soluble Tyro3 and Mer ectodomains in serum and the possibility that Gas6 might also be bound to soluble Mer ( Sather et al . , 2007; Ekman et al . , 2010c ) , antibody depletion of Gas6 from serum does not alter the gel filtration profile of either soluble Mer or Tyro3 ( Ekman et al . , 2010c ) . Elevated blood levels of soluble Axl have recently been reported to mark a variety of human disease and trauma states , including aortic aneurysm ( Ekman et al . , 2010b ) , lupus flares ( Zhu et al . , 2014 ) , pneumonia infection ( Ko et al . , 2014 ) , preeclampsia ( Liu et al . , 2014 ) , coronary bypass ( Lee et al . , 2013 ) , obesity and insulin resistance ( Hsiao et al . , 2013 ) , and limb ischemia ( Ekman et al . , 2010a ) . We suggest that the generation of a soluble Axl–Gas6 complex is triggered by the induced cellular exposure of PtdSer in many of these settings . Aberrant TAM signaling in human disease , most notably in cancer , is now a major research focus ( Leconet et al . , 2013; Meyer et al . , 2013; Schlegel et al . , 2013 ) , and small-molecule TAM tyrosine kinase inhibitors and antibodies that inhibit TAM ligand–receptor binding are in development as cancer therapies ( Holland et al . , 2010; Ye et al . , 2010; Brandao et al . , 2011; Schlegel et al . , 2013; Paccez et al . , 2014 ) . These same inhibitors have been proposed for the treatment of enveloped virus infections ( Bhattacharyya et al . , 2013; Shibata et al . , 2014 ) . Conversely , TAM activators have been proposed as possible treatments for several autoimmune indications ( Rothlin and Lemke , 2010; van den Brand et al . , 2013 ) . In each of these settings , our results indicate that it will be critical to know which TAM receptor to target . This will be particularly important in the context of cancer , since tumors profiled in the TCGA database ( https://tcga-data . nci . nih . gov/tcga/ ) that exhibit mutation or expression changes in the Tyro3 , Axl , and Mertk genes display a strong tendency towards mutually exclusive TAM changes across tumor types . At the same time , our studies here , related work in the immune system ( Zagórska et al . , 2014 ) , and prior analyses ( Scott et al . , 2001; Rothlin et al . , 2007; Seitz et al . , 2007; Burstyn-Cohen et al . , 2009 , 2012; Carrera Silva et al . , 2013 ) all demonstrate that distinct normal functions are preferentially assumed by individual TAM receptors and ligands in macrophages , dendritic cells , Sertoli cells , endothelial cells , and RPE cells . Our delineation of the rules of engagement for TAM signaling ( Figure 9 ) therefore has important implications for the therapeutic application of inhibiting and activating TAM modulators , particularly with respect to their target specificity and the avoidance of possible treatment side effects in the immune , visual , and reproductive systems . Antibodies used were as follows: anti-Tyro3 ( Santa Cruz , Dallas , TX , sc-1095 , R&D Systems , Minneapolis , MN , AF759 ) , anti-Axl ( Santa Cruz sc-1097 , R&D Systems AF854 ) , anti-Mer ( R&D Systems AF591 ) , anti-HA ( Covance , Princeton , NJ , MMS-101P , Roche , Mannheim , Germany 11 867 423 001 ) , anti-Gas6 ( R&D Systems AF986 ) , anti-Akt ( Cell Signaling , Beverly , MA , 4691 ) , anti-Gapdh ( Millipore , Billerica , MA , MAB374 ) , anti-cleaved caspase 3 ( Cell Signaling 9661 ) , and anti-phosphotyrosine ( Millipore 05–321 ) . Human Pros1 was from Enzyme Research Laboratories , South Bend , IN , and Tet-Express from Clontech , Mountain View , CA . Mice were bred and housed in the Salk Institute Animal Facility in a sterile environment under a 12-hr light/dark cycle . All experiments and procedures were conducted according to the guidelines established by the Institutional Animal Care and Use Committee ( IACUC ) . The Tyro3−/− , Axl−/− , and Mertk−/− mutants ( Lu et al . , 1999 ) , the Gas6−/− mutants ( Angelillo-Scherrer et al . , 2001 ) , and the floxed Pros1 mutants ( Burstyn-Cohen et al . , 2012 ) were all as described previously . All mice used in this paper were on a pure C57Bl/6 background , with two exceptions: the Tyro3−/−Axl−/−Mertk−/− ( TAM TKO ) triple mutant mice of Figure 8—figure supplement 1A and the young/old comparison of Figure 8—figure supplement 1B , D , both of which were on a mixed C57Bl/6 × 129sv background . Spontaneously immortalized MEF cell lines from Tyro3−/−Axl−/−Mertk−/− and Tyro3−/−Mertk−/− mutant mice were generated following a standard 3T3 protocol ( Xu , 2005 ) . Briefly , E13 . 5–15 . 5 embryos were isolated from a pregnant female of each genotype , and embryo body was subject to manual dissociation and incubation with trypsin to isolate single cells . Proliferation of isolated cells was monitored for 15–25 passages , and a growth curve was calculated for each cell passage . Immortalized MEFs were subsequently frozen and utilized for experiments . C-terminal HA-tagged Tyro3 or Axl cDNA was sub-cloned into pMX IRES retroviral vector . Retrovirus was produced in Phoenix Ampho cell lines and used to transduce TAM TKO cells . For Tet-inducible Mer cell lines , C-terminal HA-tagged Mertk cDNA was sub-cloned into a pRetroX-Tre3G vector and co-transfected with pAmpho env expression vector into a GP2-293 cell for retrovirus production according the manufacturer's protocol ( Clontech ) . Clonal populations were generated following antibiotic selection and receptor levels normalized by quantitative immunoblotting using the HA tag . cDNAs encoding either full-length mouse Gas6/Pros1 or Gas6/Pros1 lacking the Gla domain were sub-cloned into pCep4 vector ( Invitrogen , Carlsbad , CA ) . PCR primers introduced a C-terminal His6 tag for purification and the constructs were transfected into HEK293 cells . Cells were grown to confluency in DMEM supplemented with 10% FBS , 0 . 25 mg/ml G418 , 100 µg/ml hygromycin , and Pen/Strep and subsequently switched to serum-free Pro CDMa medium ( Lonza , Walkersville , MD ) supplemented with 10 µM vitamin K2 ( Sigma , St . Louis , MO ) and Pen/Strep for 72 hr . Conditioned medium was passed through a 0 . 22 µm filter , affinity purified through a nickel-NTA resin , and further purified on an anion exchange column ( Mono Q 5/5; GE Healthcare , Pittsburgh , PA ) in 20 mM Tris–HCl , pH 8 . 0 using a sodium chloride gradient . Recombinant full-length mouse Gas6 was 125I-labeled by the iodogen method according to the manufacturer's protocol ( Pierce , Rockford , IL ) . Immortalized Tyro3−/−Mer−/− MEFs were plated in a 12-well dish in triplicate and subsequently pre-starved in 3% FBS in DMEM overnight until cells reach confluency . Cells were placed in starvation medium for 3 hr before binding assays were performed . For binding experiments , cells were incubated with 125I-labeled full-length rmGas6 with increasing concentrations of either unlabeled full-length rmGas6 or unlabeled Gla-less Gas6 for 3 hr on ice . Cells were washed and subsequently lysed overnight in 0 . 5 M NaOH . Radioactive content of the cell lysates was counted in Opti-Fluor ( Perkin Elmer , Waltham , MA ) using a scintillation counter ( Beckman Coulter , Brea , CA , LS6500 ) . The half-life of the displacement curves was determined by fitting the curve with Prism software ( Graphpad ) . Cells were plated on glass coverslips , pre-starved overnight in 3% FBS/DMEM , and subsequently incubated in serum-free medium for 3 hr prior to treatment . Coverslips were briefly washed with 1× PBS , fixed for 15 min with 4% paraformaldehyde ( PFA ) /PBS , incubated in blocking buffer ( 3% BSA/0 . 05% saponin/PBS ) , primary antibody , and fluorophore-conjugated secondary antibody diluted in blocking buffer . For live labeling for surface receptor expression ( Figure 1B and Figure 6—figure supplement 1 ) , coverslips were transferred to chilled serum-free cell culture medium supplemented with anti-Tyro3 or anti-Axl antibodies ( R&D Systems ) on ice for 30 min . Coverslips were briefly washed with ice cold PBS , fixed for 15 min in 4% PFA/PBS , incubated with a fluorophore-conjugated secondary antibody , and counterstained with Hoechst solution . For live labeling for Gas6 binding assays ( Figure 4D ) , coverslips were transferred to chilled serum-free cell culture medium with and without TAM ligand or EDTA , respectively , at 4°C for 90 min . Coverslips were briefly washed with ice cold PBS , fixed for 15 min in 4% PFA/PBS , and incubated with primary antibody , fluorophore-conjugated secondary antibody , and counterstained with Hoechst solution . Following washing with 0 . 1% Tween-20/PBS , coverslips were mounted using Fluoromount G ( Electron Microscopy Sciences ) and visualized with a Zeiss LSM 710 Laser Scanning Confocal Microscope . MEFs were grown in Dulbecco's Modified Eagle's Medium supplemented with 10% FBS , and Pen/Strep , and selection antibiotic . Prior to stimulation with ligand , cells were pre-starved overnight in DMEM supplemented with 3% FBS and Pen/Strep and subsequently incubated in starvation conditions for 3 hr . For BMDM cultures , bone marrow was isolated from tibiae and femurs of 6- to 12-week old C57Bl/6 mice according to the guidelines of the International Animal Care and Use Committee ( IACUC ) . Bones were flushed with DMEM supplemented with 10% FBS and Pen/Strep , spun down at 350 × g for 6 min , incubated with ACK lysis buffer for 1 min , and hematopoietic progenitors incubated with DMEM supplemented with 10% FBS , 30% L929 conditioned medium , Pen/Strep , and glutamax . Cells were supplemented with fresh medium on day 3 and cells re-plated on day 7 in DMEM supplemented with 10% FBS and Pen/Strep for experiments . For activation assays , cells were placed under starvation conditions supplemented with 100 nM dexamethasone and Pen/Strep overnight . Following stimulation with TAM ligands at the indicated concentrations and time periods , cells were washed in PBS and lysed in lysis buffer [50 mM HEPES , pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 10% glycerol , 1% Triton X-100 , 25 mM NaF , 10 µM ZnCl2 , and protease and phosphatase inhibitors ( Roche , Sigma ) ] for 30 min on ice . For immunoprecipitation , cell lysates were incubated with anti-HA high affinity ( Roche ) , anti-Mer ( R&D System ) , or anti-Tyro3 ( Santa Cruz ) antibodies , and either protein A/protein G-Sepharose conjugated beads ( Life Technologies ) . Immunoprecipitates were washed 4× with HNTG buffer ( 20 mM HEPES , pH 7 . 5 , 150 mM NaCl , 0 . 1% Triton X , 10% glycerol , 1 mM Na3VO4 ) , separated by SDS-PAGE , transferred to PVDF membrane ( Millipore ) , incubated with primary antibodies in blocking buffer ( 1% casein block in PBS ) ( Biorad ) , and western blots were developed using an Odyssey Gel Imaging System ( Licor ) . Mice were collected at 12–14 weeks and were anesthetized with 2 . 5% avertin in saline . Mice were perfused with 20 U/ml heparin/PBS and subsequently with 4% PFA in PBS . Eyes were processed as previously described ( Burstyn-Cohen et al . , 2012 ) . Testes were collected , immersion fixed overnight at 4°C , infiltrated with 30% sucrose/PBS overnight at 4°C , and flash frozen in TBS tissue freezing medium . 10 µm serial sections were cut for light microscopy studies and immunohistochemistry , respectively . Sections were air dried overnight at room temperature and subsequently frozen at −80°C . For cCasp3 staining , sections were washed in PBS , incubated in 0 . 3% H2O2 in PBS to block endogenous peroxidase activity , blocked in 3% BSA in PBS , and incubated in anti-cCasp3 ( Cell Signaling ) diluted in blocking buffer . DAB staining was performed with an ABC Vectastain Kit ( Vector Labs ) and Peroxidase Substrate Staining Kit ( Vector Labs ) . Following washing steps in PBS + 0 . 1% Tween-20 , slides were counterstained with hematoxylin for 1 min and mounted using VectaMount ( Vector Labs ) . For quantitation , cCasp3-positive cells and tubule numbers from four testis cross-sections per animal were counted and averaged per animal . Error bars represent standard error of the mean from three independent animals . Cells were plated in 6-cm dishes and when cells reached confluency , they were pre-starved in 3% FBS/DMEM overnight and then incubated in starvation medium for 3 hr before experiment . Supernatant was collected for analysis , and adherent cells were washed with PBS , incubated with 0 . 25% trypsin briefly , and collected in FACS buffer ( 0 . 5% BSA/0 . 1% Na azide/PBS supplemented with either 2 mM CaCl2 or 10 mM EDTA ) . Supernatant and cell samples were incubated with pSiva-IANBD and Propidium iodide ( PI ) ( Imgenex ) for 30 min on ice . Samples were sorted on a Becton-Dickinson LSR II ( Salk Institute Flow Cytometry Core Facility ) and data analyzed using FloJo software ( Treestar ) . Male and female 13- to 15-week old mice were injected intravenously ( retro-orbital injection , 100 μl final volume ) with 30 μg full-length recombinant mouse Gas6 ( in saline ) , Gla-less recombinant Gas6 , or saline as a control . Mice were sacrificed at 30 min post-injection and the spleens split in half: one half was snap frozen for biochemical analysis and the second half was fresh frozen in OCT freezing medium , as indicated in Figure 5 .	Cells send out and receive signals to communicate with other cells . Detecting these signals is largely carried out by proteins called receptors that span the cell surface membrane . These proteins typically have extracellular domains outside of the cell that can bind to specific signaling molecules and an intracellular domain inside the cell that relays the information inwards to trigger a response . Three such receptor proteins are collectively known as the TAM receptors . Each day , many billions of cells in the human body die and are engulfed by other cells and broken down so that their building blocks can be reused . TAM receptors are required for this process; and the TAM receptors also help prevent the immune system from going out of control , which would damage the body's own tissues . Two different signaling proteins , called Gas6 and Protein S , can bind to and activate TAM receptors . Both of the signaling proteins can also bind to a phospholipid molecule that is found on the surface membrane of dead cells . However , it is not known if all three TAM receptors bind to both signaling proteins equally , and the importance of the phospholipid-binding domain in the signaling proteins remains unclear . To shed light on the workings of these receptors , Lew et al . created mouse cells that each only express one out of the three TAM receptors . These cells were then exposed to intact Gas6 and Protein S , or shortened versions that lacked the phospholipid-binding domain . Lew et al . found that Gas6 could trigger a response through all three TAM receptors but that Protein S was specific for only two out of the three receptors . Signaling proteins with or without their phospholipid-binding domains bound equally well to the receptors , but the maximum level of response was only triggered when both signaling proteins were intact and the phospholipid molecule was present . This is important since the phospholipid can be thought of as an ‘eat-me’ signal by which the dead cells are recognized by the TAM receptor-expressing cells that will engulf them . Using mice that only produce a TAM receptor called Mer , Lew et al . show that Protein S alone can trigger the process that engulfs and breaks down cells in a living organism . These data and previous work suggest that two TAM receptors—including Mer—are involved in the daily engulfment of dying cells , whereas the third mediates this process during infection and tissue damage . Molecules that inhibit or activate the function of TAM receptors are currently being developed to treat cancer and other diseases . By revealing which receptors respond to which signaling molecules , the findings of Lew et al . will serve to guide these efforts .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "immunology", "and", "inflammation" ]	2014	Differential TAM receptor–ligand–phospholipid interactions delimit differential TAM bioactivities
The loss of previously adaptive traits is typically linked to relaxation in selection , yet the molecular steps leading to such repeated losses are rarely known . Molecular studies of loss have tended to focus on gene sequences alone , but overlooking other aspects of protein expression might underestimate phenotypic diversity . Insights based almost solely on opsin gene evolution , for instance , have made mammalian color vision a textbook example of phenotypic loss . We address this gap by investigating retention and loss of opsin genes , transcripts , and proteins across ecologically diverse noctilionoid bats . We find multiple , independent losses of short-wave-sensitive opsins . Mismatches between putatively functional DNA sequences , mRNA transcripts , and proteins implicate transcriptional and post-transcriptional processes in the ongoing loss of S-opsins in some noctilionoid bats . Our results provide a snapshot of evolution in progress during phenotypic trait loss , and suggest vertebrate visual phenotypes cannot always be predicted from genotypes alone . The reduction and eventual loss of previously adaptive traits can be seen across the Tree of Life , and is typically linked to relaxation in selection . Within vertebrates , examples of losses include flight in birds , armor plates in sticklebacks , and the ability to synthesize vitamin C in bats ( Burga et al . , 2017; Cui et al . , 2011; Le Rouzic et al . , 2011 ) . Strikingly , many instances of trait loss occur in parallel across multiple independent lineages ( e . g . Colosimo et al . , 2004 , Drouin et al . , 2011 , and Harshman et al . , 2008 ) . There have been attempts to relate parallel trait losses to shared ecological conditions such as salinity tolerance or switches in diet , but the precise causal links are not always clear ( Marchinko and Schluter , 2007 ) . In contrast , the genetic bases of parallel trait loss are often known , with pseudogenization – whereby a non-essential gene loses some functionality – being a frequently invoked mechanism ( e . g . Cui et al . , 2011 and Protas et al . , 2006 ) . One of the best known examples of parallel phenotypic loss via pseudogenization , which can often be directly related to shifts in ecology , is that of color vision in vertebrates . Opsins encode the photoreceptor proteins of rod cells that are responsible for dim-light and cone cells responsible for color vision . Most mammals possess three visual opsins: rhodopsin ( RHO ) in rods , and opsin one long-wave sensitive ( OPN1LW ) found in L-cones , and opsin one short-wave sensitive ( OPN1SW ) found in S-cones . Reconstructions of the highly complex evolutionary history of mammalian vision suggest that there have been >20 independent losses of cone-opsins , with associated reduction in color sensitivity ( e . g . Bowmaker , 1998 , Emerling et al . , 2015 , Lucas et al . , 2003 , Porter et al . , 2012 and Yokoyama et al . , 2008 ) . This is exemplified in some cetacean and xenarthran lineages , which appear to have lost both of their cone-opsins ( Emerling and Springer , 2015; Meredith et al . , 2013 ) . Evolutionary reconstructions of color vision have nearly all been based solely on opsin gene sequences , with gene expression and protein data limited or missing for most mammalian species , including cetaceans and primates ( but see Kraus et al . , 2014 , Peichl et al . , 2017 , Schweikert et al . , 2016 and Wikler and Rakic , 1990 ) . To date , no large-scale comparative study of color vision in mammals has considered each of the steps in protein production ( e . g . , transcription , translation ) . Thus , the extent to which visual phenotypes are expressed or masked due to the modulation of protein production is currently unknown , raising the possibility of underestimating the true complexity of the evolutionary history of vertebrate color vision . This represents a major gap in our understanding of visual evolution , as mounting evidence from a range of systems reveals that complex post-transcriptional and post-translational routes shape phenotypic variation and complicate genotype-to-phenotype mapping ( Blount et al . , 2012; Csárdi et al . , 2015; Schwanhäusser et al . , 2011 ) . Such incomplete information might also lead to erroneous conclusions surrounding the adaptive significance of particular genotypes . The potential for selection to act on phenotypes at different stages of protein production may be particularly important during rapid functional trait diversification , as is often the case in visual systems . In sticklebacks , for example , the repeated colonization of lakes with different photopic environments has driven shifts in spectral sensitivity via recurrent selective sweeps in short-wave opsin genes , and changes in opsin expression ( Marques et al . , 2017; Rennison et al . , 2016 ) . Similarly , rapid shifts in the visual ecology of cichlid fishes have involved a combination of coding sequence evolution and changes in expression ( O'Quin et al . , 2010; Spady et al . , 2005 ) . However , in contrast to fishes , much less is known about the changes underpinning rapid visual adaptations in mammals and reptiles , for which relevant studies have tended to focus on ancient transitions to nocturnal , aquatic or subterranean niches ( Emerling , 2017; Emerling et al . , 2017; Jacobs et al . , 1993 ) . In this study , we investigate the molecular signatures of the repeated loss of S-opsins , and associated dichromatic and UV-vision capabilities , in bats of the superfamily Noctilionoidea ( ~200 species of New World leaf-nosed bats and allies within the suborder Yangochiroptera ) . These bats underwent ecological diversification approximately 40 million years ago ( Rojas et al . , 2012; Rossoni et al . , 2017 ) , and show marked morphological and sensory adaptations linked to their unparalleled dietary specializations ( Davies et al . , 2013; Dumont et al . , 2012; Hayden et al . , 2014; Monteiro and Nogueira , 2010; Yohe et al . , 2017 ) . Switches in feeding ecology from generalized insectivory to blood- , insect- , vertebrate- , nectar- or fruit-based diets have occurred multiple times among closely related species , making noctilionoid bats an outstanding group in which to examine the genetic basis of visual adaptations . Until recently it was thought that S-opsin , encoded by the OPN1SW gene , was likely functional across the suborder Yangochiroptera ( e . g . Butz et al . , 2015 , Feller et al . , 2009 , Marcos Gorresen et al . , 2015 , Gutierrez et al . , 2018 , Müller et al . , 2009 , Winter et al . , 2003 and Zhao et al . , 2009a ) . However , with increased taxonomic sampling of neotropical bat species this has been shown to not be the case , and multiple independent lineages with diverse ecologies ( e . g . blood feeding , plant-visiting species ) show evidence of OPN1SW pseudogenization ( Kries et al . , 2018; Li et al . , 2018; Wu et al . , 2018 ) . Notably , lineages shown to have lost their S-opsins – and thus by association UV-sensitivity – are from the Noctilionoidea superfamily . In contrast , within the other bat suborder – the Yinpterochiroptera – multiple losses of S-opsin function have previously been documented in lineages of Old World fruit bats as well as horseshoe bats and Old World leaf-nosed bats that have evolved a derived form of laryngeal echolocation ( Zhao et al . , 2009a ) . The loss of S-opsins could have profound impacts on bat visual acuity , as inferences from amino acid sequence analyses and action spectra suggest that bat short-wave opsins are sensitive to UV , and their retention is possibly related to the demands of visual processing in mesopic , or low-light , conditions ( Zhao et al . , 2009a ) , and/or plant visiting ( Butz et al . , 2015; Feller et al . , 2009; Kim et al . , 2008; Müller et al . , 2009; Müller et al . , 2007 ) . However , the limited taxonomic sampling to date has precluded clear conclusions . Similarly , while the taxonomic sampling of the recent studies of neotropical bat vision ( e . g . Gutierrez et al . , 2018 , Kries et al . , 2018 , Li et al . , 2018 , and Wu et al . , 2018 ) is considerably more extensive than previous work , it remains limited , and the functionality of OPN1SW has been based on analyses of DNA sequences and a few transcriptome samples . To determine whether patterns of adaptations and loss in cone opsins ( OPN1SW and OPN1LW ) in noctilionoid bats are associated with ecological factors such as diet shifts , we applied analyses of sequence evolution , gene expression , and immunohistochemistry across the taxonomic and ecological breadth of the clade and outgroup taxa . For the first time in mammals , our findings reveal that extensive losses of S-opsin gene function can result from disruption at all three levels of protein synthesis ( i . e . DNA open-reading frame , mRNA , and protein ) . Furthermore , we identify three putative molecular routes that may lead to disruptions of protein synthesis leading to the loss of S-opsins in key lineages . In each instance , the specific route to loss of function was seen in multiple independent lineages . Thus , across the noctilionoids we find evidence that parallel losses leading to identical phenotypes have arisen by both the same and different failures of translation . Hence , current studies both underestimate the extent of parallel losses , and might lead to an incomplete picture based on genes alone . To assess the presence or absence of OPN1SW and OPN1LW proteins we applied immunohistochemistry ( IHC ) to whole , flat-mounted retinas of adult bats ( neyes = 218 , nindividuals = 187 , nspecies = 56 ) . While the presence of a given protein does not guarantee its functionality , here we interpret the detection of protein as indicating a functional cone in the absence of contradictory evidence . Since the absence of protein is difficult to assess , we applied quality control ( see Materials and methods for the criteria to accept or reject a retina based on its condition or number of replicates ) . Surprisingly , OPN1SW was only detected in just over half of the species assayed ( n = 32 ) , including all members of the primarily frugivorous subfamily Stenodermatinae , which invariably retain their S-cones ( Figure 1 and Figure 1—figure supplements 1 and 2 ) . In contrast , OPN1SW was found to be absent in approximately one third of species assayed ( n = 18 ) , including representative species from five bat families . Thus , we not only find evidence of widespread loss of S-opsins within the noctilionoids but also find the first evidence of S-opsin loss in non-noctilionoid Yangochiroptera ( Chilonatalus micropus , Eptesicus fuscus and Molossus molossus ) . To test whether a lack of signal corresponds to a loss of opsin , we aligned the opsin gene sequences among species and confirmed that the epitope-binding site was relatively conserved and showed no correspondence with loss ( Figure 1—figure supplement 3 , Figure 1—figure supplement 4 ) . For immunohistochemistry , five bat species had replicates that were both wild-caught and from museum collections and exhibited the same phenotype , highlighting the robustness of the experiments ( Figure 1—figure supplement 5 ) . We also verified absences using multiple replicates of slides from the same individuals , and , when available , from multiple individuals from the same species ( with a minimum of two individuals; see Supplementary file 1 ) . In one species , P . quadridens , we detected evidence of polymorphism in the presence of S-opsin cones among fresh specimens , with three of the 17 specimens lacking S-cones . Samples derived from museum specimens of six species had low signal-to-background ratios in OPN1SW protein labeling ( e . g . opsin-specific staining was seemingly detected , but non-specific background staining was high , making specific staining difficult to distinguish from background ) , generating inconclusive results , and these specimens were therefore excluded from our models . In contrast to our OPN1SW results , we detected OPN1LW protein in all species examined ( Figure 1 and Figure 1—figure supplements 1 and 2 ) . We then analyzed patterns of OPN1SW and OPN1LW protein localization among cells . Consistent with cone-specific roles , we found that almost all cones expressed either OPN1SW or OPN1LW protein , with no strong evidence of co-localization of both proteins ( Figure 2 and Figure 1—figure supplements 1 and 2 ) . To investigate the underlying molecular causes of the above-detected losses of OPN1SW , we began by sequencing total mRNA isolated from the eye tissue of 39 species , assembled the short-read data and used a BLAST approach to annotate visual pigments . We found evidence of at least partial OPN1SW mRNA transcripts in a total of 34 bat species; thus , expression was only absent in five of the species assayed . Absences of OPN1SW transcripts were phylogenetically widespread , and included divergent species from three families ( Natalidae: Chilonatalus micropus; Mormoopidae: Mormoops blainvillei and Phyllostomidae: Macrotus waterhousii , Brachyphylla ( nana ) pumila and Lionycteris spurrelli ) ( see Figure 1 and Supplementary file 1 ) . In addition to losses observed in C . micropus and M . blainvillei , the three phyllostomid species each belong to separate subfamilies , and therefore , likely represent independent losses of OPN1SW expression . In comparison , we were able to recover the complete RHO and OPN1LW transcript from all taxa assessed ( n = 39; Figure 1 ) . While IHC revealed pervasive loss of S-cones across our study sample , we only detected loss of the OPN1SW transcript in a few species . Comparison of our OPN1SW transcript and protein data revealed numerous conflicts in species-specific absences , with a total of nine lineages found to possess OPN1SW transcripts but lack OPN1SW protein ( see Figures 1 , 3 and 4 ) . These species included Molossus molossus , Pteronotus parnellii , Desmodus rotundus , Trachops cirrhosus , Tonatia saurophila , Gardnerycteris crenulatum , Monophyllus redmani , Erophylla bombifrons , and Carollia brevicauda . This may also be the case in additional species , for example Tadarida brasiliensis , Eptesicus fuscus , Pteronotus davyi and Diaemus youngi , but we currently lack the complementary data to confirm this . Of the nine species lacking S-cones , but in which the presence of mRNA transcripts was detected , we further examined the nucleotide sequence of the assembled transcripts for both an intact open-reading frame ( ORF ) and completeness of transcript . Only a single partial mRNA fragment was recovered each for D . rotundus and M . redmani . The partial D . rotundus fragment ( ~300 base-pairs of exons 2–4 ) contained a premature stop codon – confirmed by PCR and the recently published common vampire bat genome . The partial M . redmani fragment ( ~240 base-pairs of exons 2–3 ) did not contain premature stop codons or indels; however , we cannot rule out the possibility that these may be present in the remaining exons not sequenced . We recovered a total of three OPN1SW transcripts for E . bombifrons from our transcriptome assembly , and the longest of these transcripts contained a putative four base-pair deletion and retained a portion of the intron between exons 2 and 3 . Therefore , D . rotundus and E . bombifrons appear to have transcribed OPN1SW pseudogenes . For the remaining six species ( M . molossus , P . parnellii , T . cirrhosus , T . saurophila , G . crenulatum and C . brevicauda ) for which the OPN1SW transcript was present but the S-cone protein was absent several alternative scenarios emerged . First , for two species , C . brevicauda and T . saurophila , our RNA-Seq assemblies recovered a single and complete OPN1SW mRNA transcript containing all five exons and the 3’ UTR , albeit with ~5 codons missing at the 5’ end in one of these taxa ( C . brevicauda ) . The individuals sequenced for M . molossus revealed five OPN1SW transcript isoforms , two for T . cirrhosus and four for G . crenulatum . While the complete transcript ( i . e . exons 1–5 , and no intronic sequences ) was detected in each of these species , we also found evidence of alternative splice variants characterized by either missing exons ( M . molossus ) , or retained introns . As a result , the reason for the apparent failure of the S-opsin translation is unclear . Finally , for Pteronotus parnellii ( n = 4 ) , we detected the most splice variation , with 2 , 4 , 5 , or 18 variants assembled per individual . Furthermore , in P . parnellii , we were unable to recover any intact mRNA isoforms among these variants . Many of the variants were repeated across individuals; for example , an entirely missing exon five ( despite the presence of the up-/down-stream sequences ) , and partial deletion of exon 1 , was seen in two individuals ( see Figure 4 ) . We thus speculate that these splice variants explain the observed failure of translation that leads to the absence of OPN1SW protein in P . parnellii . We also detected an in-frame three base-pair deletion ( Y190del ) in three of four P . parnellii individuals sequenced . A similar pattern of isoform variants , as detected in P . parnellii , was also seen in P . quadridens ( n = 1 ) . Therefore , given the polymorphic status of S-cones in P . quadridens , we speculate that the individual sequenced may not have had functional S-cones . The retention of intronic sequence , while unexpected , does not necessarily indicate a non-functional gene by itself . For example , we detected limited evidence of some intron retention in species for which the protein data suggest S-opsin presence , for example , Artibeus jamaicensis and Phyllops falcatus . Furthermore , across the 39 species , we found some evidence of OPN1LW transcript variation in 15 individuals ( 14 species ) , with instances of retained introns , missing exons and , in one case ( Artibeus planirostris ) an indel , although this latter case may arise from assembly error . For RHO , we also found some evidence of retained introns for three species ( Anoura geoffroyi , T . brasiliensis and T . cirrhosus ) . However , summing across these 16 species , and except for T . cirrhosus , we always recovered an isoform with all exons and no intronic sequences for both RHO and OPN1LW . Among the four species lacking the OPN1SW transcript and protein , we found the ORF recovered by manual PCR of exons 3–4 was intact in C . micropus ( the single sampled Natalidae species ) , as well as the unrelated phyllostomid B . ( nana ) pumila . In contrast , the genomic sequences recovered by blastn was disrupted by a number of insertions and deletions resulting in premature stop codons in Mormoops blainvillei , Macrotus waterhousii , and Lionycteris spurrelli . In the 17 species for which the S-opsin protein was detected and for which we were also able to test the presence of the transcript , we found a 1-to-1 correspondence in all but one case . The exception was Phyllonycteris poeyi , for which data from four individuals showed S-cone presence , yet the transcript of one other individual was inferred to be non-functional based on a four base-pair insertion ( confirmed via PCR ) . We also note that within the OPN1SW sequences of most species examined , the ATG start codon was found to be three codons downstream relative to that of the human orthologue ( Figure 1—figure supplement 2 ) . In contrast to the mismatches between OPN1SW mRNA and OPN1SW protein , we found complete correlation between the presence of the transcript and protein for OPN1LW in these species . To gain further insights into the molecular evolution of opsins , we performed tests of divergent selection in alignments for each of the three opsin genes ( OPN1SW , OPN1LW and RHO ) among three types of lineages: ( 1 ) those with S-cones; ( 2 ) those without S-cones but with OPN1SW transcripts and an intact OPN1SW ORF; and ( 3 ) those without either S-cones or OPN1SW transcripts ( see Figure 5—figure supplement 1 ) . We found a significantly higher ω in lineages with a pseudogenized OPN1SW in both the OPN1SW gene ( ωbackground = 0 . 13; ωOPN1SW . intact=0 . 24; ωOPN1SW . pseudo=0 . 78; χ2 ( 2 ) =70 . 99 , p=3 . 84e-16 ) , and the OPN1LW gene ( ωbackground = 0 . 08; ωOPN1SW1 . intact=0 . 08; ωOPN1SW . pseudo=0 . 19; χ2 ( 2 ) =9 . 18 , p=0 . 01 ) . In contrast , we found no differences in ω across the different lineages for RHO , indicating strong and negative selection in this gene ( Table S2 in Supplementary file 2 ) . To test the influence of diet on rates of molecular evolution , we compared ω for all three opsin genes between frugivorous and non-frugivorous lineages . We found no differences in rates for OPN1SW or OPN1LW , however background branches ( non-frugivorous ) had a significant and slightly higher ω for RHO ( ωbackground = 0 . 04; ωfugivory= 0 . 01; χ2 ( 1 ) =13 . 77 , p=2 . 07e-4; Table S3 in Supplementary file 2 ) . We compared the locations and densities of long-wavelength-sensitive cones ( or L-cones expressing OPN1LW protein ) and short-wavelength-sensitive ( or S-cones expressing OPN1SW protein ) in the whole , flat-mounted retinas of adult bats for 14 species for which we had sufficient specimen replicates ( Table S4 in Supplementary file 2 ) . We found photoreceptor density varied among examined species ( Figure 2 , Table S4 in Supplementary file 2 ) , with mean cone densities ranging from 2500 to 7500 cones/mm2 for L-cones , and 327 to 5747 cones/mm2 for S-cones ( Figure 2 , Table S4 in Supplementary file 2 ) . In every case , the density of S-cones was lower than that of L-cones . Densities of both cone types tended to be highest near the center of the retina in all species . Bat species with S-cones had significantly higher densities of L-cones , with the presence of S-cones increasing the ln-transformed density of L-cones by 43% , explaining on average ~24% of the variance in density between species ( Table 1 ) . We tested the influence of ecology on the presence of the OPN1SW ORF , mRNA , or S-cones using Bayesian hierarchical models in which the observations corresponded to species ( OPN1SW ORF nspecies = nobservations = 45 , mRNA nspecies = 39 , nspecies = 50 ) , and a phylogenetic structure of the data was included as a species-specific effect . Two types of predictor variables were analyzed: three variables for diet and one for roosting . Comparisons of the coefficients , which are all at the same scale because they are multipliers of the presence of a particular ecology , showed frugivory was the best factor for explaining the presence of S-cones ( Figure 5 , Tables S5-S7 in Supplementary file 2 ) . The predominance of fruit in the diet increases the odds of having S-cones roughly 39 times , and explains about 50% of the between-species variance in the presence of the S-cone ( Table S7 in Supplementary file 2 , Figure 5 ) . Across our study taxa , we documented molecular signatures consistent with as many as 17 instances of parallel loss in S-opsins , and equated these with a minimum of three putative routes leading to the failure of the formation of the S-opsin cones ( Figures 1 , 3 and 4 ) . First , we found evidence of multiple independent instances of pseudogenization , associated with either absent or fragmentary mRNA transcripts . Second , we found evidence of an apparently intact gene sequence that did not result in a mRNA transcript . Finally , we recovered putatively intact OPN1SW transcripts that did not result in the corresponding S-cone protein , which in some species appears to arise from aberrant isoforms . The first route of parallel loss in OPN1SW in noctilionoid bats involved disruption in opsin reading frames , a finding that has been documented in other mammals ( e . g . Emerling et al . , 2015 , Emerling et al . , 2017 , Hunt and Peichl , 2014 , Kraus et al . , 2014 and Zhao et al . , 2009a ) , including bats from both suborders ( e . g . Emerling et al . , 2015 , Kim et al . , 2008 , Kries et al . , 2018 , Müller et al . , 2007 , Wu et al . , 2018 and Zhao et al . , 2009a ) . In general , pseudogenization is thought to occur relatively frequently within mammalian genomes , and previous estimates suggest several thousand pseudogenes may be present per genome ( e . g . Torrents et al . , 2003 ) . The recent proliferation in published genomes has also led to increased efficiency in the detection of this form of gene function loss , and typically this is one of the most frequently cited mechanisms of gene loss ( e . g . Emerling et al . , 2018 and Jebb and Hiller , 2018 ) . The second inferred form of parallel OPN1SW loss in which a putatively intact open reading frame exists but appears not to be transcribed was found in two highly divergent bat species . To our knowledge such mismatches between opsin coding DNA and mRNA have not previously been documented in mammals , including bats , although this largely reflects a shortage of suitable datasets . Indeed , obtaining material with intact mRNA is challenging , and few studies have been able to test the commonly held assumption that an intact ORF equates to functionality . Additional research is thus necessary to confirm the existence , extent and underlying mechanism of ORF-transcript mismatched in bats and other groups . Possible explanations for our observed mismatches include regulatory elements and epigenetic modifications , but a lack of genomic resources for these species precludes more detailed investigations at the present time . The most widely detected form of parallel loss of S-opsins , seen in six species , was associated with the apparent failure of the expressed OPN1SW transcript to be translated into protein . Of these species , protein data for three were obtained from field specimens ( <4 years old ) , two from museum samples , and one from both field ( <3 years old ) and museum samples . Our inspection of the transcript repertoires of these affected species suggests that mismatches might arise from multiple molecular routes . Across the sampled bats , we found evidence of the expression of multiple mRNA isoforms , that in many cases contained either retained introns or skipped exons , both of which are likely to impede translation . Four individuals of P . parnellii exemplify this transcript variation , as none of many mRNA transcripts among these individuals were complete . Reports of readthrough of introns , and the skipping of exons , are becoming increasingly common ( e . g . Gaidatzis et al . , 2015 , Wen et al . , 2018 and Wong et al . , 2016 ) , and these have previously been linked to loss of gene function ( e . g . Lopes-Marques et al . , 2018 ) . Indeed it is particularly noteworthy that the OPN1SW mRNA of the blind mole-rat Spalax ehrenbergi has also been found to contain introns ( David-Gray et al . , 2002; Esquiva et al . , 2016 ) . Underlying mechanisms for these cases could potentially include mutations leading to loss of splice sites , the evolution of novel cryptic splice sites or a reduction of spliceosome efficiency ( David-Gray et al . , 2002; O'Neill et al . , 1998 ) . We note , however , that in species other than P . parnellii either a single , complete mRNA transcript was recovered , or at least one of the alternative assemblies represented the complete transcript – therefore , the ultimate molecular cause of the failure of the protein to be synthesized is unclear in these cases . Although the causal mutations or mechanisms underpinning losses of function in OPN1SW are currently not known , the observed absence of expression with putatively intact ORFs in some species , alongside the converse condition in D . rotundus , strongly indicates independent routes . Similar diversity is seen in the Mormoopidae in which we detected a disrupted ORF and no OPN1SW mRNA expression in the Mormoops lineage , but an intact ORF and mRNA expression in the Pteronotus lineages , as well as evidence of S-cones in some Pteronotus species . Given that the Mormoops and Pteronotus lineages diverged ~30 million years ago , and the taxa sampled within the Pteronotus lineages diverged ~16 million years ago ( Pavan and Marroig , 2017 ) , these patterns do not support a disruption in the common ancestor of Mormoopidae as this would have to be followed by different trajectories that led to complete gene loss in one lineage , and partial retained function in the other . The alternative scenario in which each of these cases of loss involved the same mechanism seems highly unlikely given that it would have had to have taken place independently at least four times within the family , with each of the sampled taxa from our study being at a different stage of the gene loss process . In strong contrast to the results from OPN1SW , data from proteins and transcripts revealed complete retention of OPN1LW across our study species . Such extreme differences in the conservation of color vision genes have previously been reported in other vertebrates ( e . g . Zhao et al . , 2009a and Zhao et al . , 2009b ) . Our IHC assays also revealed little evidence of co-localization of both proteins , which is consistent with cone-specific roles . This contrasts with a previous study of two noctilionoid bat species that found that almost all L-cones expressed some S-opsin ( Müller et al . , 2009 ) , although this discrepancy could have arisen from methodological differences . To document fluorescence , the previous study used epifluorescence microscopy , while our study used confocal microscopy . Through its generation of serial optical sections , confocal microscopy typically provides better resolution for co-localization studies . We also found that bat species with S-opsin cones tend to have more L-opsin cones , consistent with both types of cones serving a common functional role . We also found that S-cone retention varied among conspecifics . In Pteronotus quadridens , three of 17 individuals were found to lack S-cones . This heterogeneity could indicate the ongoing degradation of protein synthesis . Indeed , allelic variation has been reported to contribute to opsin variation in diurnal lemurs ( Jacobs et al . , 2017 ) and has previously been detected in OPN1SW in Pteronotus mesoamericanus ( Wu et al . , 2018 ) . Alongside parallel losses of shortwave-sensitive opsins in some noctilionoids lineages , we found strong conservation of S-cones , OPN1SW transcription , and protein-coding sequences in around 20 of the species studied . Thus , S-cones appear to still play an important function in these bats . Although the pseudogenization of OPN1SW , or loss of transcription had both been previously explained by the use of caves as roosts ( Gutierrez et al . , 2018; Wu et al . , 2018 ) , our phylogenetic regressions estimated the coefficient for this factor to include 0 ( Figure 5 ) . Instead , we identified the predominance of fruit consumption as the single most powerful explanatory factor explaining the variation in S-cone presence across the clade , with a similarly positive but not statistically significant effect for OPN1SW transcription and protein-coding sequences . Surprisingly , the result for plant-visiting ( which includes flower-visiting bats ) was not similarly strong and no such result was found for insectivory or cave-roosting . Diet therefore appears to be the primary selective agent for maintaining S-opsin function . We also found that while some predominantly nectarivorous species from both independent nectar-feeding lineages , have lost their S-cones , others have retained them ( e . g . Anoura geoffroyi ) . While the loss of S-opsins in flower visiting bats may seem maladaptive , behavioral assays have previously been used to infer that some nectarivorous phyllostomid species appear to be color blind , and thus may be able to utilize UV reflectance to locate flowers via an alternative rod-based mechanism ( Winter et al . , 2003 ) . This suggests that either more than one strategy to locating flowers has evolved among New World leaf-nosed bats , or other non-visual cues are used ( e . g . Gonzalez-Terrazas et al . , 2016 ) . Since fruit consumption arose as an evolutionary innovation within the Yangochiroptera , selection for this novel niche cannot explain the ancestral or present-day persistence of S-cones in non-frugivorous species . A role in light capture rather than in detecting novel visual cues might explain the density of S- and L-opsin cones in the non-frugivorous lineages sampled , as well as in ancestral bats . The signals of strong purifying selection of all three visual opsin sequences to conserve ancestral function in both frugivorous and non-frugivorous lineages further buttresses interpretation , as it implies there is no detectable relaxation of selection on non-frugivorous lineages , at least among species for which sequences were available . At the same time , divergent selection in frugivorous and non-frugivorous lineages in RHO may further support the importance of light capture , and dim light vision , in relation to novel diets in frugivorous species . As expected once pseudogenization has occurred , the main difference in molecular selection was between species with an OPN1SW ORF and a pseudogene at this locus . Instead of directly reflecting ecological covariates , the process of pseudogenization appears to represent the culmination of a longer term process that starts earlier with cone loss . This highlights post-transcriptional regulation as a more direct response to ecology than pseudogenization of the relevant opsin . Therefore , protein composition should more closely reflect visual ecology than high rates of sequence evolution and pseudogenization in the relevant opsin , as the latter only responds to long-term functional loss . We further tested this inference by modeling the presence of an OPN1SW ORF , mRNA , or protein as a function of ecological covariates , finding the strongest ecological association—estimated by higher coefficients—with the presence of S-cones ( rather than with earlier steps in protein production ) . Recent studies of color vision evolution in New World leaf nosed bats have begun to explore the complex picture of opsin gene loss in the context of selection and ecological factors ( Gutierrez et al . , 2018; Kries et al . , 2018; Li et al . , 2018; Wu et al . , 2018 ) . The detected pseudogenization of OPN1SW in infrared sensing vampire bats and in high-duty cycle ( HDC ) echolocating bats such as Pteronotus mesoamericanus ( formerly P . parnellii mesoamericanus ) have led researchers to invoke evolutionary sensory trade-offs as one factor behind the loss of color vision ( Kries et al . , 2018; Li et al . , 2018; Wu et al . , 2018 ) . An additional loss was detected in Lonchophylla mordax ( Kries et al . , 2018 ) , a nectar bat that roosts in caves , with the roosting preference taken to be driver of color vision loss in this species . In cases in which either no S-opsin losses were inferred in Yangochiroptera , or selection analyses were performed across both Old and New World species simultaneously ( Gutierrez et al . , 2018 ) , it is not possible to interpret the results solely in the context of noctilionoids . Sensory trade-offs , foraging strategy and obligate cave roosting are hypotheses that have previously been applied to loss of S-opsins in Old World bat lineages ( e . g . Zhao et al . , 2009a ) , however these traits often co-vary within species so the signal may be difficult to disentangle . By allowing us to detect previously ‘hidden’ opsin phenotypes across noctilionoid species , our approach has allowed us to identify previously undetected ecological factors , that is , fruit consumption , as an explanatory variable of S-opsin retention . Furthermore , the discovery of loss of gene function in non-HDC Mormoopidae , e . g . Mormoops blainvillei and Pteronotus davyi , also call into question the sensory trade-off hypothesis within this family . The surprising diversity in S-opsin retention recorded in this study was seen across divergent species , congeners , and even conspecifics . Although such patterns could also arise from methodological issues , some of our findings appear consistent with emerging trends . For example , within Pteronotus and the Mormoopidae family as a whole , there is increasing evidence to support an extremely complex evolutionary history of S-opsins ( Gutierrez et al . , 2018; Simões et al . , 2018; Wu et al . , 2018 ) . In comparison , methodological artifacts are less easy to rule out as causes of variation among Carollia spp . given that S-cone presence in two species was inferred from either recently collected field specimens , or a mixture of both field and museum specimens , while S-cone absence in a third species was based on museum specimens collected in 1968 and 1972 . Despite this , the utility of long-term fixed specimens for immunohistochemical staining of proteins ( e . g . vimentin and GFAP ) has been described previously ( Hühns et al . , 2015; Thewissen et al . , 2006 ) , including visual opsins in some museum specimens ( ~20 years old and using the same S-opsin antibody as used in the current study ) ( Nießner et al . , 2016 ) . In line with this , we were able to recover good IHC staining for both S- and L-cones from our oldest sampled museum specimen with a confident date , an Artibeus fraterculus collected in 1921 ( see Figure supplement S3 ) . We note , however , that our S-opsin assays were inconclusive for six species ( T . brasiliensis , P . hastatus , S . tildae , S . ludovici , P . dorsalis and C . villosum ) represented by museum samples due to low signal-to-background ratios , but L-opsin assays were successful in these species . The ability to detect protein in these , and other , museum specimens is a function of the condition of the retina and the density of the cones in question . Because L-cones are present at higher densities than S-cones ( this study and Müller et al . , 2009 ) , we were able to more readily detect L-cones even in more poorly preserved retina . Thus , while museum collections are rarely used in protein studies , relative to their use , for example , in genetic and genomic studies ( e . g . Bailey et al . , 2016; Nießner et al . , 2016 ) , they offer great potential for a range of comparative studies provided that caution is exercised ( e . g . Hedrick et al . , 2018 ) . These benefits apply particularly to groups that cannot be sampled in the wild for ethical , conservation and logistic reasons ( e . g . Russo et al . , 2017 ) . In our study , restrictions on sampling necessitated our comparisons of conspecifics collected from different countries for genetic and protein assays . For this reason , we cannot rule out geography as a source of variation , and it is noteworthy that one focal taxon , Pteronotus parnellii was recently recognized as a species complex with a strong phylogeographic divergence ( Pavan and Marroig , 2017 ) . Although other focal bat species have not been split in this way , many have wide ranges and their genetic diversity may be underestimated ( Clare et al . , 2011 ) . In another instance , our fresh specimen of M . molossus from Belize and two older museum specimens from 1968 from Uruguay all lacked S-cones , whereas a published record of an individual from an unknown geographical locality showed S-cone presence ( Nießner et al . , 2016 ) . These patterns of S-cone presence , and indeed those across the entire noctilionoid tree , suggest that losses may arise across populations of the same species . We must also consider whether our results might arise from methodological artifacts related to the short read data . For example , low gene expression can limit the number of representative reads in RNA-Seq datasets for transcript assembly ( e . g . Zhao et al . , 2011 ) and this caveat likely applies to shortwave-sensitive opsins that show low S-cone densities , inferred loss of function in some taxa , and also low expression levels based preliminary expression analyses ( data not shown ) . To cross-validate our assembled transcripts , whenever possible we compared our sequences to published data from PCR amplicons , RNA-Seq , and/or genome datasets . For example , for Monophyllus redmani , Trachops cirrhosus and Pteronotus parnellii we were able to confirm that our assembled transcripts matched published assembled mRNA contigs , PCR amplicons and genomic sequences , respectively ( Gutierrez et al . , 2018; Tsagkogeorga et al . , 2013; Wu et al . , 2018 ) . For Mormoops blainvillei and Macrotus waterhousii , the absence of OPN1SW mRNA transcripts and S-cones was supported by disrupted ORFs in published genome datasets , as well as the highly divergent and fragmented transcripts recovered by a recent study , which our visual inspections suggest may be due to cross contamination or misassembly ( Gutierrez et al . , 2018 ) . Therefore , several lines of evidence support loss of function of OPN1SW in these taxa . For C . micropus , our inferred intact ORF is based on PCR of ~100 codons ( also seen in B . pumila ) and was supported by sequence data from two closely related species from the Natalidae ( Emerling et al . , 2015; Simões et al . , 2018 ) . For several species , our transcriptomic analysis detected multiple OPN1SW mRNA transcripts variants , characterized by retained introns and missing exons . We are able to confirm the species-specific intronic sequences of several of the species due to recently available genomes and gDNA PCRs ( Kries et al . , 2018; Li et al . , 2018; Tsagkogeorga et al . , 2013; Wu et al . , 2018 ) . The observed retention of introns in OPN1SW mRNA as well as the expression of pseudogenized opsin mRNA are both supported by earlier studies ( David-Gray et al . , 2002; Schweikert et al . , 2016 ) ; however , alternative scenarios for these findings could include gDNA contamination , sequencing of immature mRNA or low-level cross contamination resulting in the assembly of highly divergent transcripts . Finally , for most species our mRNA evidence is based on one individual . However , the lack of a clear relationship between RIN score or sequencing depth and presence of OPN1SW mRNA ( data not shown ) , together with the presence of the two other visual opsins in all samples , suggests this should be sufficient . Our findings provide important insights into how parallel losses occur in response to diverse ecological demands , as well as how several alternative molecular routes may lead to the same phenotype . There are other examples of parallel loss from pelvic reduction in sticklebacks ( via repeated changes in a Pitx1 enhancer ) , color and vision in Astyanax cavefish ( via loss of function of Oca2 ) , trichomes in Drosophila spp . , and floral pigments in Iochrominae ( Chan et al . , 2010; Larter et al . , 2018; McGregor et al . , 2007; Protas et al . , 2006 ) , but few of these have examined as many species across as many steps of phenotype production . Our data and those from other recent studies on bat opsins associate independent losses of S-cones with diverse adaptations ( e . g . shifts in diet , roosting ecology and sensory traits ) , and are therefore consistent with multiple , distinct ecological demands leading to the same phenotype . Hence , our findings are also consistent with the hypothesis that UV vision represents a genetic ‘hot spot’ of evolution ( Hoekstra and Coyne , 2007; Martin and Orgogozo , 2013; Stern and Orgogozo , 2008 ) , along an evolutionary line of least resistance ( Schluter , 1996 ) . Therefore , by documenting a range of molecular routes to functional degradation , this study supports the hypothesis that vision is a highly evolvable trait that repeatedly and rapidly changes in response to diverse selective demands . In conclusion , our findings reveal that assessments of visual perception based purely on genotypic analyses of either opsin sequences or RNA transcripts can be misleading , and may even obscure the evolutionary processes and ecological agents of selection . Although variation in the complement of photoreceptors across vertebrates is usually explained by disruptions to the protein-coding sequence ( e . g . Mundy et al . , 2016 and Zhao et al . , 2009a ) , findings of mismatches between genotype and phenotype also indicate a role for transcriptional and even translational control in this process . It follows that because routes of gene loss are mainly studied at the genetic level or , in fewer cases , at the transcriptomic level , the input of changes in translation and other connections between the genetic , transcriptomic , and proteomic levels may be being underestimated . More broadly , our results highlight the importance of rapid trait loss in evolution , with apparent shifts in translation and transcription that precede pseudogenizing changes in ORFs . As genotype-centered analyses would miss important functional changes , our study also illustrates the importance of probing multiple levels of protein synthesis . We obtained eye tissue from 59 New World bat species , of which 49 were collected from the wild and 34 from the American Museum of Natural History ( AMNH ) , with 24 species common to both sources ( Supplementary file 1 ) . Our sampling was designed to maximize taxonomic coverage and include as many replicates as possible within ethical and regulatory limits . Unlike lab animals such as mice or rats , most bat species have just one offspring per year ( Wilkinson and South , 2002 ) , limiting the rate of recovery from adult mortality . All wild bats were captured with traps set in forests and/or at cave entrances , were handled , and then euthanized by isoflurane overdose , under appropriate research and ethical permits ( see Appendix 1 ) . Intact eyes were placed in RNAlater and incubated at 4°C overnight and then frozen . Total RNA was isolated using Qiagen RNeasy Mini kits with the addition of DTT and homogenization using a Qiagen TissueLyser . Following QC , total RNA from each individual was used to construct a cDNA library using the Illumina TruSeq RNA v2 kit . Pooled libraries were sequenced ( NextSeq 500 ) . Eye transcriptomes were generated for 46 individuals ( 39 species ) including biological replicates of Pteronotus parnellii ( n = 4 ) , Artibeus jamaicensis ( n = 4 ) and Phyllops falcatus ( n = 2 ) . Raw reads were trimmed , and clean reads were assembled with Trinity v . 2 . 2 . 0 ( Grabherr et al . , 2011 ) ( see Appendix 1 ) . We tested for the presence of the three focal gene transcripts ( RHO , OPN1SW and OPN1LW ) in each bat transcriptome using a reciprocal best hit blast approach against the full set ( n = 22 , 285 ) of human protein-coding genes from Ensembl 86 ( Yates et al . , 2016 ) . To confirm the absence of OPN1SW sequence , we performed additional steps in several species . First , we cut , trans-chimeras , which can prevent detection by reciprocal blast ( Yang and Smith , 2013 ) , and repeated the reciprocal blast . Second , we manually screened sequences that were initially identified as matching OPN1SW , but did not pass initial blast filtering ( see Supplementary Information ) . Recovered opsin gene sequences have been submitted to GenBank ( accession numbers MK209460 - MK209505 [RHO]; MK209506 - MK209551 [OPN1LW]; and MK209552 - MK209592 [OPN1SW] ) . Additionally , for each individual RNA dataset we manually aligned all assembled transcripts , that passed the tblastn step of the reciprocal blast for OPN1SW , together with individual exons and introns obtained from the Myotis lucifugus structural annotation downloaded from Ensembl . Finally , we obtained all OPN1SW DNA and mRNA sequences currently available for our study species from GenBank , produced by recently published studies or genomes ( Gutierrez et al . , 2018; Kries et al . , 2018; Li et al . , 2018; Wu et al . , 2018; Zepeda Mendoza et al . , 2018 ) . This data was used to confirm either our mRNA assemblies or the intronic sequences , and also to infer ORF status for species in which we had protein data for but were not able to obtain tissue for RNA-seq ( e . g . Eptesicus fuscus , Pteronotus davyi , Diaemus youngi , Phyllostomus discolor , Sturnira lilium ) The following primary antibodies were used: goat anti-OPN1SW , 1∶1000 ( RRID: AB_2158332 , sc-14363 , Santa-Cruz Biotechnologies , Heidelberg , Germany; detects S-opsin protein ) and rabbit anti-opsin red/green , 1∶750 ( RRID: AB_177456 , ab5405 , Millipore Ibérica , Madrid , Spain; detects L-opsin protein ) . Sc-14363 is an affinity-purified goat polyclonal antibody raised against a 20-amino-acid synthetic peptide mapping within amino acids 1 to 50 of human blue-sensitive opsin , and AB5405 was raised in rabbit against the last 42 amino acids of the C-terminus of recombinant human red/green opsin ( Gaillard et al . , 2009 ) . These antibodies have been used successfully in many groups , including rodents , artiodactyls , bats , and birds ( e . g . Gaillard et al . , 2009 , Müller et al . , 2007 and Nießner et al . , 2016 ) . The following secondary antibodies were used at a 1:500 dilution: donkey anti-goat Alexa Fluor 568 ( RRID: AB_2534104 ) and donkey anti-rabbit Alexa Fluor 647 ( RRID: AB_10891079 ) ( Thermofisher ) . In addition , we created amino acid alignments of the peptide regions thought to correspond to the antibody epitopes across the bat species studied to assess sequence variation . We used a strong quality control protocol to ensure that we could interpret an absence of labelling as a true loss of S-opsin protein . Given the variable age and preservation of museum specimens , we evaluated the anatomical preservation of the retina during dissection and excluded specimens ( data not shown ) if the retina was: ( 1 ) attached to the crystalline lens , poorly preserved , impossible to dissect/damaged , ( 2 ) highly fragile , poorly preserved , disintegrated/damaged upon dissection , and ( 3 ) intact or preserved in large pieces but exhibited shrinkage and/or an orange color characteristic of tissue degradation . When necessary , we also slightly modified the IHC protocol for some museum samples . Specifically , since museum samples where already permeabilized by their storage in ethanol , we reduced the number of PBS-0 . 5%Tx washes and removed the methanol permeabilization step at −70C . With these quality-control measures in place , and given the consistency of the detection of our chosen antibodies across all bats and other mammals ( Müller et al . , 2009; Müller et al . , 2007; Ortín-Martínez et al . , 2014 ) , and the number of replicates and individuals we examined , we are confident in our interpretation that no labeling indicate a true loss of the respective cone type . In addition , we created amino acid alignments of the peptide regions thought to correspond to the antibody epitopes across the bat species studied to gain a measure of the sequence variation at these points . Flat-mounted retinas were photographed using a 20X objective on a confocal microscope ( LSM710; Zeiss Microscopy ) . 564 and 633 lasers were used to excited Alexa 568 and Alexa 647 dyes , labelling S- and L- opsins , respectively . Each entire retina was completely imaged using 512 × 512 pixel tiles . For each retina , each tile was then Z-stacked and automatically counted using a 3D object counter plugin using Fiji ( ImageJ ) . The accuracy of this automatic approach was verified by manually counting three biological replicates of five bat species , by two different people . For each retina quantified , the density was calculated for each tile and then averaged for each individual ( total count was average over three individuals ) and for each species ( by averaging the average of the three individuals ) . The spatial distribution of L- and S-cone density was visualized for the following 14 species: Artibeus jamaicensis , Artibeus phaeotis , Carollia sowelli , Sturnira lilium , Monophyllus redmani , Erophylla sezekorni , Glossophaga soricina , Brachyphylla nana pumila , Desmodus rotundus , Pteronotus quadridens , Mormoops blainvillei , Macrotus waterhousii , Gardnerycteris crenulatum and Phyllops falcatus ( see Figure 2 , Table S2 in Supplementary file 2 ) . We used aligned sequences from the transcriptomes of 38 species together with those from six noctilionoid genomes ( Zepeda Mendoza et al . , 2018 ) to estimate rates of molecular evolution of visual opsin genes ( OPN1SW , OPN1LW , and RHO ) in focal bats . First , we tested for divergent selection modes among species that had S-opsin cones , lacked the S-opsin cones but had an intact mRNA sequence , and those that lacked the S-opsin cones but either did not have OPN1SW transcripts or had a pseudogenized OPN1SW sequence ( Figure 5—figure supplement 1 ) using the Branch Model 2 of codeml in PAML 4 . 8a ( Yang , 2007 ) . Second , we applied the same approach to test divergent selection modes between frugivorous and non-frugivorous bat species ( Figure 5—figure supplement 1; gene alignments have been submitted to DRYAD http://dx . doi . org/10 . 5061/dryad . 456569k ) . To determine whether cone phenotypes are explained by dietary specialization , we applied the hierarchical Bayesian approach implemented in the R packages MCMCglmm and mulTree ( Guillerme and Healy , 2014; Hadfield , 2010 ) , using a sample from the posterior distribution of phylogenies of New World noctilionoids grafted onto the phylogeny of bats ( Rojas et al . , 2016; Shi and Rabosky , 2015 ) . We modeled S-cone presence with species as observations as function of diet represented by four variables ( nspecies = nobservations = 50 ) . Since all predictor variables correspond to the presence or absence of a given diet or roosting habit , the coefficients of the resulting models were used to compare the strength of the association between the OPN1SW genotype or phenotype and the ecological covariate . Since this modeling approach neither tests against a null hypothesis of no effect , nor assumes the point estimates –in this case the mans by ecological group– are stationary , there is no requirement to adjust for multiple comparisons ( Gelman et al . , 2012 ) . Using the predictor variable from the best model for presence/absence of S-opsin cones , or the presence of S-cones as a factor , we then repeated this approach to explain L-cone density across individuals within species ( nspecies = 14 , nobservations = 33 , see Supplementary Information ) . To normalize the response data ( density ) , we transformed by taking the natural logarithm of the cone density estimate . These analyses took advantage of the hierarchical structure of observations of density replicates clustered within species , with estimates of variance between species ( corresponding to the phylogenetic regression ) , and residual variance remaining between observations . Given this data design , the estimate of the mean density per-group ( i . e . frugivory/non , or presence/absence of S-cones ) accounts for both between and within species variance . The R code for all regression models is available from DRYAD http://dx . doi . org/10 . 5061/dryad . 456569k	Bats are famous for using their hearing to explore their environments , yet fewer people are aware that these flying mammals have both good night and daylight vision . Some bats can even see in color thanks to two light-sensitive proteins at the back of their eyes: S-opsin which detects blue and ultraviolet light and L-opsin which detects green and red light . Many species of bat , however , are missing one of these proteins and cannot distinguish any colors; in other words , they are completely color-blind . Some bat species found in Central and South America have independently lost their ability to see blue-ultraviolet light and have thus also lost their color vision . These bats have diverse diets – ranging from insects to fruits and even blood – and being able to distinguish color may offer an advantage in many of their activities , including hunting or foraging . The vision genes in these bats , therefore , give scientists an opportunity to explore how a seemingly important trait can be lost at the molecular level . Sadier , Davies et al . now report that S-opsin has been lost more than a dozen times during the evolutionary history of these Central and South American bats . The analysis used samples from 55 species , including animals caught from the wild and specimens from museums . As with other proteins , the instructions encoded in the gene sequence for S opsin need to be copied into a molecule of RNA before they can be translated into protein . As expected , S-opsin was lost several times because of changes in the gene sequence that disrupted the formation of the protein . However , at several points in these bats’ evolutionary history , additional changes have taken place that affected the production of the RNA or the protein , without an obvious change to the gene itself . This finding suggests that other studies that rely purely on DNA to understand evolution may underestimate how often traits may be lost . By capturing ‘evolution in action’ , these results also provide a more complete picture of the molecular targets of evolution in a diverse set of bats .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2018	Multifactorial processes underlie parallel opsin loss in neotropical bats
Cells can , in principle , control their size by growing to a specified size before commencing cell division . How any cell actually senses its own size remains poorly understood . The fission yeast Schizosaccharomyces pombe are rod-shaped cells that grow to ∼14 µm in length before entering mitosis . In this study , we provide evidence that these cells sense their surface area as part of this size control mechanism . We show that cells enter mitosis at a certain surface area , as opposed to a certain volume or length . A peripheral membrane protein kinase cdr2p has properties of a dose-dependent ‘sizer’ that controls mitotic entry . As cells grow , the local cdr2p concentration in nodes at the medial cortex accumulates as a measure of cell surface area . Our findings , which challenge a previously proposed pom1p gradient model , lead to a new model in which cells sense their size by using cdr2p to probe the surface area over the whole cell and relay this information to the medial cortex . The fundamental process by which a cell controls its own size is not understood for any cell type . In actively dividing cells , growth , and size need to be coordinated for cells to maintain their size . In several cell types , cells have been shown to have a size threshold , in which they need to grow to a minimal cell size before committing to cell division ( Turner et al . , 2012 ) . This mechanism however requires that cells somehow monitor their own size . The molecular mechanism for how size is sensed , and what aspect of size—surface area , volume , mass , linear dimensions etc—is monitored remains unknown . The fission yeast Schizosaccharomyces pombe is an attractive eukaryotic model for cell size studies because of its highly regular dimensions , simple rod-shape , and growth patterns . During interphase , these cells grow from the cell tips at a nearly constant rate to approximately 14 µm in length before entering mitosis , when cell growth ceases until the next cell cycle ( Mitchison and Nurse , 1985 ) . Genetic analyses in fission yeast have identified a pathway of conserved protein kinases for cell size control: the DYRK kinase pom1p is an inhibitor of the SAD family kinase cdr2p , which inhibits wee1p , which in turn inhibits the cell division kinase cdk1p ( Russell and Nurse , 1987; Breeding et al . , 1998; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Loss of function of pom1 and wee1 leads to abnormally short cells , whereas loss of function of cdr2 leads to abnormally long ones . Interestingly , these factors largely localize to different sites in the cell . Pom1p localizes in cortical gradients emanating from cell tips ( Bahler and Pringle , 1998; Padte et al . , 2006; Hachet et al . , 2011; Saunders et al . , 2012 ) . Cdr2p localizes to a medial band of plasma membrane protein complexes termed ‘nodes’ , which overlie the medial nucleus ( Morrell et al . , 2004; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Wee1p , Cdk1p , and other regulators of mitotic entry localize primarily to the spindle pole body and nucleus ( Alfa et al . , 1990; Masuda et al . , 2011; Grallert et al . , 2013 ) . How this pathway may be used to sense cell size remains unclear . A current model for cell size control is based on pom1p concentration gradients as ‘rulers’ to sense cell length ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009; Vilela et al . , 2010; Tostevin , 2011 ) . This gradient model postulates that as these cells grow in length from their tips , pom1p gradients are moved away from cdr2p nodes at mid-cell , causing decreased pom1p levels at the medial cortex . This putative decrease would then allow for activation of cdr2p , leading to cdk1p activation and entry into mitosis when cells reach a critical length . Here , we use quantitative imaging and modeling to examine the relationships of pom1p and cdr2p with cell size . We find that core assumptions of the previous pom1p gradient model are not consistent with experimental findings . We further develop a novel model in which cells monitor their size using cdr2p itself as a cortical sizer molecule to probe the surface area of the cell . To test the pom1p gradient model , we quantitatively analyzed pom1p in living cells expressing a functional pom1-tomato-dimer fusion protein at near-endogenous levels ( Padte et al . , 2006; Hachet et al . , 2011; Saunders et al . , 2012 ) . Pom1p cortical gradients exhibit large cell-to-cell variability in intensity and distribution , fluctuate over time in individual cells , and show little consistent change with cell length ( Saunders et al . , 2012 ) . This variability , plus a short decay length relative to cell length , led us to question whether these gradients can function reliably as ‘rulers’ . One of the key predictions of the gradient-based model is that pom1p levels decrease on the medial cortex as cells grow . We measured pom1p concentration in a 3-µm region along the medial cortex , where cdr2p nodes are located . Using time-averaged data ( reducing fluctuations in the gradient over time [Saunders et al . , 2012] ) , we detected low but measurable intensities ( Figure 1A , Figure 1—figure supplement 1 ) . Importantly , measurements of pom1-tomato at the medial cortex showed no detectable decrease with cell length in a population of cells ( Figure 1B ) , or in individual cells imaged over time ( Figure 1C , D ) . These cortical measurements improve on previously reported pom1p measurements that integrate intensities over the whole cell ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) , which have artifacts stemming from the normal exclusion of pom1p from the nucleus ( Saunders et al . , 2012; Figure 1A , Figure 1—figure supplement 2 , 3 ) . 10 . 7554/eLife . 02040 . 003Figure 1 . Gradient distribution of pom1p is not the basis for cell size control . ( A ) Time-averaged spinning disc confocal images of fission yeast cells expressing pom1-tomato in a medial focal plane ( 60 frames over 3 min ) . Scale bar = 3 μm . Strain used: FC2054 . ( B ) Total fluorescence intensities of pom1-tomato in a medial 3-μm segment along cortical edge of interphase cells , from images like A ( n > 100 ) . See Figure 1—figure supplements 1–3 . ( C ) Time-lapse images of pom1-tomato in individual cell . Images are time averaged ( 5 frames over 25 s ) in medial focal plane . Scale bar = 3 μm . ( D ) Pom1-tomato intensities at medial cortex ( as in B ) of individual growing interphase cells . ( E ) Cells expressing pom1-GFP or the pom1-3GFP . Imaging as in A . Strains used: FC1162 , FC2685 . Scale bar = 3 μm . ( F ) Gradient profiles of pom1-3GFP , pom1-GFP and pom1-tomato ( n > 30 each strain ) . Peak absolute protein numbers in pom1-3GFP and pom1-GFP gradients were similar . Error bars not shown for clarity . See Figure 1—figure supplement 4 . ( G ) Effect of pom1p fusions on cell size as measured by length of septated cells ( n > 100 ) . Error bars: SDs . Strains used: FC420 , FC1162 , FC2685 , FC2054 . See Figure 1—figure supplement 4C , 5 . ( H ) Distribution of cell lengths at division in indicated strains . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00310 . 7554/eLife . 02040 . 004Figure 1—figure supplement 1 . Pom1p concentration at the medial cortex does not vary with cell length . ( A ) Distributions of pom1p and cdr2p intensity around the cell cortex , measured as shown in the schematic . Cells co-expressing pom1-tomato and cdr2-GFP were imaged for 3 s in a confocal section through the middle of each cell , with a 20 s interval between time points ( 30 measurements in total ) and the subsequent intensities time averaged . Average pom1p ( red ) and cdr2p ( green ) profiles for varying cell lengths are shown , normalized by the maximum value of pom1p and cdr2p time-averaged intensity recorded for an individual cell respectively . Top: n = 78 cells , middle n = 88 cells , bottom n = 32 cells . Note that cdr2p nodes reside in the low intensity region of the pom1p gradient at all cell lengths . Strain used: FC2678 . ( B ) Profile of pom1p intensity gradients at different cell lengths , based on data shown in ( A ) . The measurement region is shown in the schematic . Pom1p intensity profiles are normalized by the maximum average value in the entire set . n > 15 for each profile . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00410 . 7554/eLife . 02040 . 005Figure 1—figure supplement 2 . Pom1p concentration at the medial cortex: Comparison with previous data . Moseley et al . ( 2009 ) and Martin and Berthelot-Grosjean ( 2009 ) showed , in contrast to what we see , that pom1p levels in the middle of the cell decrease with cell length . In the measurements of these papers , pom1p fluorescence in the whole cell was collapsed onto a single line . This method of image analysis differs from our approach of measuring pom1p intensity only on the cortex , where the gradient distribution is shown in Figure 1 and Figure 1—figure supplement 1 . We plotted measurements of pom1p in the middle of the cell as a function of cell length using different methods and data sets . The data from Moseley et al . ( Figure S12 of that publication ) is presented here ( after normalization to the value for cells with length <8 . 5 μm ) as the black bars . We found a similar trend using the same whole cell analysis on our own images of pom1-tomato cells ( blue bars ) , n > 20 in each binning of cell lengths . pom1p is detectable at a low constant level all through the cytoplasm , but is not detectable in the nucleus ( Saunders et al . , 2012; Figure 1A ) . Therefore , we tested whether this difference between whole cell and cortical measurements may be due to lack of pom1p in the nucleus . We adjusted for the effect of the nucleus by filling the nucleus with the average cytoplasmic pom1p intensity in silico . This adjustment largely abrogated the length-dependent decrease of pom1p intensity ( red bars ) . We speculate that the decrease seen in the whole cell measurements may be due to the slightly larger size of the nucleus in larger cells ( Neumann and Nurse , 2007; Figure 1—figure supplement 3 ) . Thus , the decrease seen in the previous publications may ( partially ) be an artifact of including cytoplasmic and nuclear fluorescence in addition to the cortical distribution . Strain used: FC2054 . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00510 . 7554/eLife . 02040 . 006Figure 1—figure supplement 3 . Nuclear width as a function of cell length . Nuclear size was measured by the dark nuclear zone of pom1-tomato fluorescence . The width of the nucleus was determined as the maximum distance along the long axis of the cell . n = 96 cells . Error bars = SD . Strain used: FC2054 . Note that this method may provide a slight underestimate compared to measurements for instance of a nuclear envelope marker or a diffuse nuclear marker . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00610 . 7554/eLife . 02040 . 007Figure 1—figure supplement 4 . Pom1p gradients with different decay lengths do not affect cdr2p distribution . ( A ) Comparison of pom1-3GFP ( green , n = 45 ) and pom1-GFP ( red , n = 31 ) gradients . Dashed red line corresponds to the unscaled pom1-GFP absolute fluorescence values in the cell . The pom1-GFP intensity is adjusted to account for the relative intensity difference ( whole cell pom1-3GFP was 2 . 5 times more intense than pom1-GFP under the same imaging conditions ) . These data show that these gradients have similar numbers of pom1p molecules at their peaks . Strains used: FC2685 , FC1162 . Error bars not shown for clarity . ( B ) Fitted decay length of average pom1p intensity profile for three different fluorescent proteins ( pom1-3GFP ( n = 45 cells ) , pom1-GFP ( n = 31 cells ) and pom1-tomato ( n = 32 cells ) ) . Strains used: FC2685 , FC1162 and FC2054 . Note that these cells express the fusion as the only pom1p protein in the cell . Errors are estimated from the fitting of an exponential curve to the average profile for each pom1 fusion . Intensity profiles are normalized to have the same intensity at the cell centers . We have confirmed that different normalizations ( and also fitting to the raw data ) do not significantly alter the measured decay lengths ( data not shown ) . ( C ) Ratio of distributions of cdr2p nodes in cells with different pom1p gradient distributions . Maximum projection confocal images of cdr2-tomato in pom1-GFP ( n = 49 cells ) and pom1-3GFP ( n = 50 cells ) strains were acquired and nodes were specified by a thresholding approach ( ‘Materials and methods’ ) . The ratio is defined as the area of the cdr2p nodes in the pom1-GFP strain divided by the area of the cdr2p nodes in the pom1-3GFP strain . Black line is guide to the eye for ratio of one . Strain used: FC2686 . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00710 . 7554/eLife . 02040 . 008Figure 1—figure supplement 5 . Pom1p gradients with different decay lengths do not affect cdr2p node intensity or number . Maximum projection confocal images of cdr2-tomato in pom1-GFP ( n = 49 cells ) and pom1-3GFP ( n = 50 cells ) strains were acquired . Nodes were specified by a thresholding approach ( ‘Materials and methods’ ) . Strains used: FC2686 , FC2687 . ( A ) Intensity of cdr2-tomato in nodes in pom1-GFP strain . Black line is linear best fit ( r2 = 0 . 39 ) . ( B ) Number of cdr2-tomato nodes in pom1-GFP strain . Black line is linear best fit ( r2 = 0 . 61 ) . ( C ) Intensity of cdr2-tomato in nodes in pom1-3GFP strain . Note that the intensities were adjusted to be on similar scale to ( A ) . Black line is linear best fit ( r2 = 0 . 55 ) . ( D ) Number of cdr2-tomato nodes in pom1-3GFP strain . Black line is linear best fit ( r2 = 0 . 80 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 008 To further test the gradient model , we examined the effect of altering the gradient profile . Pom1-3GFP ( pom1p fused to three tandem GFPs ) produced a consistently steeper gradient profile than pom1-GFP or pom1-tomato fusions ( Figure 1E , F , Figure 1—figure supplement 4 ) . The reason for this change was not clear , as these fusion proteins displayed similar dynamics ( our unpublished observations ) . The gradient model predicts that a change in gradient distribution would lead to a significant change in cell size at division . However , we detected no differences in cell length at division between these pom1-tagged strains ( Figure 1G , H ) . Consistent with this result , there were no significant differences in the intensities or number of cdr2p nodes ( Figure 1—figure supplements 4C , 5 ) . Overall , these data are inconsistent with the gradient model . To further investigate how this regulatory pathway may sense cell size , we focused on how cell size affects cdr2p and its behavior at these medial cortical nodes . Pom1p may exert its cell size effects in part by ensuring the proper localization of cdr2p nodes to this region ( Celton-Morizur et al . , 2006; Padte et al . , 2006; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) ( see below ) . We quantitated cdr2p levels using a functional cdr2-GFP construct ( Figure 2A , Figure 1—figure supplement 1A , Figure 2—figure supplement 1; Morrell et al . , 2004 ) . Cdr2-GFP concentration in the whole cell remained approximately constant in interphase cells of various lengths ( Figure 2B ) . Interestingly , the intensity of cdr2-GFP at the medial cortex increased with cell length ( Figure 2C , Figure 2—figure supplements 2 , 3 ) . The cortical area containing the nodes also increased slightly with cell length , but the relative change was less than for the cdr2-GFP intensity ( Figure 2C; Morrell et al . , 2004 ) . Measurement of cdr2-GFP intensity within a 3-μm wide region of the medial cortex showed directly that the local cdr2p concentration in this region rises approximately twofold as cells grow through interphase ( Figure 2D , Figure 2—figure supplement 2 ) . This increase was confirmed in time-lapse analyses of individual cells ( Figure 2E , F ) . 10 . 7554/eLife . 02040 . 009Figure 2 . Cdr2p accumulates in nodes at the medial cortex as cells grow . ( A ) Fission yeast cells expressing pom1-tomato and cdr2-GFP . Left panel: single medial confocal section; right panel: maximum Z-projection through whole cell . Strain used: FC2678 . Scale bars = 3 μm . ( B ) Total cellular intensity of cdr2-GFP in cells of different lengths . Mean intensities over the whole cell from sum projection images . n = 54 cells . Black line: linear fit with r2 = 0 . 04 . ( C ) Cdr2-GFP total intensity in medial cortex ( blue ) ( Figure 2—figure supplement 3A; n = 51 ) and width of cdr2-GFP nodal region along long cell axis ( green ) as function of cell length ( n = 185 ) . Error bars = SEM . Black lines: linear fits , r2 = 0 . 90 and 0 . 89 for width and intensity respectively . See Figure 2—figure supplements 1–4 . ( D ) Total cdr2-GFP intensity in fixed a 3-μm wide medial band as function of cell length ( Figure 2—figure supplement 3F ) n = 67 . Black line: linear fit with r2 = 0 . 71 . ( E ) Time-lapse maximum projection images of a cell expressing cdr2-GFP . Scale bar = 3 μm . ( F ) Total normalized intensities of nodal cdr2-GFP in 5 cells tracked over time ( measured from images like E , using method of Figure 2—figure supplement 3A ) . See Figure 2—figure supplement 5 . ( G ) Number of cdr2-GFP nodes as function of cell length ( n = 51 ) . Black line: linear fit with r2 = 0 . 67 . Nodes identified by thresholding , using method of Figure 2—figure supplement 3C , which provides a lower-bound estimate . ( H ) Distributions of cdr2-GFP node intensities in short vs long cells . n = 89 nodes in 9 cells , n = 286 nodes in 7 cells , respectively ( nodes as determined in G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 00910 . 7554/eLife . 02040 . 010Figure 2—figure supplement 1 . Measurement of cdr2p protein number . ( A ) Protein numbers were estimated by comparison of fluorescence intensity in living cells with standard fusion proteins that have been quantitated previously . Quantification of average cdr2p molecules in the whole cell was estimated by comparing total cell fluorescent intensity of cdr2-GFP with rlc1-GFP ( regulatory light chain of myosin ) , which has been estimated around 9600 molecules/cell ( Wu and Pollard , 2005 ) ( n = 50 cells ) . Strains used: FC2688 , FC1139 . Error bars = SD . ( B ) Quantification of cdr2p molecules in each node by comparing cdr2-GFP with the E . coli flagellar protein motB-GFP expressed in bacteria , estimated to be 22 molecules/dot ( Coffman et al . , 2011 ) ( n = 200 nodes ) . Error bars = SD . ( C ) Comparison of fluorescence intensities of different cdr2-GFP species , using images of cdr2-GFP cells in a single medial slice . The cdr2-GFP mean intensities in a 12 pixel square area in a cdr2p node , cytoplasm , and dim cortical dots outside of the medial cortex were measured ( Figure 5A ) . Note that only the brighter , more discrete cortical dots were assayed . n = 20 measurements each in >5 cells . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01010 . 7554/eLife . 02040 . 011Figure 2—figure supplement 2 . Cdr2p and pom1p intensity measurements as a function of cell length . ( A ) Total intensity of cdr2-GFP and pom1-tomato in the cell as a function of cell length using maximal projection images . Black lines are linear best fits . Intensities were measured by a hand drawn region of interest ( ROI ) around the entire cell in a maximal projection of a stack of 13 confocal sections 0 . 4 μm apart . Strain used: FC2678 . n = 50 cells . ( B ) Intensity of cdr2-GFP and pom1-tomato in the medial cortex as a function of cell length . Intensities were measured from the same maximal projection as ( A ) but with a hand drawn ROI around the cortical band area . n = 50 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01110 . 7554/eLife . 02040 . 012Figure 2—figure supplement 3 . Comparison of image analysis methods for quantitating cdr2p fluorescence in the nodes . Different methods to image and analyze cdr2p intensities were compared . The details of each method are presented in the figure and described in detail in the ‘Materials and methods’ . Graphs show results of each method on cdr2p medial node intensity vs cell length in a population of cells . The intensities are normalized relative for each data set . All show a similar scaling of cdr2p node intensity with cell length . n = 49 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01210 . 7554/eLife . 02040 . 013Figure 2—figure supplement 4 . FRAP analysis of cdr2-GFP . Cdr2-GFP in the nodes was photo-bleached in the indicated regions , and fluorescence recovery was monitored over time . Cells were imaged in a single medial focal plane . Average data ( blue ) were fitted to exponential curves ( green ) . The black arrows indicate the time of 50% recovery . t1/2 was about 3 min for both sets of data . The similar rates of recovery for the full side and half side bleach patterns suggest that there is little exchange of cdr2-GFP between nodes . n = 14 cells ( A ) , n = 8 cells ( B ) . Strain used: FC1441 . Error bars = SD in both panels . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01310 . 7554/eLife . 02040 . 014Figure 2—figure supplement 5 . Cdr2p node number but not maximal intensity in each node increases with cell length . Cdr2-GFP in each node was measured by the ImageJ ‘Find Maxima’ function , and data was graphed in bins according to cell length . ( A ) shows a rise in the number of cdr2-GFP nodes with cell length . ( B ) shows the intensity of cdr2-GFP in each node does not increase on average . Note that this thresholding method underestimates the number of nodes slightly . cdr2-GFP intensity measured as defined in ( Figure 2—figure supplement 3A ) . Strain used: FC1441 . n = 51 . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 014 Time-lapse imaging also revealed dynamics of cdr2p nodes . Mature cdr2p nodes , estimated to each contain an average of ∼90 cdr2-GFP molecules ( Figure 2—figure supplement 1 ) , moved very slowly and exhibited little change over hours ( Video 1 ) . FRAP studies , however , revealed that cdr2-GFP turned over with a t1/2 of about 3 min within each node ( Figure 2—figure supplement 4 ) . With increasing cell length , the number of nodes in each cell increased ( Figure 2G , Figure 2—figure supplement 5A ) , whereas the intensities of individual nodes remained unchanged ( Figure 2H , Figure 2—figure supplement 5B ) . Thus , cell growth is accompanied by the formation of new nodes , leading to an increase in local cdr2p density . Imaging also revealed a subpopulation of less intense and more motile cortical nodes that may be newly assembling ones ( Video 1 ) . 10 . 7554/eLife . 02040 . 015Video 1 . Cdr2-GFP nodes in wild-type fission yeast cells . Fission yeast cells expressing cdr2-GFP . Spinning disc confocal images in a cortical slice , acquired every 20 s . Initial images show a brightfield/fluorescence image to show the cell outline . Scale bar: 5 µm . Strain FC2688 . Time stamp = min , sec . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 015 To determine if the cdr2p concentration is important in cell size control , we tested the effects of varying its expression level ( Figure 3A , B ) . Cdr2p was expressed from an nmt81 promoter , regulated by thiamine in the media . Mild cdr2p overexpression in the absence of thiamine ( estimated 1 . 6-fold ) caused cells to divide at abnormally short cell lengths . Consistent with previous studies ( Breeding et al . , 1998 ) , higher levels of overexpression caused cytokinesis defects and accumulation of longer cells . Conversely , decreased cdr2p expression led cells to divide at much longer lengths , similar to a cdr2 null strain ( Morrell et al . , 2004; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Thus , cdr2p is a dose-dependent regulator of cell size and mitotic entry ( Figure 3C ) . The persistence of cdr2-GFP in cells treated with the protein synthesis inhibitor cyclohexamide showed that the majority of cdr2p is highly stable in interphase cells ( Figure 3D , E ) . Together , these findings suggest that as the cell grows , cdr2p is a stable protein that is synthesized to maintain a constant concentration in the whole cell , and accumulates at the medial cortex , where it promotes mitotic entry in a concentration-dependent manner ( Figure 3C ) . 10 . 7554/eLife . 02040 . 016Figure 3 . Cdr2p is a dose-dependent regulator of cell size . ( A ) Effect of cdr2p expression level on cell size . cdr2+ was expressed at different levels using a thiamine-regulatable promoter ( nmt81-cdr2 ) . Inverted images of cells stained with cell wall dye blankofluor . Cells express cdr2p at levels on average of 1 . 6 , 1 . 0 and 0 . 3-fold relative to wild type ( top to bottom ) . Strains used: FC15 , FC2691 . Scale bar = 5 μm . ( B ) Length of cells at septation . n = 87 , 47 , 121 , 123 cells . T = thiamine . Error bars = SD , *p<0 . 0001 as determined by Kolmogorov–Smirnov statistical tests . ( C ) Model that the local concentration of cdr2p increases in the region of the medial cortical nodes as cells grow , and when it reaches a critical level , promotes entry into mitosis . ( D ) Stability of cdr2-GFP protein . Time-lapse images of cells expressing cdr2-GFP treated with 100 μg/ml cycloheximide ( Polanshek , 1977 ) . Strain used: FC2688 . Scale bar = 5 μm . ( E ) Total cell and nodal cdr2-GFP intensities in individual cells treated with cycloheximide over time . Total cell intensity was measured as in Figure 2B n = 9 cells . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 016 A critical issue in cell size regulation is whether cdr2p levels at nodes report cell size or passage of time ( Turner et al . , 2012 ) : is cdr2p a ‘sizer’ or a ‘timer’ ? To test these possibilities , we examined cdr2p behavior in cells arrested for cell growth upon treatment with an actin inhibitor Latrunculin A ( Ayscough et al . , 1997; Chang , 1999 ) . Levels of a simple timer should continue to increase over time , even without cell growth , while a sizer would not increase without cell growth . Latrunculin A-treated cells exhibited no growth and no increase in cdr2-GFP levels at nodes ( Figure 4A , B ) . Next , we compared cdr2p in cells growing at different rates . We used for3Δ ( formin ) mutants , which are defective in cell polarity regulation and exhibit highly variable growth rates ( Feierbach and Chang , 2001 ) . This mutant allowed us to measure cells in the same microscope field with identical genotype and growth conditions , but with over twofold varying growth rates ( Figure 4C , D , Figure 4—figure supplement 1 ) . The rate of cdr2-GFP accumulation at nodes strongly correlated with the rate of cell growth . ( p<10−3 see ‘Materials and methods’; Figure 4C , D , Figure 4—figure supplement 1 ) . Thus , cdr2p has properties of a ‘sizer’ not a ‘timer’ . 10 . 7554/eLife . 02040 . 017Figure 4 . Cdr2p has properties of a sizer but not a timer . ( A ) Cdr2-GFP does not accumulate at nodes over time without cell growth . Time lapse images of cells in which growth was halted by 200 μM Latrunculin A , an actin inhibitor ( Chang , 1999 ) . Scale bar = 5 μm . Strain used: FC2688 . ( B ) Mean cdr2-GFP nodal intensity over time in individual LatA-treated cells . n = 9 cells . Error bars = SD . Dotted line shows for comparison the observed average increase of nodal cdr2-GFP in untreated , growing cells ( from Figure 2F ) . ( C ) Time-lapse images of cdr2-GFP in for3Δ ( formin ) cells . Two sister cells which exhibit variable growth rates are highlighted ( yellow and orange outlines ) . Strain used: FC2690 . Scale bar = 5 μm . ( D ) Correlation between cell growth rate and the rate of accumulation of cdr2-GFP in for3Δ cells ( Figure 4—figure supplement 1 ) . Line is linear fit to the data ( r2 = 0 . 34 ) . A ‘timer’ molecule would show no correlation . n = 21 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01710 . 7554/eLife . 02040 . 018Figure 4—figure supplement 1 . Rate of cdr2p nodal accumulation correlates with the rate of cell growth . The effect of cell growth rate on cdr2p accumulation was examined by analyzing for3 ( formin ) mutant cells , which exhibit highly variable growth rates ( Feierbach and Chang , 2001 ) . Strain used: FC2690 . ( A ) Cell growth over time in a representative slow and fast-growing cell . ( B ) Cdr2-GFP nodal accumulation over time as measured by fluorescence intensity in the same two cells . Cdr2-GFP measurement method described in ‘Materials and methods’ . ( C ) Plot of cdr2-GFP increase as a function of cell length increase in the same two cells . These data suggest that cdr2p increases scale with cell size increases rather than time . See Figure 4 for results of the full data set . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 018 Our findings raise the key question of how nodal cdr2p concentration is able to scale with cell size . In further characterizing the dynamic behavior of cdr2p , we found that in addition to cdr2p in nodes and a diffuse cytoplasmic haze , it also localized to dim , dynamic dots all around the cortex ( Figure 5A , Video 2 ) . This dim cortical population has not been described previously . Interestingly , the distribution of these dim cortical cdr2-GFP dots did not vary over the cell tip , and thus did not correlate with levels of pom1p at cell tips . Thus , cdr2p is able to bind to the whole surface of the cell . 10 . 7554/eLife . 02040 . 019Figure 5 . Model for cell size-sensing by cdr2p . ( A ) Confocal time-averaged image ( 60 frames over 10 min ) in medial focal plane of cell expressing cdr2-GFP . Arrow highlights dim cdr2-GFP all around cell cortex ( Video 2 ) . Scale bar = 2 μm . Graph shows cdr2-GFP profiles on cortex around one pole at indicated time points . Cdr2-GFP appears brighter in the cytoplasm around nodes due to out-of-focus nodal fluorescence . Black arrows denote local peaks in the cdr2-GFP signal that are clearly distinct from the mean cdr2-GFP cortical signal . Strain: FC2678 . ( B ) Outline of mathematical model for cdr2p dynamics . ( C ) Equations and analytic solutions describing cortical and nodal cdr2p number . ( D ) Model parameters . ‘Measured’: deduced directly from experiment , ‘constrained’: limited by nodal cdr2p density scaling with cell length , ‘not important’: plays no role in nodal cdr2p density scaling . ( E ) Model fit to nodal cdr2-GFP density as function of cell length ( data from maximum intensity projection as described in Figure 2—figure supplement 3A , with cells binned by length at 1 µm intervals ) . Equations and parameters given in C , D . Error bars = SD . See Figure 5—figure supplement 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 01910 . 7554/eLife . 02040 . 020Figure 5—figure supplement 1 . Spatial membrane model for cdr2p distribution . ( A ) Measured cytoplasmic cdr2-GFP intensity ( n = 267 ) as function of cell length . Solid black line is best linear fit and dashed line is best fit to data assuming constant cytoplasmic cdr2p levels . Error bars are SD . ( B ) Model fit ( black line ) to experimentally measured cdr2-GFP intensity profile on the membrane ( for cells of length 11–12 μm ) ( green circles ) , where the model contributions from the nodal ( blue ) and cortical ( red ) cdr2p are also shown . Average profile obtained from individual time-averaged cdr2-GFP profiles over 90 s around nodal region in confocal section through the middle of each cell . ( C ) as ( B ) , but for shorter cells ( 9–10 μm ) . ( D ) as ( B ) but for longer cells ( 13–14 μm ) . n > 20 in B , C and D . ( E ) Model fit for the increase in total nodal cdr2p with cell growth , including cortical cdr2p diffusion , compared to experimental data . Data shown is for a subset of the experimental results shown in Figure 5E ( n = 39 ) . ( F ) Similar to Figure 5E , except model fit is for the increase in nodal cdr2p density ( defined as mean concentration in 3-μm region about cell center ) with cell growth , including cortical cdr2p diffusion . Data shown is for a binned subset of data shown in Figure 2—figure supplement 3F ( n = 39 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 02010 . 7554/eLife . 02040 . 021Figure 5—figure supplement 2 . Cdr2p-modification model . ( A ) Simplified schematic of a model for cdr2p scaling involving cdr2p-modification . Unmodified cdr2p rapidly diffuses in the cytoplasm . Cdr2p can bind ( black arrows ) to the membrane . On the membrane , cdr2p can be modified and subsequently unbinds from the membrane ( red arrows ) . This modified form of cdr2p can rapidly diffuse through the cytoplasm and bind ( dashed red arrows ) to the nodal regions on the membrane ( yellow region ) . Unbinding of modified cdr2p from the nodal region back to an unmodified form of cdr2p in the cytoplasm is denoted by dashed black arrows . For clarity we do not depict spontaneous reversion of modified cdr2p back to unmodified cdr2p in the cytoplasm . ( B ) Scaling of nodal cdr2p concentration in the cdr2p-modification model ( for parameters given in ‘Materials and methods’ ) , where nodal cdr2p concentration is normalized to 1 at cell length 7 . 5 μm . Experimental data are same as Figure 5E . Strain used: FC2678 . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 02110 . 7554/eLife . 02040 . 022Video 2 . Dynamics of cortical cdr2p in wild-type cells . Fission yeast cells expressing cdr2-GFP . Spinning disc confocal images in a medial slice , acquired every 10 s over 5 min . These images , which were taken at longer exposures and higher laser power than the other two videos , reveal dim dynamic cdr2p dots all over the cortex . Scale bar: 5 µm . Strain FC2688 . Time stamp = min , sec . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 022 Because it is not intuitively clear how these dynamic behaviors of cdr2p might cause it to concentrate in the nodal region in a cell-size-dependent manner , we developed a mathematical model to probe the mechanism responsible ( Figure 5B–D ) . Based on our experiments , this model postulates that: ( 1 ) the concentration of cdr2p in the cytoplasm is homogeneous and changes only slightly with cell length ( Figure 5—figure supplement 1A ) ; ( 2 ) cytoplasmic cdr2p molecules can bind all over the plasma membrane ( Figure 5A; Video 2 ) , and subsequently move rapidly by diffusion on the cortex; ( 3 ) cortical cdr2p molecules can transition to associate with a nodal region on the medial cortex . Note that the details of the formation and growth of individual nodes are beyond the scope of the model . Rather , we simply model the overall number of cdr2p molecules in the nodal region . ( 4 ) Both cortical and nodal cdr2p can then unbind and return cdr2p to the cytoplasm; ( 5 ) cytoplasmic cdr2p can then diffuse rapidly before rebinding to the membrane . As the timescale of cell growth ( hours ) is much slower than the timescale of the cdr2p dynamics ( minutes , Figure 2—figure supplement 4 ) , we assumed that the molecular system is , at any given time , effectively in steady state . This steady-state assumption is also consistent with experimental findings that cdr2p levels at nodes are stable over time when cells are not growing ( Figure 4A , B ) . This model ( Figure 5B ) was implemented by two mass-action equations for cortical and nodal cdr2p , and solved analytically ( Figure 5C ) . Importantly , the model reveals how the cdr2p dynamics ensure a nodal cdr2p density that scales with cell size , or more specifically , with the surface area of the plasma membrane ( Figure 5C ) . The simplicity of the model allowed us to clarify the two key elements important for this area sensing . The first is that the area of the nodal region must not scale proportionally with the total cell membrane area as the cell size increases ( Figure 2C ) . We then have one process ( cdr2p membrane association ) that scales proportionally with cell area , with a second process ( uptake of cortical cdr2p into the nodes ) , which does not . The second key element is that the nodal region receives information via cdr2p about the entire surface area of the cell . In this model , cdr2p needs to be able bind the membrane long enough to move on the membrane to reach the nodal region . The outcome is then a rising cdr2p nodal density with increasing cell area . Using the experimentally determined nodal/cortical areas , and with other parameters measured/constrained from our experiments ( Figure 5D , ‘Materials and methods’ ) , we fitted the cdr2p density in the medial nodes as a function of cell length to that measured experimentally , with good results ( Figure 5E ) . Note that wild-type S . pombe cells are rod-shaped and have an approximately constant width , so that surface area and cell length are proportional to one another . A more sophisticated version of the same underlying model , including spatially varying cdr2p on the cortex , generated similar results ( Figure 5—figure supplement 1B–F , ‘Materials and methods’ ) . In addition , alternative models in which cdr2p does not need to diffuse long distances on the plasma membrane to the nodes are also consistent with the current findings . We analyzed a model in which cdr2p was now modified ( e . g . , phosphorylated ) at the cortex and remains modified for a period even if it returns into the cytoplasm , from where it then can diffuse to and accumulate at the nodal region ( Figure 5—figure supplement 2A ) . The underlying area-sensing mechanism was nevertheless conserved in this alternative model , with similar key elements as discussed above ( Figure 5—figure supplement 2B , ‘Materials and methods’ ) . A central prediction of the modeling is that cdr2p is sensing the surface area of the cell: cdr2p at nodes should scale with surface area , and not , for instance , cell volume . To experimentally test if cdr2p scales with surface area or volume , we analyzed cdr2p levels in S . pombe mutants with different widths , so that surface area and volume are uncoupled . Rga2p and rga4p are Rho-GAPs involved in regulation of cell polarity and width ( Das et al . , 2007; Villar-Tajadura et al . , 2008; Kelly and Nurse , 2011 ) ; rga2Δ mutants are thinner while rga4Δ mutants are fatter than wild type ( Figure 6A ) . We measured surface areas and volumes in these cells ( ‘Materials and methods’; Figure 6—figure supplement 1 ) . In a group of interphase cells of similar surface area but of different volumes , nodal cdr2-GFP intensities correlated with surface area ( Figure 6B ) . Conversely , in considering cells of similar volume but with a range of different surface areas , cdr2-GFP nodal intensity correlated with surface area and not volume ( Figure 6C ) . These results thus suggest that nodal cdr2p scales with cell surface area , in agreement with the predictions of the mathematical models . 10 . 7554/eLife . 02040 . 023Figure 6 . Cdr2p and cell size at division scale with cell surface area . ( A ) Fission yeast cells expressing cdr2-GFP and pom1-tomato in wt , rga4Δ ( fat morphology ) and rga2Δ ( thin morphology ) backgrounds . Maximum Z-projection images . Cells lacking nodes are in mitosis . Strains used: FC2678 , FC2794 , FC2795 . Scale bar = 5 μm . ( B ) Comparison of measured nodal Cdr2-GFP intensity in cells of different volumes . For each cell , the surface area and volume were measured by segmentation ( ‘Materials and methods’ ) . A subset of cells whose surface area was within 10–20% of the mean surface area was selected for each cell type ( ‘Materials and methods’ ) . The graphs show the surface area , volume , and nodal cdr2-GFP intensity ( cdr2-GFP intensity measured as defined in Figure 2—figure supplement 3A ) in these selected cells . For each data type , normalization is by mean value for rga4Δ cells . Error bars = Error on the mean . n = 24 ( wt ) cells , 27 ( rga4Δ ) , 32 ( rga2Δ ) . Strains used in B and C: FC1441 , FC2792 , FC2793 . See Figure 6—figure supplement 1 . ( C ) As in B , except groups of cells were selected with similar volumes ( mean measured volume ± 10–20% ) . n = 24 ( wt ) cells , 27 ( rga4Δ ) , 27 ( rga2Δ ) . These data show cdr2-GFP scaling with surface area . The difference in surface area and cdr2-GFP intensity between the rga2Δ and rga4Δ cells is statistically significant ( p<10−3 , **p<10−4 ) . See Figure 6—figure supplement 1 . ( D ) Comparison of cell lengths , surface areas and volumes in rga4Δ , wild type and rga2Δ at time of septation ( ‘Materials and methods’ ) . The septum is not included in these measurements . Data for each set is normalized by the appropriate value for the rga4Δ cells . Error bars = SD . Strains used: FC2554 , FC2555 , FC2556 . n = 76 ( wt ) , 64 ( rga4Δ ) , 60 ( rga2Δ ) . ( E ) Quantitating differences between rga4Δ , wt and rga2Δ at time of septation . Left: probability density distributions for measured surface area ( top ) and volume ( bottom ) for wild type ( red ) , rga2Δ ( green ) and rga4Δ ( blue ) cells in ( D ) . Gray area marks the overlap region between the distributions . Error bars not shown for clarity . Right: to quantitatively compare these distributions , we calculated the Jensen–Shannon distance ( Lin , 1991 ) between the length , surface area and volume distributions for the different cell types ( where 1 corresponds to the distributions having no shared information and 0 to identical distributions , see ‘Materials and methods’ ) . This analysis shows that these cells with different shapes divide with similar surface area . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 02310 . 7554/eLife . 02040 . 024Figure 6—figure supplement 1 . Scaling of nodal cdr2-GFP intensity with surface area and volume . ( A ) Nodal cdr2-GFP intensity from maximum intensity projection for wt , rga4Δ ( fat cells ) , and rga2Δ ( thin cells ) plotted against cell surface area . ( B ) Nodal cdr2-GFP intensity from maximum intensity projection for wt , rga4Δ ( fat cells ) , and rga2Δ ( thin cells ) plotted against cell volume . Strains used: FC1441 , FC2792 , FC2793 . n = 51 ( wt ) , 54 ( rga4Δ ) , 58 ( rga2Δ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 024 These findings lead to another key prediction that cells enter mitosis at a specific cell surface area . We measured cell length , surface area and volume in wild-type , rga2Δ and rga4Δ strains in dividing cells; these dimensions are indicative of the size of the cells at entry into mitosis . These cells with different shapes entered mitosis with more similar cell surface areas but differing cell volumes and lengths ( Figure 6D , E ) . All three strains exhibited average surface areas of 150 µm2 ± 8 µm2 , while the average volumes varied from 120 µm3 to 150 µm3 and average lengths from 11 µm to 15 µm . A more rigorous analysis based on Jensen–Shannon distances ( ‘Materials and methods’ ) showed quantitatively that the distributions of surface area were more similar than those for volume or length . These findings suggest that cells monitor their size at the G2/M transition by measuring their surface area . We next examined how pom1p quantitatively affects cdr2p . In pom1Δ mutants , cdr2p is thought to be somehow more ‘active’ and promotes division at slightly shorter cell lengths than wild type ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . In pom1Δ cells , cdr2-GFP is spread in dots throughout much of the cortex , except for the growing cell tip ( Figure 7A; Celton-Morizur et al . , 2006; Padte et al . , 2006; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . The total amount of cdr2p in the cell was similar in wild-type and pom1Δ mutant cells over a range of cell lengths ( Figure 7B ) . Cortical profiles showed that in pom1Δ cells , cdr2p was still enriched over the medial cortex and that the non-growing end had levels roughly half that of the medial region ( Figure 7C ) . The fraction of cdr2p that is cortical and the area of nodal cdr2p were both substantially increased in pom1Δ cells ( Figure 7D , E ) . Interestingly , the increase of cortical cdr2p with cell length was similar in pom1Δ vs wild-type cells , but the curve was shifted slightly upwards ( Figure 7F ) . In contrast , in the medial cortical region , cdr2p levels were lower than wildtype ( Figure 7G ) . A simple interpretation is that cdr2p is able to signal to promote entry into mitosis from nodes on non-medial sites in this mutant . However , another factor to consider is that cdr2p kinase activity may also be altered in these cells . Time-lapse imaging showed that cdr2p nodes are more motile in pom1Δ cells than in WT ( Figure 7H; Videos 1 , 3 ) , suggesting a defect in the anchoring of these nodes in the membrane . At the growing end , there are also the dim cortical motile cdr2p dots , similar to those present at cell ends in WT ( Figure 5A ) . Thus , pom1p affects the distribution and mobility of cdr2p nodes . 10 . 7554/eLife . 02040 . 025Figure 7 . Cdr2p behavior in pom1Δ mutants . ( A ) Fission yeast cells expressing Cdr2-GFP in wt and pom1Δ background . Brightfield , maximum projection , and mid–focal plane images are shown . Strains used: FC1441 and FC2057 . Scale bar = 3 μm . ( B ) Comparison of total measured cdr2-GFP intensity ( from sum projection after background subtraction ) with cell length . n = 52 ( wt ) , 72 ( pom1Δ ) . ( C ) Average cdr2-GFP intensity profile around cortex of cell ( spatial direction as defined in cartoon in Figure 1F ) . pom1Δ cells are orientated such that the cell end with the higher cdr2-GFP level is defined to be at d = 0 µm . n = 52 ( wt ) cells , 72 ( pom1Δ ) . Error bars not shown for clarity . See ‘Materials and methods’ for further details . ( D ) Fraction of cdr2-GFP signal observed on the cortex compared with total measured cdr2-GFP in the medial plane . The cortical signal is calculated as the sum of measured intensity along a mask around the cortex ( see ‘Materials and methods’ for mask definition ) . The total signal is defined as the total measured cdr2-GFP intensity on and inside the mask . Error bars = SD . n = 52 cells ( wt ) , 72 ( pom1Δ ) . ( E ) Measured area of nodal cdr2-GFP region from maximum intensity projection images . Regions were measured manually for individual cells . n = 46 ( wt ) cells , 77 ( pom1Δ ) . Error bars = SD . ( F ) Accumulation of total membrane cdr2-GFP ( both nodal and cortical signal ) against cell length . n = 52 ( wt ) cells , 72 ( pom1Δ ) . See ‘Materials and methods’ for details . Lines are linear least-square fits to the data , with similar slopes . Error bars = SD . ( G ) Accumulation of nodal cdr2-GFP ( maximum intensity projection ) within 3 µm medial cortical region . n = 52 cells ( wt ) , 72 ( pom1Δ ) . Error bars = SD . ( H ) Kymograph of cortical cdr2-GFP over 5-min period in wild type and pom1Δ cells . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 02510 . 7554/eLife . 02040 . 026Figure 7—figure supplement 1 . Cdr2p behavior in cells in which pom1p is targeted all over the cortex . ( A ) Fission yeast cells expressing cdr2-GFP in wild-type and PMT-pom1C cells in which a pom1p chimera is localized throughout the cortex . Maximum Z-projection images are shown . Strains used: JM2057 , JM892 . Scale bar = 5 μm . ( B ) Total cdr2-GFP intensity in cells ( n = 64 [wt] , 54 [PMT-pom1C] ) . Method is same as in Figure 7B . ( C ) Average cortical profile of cdr2-GFP intensities from maximum intensity projection ( distance defined as in cartoon in Figure 1F ) . See ‘Materials and methods’ for mask definition . Error bars are not shown for clarity . n = 64 ( wt ) , 54 ( PMT-pom1C ) . ( D ) Intensity of cdr2-GFP in a 3-μm wide region of the medial cortex as function of cell length . Data from maximum intensity projection images . n = 64 ( wt ) , 54 ( PMT-pom1C ) . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 02610 . 7554/eLife . 02040 . 027Video 3 . Abnormal behavior of cdr2p nodes in pom1Δ cells . Fission yeast cells expressing cdr2-GFP . Spinning disc confocal images in a cortical slice , acquired every 20 s . Scale bar: 5 µm . Strain FC2057 . Note that many of these nodes appear more motile than those in wildtype cells ( Video 1 ) . Time stamp = min , sec . DOI: http://dx . doi . org/10 . 7554/eLife . 02040 . 027 We also examined the effect of disrupting pom1p localization on the cdr2p distribution . A construct in which pom1p is targeted all over the plasma membrane has been described ( PMT-Pom1C fusion , Figure 7—figure supplement 1A ) ( Moseley et al . , 2009 ) . Although cdr2p was expressed at normal levels in the whole cell , it was evenly distributed all over the cortex at a low level , and did not increase at the medial cortex with increasing cell length ( Figure 7 , Figure 7—figure supplement 1 ) . As shown previously ( Moseley et al . , 2009 ) , these cells divided at abnormally long cell lengths , similar to cdr2Δ mutants . These data show that pom1p has an inhibitory effect on cdr2p localization to nodes . These results further provide support that cdr2p needs to be present at these medial nodes in order to function effectively in cell size control . Here we propose a mechanism for cell size sensing based on a cortical sizer protein cdr2p . We provide evidence that cells sense a critical cell size by measuring cell surface area rather than , for example , cell volume or absolute length , a mechanism that could function regardless of the cell shape . As the cell grows , the concentration of cdr2p at the medial cortex increases . We have developed models explaining how cdr2p probes the surface area of the cell , and conveys this information to the medial cortex . There , cdr2p may signal to cell cycle regulators located on the nearby spindle pole body and nucleus ( see below ) . When the cell reaches a critical size , cdr2p at the nodes may reach a critical local concentration that promotes mitotic entry . Our quantitative models show how cdr2p can convey information about global cell area and deliver it in the form of a local ( nodal ) concentration . This size-sensing model shares elements with a proposed microtubule length control mechanism termed the ‘antenna model’ . In the microtubule model , longer microtubules bind more motor proteins , which then accumulate at the microtubule end in a length-dependent manner ( Varga et al . , 2006 ) . In the cell size sensing case , the whole surface area of the plasma membrane may be regarded as an ‘antenna’ . Similar to the microtubule model , the property of cdr2p to first bind to the plasma membrane ‘antenna’ ( as opposed to merely binding the nodes directly ) is critical for this mechanism to read out cell size . This membrane cdr2p must then transit to the nodal region , where the cdr2p nodal density serves as a read-out of cell area . Although cdr2p may not exhibit directed motor-driven movements , this movement can still occur by diffusion along the membrane . We also considered an alternative model , where cdr2p is modified on the membrane , but after unbinding is able to diffuse through the cytoplasm to the nodes . The modification allows information about membrane area to be preserved in the cytoplasm , from where it can be relayed to the nodes ( Figure 5—figure supplement 2A ) . Furthermore , as the amount of nodal cdr2p reflects cell size rather than time , we postulate that the system is effectively in a dynamic steady state at a given cell size , with fast cdr2p dynamics compared to the timescales of cell growth . The localization of a cdr2p sizer at cortical nodes provides several key advantages over other locales . First , it allows the local concentration of nodal cdr2p to increase as the cell grows . Previously proposed mechanisms have been based upon nuclear concentration or the nuclear/cytoplasmic ratio of a sizer , but in many cell types ( including fission yeast ) , nuclear volume also increases as cells grow ( Neumann and Nurse , 2007; Figure 1—figure supplement 3 ) . Second , we speculate that medial cortical placement of nodes surrounding the medial nucleus may allow cdr2p to communicate its local concentration to presumed targets such as wee1p and cdk1p on the nucleus . Although wee1p can be observed at some nodes upon overexpression ( Moseley et al . , 2009 ) , its localization in late G2-phase is clear in the nucleus , and at the spindle pole body ( SPB ) ( Masuda et al . , 2011 ) , a structure on the nuclear envelope situated close ( often <0 . 5 μm ) to the nodes . Cdk1/cyclin B and polo kinase are also located at the SPB and nucleus ( Alfa et al . , 1990; Masuda et al . , 2011; Grallert et al . , 2013 ) . Potentially , the SPB could detect local gradients of cdr2p ( or other molecules ) emanating from nearby cortical nodes . However , as a simple cdr2p concentration gradient in the cytoplasm is expected to be very shallow ( due to rapid diffusion ) , it is likely that additional layers of regulation such as through phosphorylation states or diffusion barriers would be needed to generate suitably steep gradients . The potential importance of the geometric relationship between the nodes and SPB/nucleus remains to be tested . The localization of these nodes to the medial cortical region involves multiple inputs . One important contributor is pom1p . Although pom1p clearly regulates cdr2p function and phosphorylation , our data indicate that the pom1p gradient distribution may not be the primary size sensing mechanism as previously proposed . Indeed , our data are consistent with a recent report that size correction still occurs in pom1Δ mutants ( Wood and Nurse , 2013 ) . Rather , a primary role of pom1p may be to ensure the medial localization of nodes . Thus , pom1p may affect cdr2p nodes in part by affecting distribution and general properties ( such as its mobility in the membrane ) of the nodes . Recent studies ( published while this work was in press ) suggest that cdr2p activity is also regulated by phosphorylation by pom1p and ssp1p protein kinases ( Bhatia et al . , 2014; Deng et al . , 2014 ) . Another important factor in cdr2p localization is likely to be the nucleus that is situated in the cell interior with roughly the same width as the nodal region . Studies on mid1p , another component of the nodes , suggest that the nucleus governs dynamic nodal localization , in a mechanism that may involve nuclear shuttling ( Paoletti and Chang , 2000; Daga and Chang , 2005; Almonacid et al . , 2009 ) . Furthermore , the organization of the cortical endoplasmic reticulum also influences nodal stability and localization ( Zhang et al . , 2010 ) . There are also likely to be additional ( or alternative ) inputs into size control ( Coudreuse and Nurse , 2010; Navarro and Nurse , 2012; Wood and Nurse , 2013 ) . Additional cell size regulators include the cell tip protein nif1p ( Martin and Berthelot-Grosjean , 2009; Wood and Nurse , 2013 ) , and skb1p , which localizes to cortical patches distinct from the nodes ( Deng and Moseley , 2013 ) . Cells expressing a cdk1p–cyclinB fusion still exhibit apparently near-normal size control in the absence of wee1p/mik1p or cdk1p-tyrosine phosphorylation control ( Coudreuse and Nurse , 2010; Navarro and Nurse , 2012; Wood and Nurse , 2013 ) , suggesting the existence of controls that are entirely independent of cdk1–tyrosine phosphorylation . Thus , this simple cdr2p-based mechanism is likely to be a core component of a larger network responsible for cell size control . Standard methods for S . pombe growth and genetics were used ( Moreno et al . , 1991 ) . In general , strains were constructed using a PCR-based homologous recombination method to insert markers in the yeast chromosome ( Bahler et al . , 1998 ) . Pom1-mGFP ( = pom1-GFP ) and pom1-3GFP strains were constructed by inserting mGFP and 3GFP constructs into the pom1+ chromosomal locus from fragments amplified from pFA6a-mGFP-kanMX6 ( monomeric GFP A206K [Zacharias et al . , 2002] ) and pFA6a-3GFP-kanMX6 ( triple tandem GFP ) ( Wu and Pollard , 2005; Martin and Chang , 2006 ) , ( from JQ Wu ) . In general , constructs were checked by PCR and sequencing , and strains were outcrossed multiple times . For live cell imaging , S . pombe cells were typically grown in exponential phase in liquid YE5S media at 25°C with shaking for 18–24 hr . In some experiments , the cells were mounted in liquid YE5S media directly on glass . For long term imaging experiments , the cells were placed in open 35-mm glass bottom dishes ( MatTek Corp , Ashland , MA ) . To stick cells to the glass , dishes were coated with lectin by drying 5 μl of 1 μg/μl lectin on the dishes; the cells in media were applied and incubated for 5 min and then 2 ml YE5S were added ( Figures 1C , D , 2E , F ) . The cells were also imaged on 1% agarose YE5S pads under a glass coverslip . For experiments to alter cdr2 levels ( Figure 3 ) , the nmt81 promoter ( Basi et al . , 1993 ) was inserted upstream of the cdr2 chromosomal locus by homologous recombination using a PCR-generated DNA fragment derived from pFA6a-kanMX6-P81nmt1 ( Bahler et al . , 1998 ) using the following primers: nmt-cdr2-F: ( 5′-TATGCTGTTCTATGAATGGGGTTTGGATTTGGCCATCACCACTTCACCGATTT ACTGGTTCTTTTGAATAGTTGAAGTGTGAATTCGAGCTCGTTTAAAC-3′ ) nmt-cdr2-R: ( 5′-TTGGCTAAACGTGATGAATTTGGTCCTCCTGATCCTAAGGAAAGACCAAGC TCCCAAGGTCCAACTTCTGAAATTGTACTCATGATTTAACAAAGCGACTATA-3′ ) . Correct insertion was verified by PCR of both sides of the construction using specific primers for the endogenous and inserted DNA . Multiple transformants showed the same cell size phenotypes . nmt81-cdr2 cells and the parental wild-type strain ( FC15 ) were grown in EMM +5 µg/ml thiamine at 25°C for 2 days , keeping the OD600 of the culture below 0 . 5 over the entire period . The cells were then washed three times by centrifugation at 2000 rpm with EMM , innoculated into EMM with or without 5 µg/ml thiamine , and then grown with shaking at 25°C for 20 hr , and then samples were collected for microscopy for cell length measurements and for RNA preparation . Relative RNA expression levels were assayed by RT-PCR . RNA was isolated from cells using the RNeasy Mini Kit ( Qiagen , Germantown , MD ) . 50 ng of RNA was used for real-time PCR using iScript One-Step RT-PCR kit with SYBR green ( Bio-Rad ) on a Bio-Rad Real-Time PCR system . The actin act1 mRNA was used as standard . Amplicons for act1 or cdr2 were generated with the following primers: act1-F ( 5′-GAAGAAGAAATCGCAGCGTTGG-3′ ) , act1-R ( 5′-CGCTTGCTTTGAGCTTCATCAC-3′ ) cdr2-F ( 5′-TGGGAGCTTGGTCTTTCCTTAG-3′ ) , cdr2-R ( 5′-TAGCCTGTTGGCTCGAAGTAAG-3′ ) . Expression levels of cdr2 in nmt81-cdr2 cells were measured as fold change relative to levels in a wild-type ( FC15 ) strain grown under the same conditions . The changes in cell lengths were consistent in multiple experiments . Cycloheximide ( Sigma , St Louis , MO ) was used at a final concentration of 100 µg/ml from a stock of 10 mg/ml stock solution in ethanol and added to exponential phase cultures in YE at 25°C ( Polanshek , 1977 ) . Latrunculin A ( LatA ) was used at a final concentration of 200 µM from a 100X stock in DMSO ( Chang , 1999 ) . LatA or cycloheximide were added to cells in a 35-mm glass bottom dish ( described above ) and imaged over time . Images were generally acquired using a spinning-disc confocal fluorescence NikonTI-based microscope system ( Nikon Instruments , Melville , NY , Yokogawa , Tokyo , Japan , Solamere Technology , Salt Lake City , UT ) with an EM CCD camera ( Hamamatsu Corp , Boston , MA ) and a 100X 1 . 4 N . A . objective with a 1 . 5X magnifier ( Saunders et al . , 2012 ) . A wide-field Nikon Eclipse 800 microscope and a 60X 1 . 4 N . A . objective was also used for some studies . FRAP studies were performed with a Zeiss 710 scanning confocal microscope . ImageJ ( NIH ) and custom MatLab ( Mathworks , Natick , MA ) software were used for analysis . Fluorescence intensity values around the cortex of cells were measured from images of cells in a medial focal plane , using custom MatLab software for the automated generation of a one-pixel wide mask around the cell cortex , followed by manual correction ( Saunders et al . , 2012 ) . Time-averaged images of pom1-fusions used average projections of 50 0 . 5 s frames over 25 s . The average pom1-tomato intensity at the medial cortex was measured in a 3-pixel wide by 3 µm long rectangle over the medial cortex , and the mean background value outside of the cells was subtracted . To measure the pom1p gradient decay lengths , cells expressing the appropriate pom1-fusion were imaged for 3 s in a single confocal section through the middle of the cell . Cells were segmented as described in Saunders et al . , 2012 . Intensities were normalized to one at the cell tip and background subtraction performed such that the different fusions had zero intensity 5 µm from the tip . Curves were then fitted to exp ( −x/λ ) , where λ is the decay length of the profile , with λ shown in Figure 1—figure supplement 4B . Cdr2-GFP intensity was quantified using six different methods ( Figure 2—figure supplement 3 ) . ( A ) Maximum projections were made of 13 slices of confocal sections taken 0 . 4 μm apart . A region of interest ( ROI ) was selected in ImageJ by hand around the cdr2p nodes , excluding as much background as possible . The area and total intensity of the ROI was recorded , and the ROI width was determined by the spread of cdr2p nodes along the long axis of the cell . ( B ) Similar to ( A ) but the maximum projection was taken from the top three slices consisting of the ‘top’ cortical section of the cell . ( C ) Similar to ( A ) , except the ROI was selected by an image analysis program in Matlab ( custom-written ) which selected only pixels over a predetermined threshold ( approximately two times the mean background intensity ) . In this case , the width was not determined . ( D ) Maximum projections were taken similar to ( A ) . We used the Find Maxima macro function in ImageJ to find the brightest pixel from a local intensity source ( likely nodes ) , counting their number and totaling their intensity to estimate total intensity levels . In this case , width was also not determined . ( E ) We used a single confocal section through the middle of the cell and acquired images over 30 s . A region was then chosen for each time-averaged data set that overlapped the nodal region ( now seen as a line on the perimeter of the cell ) in a single pixel wide line that was 3 µm long . The intensity was measured from that line and summed . ( F ) Maximum projections were made of 13 slices of confocal sections taken 0 . 4 μm apart . Individual cells were then taken and rotated so their long axis was horizontal . A rectangular ROI was fixed at 3 μm wide and 3 . 72 μm tall for all cells and placed at the center of the nodal region . The mean intensity was then recorded in this fixed area . In all these instances , the mean background intensity from an area outside of the cells was subtracted for each pixel . These different methods all resulted in the same linear increase of cdr2-GFP intensity levels in the nodal region as a function of increasing cell length . However , due to the fact that each method measured cdr2-GFP levels in different ways , the exact slope and variance of the correlation differed from method to method . The single cell analyses of cdr2-GFP in the wildtype ( Figure 2E , F ) used 13 confocal sections 0 . 4 µm apart . Intensities were measured in a hand drawn ROI that contained the majority of cdr2p nodes and the mean background outside the cells was subtracted . In Figures 3 and 4 , in analysis of LatA and cycloheximide-treated cells , and for3Δ cells , maximum projections of Z-stacks comprising 13 confocal sections 0 . 4 µm apart were used . Intensities were measured in a hand drawn ROI that contained the majority of cdr2p nodes and the mean cytoplasmic value inside the cells at each time point was subtracted . In the measurements of rates of growth and cdr2-GFP accumulation in for3 mutants ( Figure 4C , D ) , growth rates were calculated by a least squares linear fit to the cell length as a function of time ( images every 30 min ) , over 60–120 min . The rate of change in nodal cdr2p intensity with time was also calculated by a least squares linear fitting . In both the cases , the error on the fit was found for each cell . To test whether a positive correlation between growth and cdr2p accumulation rates was robust , we performed numerical simulations using the distributions of the measured rates , and their errors , to create in silico data . Corresponding to each pair of values in the measured data set , we created a new in silico pair by drawing from Gaussian distributions with widths given by the measured errors in growth rate and cdr2p accumulation rate . We then performed a linear least squares fitting on each in silico data set to find the level of correlation and test whether it was greater than zero—that is whether a positive correlation existed between cdr2p accumulation rate and cell growth rate . Repeating this process 106 times , we found a probability of ∼0 . 0005 that a positive correlation would be absent . Hence , our conclusion of a positive correlation between the cdr2p accumulation rate and cell growth rate is robust . Results shown in Figure 4D are for a single experiment ( n = 21 cells ) ; similar results were found in multiple additional experiments ( data not shown ) . Protein counts were estimated by quantitative fluorescence intensity in ratios with standard proteins that had been quantitated previously ( Wu and Pollard , 2005; Coffman and Wu , 2012 ) . GFP-MotB complexes in live bacteria were used as a standard at 22 GFP molecules/dot ( Leake et al . , 2006; Coffman et al . , 2011; Laporte et al . , 2011 ) . To calculate the width of the nodal cdr2p region , we fitted the function ( ae− ( x−x0 ) 2/2σ2+b ) to the cdr2p profile from a time-averaged ( 90 s ) confocal section through the middle of each cell ( 385 cells ) . We only analyzed cells with a good quality of fit ( so that the measured σ is meaningful ) and with σ >0 . 5 µm ( thereby excluding cells with distorted fits due to one very bright nodal region ) . This process left 237 cdr2p intensity profiles for analysis . Each cell was binned according to length ( 8–9 µm , 9–10 µm , … ) and the mean and standard deviation calculated within each bin , see Figure 2C . For the cortical Cdr2-GFP profiles shown in Figure 7C , Figure 7—figure supplement 1C , a cortical mask was extracted as described in Saunders et al . ( 2012 ) . The center of each cell was located and the angles between a chosen tip and each pixel on the mask were calculated ( so a pixel at the opposing tip would have angle π ) . Angles were then binned into 100 sectors from 0 to 2π and the mean cdr2-GFP intensity at a given angle around the cell was calculated . Angles were converted into the mean distance from the tip by assuming that in the mid plane the cell can be approximated as two semicircles connected by straight lines , using the mean cell length and radius for each cell type . For pom1Δ cells , the tip with the highest cdr2-GFP intensity was defined to be at d = 0 µm . To calculate total cortical signal the sum of the cdr2-GFP signal on the mask was used ( Figure 7F ) . For analysis of cdr2-GFP intensity in a 3-µm cortical region around the cell middle ( Figure 7G , Figure 7—figure supplement 1D ) each pixel in the cell cortical mask within ± 1 . 5 µm of the cell centre was identified and then the cdr2-GFP was summed over only these pixels . For Figure 6 , cells were grown in liquid YE5S media at 25°C , and imaged on agarose pads . Cell surface area and volumes were measured using manual segmentation . In Figure 6B , C , we used a single mid focal plane brightfield image , whereas in Figure 6D , E , we used a single mid focal plane of a fluorescent image of blankofluor-stained septated cells . Cell perimeters were manually traced , with the mean surface area , Acor and mean volume , V , calculated in Matlab assuming radial symmetry around the long axis of the cell ( as the cross-section of fission yeast cells are nearly circular ) . To compare cells of similar surface area , we selected all cells with surface areas in the range Acor ± 0 . 10–0 . 20 Acor . The range of ± 10–20% was taken to ensure we had enough cells included for statistical significance ( between 24 and 32 cells ) , but that the range was reasonably constrained . The specific range was adjusted for each subset of cells for the different cell lines such that the mean surface area ( Figure 6B ) or mean volume ( Figure 6C ) were equal to within ± 1% . The unbinned data for each cell type is shown in Figure 6—figure supplement 1 . Likewise , for comparing cells of similar volume , we included all cells with volume in the range V ± 0 . 10–0 . 20 V . For measuring cell size at septation ( Figure 6D , E ) , cells without cdr2-GFP were analyzed . The septum was not included in this analysis , as we wanted to extract cellular dimensions at entry into mitosis prior to septum formation . We also analyzed a separate data set with the cdr2-GFP strains ( n > 45 cells for each genotype ) , and a data set using brightfield images , which all showed the same behavior . The similarity of the distributions for cell length , surface area and volume at mitosis were compared in the wild-type , rga2Δ and rga4Δ mutants using the Jensen–Shannon distance ( Figure 6E ) . The Jensen–Shannon distance is a statistical measure that quantitatively compares the overlap of two or more distributions , with a distance of 1 corresponding to the distributions having no shared information and a distance of 0 to identical distributions . The Jensen–Shannon distance is the square root of the Jensen–Shannon divergence , which is defined in terms of the Shannon entropy function of the probability distributions ( see Lin , 1991 ) . A simple alternative model for area scaling involves cdr2p becoming modified ( e . g . , phosphorylated ) . We assume that unmodified cdr2p diffuses rapidly in the cytoplasm , with homogeneous density ρcyt = Ncyt/V . Unmodified cdr2p can then bind to the membrane with a binding constant β . Once present on the membrane , with a correspondingly homogeneous density ρcor = Ncor/Acor , cdr2p can unbind back into the cytoplasm at a rate ν , while at the same time becoming modified ( e . g . , phosphorylated ) . This cytoplasmic , modified form of cdr2p , with homogeneous density ρcyt = Ncyt/V can rapidly diffuse , and then bind to the nodal region on the cortex , with a binding constant α , or spontaneously become unmodified at a rate µ . Finally , modified cdr2p in the nodal region , with density ρnod = Nnod/Anod , can unbind and become unmodified cytoplasmic cdr2p at a rate η ( Figure 5 , Figure 5—figure supplement 2A ) . The corresponding steady-state equations are:0=βAcorVNcyt−νNcor0=νNcor−μNcyt∗−αAnodVNcyt∗0=αAnodVNcyt∗−ηNnod . In the case without spontaneous reversion of cytoplasmic , modified cdr2p back to its unmodified form ( i . e . , µ = 0 ) , we can solve these equations to findρnod= ( βη ) ( AcorAnod ) ρcyt . Experimentally , we observe significant scaling of ρnod with increasing Acor ( or equivalently with cell length in the wildtype ) . This is in agreement with Model II , again provided that Anod , the area of the nodal region , does not scale proportionally with Acor , the total cell membrane area , as the cell size increases . If there is spontaneous reversion of cytoplasmic , modified cdr2p back to its unmodified form , then the above solution for ρnod becomesρnod= ( βη ) ( AcorAnod ) ( 1μVαAnod+1 ) ρcyt . Provided αAnod≫μV , then the reversion process can be neglected and our cell area scaling results are unchanged . As in Model I , the larger the value of α , the stronger the scaling effect will be , as more modified , cytoplasmic cdr2p—which effectively ‘measures’ the cell area—is taken up into the nodes—which effectively ‘read-out’ the area measurement . As above with Model I , the value of β is not important as it only enters our solutions as a constant prefactor . The value of ν is also not important for the behavior of ρnod ( see above ) , though we use a value of ν = 0 . 5 s−1 to ensure low levels of cortical cdr2p , as observed experimentally . We again use the FRAP data to estimate η = 5 × 10−3s−1 . Since ρnod = ( α/η ) ρcyt , we take α = 0 . 5 µms−1 so that the concentration of modified cdr2p in the nodal region is relatively large compared to that in the cytoplasm ( required since there is no observed scaling of cytoplasmic cdr2p concentration with cell length ) . Further , we use a low rate of spontaneous modification loss µ = 0 . 03 s−1 , so that the lifetime of modified , cytoplasmic cdr2p is relatively long . The fitting of this model to the experimental scaling of cdr2p is shown in ( Figure 5 , Figure 5—figure supplement 2B ) . In this model , the mechanism of size scaling is similar to that of model I . One process , the unbinding of cortical cdr2p can occur from anywhere on the membrane , and so scales proportionally with cell area . However , a second process , in this case the uptake of modified , cytoplasmic cdr2p into the nodes , occurs over a region whose area does not scale proportionally with the total cell area as the cell grows . The outcome is again a density ( for both ρ*cyt and ρnod ) that scales with total cell membrane area . The presence of the modified form of cdr2p is vital , so that information about membrane area can be protected and relayed to the nodes without being lost into the general cytoplasmic cdr2p population , which does not show size scaling characteristics ( Figure 5—figure supplement 1A ) . The model does not explicitly include pom1p . However , this does not mean that pom1p is unimportant in the regulation of nodal cdr2p . As the pom1Δ experiments demonstrate , without pom1p acting as a tip inhibitor for cdr2p , nodal cdr2p can form in an extended part of the cell , with such cells observed to divide at shorter lengths . Conversely , in pom1p mutants where pom1p is mistargeted all over the plasma membrane , cdr2p does not form localized regions of high nodal concentration , with such cells dividing at longer lengths . Therefore , pom1p appears to play an important role in defining the region of nodal cdr2p accumulation , without which the cdr2p-dependent control of cell length is perturbed . Accordingly , pom1p is implicitly included in the model by defining a spatially limited region that can be occupied by the cdr2p nodes .	Although different types of cells come in a variety of shapes and sizes , most cells are able to maintain a fairly consistent size and shape as they grow and divide . For example , the rod-shaped cells of the fission yeast S . pombe grow to be 14 microns long before dividing in the middle to form two new cells . This prevents any single cell becoming too large or small . A similar phenomenon has been observed in other types of cells , so it is clear that cells must be able to measure their own size , and then use that information to trigger cell division . A number of proteins that regulate cell size and cell division in fission yeast have now been identified . These proteins form a pathway in which a protein called pom1p inhibits another protein , cdr2p , which in turn causes a third protein , cdk1p , to start the process of cell division . However , the details of the measurement process and the property that the cells are actually measuring—surface area , volume , mass or something else—remain mysterious . Pan et al . have now used imaging techniques and mathematical modeling to probe the distribution and movements of proteins in fission yeast cells . Their results do not support a previous model in which the cell uses the gradient of pom1p as a ruler to measure cell length . Rather , Pan et al . propose a new model in which the level of cdr2p is used to sense the size of the cell . Individual molecules of cdr2p come together to from clusters called nodes on the cell membrane . As the cell grows larger , more and more cdr2p proteins accumulate in these nodes , which are found in a band around the middle of the cell . When the cells reaches a critical cell size , the increased concentration of cdr2p at these nodes may help to trigger the start of cell division . By examining cells that grow at different rates , Pan et al . show that the rate of accumulation of cdr2p in the nodes depends on how big the cells are , rather than on the length of time that has elapsed . Analysis of fission yeast cells of different shapes shows that cell division starts when the surface area of the cell grows to a certain value , as opposed to starting when the volume or length reach a given value . Pan et al . also show that cdr2p binds to all parts of the cell membrane , not just to the nodes near the middle , and go on to provide a simple mathematical model showing how this property can allow cells to measure their surface area . However , as Pan et al . point out , this is probably just one component of a larger mechanism that tells cells when they need to divide .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Cortical regulation of cell size by a sizer cdr2p
Plant meristems carry pools of continuously active stem cells , whose activity is controlled by developmental and environmental signals . After stem cell division , daughter cells that exit the stem cell domain acquire transit amplifying cell identity before they are incorporated into organs and differentiate . In this study , we used an integrated approach to elucidate the role of HECATE ( HEC ) genes in regulating developmental trajectories of shoot stem cells in Arabidopsis thaliana . Our work reveals that HEC function stabilizes cell fate in distinct zones of the shoot meristem thereby controlling the spatio-temporal dynamics of stem cell differentiation . Importantly , this activity is concomitant with the local modulation of cellular responses to cytokinin and auxin , two key phytohormones regulating cell behaviour . Mechanistically , we show that HEC factors transcriptionally control and physically interact with MONOPTEROS ( MP ) , a key regulator of auxin signalling , and modulate the autocatalytic stabilization of auxin signalling output . The evolutionary success of multicellular organisms is based on the diversification of cellular identities and the division of labour among cell types . To orchestrate this diversity , complex signalling systems have evolved to guide stem cell differentiation based on hard-wired developmental programs and environmental signals ( reviewed in [Pfeiffer et al . , 2017] ) . Plants represent particularly attractive models to study the molecular mechanisms underlying the transition from stem cell to differentiated cell fate: Firstly , plants employ a postembryonic mode of development , which is based on the continuous activity of pluripotent stem cells embedded in specialized tissues , called meristems . Secondly , plant development is modular and thus the same set of organs is initiated repeatedly from a stem cell system , greatly facilitating in vivo analysis of cell-decision-making . Thirdly , due to the encasement by a cell wall , plant cells are immobile and thus their identity is determined by position , rather than lineage and can change multiple times during their development until terminal differentiation . In the shoot apical meristem ( SAM ) , the stem cell system responsible for the generation of all above ground structures , two major fate transitions can be identified: From stem cells in the central zone ( CZ ) to transit amplifying cells in the peripheral zone ( PZ ) and further on into organ primordia , which will give rise to fully differentiated lateral structures , such as leaves or flowers ( reviewed in [Gaillochet et al . , 2015] ) . At the molecular level , cell fate trajectories are instructed by an intertwined communication system between local transcriptional networks and non-cell autonomous phytohormone signals ( Brand et al . , 2000; Gordon et al . , 2009; Jasinski et al . , 2005; Leibfried et al . , 2005; Schoof et al . , 2000 ) . Stem cell fate in the SAM is dependent on the homeodomain transcription factor WUSCHEL ( WUS ) , whose RNA is expressed in the organising centre ( OC ) , located below the stem cells . WUS protein moves apically through plasmodesmata into the overlying cells , where it is required to maintain stem cell identity ( Daum et al . , 2014; Yadav et al . , 2011 ) . Stem cells in turn express CLAVATA3 ( CLV3 ) , a short , secreted peptide that acts to limit WUS expression via the CLV1 , CLV2 , CORYNE ( CRN ) , BARELY ANY MERISTEM ( BAM ) receptors system ( Bleckmann et al . , 2010; Clark et al . , 1997; Fletcher et al . , 1999; Nimchuk et al . , 2015; Ohyama et al . , 2009 ) . The resulting negative feedback loop represents the core module of SAM regulation and couples the size of the OC with that of the CZ ( Brand et al . , 2000; Schoof et al . , 2000 ) . In parallel , the KNOTTED-like homeobox transcription factor SHOOT MERISTEMLESS ( STM ) is required throughout the SAM to inhibit differentiation by stimulation of cytokinin production and repression of gibberellic acid ( GA ) biosynthesis ( Jasinski et al . , 2005; Long et al . , 1996 ) . Transcriptional and hormonal regulation interact to control SAM activity in spatially distinct sub-domains: In the centre of the SAM , WUS represses A-Type ARABIDOPSIS RESPONSE REGULATOR ( ARR ) genes , which encode for negative feedback factors in cytokinin signalling . In essence , WUS acts to sensitize the cellular environment to cytokinin , which in turn promotes WUS expression thus creating a positive feedback loop that helps to establish and maintain the OC ( Gordon et al . , 2009; Leibfried et al . , 2005 ) . Cytokinin also plays a key role in controlling cell proliferation in the SAM by stimulating CYCLIN D3 expression ( Riou-Khamlichi et al . , 1999 ) . In addition to its role in integrating hormonal signals , WUS directly represses the expression of HECATE1 ( HEC1 ) , which encodes a bHLH transcription factor that redundantly functions with its closest paralogs HEC2 and HEC3 in various developmental contexts ( Gremski et al . , 2007; Schuster et al . , 2015; Schuster et al . , 2014; Zhu et al . , 2016 ) . In line with this regulation , HEC1 mRNA is expressed throughout the SAM , but excluded from the OC ( Figure 2—figure supplement 1; Schuster et al . , 2014 ) . This pattern is faithfully translated into protein accumulation , since HEC1 protein displays limited intercellular movement ( Daum et al . , 2014; Schuster et al . , 2014 ) . Importantly , the precise spatial control of HEC1 activity by WUS is essential for SAM function , since uncoupling HEC1 expression from WUS in the OC leads to SAM termination , whereas enhancing HEC1 activity in stem cells leads to massive over-accumulation of these cells followed by a progressive repression of the core WUS/CLV3 feedback system ( Schuster et al . , 2014 ) . Mechanistically , this function is mediated by the formation of a protein complex between HEC and the bHLH transcription factor SPATULA ( SPT ) ( Schuster et al . , 2014 ) . In contrast to the long term maintenance of stem cells in the centre of the SAM , lateral organ primordia are continuously initiated at the periphery and cells are guided towards differentiation in restricted domains defined by the accumulation of auxin ( reviewed in [Weijers and Wagner , 2016] ) . Local auxin signalling maxima are dynamically formed through a combination of different mechanisms that include auxin biosynthesis and the controlled intracellular polarization of the auxin transporter PIN-FORMED1 ( PIN1 ) by the activity of the protein kinase PINOID ( PID ) ( Benjamins et al . , 2001; Pinon et al . , 2013; Reinhardt et al . , 2003 ) . Dynamic auxin maxima are translated into robust auxin signalling output by AUXIN RESPONSE FACTOR transcription factors , such as ARF5/MONOPTEROS ( ARF5/MP ) ( Bhatia et al . , 2016; Hardtke and Berleth , 1998 ) . MP associates with the chromatin remodelling factors SPLAYED ( SYD ) and BRAHMA ( BRM ) , forming a regulatory protein complex sufficient to promote floral fate over undifferentiated SAM fate ( Wu et al . , 2015 ) . In addition to primordia initiation , a boundary zone ( BZ ) , which surrounds the entire SAM like a ring is required for proper spatial separation of organs from the SAM ( reviewed in [Žádníková and Simon , 2014] ) . The BZ is dependent on the activity of the CUP SHAPED COTYLEDONS ( CUC ) and LATERAL ORGANFUSION ( LOF ) genes and loss of their function results in extensive fusions of lateral organs with the active SAM ( Aida et al . , 1999; Gaillochet et al . , 2015 ) . Importantly , the activity of the central domain of the SAM is tightly coupled with the specification of lateral organs at the periphery by inter-domain communication systems . First , MP relays auxin signals to the core WUS/CLV3 feedback loop by negatively regulating ARR7 ( Zhao et al . , 2010 ) . Second , organ primordia produce the short peptide CLE27 , which signals in the centre of the SAM through AtFEA3 and represses WUS expression ( Je et al . , 2016 ) . As a consequence of these multiple regulatory loops , the rate of lateral organ initiation correlates with the size of the SAM and larger SAMs tend to produce more organs per unit of time ( Landrein et al . , 2015 ) . Given the intertwined nature of the regulatory feedbacks , and their spatio-temporal deployment , the interpretation of the outcome for a particular regulatory interaction may not be intuitive . To overcome this limitation , several studies have used an iterative approach combining experimental quantifications and computational modelling that allowed to predict and reveal new regulatory nodes mediating SAM activity ( Besnard et al . , 2014; Chickarmane et al . , 2012 ) . From a dynamic point of view , cells produced in the central stem cell domain are continuously displaced towards the periphery by the divisions of cells located more centrally until they are incorporated into an organ primordium and finally differentiate . Hence , they undergo two major fate transitions: First , the transition from a slowly dividing stem cell to a more rapidly dividing transit amplifying cell and second , the transition from a non-committed amplifying cell to primordium cell by passing the peripheral zone . Despite the identification of key regulators controlling shoot stem cell activity , and patterning different domains of the SAM , our current understanding of how cell fate progression is modulated remains largely elusive . We used an integrated approach combining live-cell imaging , computational modelling and genome-wide profiling to show that HEC function modulates the stepwise fate transition from CZ cell to PZ and from PZ to lateral organ cell fate by coordinating the balance between cytokinin and auxin responses . In order to dynamically trace cell-behaviour at the SAM , we established an image analysis pipeline that allows identification and quantification of cells within specific sub-domains of the SAM . When we applied this tool to plants that lack HEC function ( hec1 , 2 , 3 triple mutants ) , we found that their SAMs were smaller and displayed a reduced cell number , in line with our previous results ( Figure 1A–D ) ( Schuster et al . , 2014 ) . Average cell size was not affected in SAMs of hec1 , 2 , 3 triple mutants , supporting that reduction in meristem size results from decreased cell number ( Figure 1E–F ) . Surprisingly , the average time interval between the initiation of successive lateral organs along the stem ( plastochron ) was substantially reduced in hec1 , 2 , 3 plants , despite the smaller SAM ( Figure 1G–H ) . The conflicting phenotype of a smaller stem cell system that produced more organs per time led us to hypothesise that rather than exclusively acting on stem cells ( Schuster et al . , 2014 ) , HEC genes may have a broader function to coordinate the acquisition of cellular identity and thus cell behaviour across different regions of the SAM . Since the regulatory network underlying SAM activity is strongly stabilized by feedbacks , whereas cell fate transitions are inherently dynamic , we decided to study the function of HEC genes by time-resolved live-imaging . To this end , we created lines , which allowed us to experimentally control HEC activity in space and time by fusing the coding sequence of HEC1 to the glucocorticoid receptor ( GR ) domain from rat via a flexible linker . The resulting HEC1-linker-GR protein remained trapped in the cytoplasm and its activity could be induced by treatment with the steroid dexamethasone ( dex ) . Given the limited capacity of HEC-linker-GFP for lateral cell-to-cell movement in the SAM ( Figure 2—figure supplement 1; Daum et al . , 2014 ) , inducing the even larger HEC1-linker-GR protein allowed us to assess HEC function locally in different domains of the meristem . First , we analysed HEC1 function in the CZ using lines carrying pCLV3:HEC1-linker-GR and imaged plants daily after dex or mock treatment . In contrast to the mock control , dex treatment caused a gradual enlargement of the SAM over time , which was in line with our observation of constitutive pCLV3:HEC1-linker-GFP lines ( Figure 2—figure supplement 2A , B ) . To investigate the regulatory mechanisms underlying the observed increase in SAM size , we quantified expression intensities and domain sizes of the key stem cell regulators WUSCHEL ( WUS ) and CLAVATA3 ( CLV3 ) by fluorescent markers , which faithfully label niche ( pWUS:3xYFP-NLS ) and stem cells ( pCLV3:mCherry-NLS ) . Interestingly , the increase in SAM size caused by stem cell-specific activation of HEC1 was accompanied by a transient enlargement of both the niche , or organising centre ( OC ) , and the CZ ( Figure 2A–G; Figure 2—figure supplements 3–5 ) . After an initial expansion of expression domains , we found that the intensities of WUS and CLV3 reporters were strongly reduced at day 4 , likely as an effect of a feedback mechanism restricting CZ and OC fate ( Figure 2—figure supplement 4 ) . This indirect negative effect of HEC1 on WUS and CLV3 expression explained our previous findings using stably expressing transgenes , which had shown a reduction in WUS and CLV3 expression ( Schuster et al . , 2014 ) and revealed that HEC1 promotes SAM fasciation via transient stimulation of WUS ( Figure 2A–C , Figure 2—figure supplements 3 and 4B ) . The observed short-term increase in stem cell number could result from different HEC1-dependent mechanisms: ( 1 ) a local increase of stem cell proliferation; ( 2 ) re-specification of early PZ cells into stem cells; and ( 3 ) a reduction in the differentiation rate between stem cells and the PZ . To discriminate between these scenarios , we developed a novel imaging tool combining a fluorescent timer protein ( Subach et al . , 2009 ) driven from a cell cycle dependent promoter ( pKNOLLE:fast-mFluorescentTimer-NLS ) with an ubiquitously and homogeneously expressed GFP ( pUBQ10:3xGFP-NLS ) . The timer protein was exclusively expressed during cytokinesis and then slowly matured from a blue form to a form that exhibits red fluorescence . Therefore , the ratio of the fading blue to stable green signal could be used as a readout for time passed since the last division , allowing us to assess the age of cells and thus served as a proxy for division frequency . Interestingly , we observed that although the SAM expanded , cell division activity after stem cell specific induction of HEC1 was decreased at the centre of the SAM and mostly confined to the PZ as in mock-treated plants ( Figure 2I–L; Figure 2—figure supplement 6; Materials and methods section ) . This demonstrated that HEC1 did not locally promote stem cell division . In contrast , we observed an increase in the mitotic index of the PZ at later stages ( day 3 and day 4 ) despite the fact that HEC1 protein is largely unable to move from cell to cell , suggesting that cell proliferation in the periphery was stimulated non-cell autonomously ( Figure 2L; Figure 2—figure supplement 6C ) . Next , we addressed fate re-specification of early PZ cells into stem cells . To this end , we analysed the ratio between CLV3 positive cells and all SAM cells after stem cell specific expression of HEC1 ( Figure 2E–H , Figure 2—figure supplement 5 ) . Interestingly , although the number of CLV3 positive cells strongly increased after dex treatment , the ratio between CLV3 and all SAM cells remained essentially unchanged ( Figure 2G–H ) . This argued against reprogramming of early PZ progenitor to stem cells by HEC1 , but rather supported the idea of coordinated cell behaviour in CZ and PZ ( Figure 2—figure supplements 3–5 ) . Taken together , these two lines of experiments excluded that cell proliferation in the CZ or re-specification of early PZ cells were the main drivers of HEC1-mediated meristem expansion and left us with a model in which HEC1 activity would locally inhibit the transition from stem cell to PZ fate , giving rise to a larger stem cell domain and consequently to an enlarged shoot meristem . This idea was in line with the hec1 , 2 , 3 mutant phenotype and suggested that in these plants , stem cells would transit more quickly from stem cell to PZ fate and further on to organ fate . To test this model , we next asked whether HEC1 could also interfere with the PZ to organ transition . Therefore , we increased HEC1 activity at the periphery and boundary zone ( BZ ) either by stable pCUC2:HEC1-linker-GFP or transient pCUC2:HEC1-linker-GR expression ( Figure 3A–B; Figure 3—figure supplement 1 ) . Strikingly , HEC1 induction at the periphery and BZ gradually supressed the emergence of lateral organs , eventually resulting in the formation of pin-like inflorescences , demonstrating that HEC1 can potently interfere with incorporation of cells into organs when expressed at the BZ ( Figure 3C , Figure 3—figure supplement 1G ) . To test the biological relevance of these results , we combined domain specific activation of HEC activity with constitutive loss of function of the interacting partner SPT , which we had shown to be required for HEC1 output in stem cells and flowers ( Gremski et al . , 2007; Schuster et al . , 2014 ) . Strikingly , the developmental phenotypes observed in distinct domains of the SAM were fully suppressed in the spt mutant background , including the formation of pin-like inflorescences after activation of HEC1 at the periphery ( Figure 3D–G ) . These results underlined a relevant function of the HEC-SPT protein complex in controlling SAM dynamics ( Figure 3D–G ) . Taken together , HEC function appeared not only to control stem cell to PZ , but also to be required and sufficient for PZ to organ fate transitions together with its partner SPT and thus acted as a central gatekeeper for cell fate progression throughout the SAM . Our results from live-cell imaging had shown that HEC genes were able to control CZ to PZ as well as PZ to organ primordia cell fate transitions , and promote cell proliferation at the SAM periphery . While consistent with our observation of hec1 , 2 , 3 mutants having smaller meristems that initiate more lateral organs , these findings were insufficient to formally explain the changes in cell behaviour underlying HEC loss-of-function . Due to the static nature of the triple mutant , which precluded time resolved experimental analysis , we developed a computational model to elucidate the role of HEC factors in regulating the dynamics of cell fate progression at the SAM by simulations . First , we established and calibrated a cell population model by defining model parameters based on published data and quantitative in vivo imaging results of stem cell , peripheral cell and organ numbers ( Figure 4A ) . We specified primordia initiation rate ( 2 . 3 per day; [Figure 1H] ) , CZ and PZ proliferation rate ( 39 . 8 and 18 . 3 hr respectively; [Reddy et al . , 2004] ) and primordia separation time after their initiation ( 2 . 2 days; [Besnard et al . , 2014] ) . Using these empirically derived parameters , the calibrated model was robust to perturbations and converged to a unique dynamic state of balanced cell proliferation , organ formation and separation ( Figure 4B; Figure 4—figure supplement 1A ) . We then simulated the HEC loss-of-function scenario by increasing the differentiation rate between the CZ , PZ and primordia and compared the resulting dynamics with time-resolved data obtained from in vivo SAM imaging . Although the resulting CZ and total SAM cells number fitted our experimental measurements , the cumulated number of primordia did not increase as observed in hec1 , 2 , 3 mutant plants ( Figure 4C ) . Thus , we further tested the impact of modulating the primordia initiation rate on SAM cell behaviour . By combining an increased CZ to PZ transition with an increased primordia initiation rate , our simulations could reproduce experimental measurements for the number of CZ cells , SAM cells and for the cumulated number of lateral organs ( Figure 4D ) . Furthermore , these model simulations predicted that increasing initiation rate would lead to a larger number of unseparated organ primordia at any given point in time , which could experimentally be tested by assessing the number of auxin output maxima as a proxy . Therefore , we introduced the auxin output reporter pDR5v2:3xYFP-NLS ( Liao et al . , 2015 ) into the hec1 , 2 , 3 triple mutant and quantified the number of auxin output maxima ( Figure 4E ) . However , in contrast to the model prediction , we did not observe a major increase in the number of DR5 positive domains in SAMs of hec1 , 2 , 3 compared to wild type ( 5 . 25 in wild type; 5 . 70 in hec1 , 2 , 3; n > 16 ) ( Figure 4E´´´ ) . While the difference was statistically significant , it was substantially smaller than predicted by the model and insufficient to explain the increase in organ formation ( Figure 4D ) . Hence , we needed to test additional factors in our model for their contribution to the hec loss-of-function phenotype . Having experimentally assessed the contribution of organ initiation rate in hec1 , 2 , 3 now allowed us to fix this important parameter for wild type and mutant plants and to further explore the parameter space for the time organs take to separate from the SAM after initiation . Indeed , reducing the time from initiation to separation of organs from 52 hr in wild type to 42 hr in hec1 , 2 , 3 combined with a slight increase in the rate of organ initiation in the mutant resulted in simulations fitting all our experimental results ( Figure 4F ) . Importantly , our simulations not only qualitatively captured the dynamics of HEC loss-of-function meristem but also allowed us to compute that in hec1 , 2 , 3 , primordia were initiated at a 15% higher rate and separated 10 hr earlier from the SAM than in wild type , supporting the idea that HEC function modulates the dynamics of stem cell differentiation . To further test our model , we next simulated HEC1 gain-of-function experiments in the CZ ( Figure 4—figure supplement 1B–E ) . In line with our hypothesis and with the results of the quantified mitotic index , combining a reduction in stem cell to peripheral fate transition with an increase in peripheral cell proliferation was sufficient to recapitulate our experimental observations ( Figure 4—figure supplement 1F ) . On the other hand , delaying CZ to PZ transition only , increasing the proliferation rate in the PZ only or introducing a re-specification of PZ cells into CZ cells did not reproduce our in vivo data ( Figure 4—figure supplement 1B–E ) . Taken together , applying reiterative cycles of experimentation and modelling allowed us to derive a quantitative framework of HEC function in the SAM . Simulating loss- and gain-of-function experiments , our model faithfully captured the dynamics of the SAM and quantitatively supported our hypothesis that HEC factors modulate the rate of cell differentiation at the shoot meristem at multiple levels . After having shown that HEC genes control SAM cell behaviour , we wondered what the underlying molecular mechanisms might be . The observation that HECs locally inhibited cell fate progression and non-cell autonomously stimulated proliferation suggested that they could modulate cell-to-cell communication . Thus , we first analysed the response of the auxin and cytokinin systems , two key phytohormones controlling cell fate and cell proliferation at the SAM ( reviewed in [Gaillochet et al . , 2015] ) . To investigate the involvement of cytokinin , we created gain- and loss-of-HEC-function plants that carried the pTCSn:erGFP cytokinin output reporter ( Zürcher et al . , 2013 ) . In contrast to WT , hec1 , 2 , 3 triple mutants displayed a substantial reduction in pTCSn:erGFP signal specific to the SAM , whereas TCS activity in the root was unchanged ( Figure 5A–C; Figure 5—figure supplement 1A–B ) . Conversely , increased HEC1 activity in stem cells led to a significant expansion of the central cytokinin-signalling domain , which was concomitant with SAM enlargement ( Figure 5D–F; Figure 5—figure supplement 1C ) . Induction of HEC1 at the PZ and BZ did not locally promote cytokinin signalling , highlighting the domain-specific activity of HEC factors ( Figure 5—figure supplement 1D–E ) . To further investigate the functional interaction between HECs and cytokinin , we tested whether promoting cytokinin signalling was sufficient to rescue the reduction in SAM size observed in hec1 , 2 , 3 plants . Therefore , we enhanced cytokinin signalling either chemically or genetically by 6-Benzylaminopurine ( BA ) treatment or removal of AHP6 , a negative component of cytokinin signal transduction ( Besnard et al . , 2014; Mähönen et al . , 2006 ) , respectively . In our growth conditions , both chemical and genetic stimulation largely suppressed SAM size defects of hec1 , 2 , 3 SAMs ( Figure 5G–H ) . Permanent inactivation of AHP6 by the ahp6-1 mutation caused SAM expansion in both wt and hec1 , 2 , 3 , but the mutants responded more strongly , as shown by the reduced SAM size difference between WT and hec1 , 2 , 3 ( Hedges´ g coefficient decreased from 0 . 20 to 0 . 15 ) ( Figure 5H ) . Treatment with 50 µM BA for 8 days showed a similar trend with SAM expansion in both genotypes and a more pronounced response in hec1 , 2 , 3 ( Hedges´ g coefficient decreased from 0 . 11 to 0 . 06 ) ( Figure 5G ) . While after mock treatment mutants exhibited SAMs of 93% the wt size , BA treatment almost fully supressed this phenotype and hec1 , 2 , 3 SAMs were now increased to 97% , which was not significantly different from treated wt apices ( Figure 5G ) . Together , these results showed that HEC function was sufficient and required to promote cytokinin signalling , which subsequently affected SAM size . This also suggested a mechanism for the expansion of the OC and CZ after stem cell specific induction of HEC1: A non-cell autonomous stimulation of cytokinin signalling by HEC1 could trigger the activation of WUS , which in turn would promote stem cell fate . To analyse the interplay between HEC activity and auxin signalling , we next monitored auxin sensing and downstream transcriptional output using the R2D2 and pDR5v2:3xYFP-NLS reporters ( Liao et al . , 2015 ) , respectively , in HEC gain- and loss-of-function plants . Our analysis showed that the topology of auxin signalling input and output in the SAM was only mildly changed in hec1 , 2 , 3 compared to WT ( Figure 4E; Figure 6—figure supplement 1 ) , suggesting that auxin signalling does not critically depend on HEC function . However , the observation of a small but significant increase in the number of DR5v2 positive auxin maxima ( Figure 4E´´´ ) , led us to hypothesise that HEC factors might impinge on the auxin feedback system and thus may quantitatively modulate signalling output . To test this hypothesis , we recorded auxin responses after induction of HEC1 at the SAM periphery using pCUC2:HEC1-linker-GR . Consistent with the increase in auxin output observed in hec1 , 2 , 3 mutants , boosting HEC activity at the periphery led to a substantial reduction in the number of DR5 signal maxima , eventually bringing about a complete collapse of lateral organ initiation ( Figure 6A–C; Figure 6—figure supplement 2A–C ) . Importantly , inducing HEC1 in stem cells repressed auxin perception locally but did not change auxin responses at the site of primordia initiation , demonstrating that HEC1 controls auxin signalling strictly cell autonomously ( Figure 6—figure supplement 2D–G ) . Taken together , these results showed that HEC function regulates cellular sensitivity to cytokinin and auxin signals in a domain-specific manner . In addition to their critical role in the SAM , auxin and cytokinin determine cell fate acquisition in the root apical meristem ( RAM ) . However , in contrast to the SAM , auxin promotes stem cell fate and cytokinin signalling marks the entry into differentiation ( reviewed in [Gaillochet and Lohmann , 2015] ) . Since HEC genes are only very weakly expressed in the RAM ( Figure 6—figure supplement 3A ) ( Li et al . , 2016 ) , the root is ideally suited to test the ability of HEC1 to control cell fate transition through the modulation of auxin and cytokinin signalling independent of SAM specific feedback systems . To this end , we generated lines that combined inducible expression of HEC1 in proliferative tissues ( Figure 6—figure supplement 3B ) , including the root tip ( p16:HEC1-linker-GR ) , and the output reporters for cytokinin and auxin , pTCSn:erGFP and pDR5v2:3xYFP-NLS , respectively ( Figure 6—figure supplement 3C–R ) . Strikingly , we observed a reduction of RAM size and meristem cell number after induction of p16:HEC1-linker-GR , while cell division activity and cell size were unaffected ( Figure 6—figure supplement 3C–L ) . These developmental changes in the RAM correlated well with increased cytokinin and decreased auxin signalling at the transition zone ( Figure 6—figure supplement 3M–P ) . Furthermore , the elongation zone was substantially reduced as marked by the development of root hairs and we frequently observed ectopic periclinal divisions within the cortical layer two days after induction ( Figure 6—figure supplement 3F and Q–R ) . Taken together , these results demonstrated that HEC activity was sufficient to modulate phytohormonal balance in diverse cellular contexts independent of the regulatory environment and underlined its central role in regulating the crosstalk between auxin and cytokinin responses . After having shown that HEC1 likely works via modulation of auxin and cytokinin pathways at the SAM , we next aimed at dissecting the transcriptional regulatory network orchestrated by HEC factors , using HEC1 as a proxy . To this end , we used genome-wide profiling to identify early HEC1 response genes ( Figure 7—figure supplement 1A; Supplementary file 1 ) . First , we recorded HEC1 DNA binding pattern using ChIP-seq on a functional p35S:HEC1-linker-GFP line . We found 6930 binding regions of HEC1 in 5250 unique genes with 74 . 5% of the events located within 3 kb upstream of transcriptional start sites ( Figure 7A ) . The HEC1 DNA binding pattern was distinct from those of other bHLH transcription factors , suggesting that our ChIP-seq had indeed captured the chromatin binding universe of HEC1 ( Figure 7—figure supplement 1C ) . To complement the binding data , we recorded inflorescence-specific HEC1 response genes by RNA-seq analysis using micro-dissected shoot apices of our p16:HEC1-linker-GR line . We identified 957 significantly regulated genes after three hours of HEC1 induction by dex and 815 transcripts after induction by dex and simultaneous inhibition of protein biosynthesis by cycloheximide ( cyc ) ( p<0 . 05 ) ( Figure 7—figure supplement 1A; Supplementary file 1 ) . We were able to confirm the direct regulation of PIN3 , which we had previously shown , suggesting that our experimental strategy was successful ( Figure 7—figure supplement 1B ) ( Schuster et al . , 2015 ) . Surprisingly , we only found a few canonical components of the auxin or cytokinin signalling circuitries among the direct targets , suggesting that HEC regulators do not have switch-like properties for these pathways ( Table 6 in Supplementary file 1 ) . However , we found a significant overlap between HEC1-response genes and genes responsive to cytokinin , suggesting that these two regulatory pathways also converge at the molecular level ( Figure 5; Figure 7—figure supplement 1D ) . Next , we carefully analysed the identified binding regions and in line with the quality of our data-set , one of the most highly enriched DNA motif in HEC1 binding regions was a G-Box , the sequence known to be the preferentially bound by bHLH transcription factors ( E-value = 1 . 4 e-141 ) ( Figure 7B ) ( Lau et al . , 2014; Pfeiffer et al . , 2014 ) . Interestingly , auxin response elements ( ARE ) , the DNA cis-regulatory motifs targeted by ARF transcription factors to regulate auxin dependent gene expression , were also significantly over-represented under HEC1 peaks ( ARE , E-value = 5 . 0 e-61 ) . In contrast , GARP elements , bound by type B-ARRs , the cytokinin output transcription factors , were only mildly enriched ( E-value = 2 . 4 e-2 ) , which suggested a specific association between HEC1 and the promoter of auxin responsive genes ( Hosoda et al . , 2002 ) ( Figure 7B ) . To investigate the relevance of these interactions for meristem regulation , we analysed the promoters of all ARF and Aux-IAA factors expressed in the SAM ( Figure 7C; Figure 7—figure supplement 2; Supplementary file 1 ) ( Vernoux et al . , 2011 ) . Strikingly , we found that HEC1 bound to 23 out of 25 promoters of auxin signalling components known to be active in the SAM , a rate significantly higher than expected by chance ( Figure 7C; Figure 7—figure supplement 2 ) . Consistently , we observed that G-boxes and AREs found in HEC1 binding regions were significantly more closely spaced than expected from their relative positions across the whole genome ( Figure 7D ) . Along these lines , we found that the most frequently occurring distance between G-Boxes and AREs in HEC1 binding regions was less than 50 bp and in immediate proximity to the peak summit ( Figure 7D; Figure 7—figure supplement 1E ) . Examples for a promoter that exhibited such a close distance of HEC and ARF binding sites included the regulatory region of the auxin receptor TRANSPORT INHIBITOR RESPONSE 1 ( TIR1 ) ( Gray et al . , 2001 ) , ( Figure 7—figure supplement 1F ) . Furthermore , we found a significant enrichment of HEC1-regulated genes carrying an ARE and a G-box in their promoter ( Figure 7—figure supplement 1G–I ) . Taken together , these results indicated that HEC1 and ARFs bind to the same genomic regions , either in competition , independently , or as a complex . Given the strong but slow negative effect of HEC1 on auxin transcriptional output , we hypothesised that HEC1 could interfere with the positive feedback in auxin signalling ( Bhatia et al . , 2016 ) either by binding site competition or direct physical interaction with ARFs . To test this hypothesis , we analysed the potential for interaction between HEC transcription factors and MP , the key ARF orchestrating primordia initiation at the SAM ( Yamaguchi et al . , 2013 ) . Both Yeast-2-Hybrid and Bimolecular-Fluorescence Complementation assays robustly demonstrated a physical interaction between MP and HEC1 , HEC2 , or HEC3 , respectively and thus suggested that HEC factors could act as transcriptional modifiers for ARF activity ( Figure 8A–B ) ( Simonini et al . , 2016 ) . In line with these results , HEC1 and HEC2 , as well as their key cofactor SPATULA ( SPT ) were able to physically interact with BRAHMA in Yeast-two-Hybrid assays ( Figure 8C ) ( Efroni et al . , 2013 ) . Taken together , these results showed that HEC transcription factors genomically associate with ARF targets and suggested that in the context of the SAM , HECs might be part of a higher order protein complex that modulates MP activity . Given the slow repressive activity of HEC1 on auxin signalling and its association with MP , we next investigated what regulatory changes could mediate its impact on the auxin feedback system . To test this , we performed RNA-seq on micro-dissected shoot apices 14 hr after dex induction of p16:HEC1-linker-GR plants . In contrast to the set of early targets , we found that transcript levels of essential components for primordia initiation , such as MP itself , but also PID , a kinase required for proper polar localisation of PIN auxin transporters , were substantially reduced in this dataset ( Figure 8D ) . In line with this finding , we observed a dramatic decrease in MP-GFP protein accumulation at the boundary zone after HEC1 induction in this domain , demonstrating the relevance of this interaction in the SAM ( Figure 8E–F; Figure 8—figure supplement 1A ) . Importantly , the expression of the YUCCA auxin biosynthetic genes required for flower primordia formation was not affected by HEC1 , suggesting that HEC factors likely do not work via the modulation of auxin production ( Supplementary file 1; Cheng et al . , 2006 ) . Consistent with the decrease of MP-GFP accumulation and the global collapse of auxin output observed by DR5v2 , we found that PIN1-GFP polarity was severely disturbed after HEC1 induction ( Figure 8—figure supplement 1C–E ) . In contrast , PIN1-GFP expression levels remained stable or even increased ( Figure 8—figure supplement 1C–E; Bhatia et al . , 2016 ) . Consistently , the activity of the R2D2 auxin input sensor was also changed and domains of low auxin perception , usually restricted to the boundaries of the SAM , expanded substantially towards the centre of the SAM over time ( Figure 8G–H; Figure 8—figure supplement 1B ) . To test whether these dramatic alterations in auxin signalling caused by HEC1 also translated into stable modifications of cell fate , we analysed the expression of the boundary zone marker CUC2 after HEC1 stimulation , since the boundary is marked by a small , but stable local auxin minimum ( Bhatia et al . , 2016; Heisler et al . , 2005 ) . In line with the idea that HEC1 potently interfered with cell fate decisions at the periphery by disruption of localized auxin signalling , we found a massive expansion of pCUC2:3xGFP-NLS expression ( Figure 8I–J ) . The slow effect of HEC1 on the auxin input and output patterns in the SAM and the absence of auxin biosynthetic and signalling components among the direct targets were consistent with the hypothesis that HEC1 may interfere with auxin signalling via its positive feedback loop ( Bhatia et al . , 2016 ) . To test this more directly , we aimed at establishing epistasis between the auxin signalling system and HEC activity by using recent evidence that auxin can promote the expression of MP and PIN1 , which in turn leads to a stabilization of the signalling system ( Bhatia et al . , 2016; Heisler et al . , 2005 ) . Therefore , we stimulated auxin signalling by chemical treatment while at the same time boosting HEC1 expression and scored for expression of a key auxin signalling component , as well as for SAM phenotypes over time . In line with our RNA-seq data , we observed a significant reduction of MP expression 24 hr after HEC1 induction ( Figure 9A ) . However , co-treatment with auxin rescued MP mRNA expression levels , indicating that our approach to stabilize the auxin feedback system was successful ( Figure 9A ) . We next analysed the phenotypic outcome of this double perturbation . Strikingly , although pCUC2:HEC1-linker-GR and p16:HEC1-linker-GR inductions alone inhibited primordia initiation , auxin co-treatment suppressed this phenotype ( Figure 9C–J , Figure 9—figure supplement 1E ) . Importantly , this suppression was neither the result of reduction of the key cofactor SPT , nor from the inhibition of CUC2 promoter activity driving HEC1-linker-GR , further supporting that auxin acts downstream of HEC function during primordia initiation ( Figure 9B; Figure 9—figure supplement 1A–D ) . Together , these results demonstrated that HEC function is able to control the cell fate switch from peripheral meristematic cell to organ cell identity by locally interfering with auxin signalling , likely via its feedback system . Having delineated a core regulatory system for controlling the timing of cell fate transitions in the shoot meristem , we wondered about the developmental relevance for this layer of control . Since plant cell fate is strictly determined by position , the timing of cellular transitions is intrinsic to the system under stable and optimal growth conditions . However , under changing environments , the regulatory system needs to adapt the morphogenetic output to the available resources , while at the same time conserving the functional pattern of the SAM ( reviewed in [Pfeiffer et al . , 2017] ) . Thus , we hypothesized that HEC genes could contribute to the modulation of SAM activity and growth in response to the environment . To test this hypothesis , we challenged the nutritional status of wild type and hec1 , 2 , 3 mutants by shifting plants for 14 days to low light conditions ( 15 µmol m−2 s−1 ) just after bolting and assessed developmental responses at the SAM ( Jones et al . , 2017 ) . In line with previous studies , we observed that wild type plants displayed a substantially smaller SAM under low light conditions ( Jones et al . , 2017 ) , and additionally observed a three-fold reduction in cytokinin responses compared to plants grown under normal light intensity ( Figure 10A , C , E , F , H , J ) . Importantly , this reduction in the SAM size did not result from changes in cell size , suggesting that meristematic activity was reduced in these plants ( Figure 10—figure supplement 1A–F ) . In contrast to wild type , we did not observe significant changes in the size of SAMs in hec1 , 2 , 3 plants under low light , demonstrating that HEC genes are required for SAM adaptation to environmental changes ( Figure 10B , D , E ) . Interestingly , even in hec1 , 2 , 3 , cytokinin signalling was responsive to nutritional status and we observed a decreased TCSn reporter activity similar to wild type plants ( Figure 10F–J ) . This suggested that cytokinin signalling and SAM size might be uncoupled in the hec1 , 2 , 3 mutants . Taken together , this experiment revealed that HEC genes may have a key role during SAM adaptation to environmental challenges and suggested that regulating the timing of cell fate transitions might be important for developmental plasticity . Cellular fate decisions occurring at the shoot apical meristem have important implications for the establishment and maintenance of plant architecture . Using precise spatio-temporal perturbations of gene expression , quantitative live-cell imaging and computational modelling , we revealed that HECATE bHLH transcription factors modulate cell fate transitions and coordinate the dynamics of cell fate decisions across key developmental domains of the SAM by balancing cytokinin and auxin phytohormonal signals ( Figure 11 ) . In contrast to our earlier findings where using steady state end-point phenotypes we concluded that HEC function act partially independently of the WUS/CLV3 system and repress cytokinin signalling by promoting type-A ARRs expression ( Schuster et al . , 2014 ) , we now show that HEC function in the centre of the SAM not only interferes with the core WUS/CLV3 regulatory system , but also promotes the enlargement of the OC , CZ and the cytokinin signalling domain . These contradicting results can be reconciled by considering the dynamics of HEC function . Early HEC activity promotes CLV3 , WUS and cytokinin domain expansion and consequently elevates type-A ARR expression , which are primary targets of cytokinin signalling ( Bhargava et al . , 2013 ) . In turn , A-type ARRs act as negative regulators of cytokinin signalling and constitute a negative feedback dampening WUS expression ( Schuster et al . , 2014 ) . Furthermore , the additive regulation of SAM size by HEC and WUS function ( Schuster et al . , 2014 ) , together with the ability of HEC1 to ectopically promote cytokinin signalling in the root meristem , where WUS is not expressed , suggests that HEC function primarily acts on cytokinin signalling at the SAM and in turn promotes WUS expression . However , given the indirect regulation of both WUS and cytokinin , it will be important in future studies to further clarify the network topology and to identify the intermediate regulatory components mediating HEC regulatory function . These results highlighted the power of time resolved analyses coupled to transient perturbations in studying the SAM , which now allowed us to discriminate direct from indirect effects arising as a consequence of feedback mechanisms . Since traditional loss-of-function mutations are inherently stable and thus do not lend themselves to this type of approach , we employed computational modelling to test divergent regulatory scenarios , which could not be analysed experimentally . The model not only allowed us to identify processes that were sufficient to explain the experimental observations , but also helped to rule out alternative hypotheses , such as stem cell re-specification , for which simulations could not reproduce in vivo data . Modelling and experimentation suggested that the combination of a reduced stem cell system with increased organ initiation rate observed in hec1 , 2 , 3 triple mutants likely was caused by a faster differentiation of stem cells . This occurred at least two levels , namely an increased rate of primordia initiation , as well as a reduced time for the primordium to grow and separate from the SAM . While the regulation of cell fate transition dynamics has not received much attention in the plant field with the exception of the stomatal lineage ( Simmons and Bergmann , 2016 ) , a large body of work from animal model systems has addressed this issue ( Busch et al . , 2015; Maduro , 2010; Marciniak-Czochra et al . , 2009 ) . It has emerged that cells within a developmental lineage undergo specific phenotypic steps on their trajectory towards terminal differentiation , however whether fate decisions occur deterministically or rather stochastically is still unresolved and might strongly depend on the cell type and the developmental context ( Moris et al . , 2016 ) . Given the purely position-dependent fate regulation observed in plant shoots , cells do not progress along a deterministic cell fate trajectory , but rather acquire alternative identities until they reach their final position in an organ and fully differentiate accordingly . Consequently , the transition from stem cell to transit-amplifying cell and further on to primordium founder cell mainly pertains to a timing of differentiation rather than providing intrinsic information on final cell fate . We have found that HEC transcription factors act in accordance with this idea and modulate the relative speed of the successive cell fate transitions at the SAM rather than specifying a specific developmental outcome . Previous theoretical studies on plant stem cell systems have focused on pattern formation ( Chickarmane et al . , 2012; Espinosa-Soto et al . , 2004; Robinson et al . , 2011b; Yadav et al . , 2013 ) , morphogenesis ( Kierzkowski et al . , 2012; Kuchen et al . , 2012 ) or cell division behaviour ( Louveaux et al . , 2016 ) , and only few studies investigated how local signals can coordinate the growth of different functional domains ( Grieneisen et al . , 2007; Mähönen et al . , 2014 ) . While lacking 3-dimensional resolution for the sake of simplicity , our 2D cell population model has allowed us to provide a theoretical framework on how fate decision events are coordinated along stem cell differentiation trajectories and how affecting key transition checkpoints during this process quantitatively modulates dynamics of the stem cell system . In line with a multi-step model , we previously showed that HEC1 is expressed in all relevant domains of the SAM and that its expression is under direct control of WUS ( Figure 2—figure supplement 1A ) ( Schuster et al . , 2014 ) . Thus , in addition to specifying stem cell fate , WUS may play an additional role in facilitating the transition from CZ to PZ fate via transcriptional repression of HEC1 ( Schuster et al . , 2014 ) . Using local modulation of HEC activity , and given the low HEC1 protein mobility in the SAM , we also revealed that HEC factors do not promote cell proliferation locally but rather non-cell autonomously ( Daum et al . , 2014; Schuster et al . , 2014 ) . Interestingly , these changes in cell behaviour were reminiscent of the increased PZ cell number observed after local perturbation of the WUS/CLV3 feedback system in the CZ ( Reddy and Meyerowitz , 2005; Yadav et al . , 2010 ) . Although the mechanisms responsible for the communication between CZ and PZ are still unresolved , the convergence of HEC and WUS function in controlling cytokinin signalling , and its role in promoting cell cycle progression ( Riou-Khamlichi et al . , 1999 ) , points towards a potential function of cytokinin signalling in mediating this inter-domain communication . In addition to their role in regulating cytokinin , we found that HEC proteins modulate the auxin regulatory loop . One mechanism , which could be responsible for this effect , could be the physical interaction with the auxin response factor MP . MP plays a central role in stabilizing the auxin feedback system via non-cell autonomous control of PIN1 polarity towards the site of MP accumulation ( Bhatia et al . , 2016 ) . This self-reinforcing regulatory system dynamically builds auxin maxima and generates sites of high MP accumulation which subsequently trigger a switch to primordia fate ( Bhatia et al . , 2016 ) . Importantly , protein-protein interaction data suggest that HEC1 , HEC2 and SPT physically interact with the SWI/SNF chromatin remodelling ATPase BRAHMA ( BRM ) , which also operates in a protein complex with MP during primordia initiation ( Figure 8C; Wu et al . , 2015 ) . Although the mechanistic details of their interaction still remain elusive , we propose that the HEC-SPT complex could modify MP-BRM function by direct physical interaction and thereby could modulate the dynamics of the entire auxin feedback system . Consistently , a reduction of HEC function would result in increased MP-BRM activity , which in turn enhances the auxin feedback system to instruct the initiation of flower primordia at a higher rate . Alternatively , HEC-SPT complex could indirectly regulate the expression of key components of the auxin feedback system , independently of the physical interaction with MP or locally reduce the levels of available auxin . Both models could explain the fairly slow , progressive changes observed in the dynamics of auxin perception , transport and response after promoting HEC activity at the periphery of the SAM . It will therefore be important in the future to mechanistically dissect the function of the physical interaction between HEC factors and MP-BRM complex to further reveal how HEC function impacts on the auxin feedback dynamics . Similarly to the shoot stem cell system , the balance between auxin and cytokinin is essential to control the dynamics of stem cell differentiation at the root apical meristem ( Di Mambro et al . , 2017; Dello Ioio et al . , 2008 ) . Our finding that HEC function can ectopically shift this hormonal balance and can impact on the dynamics of RAM differentiation suggests that plant cells can read out hormonal inputs and integrate this information to specify their identity along their differentiation trajectory . Along the same lines , auxin and cytokinin are essential for several aspects of cambial activity , including restriction of stem cell fate , cambial cell proliferation , and xylem differentiation ( Bhalerao and Fischer , 2014; Brackmann et al . , 2017; Hejatko et al . , 2009 ) . However , in this context , the exact function of individual hormones on the progression of cell fate acquisition and the nature of their interaction still remains elusive . In addition to the integration of hormonal and transcriptional signals to control cell fate decisions , the SAM adjusts its activity in response to environmental signals including light or nutritional status ( Pfeiffer et al . , 2016; Jones et al . , 2017 ) . Although this dynamic process is crucial to understand the molecular basis for plant developmental plasticity , the regulatory network mediating SAM homeostasis remains poorly characterized . Our findings that HEC function is required to adjust SAM size in response to low light suggests that it defines a regulatory hub in integrating environmental cues at the SAM . It will be important to further characterize the molecular mechanism underpinning this response and to unravel how HEC function interact with light signalling components during SAM activity ( Zhu et al . , 2016 ) . Furthermore , it will be important in the future to systematically test the role of known stem cell regulators and assess their regulatory function in the SAM upon various environmental challenges including temperature , nutrients , light intensity or biotic interactions . These experiments could reveal the mechanisms of developmental plasticity and how the regulatory landscape of the shoot stem cell system adjusts and rewires in response to the environment . pCLV3:HEC1-linker-GR , pCUC2:HEC1-linker-GR and p16:HEC1-linker-GR constructs were generated by ligation of HEC1 coding sequence ( CDS ) with a 33 aa Serine-Glycin linker and GR tag into pDONOR221 vector and recombined in pGreenIIs constructs ( Schuster et al . , 2014 ) . pKNOLLE:fast-mFluorescentTimer-NLS was generated from fast m-FluorescentTimer CDS fused to N7 NLS and introduced by subsequent BP and LR reactions ( Thermo Fischer Scientist , Waltham , Massachusetts , USA ) in a pGreenIIs vector containing 2 . 1 kb of genomic sequence upstream of the KNOLLE start codon . The CUC2 promoter used correspond to the 3 . 2 kb genomic sequence upstream the ATG . pCLV3:HEC1-linker-GFP , pCUC2:HEC1-linker-GFP , pCUC2:3xGFP-NLS and p35S:HEC1-linker-GFP were cloned using the Green Gate system ( Lampropoulos et al . , 2013 ) . N7-NLS was used as NLS tag ( Daum et al . , 2014 ) . For Yeast-two-Hybrid and Bi-Fluorescence complementation ( BiFC ) assay , HEC1 , HEC2 , HEC3 , MP , BDL and PEP CDS were PCR amplified using Phusion Taq-polymerase ( New England Biolabs , Inc . , Massachusetts , USA ) and subsequently cloned by Gibson assembly in pGILDA/pB42AD ( Yeast-two-Hybrid ) ( Gibson , 2011 ) or ligated in pGreenII0179 ( SPYCE constructs ) or pGreenII0229 ( SPYNE cassettes ) via NotI ( BiFC complementation ) ( Rodríguez-Cazorla et al . , 2015; Waadt et al . , 2008 ) A detailed list of primers used in this study can be found in Supplementary file 2 . Plants were transformed with Agrobacterium tumefaciens ASE strain by floral dipping according to standard protocols . All HEC1-GR inducible lines are homozygous for a single T-DNA insertion and were subsequently crossed to the corresponding reporter lines . pWUS:3xYFP-NLS_pCLV3:mCherry-NLS ( Pfeiffer et al . , 2016 ) , pTCSn:erGFP ( Zürcher et al . , 2013 ) , pPIN1:PIN1-GFP ( Heisler et al . , 2005 ) , hec1 , 2 , 3 ( Schuster et al . , 2014 ) , pDR5v2:3xYFP-NLS , R2D2 ( Liao et al . , 2015 ) , wus-1_pWUS:WUS-GFP ( Daum et al . , 2014 ) were previously described . hec1 , 2 , 3 were PCR genotyped as described in ( Schuster et al . , 2014 ) . All plant lines generated in this study are in the Col-0 background . Plants were grown at 23°C , 65% humidity under long day conditions ( 16 hr light/8 hr dark ) with LED lights or white lights ( Philips , Amsterdam , Netherlands ) at approximately 200 µmol m−2 s−1 . For the light shift experiments , plants were grown under white light at 200 µmol m−2 s−1 until flowering transition . As soon as the first flower primordia were observed , plants were transferred to low light intensity regimes ( approximately 15 µmol m−2 s−1 ) and kept for 14 days before imaging . For dexamethasone treatment ( Sigma_D4903 , St . Louis , Missouri , United States ) of the shoot apices , a solution of 10 µM dex , 0 . 01% ethanol and 0 . 015% Silwet were manually sprayed and applied on top of the inflorescence meristem of 25–30 DAG plants . Mock treatment ( 0 . 01% ethanol and 0 . 015% Silwet ) was conducted similarly . For 1-Naphtalenacetic acid ( NAA , Sigma , St . Louis , Missouri , United States ) treatment , 1 mM was applied according to previous studies ( Heisler et al . , 2005 ) . Shoot apices were treated once at the first day of the experiment . Cytokinin treatment was performed by treating inflorescences with a solution of 50 µM 6-Benzylaminopurine ( BA , Sigma , St . Louis , Missouri , United States ) supplemented with 0 . 015% Sylwet once every 5 days . Inflorescence meristems were analysed after 8 days . The cumulated silique number was measured by counting over time the total number of flower above stage 15 emerging after plant bolting . The inflorescence plastochron was then obtained by calculating the average time separating the emergence of 2 successive siliques . The number of flower primordia at a given time point were counted up to flower stage 2 . For root meristem analysis , plants were grown vertically on 0 . 5x MS ( Duchefa , Haarlem , The Netherlands ) , 0 . 8% Phytoagar plates under long day conditions ( 16 hr light/8 hr dark ) with white lights ( Philips , Amsterdam , Netherlands ) . 3DAG seedlings were subsequently transferred and grown vertically on media supplemented with ethanol ( 0 . 01% ) and with or without dex ( 10 µM ) . The EGY48 yeast strain ( -Ura ) was cotransformed with the combination of pGilda and pB42AD constructs under study or with the corresponding empty vector or negative control . Positive colonies were selected on solid media ( –Ura , -His , -Trp + glucose ) and verified by PCR . Induction for testing protein-protein association and colorimetric assays on plates were assayed as decribed in ( Ripoll et al . , 2015 ) Empty versions of pGreen0179 + SPYCE or pGreenII0229 + SPYNE were used as controls and T-DNA binary vectors were transformed into the Agrobacterium strain AGL-0 . Nicotiana benthamiana leaves were infected and YFP fluorescence assayed 72 hr after inoculation under a Nikon Eclipse TE2000-U epifluorescence microscope . The reciprocal assays for all the BiFC interactions shown in this study were performed obtaining the same results as presented in Figure 7A–B ( data not shown ) . Tobacco leaves used for these assays were also co-injected with the Agrobacterium strain expressing the viral suppressor p19 ( Voinnet et al . , 2003 ) . For Chromatin immuno-precipitation followed by sequencing , we used functional homozygous p35S:HEC1-linker-GFP in the hec1 background and performed individual ChIP as previously described ( Schuster et al . , 2015 ) . Individual samples were then pooled ( 10 to 12 individual ChIP/replicate ) by precipitation with 50 µl NaAc , 10 µl Acrylamid and 1 ml ethanol and incubated at −80°C for overnight . Samples were then centrifuged for 1 hr at 4°C and air-dried in a sterile bench before being resuspended in sterile water . Biological duplicates were analysed . Raw data has been deposited at NCBI GEO under the series number GSE94311 . RNA extraction was performed using the RNAeasy plant mini kit according to manufacturer instructions ( Qiagen , Hilden , Germany ) , on individually micro-dissected and pooled inflorescence meristems ( till flower stage 3–4; 15–20/replicate ) . For qRT-PCR , cDNA was prepared using cDNA synthesis kit after DNase treatment ( Thermo Fischer Scientist , Waltham , Massachusetts , USA ) , qPCR was performed using SYBR Green kit ( EurX , Gdańsk , Poland ) . For RNA-seq and Chip-seq , libraries and next-generation sequencing were performed according to standard protocols ( core sequencing facility , Bioquant , Heidelberg University ) . For RNA-seq , biological triplicates were analysed . Raw data has been deposited at NCBI GEO under the series number GSE94311 . All confocal pictures were acquired using Nikon ( Minato , Tokyo , Japan ) A1 Confocal with a CFI Apo LWD 25x water immersion objective . For time series analysis , settings were established at the first day on mock samples and were kept during the course of the experiment . Shoot meristems were manually dissected by cutting of the stem , removing the flowers and were counterstained with 1 mg/ml DAPI . Root meristems were counterstained with propidium iodide ( Sigma , St . Louis , Missouri , United States ) at 0 , 1 mg/ml . Individual populations of 5 to 15 plants were analyzed daily . To quantify pWUS:3xYFP-NLS , pCLV3:mCherry-NLS , pTCS:erGFP intensity plot profiles , we used Fiji software ( Schindelin et al . , 2012 ) . Z-stacks were first averaged using gaussian blur , and used to create a maximum projection picture . Intensity plot profiles were then generated using a ROI ( line ) crossing the SAM on its centre ( Figure 2—figure supplement 3 ) . To quantify WUS , CLV3 and TCS domain size , intensity plot profiles from maximum projection pictures previously generated were analysed . The size of the domain was obtained by measuring the distance including all points displaying an intensity value higher than one quarter of the maximum intensity value . SAM size was measured using the Nikon A1 software , by averaging three diameter segments starting from primordia 1 ( P1 ) , P2 and P3 and crossing the meristem at its centre . For quantifying the number of cells in the SAM , pre-processing , segmentation and data analysis was done using a customized workflow for the KNIME Image Processing platform ( KNIP ) ( Berthold et al . , 2008 ) . 3D visualization of analysed image stacks was done using the Fiji 3D viewer ( Schindelin et al . , 2012 ) . To obtain overall cell numbers a ubiquitously expressed reporter ( pUBQ10:3xGFP-NLS ) was used for imaging . To count numbers of CLV3-expressing cells , this reporter construct was combined with a respective stem cell-specific transcriptional reporter pCLV3:mCherry-NLS . For image processing , meristem image volumes were background subtracted and segmented by a 3D seeded watershed algorithm provided by the KNIP package ( Berthold et al . , 2008 ) . Using different pre-processing the same image stack was used to create a 3D mask for the whole meristem using again 3D seeded watershed . Borders between meristem and emerging flower primordia were marked manually to obtain a 3D mask that was used to filter out nuclei residing outside of the inflorescence meristem . To quantify numbers of CLV3-expressing stem cells nuclear segments from the pUBQ10:3xmCherry-NLS channel were used to obtain mean intensities in the CLV3 channels . Cells were considered to be positive for CLV3 if their respective mean nuclear intensity was higher than 35% of the maximum mean value . To analyse cell proliferation , a cell cycle-regulated transcriptional reporter was constructed ( pKNOLLE:fast-mFluorescentTimer-NLS ) and combined with a ubiquitously expressed reporter ( pUBQ10:3xGFP-NLS ) . Nuclear segments were obtained as described before and mean intensities for the timer ( blue channel ) and the GFP channel were measured for each segment . The ratio between the blue and the green mean intensities were normalized and assigned to four different classes ( Class1: 0–0 . 3 , Class2: 0 . 3–0 . 53 , Class3: 0 . 53–0 . 76 , Class4: 0 . 76–1 ) . Cells with the highest ratio ( Class4 ) represent young cells with a very recent cell division ( <4 hr , data not shown ) . To compare proliferation rates in the inner ( central ) and outer ( peripheral ) domain of the meristem a sphere of radius r was fitted through the centroids of the L1 cells of the meristem summit ( all L1 cells with a distance of <= 35 µm from a manually selected center point P of the meristem ) using a Matlab function . To adjust the size of the inner ( central ) domain to overall meristem size all cells with a distance to P smaller then 0 . 33 * r were considered to belong to this central domain ( defined from the size of CLV3 domain in other plant lines ) , whereas cells with a distance larger then 0 . 33 * r were classified as peripheral cells . Root cortex cell number was quantified as described in ( Dello Ioio et al . , 2007 ) . Cumulative cortex cell length was quantified as described in ( Kang et al . , 2017 ) . The quality of the sequence files quality was first confirmed using FastQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc ) . For RNA-seq analysis , read alignment and peak calling was then performed using TOPHAT2 algorithms using default settings ( Kim et al . , 2013 ) . BAM files were converted to SAM files using samtools and read tables were constructed using HT-seq ( Anders et al . , 2015; Li et al . , 2009 ) . Next , individual HT-seq tables were combined in a common table and differential gene expression was calculated using EdgeR with p<0 , 05 as a cut-off for differentially expressed genes ( Robinson et al . , 2010 ) ( Supplementary file 1 ) . For ChIP-seq analysis , reads were aligned using BOWTIE2 algorithms using default settings and peak calling was performed with macs2 ( Langmead and Salzberg , 2012; Zhang et al . , 2008 ) . During peak calling , we limited the number of duplicated reads to 2 ( ‘keepdup 2’ ) . Next , 200 bp regions were defined around the peak summits and the overlapping intervals of the two biological replicates were intersected using bedtools "multiinter" function ( Quinlan and Hall , 2010 ) . The overlapping peaks were then annotated using Homer and used to locate the peaks in relation to gene model ( Heinz et al . , 2010 ) ( Supplementary file 1 ) . ChIP peaks were visualised using the Integrative Genomic Viewer ( IGV ) ( Robinson et al . , 2011a ) . For de novo motif identification , a 500 bp region around the overlapping peak summits was defined and used for de novo motif discovery using MEME-ChIP with JASPAR core 2016 as motif input ( Bailey et al . , 2009 ) . For comparing DNA-binding regions of HEC1 , SPCH ( Lau et al . , 2014 ) , PIF3 , PIF5 ( Pfeiffer et al . , 2014 ) , KAN1 ( Merelo et al . , 2013 ) and LFY ( Moyroud et al . , 2011 ) , 50 bp region were centered on peak summits . Regions with 80% or more overlap were defined as shared binding regions . To obtain the position of G-boxes and ARE across the genome , bed files were generated using IGV ( Robinson et al . , 2011a ) . To measure the relative distribution between G-box and ARE on the entire genome , the closest ARE for each G-box were detected using only open chromatin regions ( Zhang et al . , 2012 ) and distances were reported using the bedtools ‘closest’ function with –d option . To measure their distribution under HEC1 binding regions , G-boxes under HEC1 peaks were identified using bedtools ‘intersect’ function . For each G-box , the closest ARE in open chromatin was next detected ( Zhang et al . , 2012 ) . Distances were reported using bedtools ‘closest’ function and histograms were constructed using R ( https://www . r-project . org/ ) . To assess the percentage of HEC1 target genes carrying G-box , ARE or both and the distribution of these motifs in open chromatin regions , respective bed files were first used to generate lists of annotated genes using the homer function ‘annotate . peaks . pl’ . List of genes regulated by HEC1 ( p<0 . 05 ) were next used and intersected with the list of genes carrying G-box , ARE or both in open chromatin regions using Microsoft Excel ( Heinz et al . , 2010; Zhang et al . , 2012 ) . Significance tests were performed using two-sided Fisher's exact test . We developed a cell population model that takes into account the simplified geometry of the SAM . The model describes the evolution in time of different SAM structures , i . e . , CZ , incipient primordia and unspecified PZ cells . Cells are continuously displaced towards the periphery by the divisions of cells more centrally located , with cells located at the outer boundary of the CZ transiting to PZ , depending on local WUS concentrations . Incipient primordia are initiated near the central boundary of the PZ and primordia separate from the meristem at fixed time after their initiation . PZ cells that do not contribute to primordia contribute to longitudinal growth of the plant . The following key processes were considered: ( i ) proliferation of cells in CZ and PZ , ( ii ) fate transition from the CZ to the PZ and from the PZ to the organs , ( iii ) initiation of incipient primordia , ( iv ) separation of primordia from the meristem and ( v ) contribution of meristem cells to longitudinal growth . Different feedbacks were included in the model . As suggested by our in vivo data , the initiation frequency of incipient primordia and the transition from the CZ to the PZ , both depend on the CZ cell number . The model is formulated using a system of ordinary differential equations , which number may vary in time . A detailed description of the model is found in Supplementary file 3 .	Unlike animals , plants continuously generate new organs that make up their body . At the core of this amazing capacity lie tissues called meristems , which are found at the growing tips of all plants . Meristems contain dividing stem cells . The daughters of these stem cells pass through nearby regions called transition domains . Over time , they change – or differentiate – to go on to become part of tissues like leaves , roots , stems , shoots , flowers or fruits . Stem cell differentiation has a direct impact on a plant’s architecture and eventually its reproductive success . For crops , these factors determine yield . This means that understanding this aspect of plant development is central to basic and applied plant biology . Many factors required for shoot meristem activity have been identified , with a focus so far on the processes that control the identity of the cells produced . Now , Gaillochet et al . have asked which genes are responsible for controlling when stem cells in meristems differentiate . The analysis focused on the meristem that makes all the above ground parts of model plant Arabidopsis thaliana – the shoot apical meristem . Gaillochet et al . found that HECATE genes ( or HEC for short ) control the timing of stem cell differentiation by regulating the balance between the activities of two plant hormones: cytokinin and auxin . These genes promote cytokinin signals at the centre of the meristem , and dampen auxin response at the edges . This acts to slow down cell differentiation in two key transition domains of the shoot meristem . These new findings provide a molecular framework that now can be further investigated in crop plants to try to improve their yield . The findings also lay the foundation for studies of animals that may define common principles shared among stem cell systems in organisms that diverged over a billion years ago .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "developmental", "biology" ]	2017	Control of plant cell fate transitions by transcriptional and hormonal signals
Spontaneous slow oscillation-associated slow wave activity represents an internally generated state which is characterized by alternations of network quiescence and stereotypical episodes of neuronal activity - slow wave events . However , it remains unclear which macroscopic signal is related to these active periods of the slow wave rhythm . We used optic fiber-based calcium recordings of local neural populations in cortex and thalamus to detect neurophysiologically defined slow calcium waves in isoflurane anesthetized rats . The individual slow wave events were used for an event-related analysis of simultaneously acquired whole-brain BOLD fMRI . We identified BOLD responses directly related to onsets of slow calcium waves , revealing a cortex-wide BOLD correlate: the entire cortex was engaged in this specific type of slow wave activity . These findings demonstrate a direct relation of defined neurophysiological events to a specific BOLD activity pattern and were confirmed for ongoing slow wave activity by independent component and seed-based analyses . Slow oscillation-associated slow wave activity is characterized by a waxing and waning of spontaneous neuronal firing arising from neuronal populations in deeper cortical layers ( Chauvette et al . , 2010; Sanchez-Vives and McCormick , 2000; Stroh et al . , 2013 ) , influencing neuronal excitability as well as stimulus-response properties of neuronal networks throughout the brain ( McGinley et al . , 2015b; Petersen et al . , 2003; Steriade et al . , 1993c; 1993b ) . The slow wave rhythm as a ‘default mode’ of cortical network activity ( Sanchez-Vives and Mattia , 2014 ) stands in contrast to the rather desynchronized , persistent state during rapid-eye-movement ( REM ) sleep and active wakefulness , dominated by low amplitude and high frequency cortical activity ( Constantinople and Bruno , 2011; McGinley et al . , 2015a; Steriade et al . , 2001 ) . A similar persistently active neuronal population activity can also be maintained by sedation ( Constantinople and Bruno , 2011 ) . During slow wave activity , the corresponding low frequency ( <1 Hz ) component of field potential recordings reflects bimodality: active phases , in which cells are depolarized and silent periods with rather hyperpolarized membrane potentials ( Timofeev et al . , 2001 ) , likely influenced through neuromodulatory pathways ( Eggermann et al . , 2014 ) . This oscillation is highly similar across cortical regions ( Ruiz-Mejias et al . , 2011 ) , and also across species , from lizards to humans ( Achermann and Borbély , 1997; Buzsáki et al . , 2013; Eschenko et al . , 2006; Mölle et al . , 2002; Shein-Idelson et al . , 2016; Steriade et al . , 2001 ) , suggesting well-defined conserved roles ( Destexhe et al . , 2007 ) . It has been shown that slow waves are propagating , most of the times in anterior-posterior direction over the cortical surface , eventually recruiting large cortical areas ( Busche et al . , 2015; Massimini et al . , 2004; Stroh et al . , 2013 ) , resembling a travelling wave , likely mediated by recurrent excitatory activity ( Luczak et al . , 2007; Massimini et al . , 2004; Sanchez-Vives et al . , 2017; Stroh et al . , 2013 ) . In addition , slow wave activity – correlated to cortical slow wave events - can also be found in sub-cortical regions such as the thalamus ( Stroh et al . , 2013 ) and the hippocampus ( Busche et al . , 2015; Hahn et al . , 2006 ) , probably mediated by long-range excitatory connections ( Leong et al . , 2016 ) . However , it remains to be established whether slow wave dynamics monitored in a local neural population are correlated with global functional network dynamics captured by whole-brain measures . Slow wave activity of a spatially confined neural population can be detected by optic fiber-based calcium recordings ( Adelsberger et al . , 2014; Grienberger et al . , 2012; Stroh et al . , 2013 ) . Upon staining of cells with fluorescent calcium indicators such as the synthetic dye Oregon Green BAPTA-1 ( OGB-1; Stosiek et al . , 2003 ) or by transduction of neurons with the genetically encoded GCaMP6 ( Chen et al . , 2013a ) , these optical recordings exclusively monitor activity of stained or transfected cells . Besides neurons , OGB-1 also stains astrocytes ( Garaschuk , 2013 ) , yet population-based recordings are dominated by neuronal action potential related calcium influx ( Grienberger et al . , 2012; Grienberger and Konnerth , 2012 ) . Calcium transients acquired by the means of an optic fiber result from synchronous spiking of at least 30 cells at the recording site ( Grienberger et al . , 2012 ) and represent the integrated signals obtained mainly from neurons and the surrounding neuropil , as demonstrated by simultaneous two-photon calcium imaging ( Grienberger et al . , 2012; Schulz et al . , 2012 ) . Consequently , optic fiber-based calcium recordings are well suited for the recording of local slow wave activity . BOLD fMRI provides a measure of brain-wide hemodynamic signals . The BOLD signal reflects changes in blood oxygenation , regional cerebral blood flow , and regional cerebral blood volume , linking these hemodynamic parameters to neuronal activity by neurovascular coupling . Resting state fMRI ( Biswal et al . , 1995; Fox and Raichle , 2007; Hutchison and Everling , 2014 ) is increasingly employed to study large-scale correlations of brain activity fluctuations in low frequency bands . Neuronal correlates of spontaneous fluctuations in the BOLD signal have been identified by combining fMRI and LFP recordings at rest , subsequently correlating these two signals ( Chang et al . , 2013; Hsu et al . , 2016; Magri et al . , 2012; Pan et al . , 2013; Shmuel and Leopold , 2008; Thompson et al . , 2014 ) . However , those studies used resting state fMRI methods to spatially map signal fluctuations which correlate to infra-slow fluctuations of the LFP signal , but they did not detect fMRI signal responses directly related to individual slow wave events . Others correlated hemodynamic fluctuations with wide-field calcium imaging of excitatory neurons , convolving these two signals ( Ma et al . , 2016b ) or compared functional connectivity during different , temporally transient patterns of calcium activity ( Matsui et al . , 2016 ) . While these results support the notion of resting-state hemodynamics being coupled to underlying patterns of excitatory neuronal activity , particularly lower-frequency hemodynamic fluctuations were not well-predicted ( Ma et al . , 2016b ) . Evidence was found that globally detected calcium signals are linked to hemodynamic functional connectivity and that the spatial information of the cortical network functional connectivity may be embedded in the phase of global calcium waves ( Matsui et al . , 2016 ) . Nevertheless , it has not been investigated until now , which cortical network activity reflected by the hemodynamic BOLD signal underlies the active phase of the slow wave rhythm – the slow wave event . In order to investigate brain-wide BOLD correlates of locally occurring slow waves , these two signals have to be recorded simultaneously , and slow waves have to be detected in a spatiotemporally precise manner and directly related to BOLD activity . To achieve this aim , optic fiber-based calcium recordings are well-suited as they can be performed unperturbed by the magnetic field of the MR scanner ( Schmid et al . , 2016; Schulz et al . , 2012 ) . In this study , we examine BOLD hemodynamic responses upon individually detected , optically recorded slow calcium waves by employing event-related fMRI analysis which can be used to identify BOLD changes in response to individual events . The key concepts of event-related analyses are time-locking and signal averaging . Although originally intended to reveal transient changes in brain activation associated with the presentation of discrete sensory stimuli , event-related analysis can detect BOLD activity upon any type of previously defined events ( Huettel et al . , 2009; Josephs et al . , 1997 ) , in our case the active phases of the optically recorded slow calcium wave activity . In addition to the general linear model ( GLM ) based event-related method , we employ two multivariate exploratory approaches for fMRI data: seed-based and independent component analysis ( ICA ) . Here we show that a change of excitability state of a small population of neurons can be indicative for rather global network states being reflected by whole-brain BOLD activity patterns , demonstrating the particular relevance of considering readouts of local population activity for fMRI measurements . We reveal the interrelation of a neurophysiological defined slow wave event and a macroscopic , network organizing signal , and find a cortex-wide BOLD fMRI correlate . We employed an optic fiber-based approach to detect calcium transients simultaneously to fMRI scans within a 9 . 4 T small animal MR scanner ( Figure 1A ) ( Schmid et al . , 2016; Schulz et al . , 2012 ) . To monitor intracellular calcium as proxy of spiking , the synthetic dye Oregon Green 488 BAPTA-1 ( OGB-1 ) was injected stereotactically resulting in a column-like stained region with a diameter of about 600 µm using the multicell bolus loading technique ( Stosiek et al . , 2003 ) ( Figure 1B , C ) . The tip of the optical fiber with a diameter of 200 µm was implanted dorsal to the stained region ( Figure 1B ) in primary somatosensory cortex ( S1 front limb , S1FL ) . As mentioned above , OGB-1 stains both neurons and astrocytes , as confirmed by co-staining with the astrocytic marker SR101 ( Nimmerjahn et al . , 2004 ) ( Figure 1D ) . In addition , we employed the genetically encoded calcium indicator GCaMP6f under control of a pan-neuronal promoter hSyn ( Figure 1E ) and the CamKII promoter limiting expression to excitatory neurons ( Figure 1F , G , Figure 1—figure supplement 1 ) . Experiments were conducted under isoflurane anesthesia ( 1 . 1–1 . 8% ) inducing and maintaining slow wave activity with recurrent slow wave events at similar frequencies ( Chen et al . , 2013b; Grienberger et al . , 2012; Lissek et al . , 2016; Luczak et al . , 2007; Stroh et al . , 2013 ) . Calcium waves were reliably recorded during simultaneous fMRI recordings ( 8 out of 10 experiments in 7 out of 8 animals ) , occurring at frequencies ranging between 9 and 15 events/min ( mean 10 . 9 ± 1 . 3 events/min ) , in line with previous in vivo recordings ( Busche et al . , 2015; Kerr et al . , 2005; Stroh et al . , 2013 ) ( Figure 1H , I ) . For OGB-1 , the typical and rather uniform calcium waves exhibited sharp rise times ( 72 ± 14 ms , and rather variable durations and amplitudes of 1356 ± 85 ms and 2 . 6 ± 0 . 1 Δf/f , respectively ( n = 30 traces , five animals ) . For GCaMP6f ( hSyn ) the rise times exhibited slower kinetics ( 163 ± 14 ms , n = 30 traces , two animals , Figure 1J ) presumably due to four calcium binding sites of the calmodulin , which need to be occupied to reach maximum fluorescence change , while similar values for durations and amplitudes of slow waves were observed as compared to OGB-1 ( Figure 1K , L , M ) . To test , whether indeed OGB-1 and GCaMP6f both reveal the same slow wave events , we conducted two-fiber experiments outside of the MR scanner , in two animals expressing GCaMP6f under control of the CaMKII promoter in S1FL , and being injected with OGB-1 2 mm posterior to that site on the same hemisphere . ( Figure 1N ) . Indeed , we found the associated slow waves to be highly correlated ( Figure 1N , O ) . The calcium waves were separated by periods of network quiescence of variable durations ( Figure 1H , I , N , P ) , equivalent to down states in electrophysiological recordings ( Poulet and Petersen , 2008; Seamari et al . , 2007; Steriade et al . , 1993c; 1993b ) . In line with previous results ( Busche et al . , 2015; Stroh et al . , 2013 ) , the observed population calcium transients were correlated with electrically recorded slow oscillations in the local field potentials ( LFPs; Figure 1P , Q ) . Calcium slow wave events were identified using an adapted version of an algorithm established for the detection of slow oscillations in electrophysiological recordings in vitro and in vivo ( Seamari et al . , 2007 ) , separating slow oscillatory activity and silent periods based on exponential moving average ( EMA ) filters . We adapted and optimized this procedure for the identification of slow wave activity in optical calcium recordings in vivo . Applying this detection algorithm yielded an average number of 298 ± 20 slow wave events per 30 min experiment ( n = 7 animals; Figure 2B ) . The resulting onsets and durations of calcium slow waves were encoded as a binary array of events ( one ) and silent periods ( zero ) . These ‘slow wave vectors’ ( Figure 2A , C ) were used as a condition in an event-related fMRI analysis and were convolved and correlated with the fMRI signal to detect BOLD responses following the specified events defined in a design matrix . The temporal characteristics of the hemodynamic response ( HR ) elicited by slow wave activity remains insufficiently explored and due to the specific slow wave network activity may differ from HRs elicited e . g . by sensory stimulation . Therefore we first examined the shape of typical HRs elicited by individual slow wave events in cortical ROIs on left and right hemisphere ( Figure 2D , Figure 2—figure supplement 1 ) . Timecourse of averaged HRs exhibited a sharp rise from baseline to maximum during 0 . 7 s ± 0 . 4 s after the onset of a slow calcium wave , reaching the maximum signal amplitude of 0 . 17%±0 . 03% ( ΔSA ) at 6 . 5 s ± 0 . 4 s ( TTP ) with a half maximum duration ( HMD ) of 6 . 8 s +- 0 . 5 s . Next , we extracted HRs following the onsets of slow calcium waves from the 30 most active voxel of each measurement using a model-free approach based on a finite impulse response method ( FIR ) ( Dale , 1999; Glover , 1999 ) which approximates the underlying timecourse of HRs without any a priori assumptions ( Jansma et al . , 2013 ) ( Figure 2E ) . Subsequently , we used a leave-n-out approach analyzing datasets with averaged HRs extracted solely from the remaining animals , to avoid circular analysis ( Kriegeskorte et al . , 2009 ) ( Figure 2F ) . For this purpose , the resulting HRs were averaged and used as hemodynamic response functions ( HRFs ) convolved with the event arrays generated from the calcium waves to create activation maps reflecting ongoing BOLD activity during slow oscillation-associated slow wave activity ( Figure 2G ) . We used event-related analysis ( Josephs et al . , 1997 ) , based on either the extracted HRF or the FIR-derived HRF ( see Materials and Methods ) to correlate onsets and durations of locally detected slow calcium waves ( Figure 3A , B ) with the simultaneously acquired fMRI BOLD signal . Event-related fMRI analysis accounts for responses to each predefined event which is specified in terms of its onset and duration ( Huettel et al . , 2009 ) . Using the detected individual slow wave events in the calcium signal in an event-related analysis based on extracted HRFs revealed BOLD activation of nearly the entire cortex during slow wave activity ( Figure 3C , D; Figure 3—figure supplement 1 ) in 6 of 8 experiments ( n = 7 rats ) . Although cluster sizes varied over animals and experiments , the activation clusters were spanning almost all slices for all experiments showing BOLD activation ( Figure 3C , D; Figure 3—figure supplement 2 ) . While activation patterns were occasionally patchy ( Figure 3—figure supplement 1C; Figure 3—figure supplement 2 ) , each cortical area was showing at least partial BOLD activation predicted by slow calcium waves in every experiment , i . e . no cortical area was excluded from BOLD activity . We next asked , whether the BOLD activity quantitatively differs between cortical areas . Therefore , we analyzed bilateral ROIs drawn on the F-maps derived from the event-related analysis employing HRFs obtained from a FIR analysis previously performed in a different set of animals , both in somatosensory and visual cortex ( Figure 3E ) . The F-values of the two regions of the cortex showed no significant differences in mean values ( Figure 3F ) . In addition , we extracted the HR of the two regions from the beta-maps , no apparent differences in the shape of the HRs could be observed ( p=0 . 23 , Wilcoxon rank-sum test ) , indicative of a rather uniform activation profile throughout the cortex . These results suggest that locally recorded slow calcium wave events have a cortex-wide correlate in the fMRI BOLD signal representing global slow wave activity . To confirm that the resulting HRs do not differ depending on the dataset used for extraction , we performed a cross-validation procedure and analyzed datasets with two different previously extracted HRs and compared cluster sizes and T-values ( Figure 2—figure supplement 3 ) . For this analysis we used mean HRs extracted from at least two other animals . Next , we compared our individualized HRF estimation based on the experimental slow wave data with the canonical HRF , routinely employed in fMRI data both in humans ( Glover , 1999; Lindquist and Wager , 2007 ) and rats ( Amirmohseni et al . , 2016 ) , by applying the SPM standard procedure which uses a continuous canonical HRF model ( Figure 3—figure supplement 2A , B ) . We compared T-values in the previously defined ROIs and the approaches yielded comparable results ( Figure 3—source data 1 ) . To further test for the specificity and interrelation of a recorded calcium event array to the respective BOLD activation and to rule out a general effect of specifically spaced event-timing causing the reported effects , we applied two control procedures for this analysis: swapping regressors between the animals ( Figure 3—figure supplement 3A , B ) and temporally reverting the event arrays ( Figure 3—figure supplement 3C , D ) resulted in no BOLD activation in any of the data sets , thus confirming the specificity of the observed activations . As mentioned , slow oscillations-associated calcium waves recruit not only the cortex , but also the thalamus . We therefore conducted optic fiber-based calcium recordings in the posterior medial nucleus of the somatosensory thalamus ( POm; Figure 4A ) . As expected , we were able to record typical slow wave events , albeit at a lower SNR ( Figure 4B ) in two animals , with similar characteristics compared to the cortically recorded slow waves , in line with a previous study ( Stroh et al . , 2013 ) . We then extracted the HR as described earlier . Using those thalamic calcium waves as events , we found BOLD activation in the cortex only ( Figure 4E , F; 2 experiments in two animals ) , similar to the cortical recordings . This confirmed the tight synchronicity between cortical and thalamic calcium waves . The fact that no BOLD signal in thalamus was detected is most likely due to using a MR surface coil which results in lower detection sensitivity in subcortical regions . GLM-based fMRI analysis identifies voxel correlated with the model timecourse . Model-free analysis approaches do not require a priori knowledge about activation patterns and do not consider voxel individually to reveal potential voxel interactions . Consequently , in addition to the event-related GLM method , we employed independent component analysis ( ICA ) for ongoing slow wave activity . ICA aims to recover a set of maximally independent sources from their observed mixtures without knowledge of the source signals or the mixing parameters ( Hyvärinen and Oja , 2000 ) . With this analysis , during ongoing slow wave activity , we found ICA components mirroring the results obtained by using the slow calcium waves as regressor in the same animals ( Figure 5A ) . Notably , the timecourse of the ICA component revealing cortex-wide activation is highly correlated with the GLM regressor obtained from taking into account slow wave onsets detected in the calcium recordings ( R = 0 . 4 ± 0 . 05 , n = 9 animals; Figure 5B ) , further confirming the neurophysiological basis of this component . To test for specificity of those results we also calculated the same cross-correlation of the ICA component timecourse with the GLM-regressor , but with the GLM-regressor obtained from the temporally mirrored slow wave vector ( see Figure 3—figure supplement 3C ) of the same animal , which leads to a complete absence of correlation between the ICA component timecourse and the event-related GLM-regressor ( Figure 5C ) . As ICA decomposition can produce a substantial number of noise components , we extracted the component set ( number of components identified by minimum description length criterion ( MDL ) , see Materials and Methods ) for a group ICA ( Figure 5—figure supplement 1A ) , as well as for the corresponding single datasets ( Figure 5—figure supplement 1B ) . Notably , we could again identify pan-cortical activation patterns in both the group- and the single-subject based results ( ICs #3 , #6 in Figure 5—figure supplement 1 ) . In addition , with this analysis we could identify components related to typical resting state networks including auditory ( IC #1 ) and visual ( IC #8 ) cortex , striatum ( IC #7 ) and hippocampus ( IC #20 ) in line with others ( Jonckers et al . , 2011 ) . A comparison of the power spectra related to the timecourses of the pan-cortical and two other components shows that the pan-cortical component exhibits higher power in the <0 . 1 Hz range ( Figure 5—figure supplement 2A–D ) . Regressing out the timecourse derived by the slow wave vector obtained from the calcium transients ( see Materials and Methods ) , leads to complete absence of pan-cortical components in the ICA ( Figure 5—figure supplement 2E , F ) . Next , we asked , whether the cortex-wide BOLD activation is indeed specifically related to individually detected slow wave events . Therefore , we contrasted this activity , maintained by isoflurane anesthesia , with persistent activity , maintained by medetomidine sedation . This persistent activity is lacking slow calcium wave activity , i . e . bimodality ( Figure 5—figure supplement 3 ) . We confirmed the differential population activity under the two conditions by assessing both spontaneous as well as sensory-evoked calcium transients as previously described separately for both conditions ( Schmid et al . , 2016; Schulz et al . , 2012; Stroh et al . , 2013 ) . During persistent activity , we did not find any component indicative of a synchronous cortical activation ( n = 6 ) , but we found resting state networks as previously described ( Figure 5—figure supplement 4A , B ) ( Hsu et al . , 2016; Jonckers et al . , 2011; Kalthoff et al . , 2013; Lu et al . , 2012; Ma et al . , 2016b ) . To complement the ICA approach , we employed seed-based analyses . Using the somatosensory cortex as seed-ROI , during ongoing slow wave activity , we found a correlating cortex-wide BOLD activation ( Figure 6A ) . Furthermore , the average signal in the somatosensory seed-ROI was strongly correlated with the normalized signal of the pan-cortical component previously revealed by the ICA ( Figure 6B; Figure 6—source data 1 ) . During persistent activity this pan-cortical activity was absent ( Figure 6—figure supplement 1; Figure 6—source data 1 ) , but bilateral somatosensory activity , potentially reflecting inter-hemispheric connectivity between the somatosensory cortices ( Hodkinson et al . , 2016 ) could be detected ( Figure 6—figure supplement 1 ) . As we occasionally observed hippocampal BOLD ( Figure 3—figure supplement 1C; Figure 3—figure supplement 2A , B ) we applied the same seed-based approach to the hippocampal formation . Again , only during ongoing slow wave activity we found pan-cortical BOLD activation related to the hippocampal seed-ROI and also found a correlation to the pan-cortical ICA component ( Figure 6—figure supplement 2A–C , Figure 6—source data 2 ) . Finally , we conducted an optical calcium recordings-informed correlation analysis ( Figure 6C ) . When HRs upon slow calcium wave onsets in a ROI in somatosensory cortex are correlated with every remaining voxel in the brain , again a cortex-wide network becomes apparent . Independent of the location of the somatosensory ROI ( right or left hemisphere ) this correlation analysis revealed highest r-values for the entire cortex . Additionally , correlation values reveal that slow wave related BOLD activity recruits almost all cortical areas with nearly synchronous time-to-peak values for HRs ( Figure 6D; Figure 6—figure supplement 3 ) . Our results provide evidence for large-scale recruitment of cortical networks upon a specific type of neural activity – slow waves – which have previously been shown to be present under different types of anesthesia ( Busche et al . , 2015; Chauvette et al . , 2011; Petersen et al . , 2003; Sanchez-Vives et al . , 2017; Sanchez-Vives and McCormick , 2000; Seamari et al . , 2007; Stroh et al . , 2013; Zucca et al . , 2017 ) , and in natural slow wave sleep ( Massimini et al . , 2004; Sanchez-Vives et al . , 2017; Steriade et al . , 1993a; Steriade et al . , 1993b; Steriade et al . , 1993c; 1993b; Steriade and Timofeev , 2003; Steriade et al . , 2001 ) . Although the main features appear similar to those of the SWS oscillation ( Destexhe et al . , 2007; Sanchez-Vives et al . , 2017 ) , under anesthesia , slow waves appear more rhythmic and more synchronous across the cortex and also show longer silence periods ( Busche et al . , 2015; Chauvette et al . , 2011 ) . During natural sleep , spindle activity - which is related to memory consolidation - is grouped by slow waves ( Mölle et al . , 2002 ) , but this is not the case for anesthesia induced slow waves ( Murphy et al . , 2011 ) . Nevertheless sleep slow waves and anesthesia slow waves may recruit the same cortical and subcortical structures ( Murphy et al . , 2011 ) . Although cortically generated ( Steriade et al . , 1993c; Timofeev et al . , 2000 ) slow waves also engage the thalamus ( Sheroziya and Timofeev , 2014; Steriade et al . , 1993a; Stroh et al . , 2013 ) and hippocampus ( Busche et al . , 2015; Ji and Wilson , 2007; Sirota et al . , 2003 ) , suggesting that slow wave associated excitation plays a synchronizing role in functional coupling of remote brain regions ( Hahn et al . , 2006 ) . Indeed we occasionally observed hippocampal BOLD activation during slow waves and therefore conducted a seed-based analysis for an atlas-based hippocampal ROI , showing that voxel in the hippocampus correlate well with cortical voxel , again spanning the entire cortical surface during ongoing slow wave activity . This correlation of hippocampal activity to slow-wave related cortical BOLD ( Chan et al . , 2017 ) might provide a mechanism of coordination for reactivation and redistribution of hippocampus-dependent memories to neocortical sites ( Buzsáki , 1996; Mitra et al . , 2016 ) . Although this is unlikely the case for anesthesia related slow waves ( Murphy et al . , 2011 ) , they might recruit the same structures and thereby provide a spatiotemporal mechanism for such plasticity-relevant cortico-hippocampal functions ( Logothetis et al . , 2012 ) . Although the thalamus is being recruited during slow wave activity ( Sheroziya and Timofeev , 2014; Stroh et al . , 2013 ) , we did not observe robust thalamic BOLD activity , despite of optically recorded thalamic slow calcium waves . This might be due to a lower density of neurons carrying the slow wave rhythm , reflecting the lower overall cell density in the thalamus in comparison to the cortex ( Meyer et al . , 2013 ) or due to reduced sensitivity of the MR surface coil in deeper brain areas . Intracortically , slow waves can propagate as a travelling wave ( Busche et al . , 2015; Matsui et al . , 2016; Stroh et al . , 2013 ) , likely by relays of local populations of cells ( Destexhe et al . , 2007 ) . Depending on excitability state of the network , a slow wave event could either remain local ( Nir et al . , 2011; Vyazovskiy et al . , 2011 ) or could spread ‘smoothly like an oil-spot’ ( Massimini et al . , 2004; Sanchez-Vives et al . , 2017 ) . The recruitment of sub-cortical regions is discussed to rather involve long-range excitatory projections ( Leong et al . , 2016 ) . Although slow wave activity is a multiscale phenomenon ( Jercog et al . , 2017; Sanchez-Vives et al . , 2017 ) , previous work has focused either on local and cellular ( Chen et al . , 2013b; Grienberger et al . , 2012; Kerr et al . , 2005; Sanchez-Vives and McCormick , 2000; Stroh et al . , 2013 ) or on large-scale characteristics ( Massimini et al . , 2004 ) of slow waves . Here , we bridge these mentioned levels of observation by relating the mesoscopic representation of slow waves in a defined cortical or thalamic population - in a spatio-temporally precise manner - to the whole-brain hemodynamic BOLD signal . We contrast our findings obtained during slow wave activity to a network activity , maintained by light sedation , in which slow calcium wave events are absent . Indeed , the calcium recordings reveal distinct response properties of local sensory networks , matching earlier studies under the same condition ( Schmid et al . , 2016; Schulz et al . , 2012 ) . Here , seed-based analysis and ICA revealed typical default mode and resting state networks as described previously ( Hsu et al . , 2016; Lu et al . , 2012 ) . Notably , we did not find any component resembling pan-cortical activation as identified for slow wave activity , speaking for a direct relation of slow waves to this BOLD pattern . Furthermore , during slow wave activity , we could demonstrate that the ICA component showing pan-cortical activation is highly correlated to the GLM timecourse derived from optically detected calcium waves . Exclusively this pan-cortical component is related to higher power in the ultra-low frequency range . Such low frequency components were related to slow wave activity in spontaneous EEG/fMRI signal fluctuations in earlier studies and have been investigated in different species ( Leopold et al . , 2003; Schölvinck et al . , 2010; Steriade and Amzica , 1998; Wang et al . , 2012; Wen and Liu , 2016 ) ) including humans ( Achermann and Borbély , 1997; Boly et al . , 2012; Dang-Vu et al . , 2008; Hiltunen et al . , 2014; Marshall et al . , 2006; Massimini et al . , 2004; Wen and Liu , 2016 ) . Notably , if the timecourse derived from the slow wave vector is regressed out of the data , the ICA is devoid of any components showing pan-cortical activation . Nonetheless , also during ongoing slow wave activity , we can identify components which likely resemble resting state networks , which seems reasonable , given that these networks are discussed to mirror basic functional connectivity . Conventional model-based analysis for fMRI using the general linear model ( GLM ) solely identifies individual voxel significantly correlated with the chosen model timecourse . Data-driven , or model-free analysis as ICA , do not require a priori knowledge about activation patterns and furthermore consider interactions between voxel , as voxel are not considered individually . But it has to be noted that especially the ICA approach bears several significant drawbacks . For example this analysis can produce a large number of noise components . We used the minimum description length criterion ( Li et al . , 2007 ) to estimate the number of components , but also alternative techniques , as e . g . Bootstrap Stability Analysis ( Majeed and Avison , 2014 ) are available . Still , this approach has to deal with the trade-off between extracting enough components to actually reveal all signals of interest and adding up spurious noise components . We used ICA to extract additional components which would be meaningful for our data , but were nevertheless able to identify the pan-cortical component of interest , as only this component showed high correlation to the timecourse revealed by the locally measured slow calcium wave signal . We furthermore used a seed-based approach to confirm our event-related GLM-based results . Ongoing changes in network activity can contribute to the variability of experimental results and may appear as noise if they are disregarded in the statistical modeling of neural signals ( McGinley et al . , 2015b ) . A direct readout of neurophysiological measures may also be beneficial to monitor and control for homogenous network activity states during BOLD measurements , as depth of anesthesia may vary depending on many physiological factors , as e . g . cortical temperature ( Reig et al . , 2010; Schwalm and Easton , 2016; Sheroziya and Timofeev , 2015 ) . From this perspective it is of high importance to relate local network activity to the global BOLD response , especially regarding the potential translational value of such studies . Including ongoing signals revealed by local neurophysiological readouts as an independent variable , i . e . covariate , regressor , or entirely new factor depending on the type of analysis , could explain a considerable amount of variance in neuronal responses , leading to more reliable readouts , especially in global measures as fMRI . We investigated the functional recruitment of brain areas by slow waves based on the reasoning that , if there is a global participation of cortical areas in slow wave rhythmicity , the activity of the cellular population as a whole should produce a macroscopic signal detectable by functional neuroimaging ( Doeller et al . , 2010 ) . This study illustrates the power of combining cellular measures of neural activity with fMRI in systems neuroscience . Our study directly relates a neurophysiological event – the calcium slow wave – to its potentially present BOLD correlate by employing a specific type of fMRI analysis ( event-related ) , a straightforward and easy-to-apply approach as it is an inbuilt type of analysis of many standard fMRI analysis toolboxes . Results acquired with this method are in agreement with model-free and seed-based analysis approaches revealing a similar extension of network activity spanning almost the entire neocortex , also during ongoing slow wave activity . Our results provide a framework for further analysis of slow wave brain activity within fMRI experiments in anesthetized rodents or for EEG/fMRI studies at a translational level ( Tagliazucchi et al . , 2012 ) , since spatio-temporal dynamics and behavioral correlates of slow waves in humans and animal models are highly preserved ( Buzsáki et al . , 2013 ) . For example , there is growing evidence of impaired slow oscillations in models of Alzheimer disease ( Busche et al . , 2015; Menkes-Caspi et al . , 2015 ) and resting-state functional connectivity in Alzheimer patients ( Zhou et al . , 2015 ) . Such studies could refine slow wave activity as potential disease marker ( Tagliazucchi and Laufs , 2014 ) or specifically target impaired slow oscillations and their potential restoration . Even though calcium measures provide advantages in terms of spatial specificity ( Kajikawa and Schroeder , 2011 ) and lack signal distortions by the varying magnetic fields of the scanner in comparison to electric population recordings , our event-related approach could also be applied for combined EEG-fMRI measures feasible in human experiments . Experiments were performed on 36 adult female Fisher rats ( >12 weeks old , 160–180 g ) , of which optic fiber implantation in the cortex was performed in five animals ( C1-C5 ) and optic fiber implantations in the thalamus in three animals ( T1 , T2 , T3 ) . Dual optic fiber implantation had been performed in two animals ( DF1 , DF2 ) . LFP recordings were carried out in three animals ( LFP1-LFP3 ) and resting state data acquisition without optic fiber implantation in 10 animals ( M1-M10 ) . Immunohistochemistry was done in four animals ( H1-H4 ) and quantifications ( amplitude , duration , rise time and probability of induction ) of calcium responses and for the two different indicators were performed respectively in 8 ( Q1-Q8 ) and in 2 ( C3 and Q9 ) animals . Animals were housed under a 12 hr light–dark cycle and provided with food and water ad libitum . Animal husbandry and experimental manipulation were carried out according to animal welfare guidelines of the Westfalian Wilhelms-University Münster and the Johannes Gutenberg-University Mainz and were approved by the Landesamt für Natur- , Verbraucher- und Umweltschutz Nordrhein-Westfalen , Recklinghausen , Germany , and the Landesuntersuchungsamt Rheinland-Pfalz , Koblenz , Germany , respectively . For surgical procedures including staining with fluorescent calcium indicator and optic fiber implantation , rats were anesthetized with isoflurane ( maintenance 2–3% , Forene , Abbott , Wiesbaden , Germany ) , treated with subcutaneous injection of 1 mg/kg Metacam for analgesia , placed on a warming pad ( 37° C ) , and fixed in a stereotactic frame with ear and bite bars . The skull was exposed , dried from blood and fluids , and leveled for precise stereotactic injections . Under a dissection microscope a small craniotomy was performed with a dental drill ( Ultimate XL-F , NSK , Trier Germany , and VS1/4HP/005 , Meisinger , Neuss , Germany ) . The fluorescent calcium sensitive dye Oregon Green 488 BAPTA-1 ( OGB-1 , Invitrogen , Life Technologies , Carlsbad , CA , USA ) was prepared as described previously ( Garaschuk et al . , 2006 ) , filtered , and injected into primary somatosensory cortex front limb area ( S1FL; 0 mm AP , +3 . 5 mm ML , −0 . 5 , –0 . 7 and 0 . 9 mm DV; n = 4 animals ) or into the posterior thalamic nucleus ( POm; three animals ) at AP −3 . 3 mm , ML +1 mm , 12° , dorso-ventral 4 . 6 , 4 . 9 , 5 . 2 mm according to stereotactic coordinates ( Paxinos and Watson , 2006 ) . For co-staining of astrocytes , 10 mM of sulforhodamine 101 ( SR101 ) was added to the OGB-1 solution in two animals ( Nimmerjahn et al . , 2004 ) . Dye solution was delivered using a glass micropipette with an outer tip diameter of 45 µm and an inner diameter of 15 µm connected to a 10 mL syringe . Approximately 0 . 3 µL of the solution were slowly released at each depth by gentle manual pressure . After injection , the pipette was held in place for 2 min before slowly retracting it from the tissue . Alternatively , the genetically encoded calcium indicator GCaMP6f ( UPenn Vector Core , PA , USA ) was expressed via viral gene delivery four weeks prior to calcium/fMRI measurements . two animals received injections of approximately 1 . 0 µL AAV1 . Syn . GCaMP6f . WPRE . SV40 in S1FL ( 0 . 0 mm AP , +3 . 0 mm ML , −1 . 2 mm DV ) with an 35° injection angle . Two animals received approximately 1 . 0 µL AAV1 . CamKII . GCamP6f . WPRE . SV40 at AP +0 . 2 , ML +2 . 4 , DV −1 . 6 with an 34° injection angle . A craniotomy ( AP +0 . 2 , ML +3 . 3 ) was performed on the day of the experiment . For the dual-fiber experiments a second craniotomy was performed at AP −1 . 8 , ML +2 . 3 and approximately 0 . 3 µL OGB-1 was injected as described above . After removing the cladding from the tip , an optic fiber was inserted perpendicular to the dura above the OGB-1 stained or GCaMP6f expressing region , typically at −300 µm ( cortex ) and −4 . 4 mm ( POm ) , respectively and fixed to the skull with UV glue ( Polytec , PT GmbH , Waldbrunn , Germany ) . The amount of glue for holding the fiber in place was kept at a necessary minimum to reduce MR image distortions . After experiments rats were transcardially perfused with 4% paraformaldehyde ( PFA ) under deep isoflurane anesthesia . Coronal sections with 1000 µm slice thickness of acute brain slices were prepared from OGB-1/SR101 animals . For GCaMP6f animals 20 µm microtome slices ( Leica CM 1850 , Leica microsystems ) of brains fixed overnight in 4% PFA , 30% sucrose , and embedded in Tissue Tec ( Sakura Finetec Europe , NL ) were prepared . Precise location of OGB-1 injections and GCaMP6f expressions were validated under a fluorescence microscope . Coronal brain sections of 70 µm thickness of previously perfused animals were prepared using a vibratome ( Leica , Wetzlar , Germany ) . For permeabilization/blocking , slices were incubated with 0 . 1% Triton X-100% and 5% normal donkey serum ( Invitrogen , Life Technologies , Carlsbad , CA ) in phosphate buffer solution for 90 min . Slices were incubated with rabbit anti-GAD 65/67 ( 1:200 , Swant , Marly , Switzerland ) or rabbit anti-CamKII ( 1:200 , Epitomics , Burlingame , CA ) at 4° C overnight . On the next day , slices were incubated with the secondary antibody Cy-3 donkey anti-rabbit ( 1:500 , Jackson Immuno Research , West Grove , PA ) . Slices were mounted using antiquenching Vectashield ( Vector Laboratories , Burlingame , CA ) , and confocal imaging was conducted using a Leica SP8 ( Leica , Wetzlar , Germany ) confocal microscope . A custom-built setup was used for optic-fiber based calcium recordings ( Schmid et al . , 2017; Schmid et al . , 2016 ) . The light for excitation of the calcium indicators was delivered by a 20 mW solid state laser ( Sapphire , Coherent , Dieburg , Germany ) with a wavelength of 488 nm . To control laser power , an acousto-optic modulator ( AOM 3080–125 , Crystal Technology , Palo Alto , CA ) was used . After deflection by a dichroic mirror , the beam was focused by means of a collimator into one end of a multimode fiber ( Thorlabs , Grünberg , Germany ) with a diameter of 200 µm and a numerical aperture of 0 . 48 . The emitted fluorescent light was guided back through the same fiber and focused on a photodetector containing an avalanche photodiode ( LCSA500-01 , Lasercomponents GmbH , Olching , Germany ) with an aperture of 0 . 5 mm . The recorded fluorescence signals were digitized with a sampling frequency of 2 kHz using a multifunction I/O device ( PCI 6259 , National Instruments , Austin , TX ) and custom-written LabVIEW-based software ( LabVIEW , National Instruments ) . Calcium data preprocessing was performed with Igor analysis software ( WaveMetrics , Portland , OR; RRID:SCR_000325 ) . All traces represent relative changes in fluorescence intensity ( ∆f/f ) and were low-pass filtered before subsequent event detection . Rise time , duration and amplitude of calcium transients were assessed and statistically compared in animals measured upon OGB-1 injections ( n = 8 animals ) as well as in GCaMP6f transduced rats ( n = 2 animals ) . Rise time was defined as time between onset of the transients and 50% of maximum intensity peak . Duration of the transients was defined as the time between the first and the last intensity value exceeding 50% of maximum intensity peak . Amplitude was determined as the intensity difference ( ∆f ) from the baseline and the highest intensity value of the transients . For details see Source Data files 1 and 2 linked to Figure 1 . Data were tested for normal distribution using the Lilliefors test , an assumption-free adaptation of the one-sample Kolmogorov-Smirnov test . In cases where normal distribution could be assumed ( p>0 . 05 ) , the parametric two-tailed Student’s t-test was employed to compare means . Where normal distribution could not be verified , the non-parametric Wilcoxon rank-sum test for equal medians was used to test median differences . All quantifications are presented as mean ± SEM , unless stated otherwise . In cases of an unstable baseline or absence of characteristic signal dynamics of slow wave-associated calcium transients ( Grienberger et al . , 2012; Stroh et al . , 2013 ) calcium data was excluded from analysis ( n = 2 animals ) . Data was read into Matlab ( The Mathworks , Inc . , Natick , MA; RRID:SCR_001622 ) and downsampled from 2 kHz to 1 kHz by averaging two adjacent values . The baseline of the calcium signal was determined and corrected by using the Matlab function msbackadj estimating the baseline within multiple shifted windows of 2500 datapoints width , regressing varying baseline values to the window’s datapoints using a spline approximation , then adjusting the baseline in the peak range of the calcium signal . This method provides a correction for linear and low-frequency drifts while avoiding contributions of signal-of-interest events . We employed a procedure based on exponential moving average ( EMA ) filters , originally established to identify slow wave activity in electrophysiological recordings ( Seamari et al . , 2007 ) . According to this algorithm , to detect slow waves , an EMA filter with a window of 25 ms was applied . Then , the datapoints of the filtered signal exceeding or falling below a threshold were used as preliminary transition points between slow calcium waves and silent periods . To determine this threshold , amplitudes of all datapoints were sorted into a histogram . Onsets and durations for each threshold ( in percent of maximum ) were derived and the resulting numbers of active voxel of SPM-evaluations using these onsets and durations showed an inflection point at 70% , therefore the threshold was set to 70% for further evaluations ( Figure 2—figure supplement 2 ) . Onsets of slow calcium waves were defined as signal timepoints exceeding the threshold ( 70% ) , termination of calcium slow waves was defined as signal timepoints dropping below 50% of the threshold value . As EMAs are optimized to detect onsets of events but provide less precision regarding event terminations , these were corrected with regard to the noise level . Further , the detected activity was post-processed in the following order: ( 1 ) calcium waves separated by a time-interval below 100 ms were interpreted as one wave; ( 2 ) calcium waves with a duration of less than 600 ms were discarded; ( 3 ) activity not reaching 90% of the histogram-based cumulative signal intensity was discarded . Frequencies of calcium slow wave events were evaluated by separating the measurements into bins comprising of 100 s duration of measurement time . For each bin , the number of detected events was counted and the frequency of these events was recalculated in Hz . Onset and duration time points of calcium slow waves were extracted and subsequently used for the fMRI analysis described below . fMRI data acquisition under isoflurane ( 1 . 2–1 . 8% in 80% air and 20% oxygen ) was performed on a 9 . 4 T small animal imaging system with a 0 . 7 T/m gradient system ( Biospec 94/20 , Bruker Biospin GmbH , Ettlingen , Germany ) . The optic fiber-implanted animals were mounted on a heated MRI cradle . The skull was covered with a 1–2 mm thick layer of dental alginate ( Weiton , Johannes Weithas dental-Kunststoffe , Lütjenburg , Germany ) or 1% agar , to reduce susceptibility artifacts which may occur at the tissue air boundary . The fiber was guided through a lead-through in the RF surface head coil and local shimming was applied ( shimming volume 0 . 6–2 . 2 cm³ , first order Mapshim , Bruker ) . Anatomical images were acquired with a T2-weighted 2D RARE sequence , TR/TE 2000/12 . 7 ms , RARE factor 8 , 256 matrix , 110 × 100 µm² spatial resolution and slice thickness 1 . 2 mm , 6–16 contiguous slices prior to fMRI scans to check for tissue damage and validate the fiber’s location . For BOLD fMRI measurements , T2*-weighted images were acquired with a single-shot gradient echo EPI sequence with TR = 1 s ( 1 . 5 s for animals M1-M10 ) , TE = 18 ms ( 14 ms for animals M1-M10 ) , and FA 65° , 350 × 325 µm² spatial resolution ( 320 × 290 µm² for animals M1-M10 ) , slice thickness 1 . 2 mm ( 0 . 8 mm for animals M1-M10 ) , 9–16 continuous slices ( 34 for animals M1-M10 ) , 2400 ( n = 3 ) , 1800 ( n = 5 ) or 1200 ( n = 10 ) acquisitions resulted in a scan time of 40 , 30 or 20 min , respectively . Experiments were performed at least 1 hr after OGB-1 injection ( Stosiek et al . , 2003 ) under isoflurane anesthesia ( 1 . 1–1 . 8% ) inducing slow wave activity ( Grienberger et al . , 2012; Stroh et al . , 2013 ) . Body temperature and respiration rate were routinely monitored during experiments . fMRI data acquisition under medetomidine ( n = 5 ) was performed under the same parameters as under isoflurane . For the recordings animals were anesthetized with 1 . 5% isoflurane and mounted in on a heated MRI cradle in the scanner . Anatomical images with the same parameters as the previous experiment were acquired . During the acquisition of the T2 images a bolus of 0 . 04 mg/kg of medetomidine was administrated to the animals . Five minutes later , isoflurane anesthesia was turned off and medetomidine 0 . 08 mg/kg/h was perfused until the end of the experiment . In a series of control experiments ( n = 3 ) simultaneous local field potential ( LFP ) recordings and calcium measurements ( as described above ) were performed outside the MR scanner . A pipette with a tip resistance of 0 . 2 MΩ filled with Phosphate Buffered Saline ( Sigma , Munich , Germany ) was inserted through a second craniotomy at a depth of 300 µm , lateral to the optic fiber insertion site . A reference electrode was inserted into the cerebellum , 1 mm posterior from Lambda and signals were amplified using an extracellular amplifier ( EXT-02F/2 , npi Electronics , Tamm , Germany ) . Signals were filtered at 300 Hz ( low pass ) , digitized at 2 kHz and recorded together with the optical signals using LabView ( RRID:SCR_014325 ) .	When a person is in a deep non-dreaming sleep , neurons in their brain alternate slowly between periods of silence and periods of activity . This gives rise to low-frequency brain rhythms called slow waves , which are thought to help stabilize memories . Slow wave activity can be detected on multiple scales , from the pattern of electrical impulses sent by an individual neuron to the collective activity of the brain’s entire outer layer , the cortex . But does slow wave activity in an individual group of neurons in the cortex affect the activity of the rest of the brain ? To find out , Schwalm , Schmid , Wachsmuth et al . took advantage of the fact that slow waves also occur under general anesthesia , and placed anesthetized rats inside miniature whole-brain scanners . A small region of cortex in each rat had been injected with a dye that fluoresces whenever the neurons in that region are active . An optical fiber was lowered into the rat’s brain to transmit the fluorescence signals to a computer . Monitoring these signals while the animals lay inside the scanner revealed that slow-wave activity in any one group of cortical neurons was accompanied by slow-wave activity across the cortex as a whole . This relationship was seen only for slow waves , and not for other brain rhythms . Slow waves seem to occur in all species of animal with a backbone , and in both healthy and diseased brains . While it is not possible to inject fluorescent dyes into the human brain , it is possible to monitor neuronal activity using electrodes . Comparing local electrode recordings with measures of whole-brain activity from scanners could thus allow similar experiments to be performed in people . There is growing evidence – from animal models and from studies of patients – that slow waves may be altered in Alzheimer’s disease . Further work is required to determine whether detecting these changes could help diagnose disease at earlier stages , and whether reversing them may have therapeutic potential .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Cortex-wide BOLD fMRI activity reflects locally-recorded slow oscillation-associated calcium waves
Laetoli is a well-known palaeontological locality in northern Tanzania whose outstanding record includes the earliest hominin footprints in the world ( 3 . 66 million years old ) , discovered in 1978 at Site G and attributed to Australopithecus afarensis . Here , we report hominin tracks unearthed in the new Site S at Laetoli and referred to two bipedal individuals ( S1 and S2 ) moving on the same palaeosurface and in the same direction as the three hominins documented at Site G . The stature estimates for S1 greatly exceed those previously reconstructed for Au . afarensis from both skeletal material and footprint data . In combination with a comparative reappraisal of the Site G footprints , the evidence collected here embodies very important additions to the Pliocene record of hominin behaviour and morphology . Our results are consistent with considerable body size variation and , probably , degree of sexual dimorphism within a single species of bipedal hominins as early as 3 . 66 million years ago . Laetoli ( Figure 1A , B ) is one of the most important palaeontological localities in Africa . It lies within the Ngorongoro Conservation Area at the southern edge of the Serengeti Plains . The region includes sites such as Olduvai Gorge , Lake Ndutu and Laetoli itself and provides a long sequence of Plio-Pleistocene , mostly volcano-sedimentary , deposits that are rich in archaeological and paleontological remains ( Hay , 1987 ) , overlying Precambrian metamorphic rocks . The paleoanthropological significance of the whole area has been known since the mid 1930s ( Reck and Kohl-Larsen , 1936; Kohl-Larsen , 1943 ) , whereas Laetoli became known worldwide in the 1970s for stimulating discoveries , such as the holotype and other remains of Au . afarensis ( Leakey et al . , 1976; Johanson et al . , 1978 ) and remarkable evidence of the earliest bipedal hominin tracks ( Leakey and Hay , 1979; Leakey and Harris , 1987 ) dated to 3 . 66 million years ago ( Ma ) ( Deino , 2011 ) . 10 . 7554/eLife . 19568 . 003Figure 1 . Geographical location and site map . ( A ) Location of the study area in northern Tanzania . ( B ) Location of Laetoli within the Ngorongoro Conservation Area , about 50 km south of Olduvai Gorge . ( C ) Plan view of the area of Laetoli Locality 8 ( Sites G and S ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 003 Mammal , bird and insect prints and trails have been identified in 18 sites ( labelled from A to R ) out of 33 total palaeontological localities in the Laetoli area ( Leakey , 1987a; Musiba et al . , 2008; Harrison and Kweka , 2011 ) . Footprints occur in 10 sublevels within the so-called Footprint Tuff , corresponding to the lower part of Tuff 7 in the Upper Laetolil Beds stratigraphic sequence ( Hay , 1987 ) . These hominin trackways were found in 1978 at Site G ( Locality 8 ) and were referred to three individuals ( G1 , G2 , G3 ) of different body size: the smallest individual , G1 , walked side by side on the left of the largest individual , G2 , while the intermediate-sized individual , G3 , superimposed its feet over those of G2 ( Leakey , 1981 ) . The trackways are usually ascribed , not without controversy ( Tuttle et al . , 1991; Harcourt-Smith , 2005 ) , to Au . afarensis ( White and Suwa , 1987 ) , which is the only hominin taxon found to date in the Upper Laetoli Beds ( Harrison , 2011 ) . The new Site S ( situated within Locality 8 ) is located about 150 m to the south of Site G ( Figure 1C ) , on the surface of the same morphological terrace . It was discovered during systematic survey and excavation activities ( Cultural Heritage Impact Assessment ) aimed at evaluating the impact of a proposed new field museum at Laetoli , in the area of Locality 8 . Sixty-two 2 × 2 m test pits were randomly positioned within a grid and were carefully excavated down to the Footprint Tuff and sometimes deeper . In 2015 , fourteen hominin tracks always associated with tracks of other vertebrates ( see Results ) were unearthed in three test-pits , respectively labelled L8 , M9 and TP2 from north to south ( see Materials and methods ) ( Figures 1C and 2 ) . Seven bipedal tracks in different preservation state ( see below ) were exposed in L8 ( Figure 2; Figure 2—figure supplement 1 and Figures 3–4 ) and four in M9 ( Figure 2—figure supplement 2 and Figure 5 ) . Two additional tracks of the same individual were found in the eastern part of TP2 ( Figure 6 ) . All these prints are clearly referable to a single individual trackway , with an estimated total length of 32 m and trending SSE to NNW ( i . e . , 320–330° ) , approximately parallel to the G1 and G2/3 trackways . Following the code used for the Site G prints ( Leakey , 1981 ) , we refer to the new individual as S1 ( footprint numbers S1-1–7 in L8 , S1-1–4 in M9 and S1-1–2 in TP2 ) . At the end of the September 2015 field season , we discovered one more track referable to a second individual ( S2 ) , in the SW corner of TP2 . Conversely , we exposed only non-hominin footprints in test-pit M10 ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 19568 . 004Figure 2 . Plan view of the four test-pits excavated at Laetoli Site S . Dashed lines indicate uncertain contours . Some of the most interesting tracks are coloured: hominins in orange ( heel drags in dark grey ) , equid in dark green ( M9 ) , rhinoceros in red ( M9 ) , giraffe in light brown ( M10 ) , and guineafowl in blue ( M10 ) . Large roots and the bases of trees are in light green ( L8 ) . The main faults/fractures are indicated by brown lines . Raindrop impressions occur in the northern part of L8 ( dotted areas ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00410 . 7554/eLife . 19568 . 005Figure 2—figure supplement 1 . Orthophotos of selected hominin tracks from test-pit L8 at Site S . ( A ) L8/S1-1 . ( B ) L8/S1-2 . ( C ) L8/S1-3 . ( D ) L8/S1-4 . From left to right: textured models , textured and shaded models , shaded models , and shaded coloured models . Colours represent the density of the point clouds obtained by determining the distance to the nearest neighbour . The surface density is the number of neighbours divided by the neighbourhood surface = N/ ( πR2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00510 . 7554/eLife . 19568 . 006Figure 2—figure supplement 2 . Orthophotos of selected hominin tracks from test-pit M9 at Site S . ( A ) M9/S1-2 . ( B ) M9/S1-3 . Details as in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00610 . 7554/eLife . 19568 . 007Figure 2—figure supplement 3 . Orthophotos of selected tracks from test-pit M10 at Site S . ( A , B ) Small bovid ( ? Madoqua ) and bird ( ? Numida ) tracks . ( C ) Two giraffe tracks surrounded by small bovid and bird tracks . Details as in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00710 . 7554/eLife . 19568 . 008Figure 3 . Shaded 3D photogrammetric elevation model of the L8 trackway . Colour renders heights as in the colour bar . The empty circles indicate the position of the targets of the 3D-imaging control point system ( see Materials and methods for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00810 . 7554/eLife . 19568 . 009Figure 4 . Shaded 3D photogrammetric elevation model of test-pit L8 and close-up of the best-preserved tracks with contour lines . Colour renders heights as in the colour bar; distance between elevation contour lines is 2 mm . The empty circles indicate the position of the targets . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 00910 . 7554/eLife . 19568 . 010Figure 5 . Shaded 3D photogrammetric elevation model of the central portion of test-pit M9 and close-up of the best-preserved tracks with contour lines . Colour renders heights as in the colour bar; distance between elevation contour lines is 2 mm . The empty circles indicate the position of the targetsDOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 01010 . 7554/eLife . 19568 . 011Figure 6 . Shaded 3D photogrammetric elevation model of test-pit TP2 and close-up of the three hominin tracks with contour lines . Colour renders heights as in the colour bar; distance between elevation contour lines is 2 mm . The empty circles indicate the position of the targets . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 011 The preservation state of the tracks varies considerably along the trackway , depending on the depth of the Footprint Tuff from the surface . In L8 , the Tuff is very shallow , not deeper than 20 cm to the south , whereas it even crops out on the scarp of the terrace on the opposite side . Consequently , the Tuff is overlain here only by reworked loose soil , and the tracks are not filled up with compact and/or cemented sediment . Preservation issues arise from this situation , because the tuff tends to be rather altered and dislodged along the natural fractures ( Figure 7 ) . The first four tracks in the L8 trail are the best preserved , whereas the state of preservation of the footprint-bearing surface is particularly critical in the northern part ( Figure 8 ) , where the surface appears very damaged by cracks of different size and by plant roots . Some parts of the surface even subsided into micro-grabens developed along the main faults . Consequently , the anterior portion of the track L8/S1-6 is no longer visible because it is situated in one of these lowered parts ( Figure 3 ) . Moreover , a zigzag channel , probably formed by a large root , crosses the northern half of this test-pit from SE to NW , so that L8/S1-5 is virtually indiscernible ( Figure 3 ) . In the western portion of L8 , three large rounded holes ( green circles in Figure 2 ) originated from roots of acacia trees that grew on the surface . Raindrop imprints are visible to the northern edge of the test-pit ( Figure 2 ) on two relatively well-preserved portions of the tuff surrounded by weathered and lowered areas . These features have also been described in several other footprint-bearing sites at Laetoli ( Leakey , 1987a ) . 10 . 7554/eLife . 19568 . 012Figure 7 . Southern part of the hominin trackway in test-pit L8 . Footprints L8/S1-1 , L8/S1-2 , L8/S1-3 and L8/S1-4 are visible from left to right . The heel drag mark is well visible posteriorly to L8/S1-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 01210 . 7554/eLife . 19568 . 013Figure 8 . Test-pit L8 at Laetoli Site S . In the northern part of the test-pit ( at the top ) , the Footprint Tuff is particularly altered , damaged by plant roots and dislodged along natural fractures . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 01310 . 7554/eLife . 19568 . 014Figure 9 . Central part of the hominin trackway in test-pit M9 . Tracks M9/S1-3 and M9/S1-2 are visible from left to right . The two tracks are crossed by some fractures filled by hard calcite veins , which were not removed . In M9 , the Footprint Tuff is in almost pristine condition , and most of the tracks are still filled by compact sediment . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 014 The situation is different in M9 , where about 72 cm of grey soil and unaltered sediments overlie the Footprint Tuff . Here , the tracks are sealed by the upper , laminated part of Tuff seven and filled with strongly cemented sediment . The tuff is here in reasonably good condition , even if it is crossed by old tectonic fractures re-cemented by calcite ( Figures 5 and 9 ) . Moreover , deeply expanding roots penetrate preferentially into the subhorizontal fissures situated between bedding planes , dislodging the rock and fostering carbonate dissolution . The taphonomic state of the Footprint Tuff and of the tracks is very similar in M10 , which is about 80 cm deep . In M9 , the infilling matrix was removed from two hominin tracks ( M9/S1-2 and M9/S1-3 ) ( Figures 5 and 9 ) in order to examine their inner morphology . Small amounts of water were used during the excavation , in order to soften the sediment and darken its hue to better distinguish it from the surrounding tuff . The infill was finally removed by small dental tools , trying not to damage the very thin calcite film covering the original footprint surface ( White and Suwa , 1987 ) . Unfortunately , some vertical crisscross fractures filled by hard calcite veins ( Figures 5 and 9 ) preclude a detailed morphological study of the two footprints . An about 4-cm-thick layer of tuff was removed from a footprint-free area of the M9 SW corner , putting into light a deeper horizon containing bovid tracks ( Figure 2 ) . In TP2 , the preservation state of the ~66-cm-deep printed tuff is intermediate between the L8 and M9/M10 ones . The southern part is in better condition: the hominin track TP2/S1-1 is rather well preserved and some of the other animal prints are still filled by the sediment of the overlying unit . Unfortunately , the SW portion of the test-pit is crossed longitudinally by north-running roots that cross TP2/S2-1 , partially damaging it ( Figures 2 and 6 ) . On the contrary , the northern part of the test-pit is poorly preserved because of a micro-graben developed along an EW-trending fault , which also crosses TP2/S1-2 , causing the lowering of its anterior portion ( Figures 2 and 6 ) . The assessment of the Laetoli Site S sequence within the wider framework of the Eyasi Plateau formations is crucial to understand the stratigraphic relationships between the footprint-bearing units of the newly discovered Site S and those of the historical Site G . These relationships can be discussed at two levels of increasing detail , each one affecting different and similarly more detailed aspects of the study of the tracks . The first – and most relevant – level regards verifying whether the unit bearing the new tracks corresponds to the Footprint Tuff , part of Tuff 7 together with the overlying Augite Biotite Tuff ( Leakey and Hay , 1979 , p . 317; Hay , 1987 , p . 36 ) , where the Site G tracks were printed . This would imply that the trackways are contemporaneous from a geological/geochronometric point of view . Moreover , considering that Tuff 7 includes a sequence of several sublevels originated by distinct eruptions closely spaced in time , and that its overall deposition time was estimated in weeks ( Hay and Leakey , 1982 , p . 55; Hay , 1987 , p . 36 , it can be concluded that all the tracks belong to the same general population of hominins . Secondarily , stratigraphic relationships can be explored at higher detail , in order to assess whether the tracks of Site S were printed on exactly the same sublevel of the Footprint Tuff as those in Site G . This aspect would mostly concern the behavioural aspects of a hypothetical single group of hominins , but it must be pointed out that extra-fine correlation between outcrops , even in a depositional environment with moderate lateral variability like the Footprint Tuff deposition area , can be affected by major uncertainty . The eye-scale characteristics of the profiles exposed in the test-pits are reported here from the top downwards . Tracks and trackways of mammals , birds and insects , as well as raindrop impressions , are recorded from 18 sites at Laetoli , named alphabetically from A to R . Sites from A to P were listed and geographically located by Leakey ( 1987b ) , who also described in detail the ichnological record of the most important exposures . Sites Q and R were discovered and described by Musiba et al . ( 2008 ) . More than 11 , 300 single footprints are recorded from Sites A–R . These tracks testify to a very rich ichnofauna , although a very high percentage of them ( more than 88% ) can be ascribed to small mammals such as lagomorphs and/or Madoqua-like bovids ( Leakey , 1987a; Musiba et al . , 2008 ) . Numerous footprints were discovered in the new exposures ( test-pits L8 , M9 , TP2 and M10 ) of the Footprint Tuff at Site S in Locality 8 ( Figure 2 ) . A total of 529 footprints of mammals ( excluding hominins ) and birds ( Table 1 ) were recorded during the September 2015 field season . The prints were carefully cleaned using soft brushes to reveal detailed features , measured , photographed , traced , mapped and identified in a preliminary study . 10 . 7554/eLife . 19568 . 016Table 1 . Number of individual tracks ( excluding hominins ) at Laetoli Site S . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 016Taxon L8 M9 TP2 M10 Total Numididae ( ? Numida ) -4-913Bovidae , small size ( ? Madoqua ) 1073916211373Bovidae , medium size ( ? Gazella ) 399-2179Equidae ( ? Hipparion ) 12--3Giraffidae---44Lagomorpha ( ? Lepus ) 8--412Rhinocerotidae-1--1Unidentified micromammals-27-1744Total1558226266529 Mammal tracks – mostly of small and medium-size bovids – are very abundant in M10 , L8 and M9 and occur less frequently in TP2 . Their size ( 30–40 mm long and 20–36 mm wide ) and morphological features suggest that most of them can be ascribed to the genus Madoqua ( Figure 2 and Figure 2—figure supplement 3 ) . Some slightly larger prints ( 60–80 × 40–60 mm ) can be referred to medium-sized bovids such as Gazella , Eudorcas or Nanger . It is very difficult to distinguish the footprints of Madoqua-like bovids from lagomorph footprints because of their very similar morphology and size ( Leakey , 1987a ) . Consequently , we decided to ascribe to Lagomorpha only trails that clearly include at least four footprints arranged in the normal hare gait pattern , i . e . two single prints left by the front feet followed by a couple of prints made by the hind feet in the direction of gait . Each single trail ( i . e . , four footprints ) is approximately 200 mm long and 100 mm wide . We identified very few prints of giraffids ( about 170 × 125 mm ) in M10 , equids ( about 50–95 × 45–70 mm ) in L8 and M9 and rhinoceroses ( about 150–135 mm ) in M9 ( Figure 2 and Figure 2—figure supplement 3C ) . In M9 and M10 , some avian prints ( about 60 × 75 mm ) often organised in trails , can be referred to Galliformes of the family Numididae , such as the guinea fowl ( genus Numida ) ( Figure 2 and Figure 2—figure supplement 3A , B ) . Finally , we report some very small ( about 10 × 10 mm ) tracks of unidentified animals , probably micromammals , in M9 and M10 . The above-mentioned assemblage of terrestrial mammal and bird footprints suggests that the local palaeoenvironment was characterised by a mosaic of dry tropical bushland , woodland , open grassland and riverine forest similar to the extant one . The morphology of the S1 tracks can be described in detail , but unfortunately the only preserved track of S2 shows an abnormal widening of the anterior part . This enlarged morphology is possibly due to a lateral slipping of the foot before the toe-off; alternatively , it could be due to taphonomic factors as a thick root crossing the footprint longitudinally may have altered its original morphology . The overall morphology of the S1 tracks matches those at Site G ( Figure 11 ) and is similar in particular to the prints of the larger individual , G2 ( Robbins , 1987 ) : the heel has an oval shape and is pressed deeply into the ground; the medial side of the arch is higher than the lateral one; the ball region is oriented at an angle of about 75° with respect to the longitudinal axis of the foot and is delimited anteriorly by a transversal ridge , formed when the toes gripped the wet ash and pushed it posteriorly . No clear distinction among the toes is visible . The adducted hallux extends more anteriorly than the other toes in all visible footprints . In TP2/S1-1 , the hallux apparently shuffled anteriorly when the foot was lifted from the ground . Some tracks ( especially L8/S1-3 , M9/S1-2 , M9/S1-3 and TP2/S1-1 ) are characterised by a posterior drag mark about 100 mm long ( Figures 4–7 and Figure 2—figure supplements 1 and 2 ) . These marks were possibly left by the heel shuffling on the ash before being firmly placed into the soil . The two latter features were also recognised in some of the G2 prints ( Robbins , 1987 ) and suggest that the feet were probably lifted above the ground at a low oblique angle . The depth distribution pattern indicates that the weight transfer of S1 was similar to that described for G1–3 ( Robbins , 1987 ) : starting from the heel , the weight was transferred along the lateral part of the foot ( note the steep slope of the lateral wall of the tracks compared to that on the medial side ) up to the distal metatarsal region , and from here to the toes . In some of the S1 tracks ( L8/S1-1 , L8/S1-3 and TP2/S1-8 , all of the right side ) , however , the area of maximum depth is located beneath toes 2–5 . This may suggest a somewhat asymmetrical walking , in which the weight was sometimes loaded on the anterolateral part of the foot before the toe-off . Alternatively , this pattern may be indicative of a rotation of the upper body during the gait ( Schmid , 2004 ) . The angle of gait ranges approximately from 2° to 11° , without any particular difference between the right and left sides . Regarding this aspect , S1 resembles more G2/3 , for which very low average angles are reported , whereas G1 shows instead wide asymmetrical angles ( Tuttle , 1987 ) . 10 . 7554/eLife . 19568 . 017Figure 11 . Shaded 3D photogrammetric elevation model of a cast of the southern portion of the Site G trackway with close-ups of selected hominin tracks with contour lines . Colour renders heights as in the colour bar; distance between elevation contour lines is 2 mm . The empty circles and squares indicate the position of the targets . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 01710 . 7554/eLife . 19568 . 018Figure 11—figure supplement 1 . Orthophotos of selected hominin footprints from a cast of the southern portion of the Site G trackway . ( A ) G2/3–29 . ( B ) G1-34 , G1-35 , G2/3–25 , G2/3–26 . ( C ) G2/3–18 . From left to right: textured models , textured and shaded models , shaded models , and shaded coloured models . Colours represent the density of the point clouds obtained by determining the distance to the nearest neighbour . The surface density is the calculation of number of neighbours divided by the neighbourhood surface = N/ ( πR2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 018 The main dimensional parameters of the tracks at Site S are presented in Table 2 ( the single measurements are explained in Materials and methods ) . 10 . 7554/eLife . 19568 . 019Table 2 . Dimensional parameters measured and derived from the Laetoli Site S tracks and stature and body mass estimates for S1 and S2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 019Footprint Side Length ( mm ) Max width ( mm ) Foot index ( % ) Heel width ( mm ) Angle of gait ( degrees ) Estimated stature ( cm ) Estimated body mass ( kg ) H . sapiens§ H . sapiens° Au . afarensis‡ H . sapiens° Au . afarensis‡ TP2/S1-1right27110137 . 2836194–170 175 . 4167–175 53 . 842 . 9–50 . 0 TP2/S1-2left2719936 . 6814193–169 175 . 1167–175 53 . 142 . 8–49 . 8 M9/S1-1left25010240 . 6732179–156 167 . 5154–161 51 . 639 . 6–46 . 0 M9/S1-2right26410539 . 7803189–165 172 . 8163–171 54 . 241 . 8–48 . 7 M9/S1-3left26811141 . 2914192–168 174 . 3166–173 56 . 342 . 5–49 . 4 M9/S1-4right24510141 . 2714175–153 165 . 6151–158 50 . 938 . 8–45 . 1 L8/S1-1right24510442 . 4788175–153 165 . 6151–158 51 . 738 . 8–45 . 1 L8/S1-2left26510640 . 08211189–166 173 . 1164–171 54 . 541 . 9–48 . 8 L8/S1-3right26010339 . 6773186–163 171 . 3161–168 53 . 141 . 2–47 . 9 L8/S1-4left27410638 . 68110196–171 176 . 5169–177 55 . 643 . 4–50 . 5 L8/S1-5right----------L8/S1-6left---863-----L8/S1-7right25811042 . 7908184–161 170 . 3159–166 54 . 840 . 7–47 . 4 Average S1 - 261 104 40 . 0 81 6 184–163 171 . 6 161–168 53 . 6 41 . 3–48 . 1 TP2/S2-1right23112051 . 986-165–144 160142–149 46 . 736 . 5–42 . 4 Step length Stride length Footprints Side Step length ( mm ) Footprints Side Stride length ( mm ) TP2/S1-1 → 2right → left553M9/S1-1 → 3left1044M9/S1-1 → 2left → right548M9/S1-2 → 4right1069M9/S1-2 → 3right → left505L8/S1-1 → 3right1140M9/S1-3 → 4left → right571L8/S1-2 → 4left1159L8/S1-1 → 2right → left552L8/S1-4 → 6left1284L8/S1-2 → 3left → right587Average right1105L8/S1-3 → 4right → left573Average left1162L8/S1-6 → 7left → right660Average1139Average right → left545Average left → right591Average568Values overestimated because of the enlarged morphology of the only preserved track of S2 . §Estimation based on the relationship between foot length and stature in Homo sapiens ( Tuttle , 1987 ) . °Estimation based on the relationship between footprint length and stature/body mass in H . sapiens ( Dingwall et al . , 2013 ) . ‡Estimation based on the relationship between foot length and stature/body mass in Au . afarensis ( Dingwall et al . , 2013 ) . See Materials and methods for details . Speed estimates for S1 and G1–3 were computed starting from stride length ( Figure 3 ) ( see Materials and methods ) . The obtained values ( Table 3 ) show that these hominins were all walking at similar low speed ( about 0 . 44 to 0 . 9 m/s , depending on the analysis method ) . 10 . 7554/eLife . 19568 . 020Table 3 . Data and estimates for the five Laetoli track-makers from Sites S and G . Limited to S1 , mean values , standard deviation and range are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 020Trackway S1 S2 G1 G2 G3 Number of measurable footprints111928Average footprint length ( mm ) 261 ± 10 . 5 ( 245–273 ) 231180225209Average footprint max width ( mm ) 104 ± 3 . 7 ( 99–111 ) 1207911785Average foot index ( % ) 40 . 0 ± 1 . 9 ( 36 . 6–42 . 7 ) 51 . 943 . 848 . 041 . 5Average step length ( mm ) 568 ± 44 . 3 ( 505–660 ) -416453433Average stride length ( mm ) 1139 ± 94 . 0 ( 1044–1284 ) -829880876Estimated stature ( cm ) H . sapiens§163–186 144–165 113–129 141–161 130–149 H . sapiens°171 . 6 ± 5 . 4 160 ± 5 . 4 141 . 4 ± 5 . 4 158 . 2 ± 5 . 4 152 . 2 ± 5 . 4 Au . afarensis‡ 161–168 142–149 111–116 139–145 129–135 Estimated body mass ( kg ) H . sapiens°53 . 6 ± 3 . 7 46 . 7 ± 3 . 8 39 . 3 ± 3 . 7 52 . 6 ± 3 . 7 43 . 2 ± 3 . 7 Au . afarensis‡ 41 . 3–48 . 1 36 . 5–42 . 4 28 . 5–33 . 1 35 . 6–41 . 4 33 . 1–38 . 5 Walking speed ( m/s ) 0 . 47–0 . 55 ( 0 . 93 ) –0 . 43–0 . 50 ( 1 . 00 ) 0 . 36–0 . 42 ( 0 . 79 ) 0 . 39–0 . 46 ( 0 . 88 ) Relative speed ( s−1 ) 0 . 25–0 . 34 ( 0 . 54 ) –0 . 33–0 . 44 ( 0 . 71 ) 0 . 23–0 . 30 ( 0 . 50 ) 0 . 26–0 . 35 ( 0 . 58 ) Values overestimated because of the enlarged morphology of the only preserved track of S2 . §As in Table 2 . °As in Table 2 . ‡ As in Table 2 . For walking speed and relative speed , values outside the brackets are based on the method of Alexander ( 1976 ) , those inside the brackets are based on the method of Dingwall et al . ( 2013 ) . See Materials and methods for details . The average length of the tracks in the S1 trackway is 261 mm ( range 245–274 ) . Lower values were measured for the three individuals at Site G . The average lengths are 180 mm for G1 , 225 mm for G2 and 209 mm for G3 ( Leakey , 1981; Tuttle , 1987 ) ( Table 3 ) , although a digital analysis-based study ( Bennett et al . , 2016 ) of some Site G footprint casts suggests higher values for G1 ( 193 mm ) and G3 ( 228 mm ) . The main metrical features of the S1 and S2 tracks ( footprint length and width , step and stride lengths ) are larger than the G1–3 equivalents ( Table 3 ) . The stature and mass of the Laetoli print-makers were estimated following the relationships between foot/footprint size and body dimensions ( Tuttle , 1987; Dingwall et al . , 2013 ) . It must be pointed out that stature and body-mass estimates obtained by linear regressions from modern humans ( Tuttle , 1987; first method by Dingwall et al . , 2013 are probably exaggerations , as the body proportions of modern Homo sapiens are considerably different from those of the Laetoli putative track-makers . Consequently , we focused our interpretations on the more appropriate predictions inferred from the relationship between foot size and body dimensions in Australopithecus ( second method by ( Dingwall et al . , 2013; see Materials and methods for details ) . The data in Tables 2–3 indicate that stature and mass estimates for S1 and S2 ( about 165 cm and 44 . 7 kg , and 146 cm and 39 . 5 kg , respectively ) are higher than those obtained for G1 , G2 and G3 ( with S2 partly overlapping the higher estimates for G2 ) . Site S is situated on an almost level or very gently dipping surface , situated at the foot of the left ( southern ) side of the Garusi River valley . Site G is situated about 150 m to the north , on the same surface but 1 . 5–2 m lower than Site S . Several shallow gullies dissect this surface , producing a complexly terraced morphology: consequently , there is no observable stratigraphic continuity between the two sites . However , the gullies put into light about 2–3 m of the underlying sequence , whose units are horizontally layered and characterised by almost constant thickness . Only a shallow depression elongated E-W can be observed between the sites; this is probably an ancient erosion channel filled by a constant thickness of the Site S footprint-bearing tuff . Even if the area of possible outcrop of the Footprint Tuff on gully sides close to Site S is covered by debris , the correlation between G and S is in general straightforward . All previous literature describing the original stratigraphic setting at Laetoli ( Leakey and Hay , 1979; Hay and Leakey , 1982; Hay , 1987 ) indicates that the Footprint Tuff can be divided into two main units – the lower and the upper one – which can be subdivided into 14 and 4 sublevels , respectively . Footprints occur on several sublevels of each unit all over the Laetoli area: eight within the lower one ( mostly on sublevel 9 and on the topmost sublevel 14 ) , and two within the upper one ( sublevels 1 and 2 ) . Leakey and Hay ( 1979 , pp . 317–318 and fig . 4 ) provided a brief description of the type-sequence of the Footprint Tuff at Locality 6 ( Site A ) , where a short trackway of human-like footprints – later referred to an ursid ( Tuttle , 2008 ) – was also found . Later , Hay and Leakey ( 1982 , p . 55 and White and Suwa ( 1987 , p . 488 specified that the hominin tracks at Site G are situated on the top of horizon B , i . e . on the top of sublevel 14 within the lower unit of the Footprint Tuff . Eventually , Hay ( 1987 , pp . 34–35 and fig . 6 ) provided a generalised columnar profile of the Footprint Tuff; this is by far the most accurate description available , but is averaged over all the Laetoli area sites . Although the stratigraphic descriptions above are very accurate , they do not provide details about the eye-scale characteristics of the tuffs , i . e . colour , texture , limits , and so on , and no photographs of the sequence have been published . The Site S sequence does not fit the aforementioned descriptions perfectly , at least not within the observed area , which is rather narrow . The grey augite-rich tuff of Site S largely matches the description of the Augite Biotite Tuff described by Hay ( 1987 , p . 34 and following , level 4 in fig . 2 . 6 , p . 35 ) . Regarding the Footprint Tuff , the upper unit corresponds to Site S Laminated Grey Tuff , but the sublevels here are layered rather crudely , whereas the most evident sedimentary structure is a very fine and almost continuous lamination , which makes the subdivision rather problematic . Energy-sorting of denser grains is apparently a relevant aspect of the depositional processes . The Finely Layered Grey and White Tuff of Site S corresponds to the lower subunit of the Footprint Tuff; 14 sublevels are apparent as in the standard description , but this number may be imprecise ( or evaluated differently ) because some of them are extremely thin and apparently discontinuous; in fact , some of the thinner ( and darker ) ones look more like concentrations of gravity-sorted coarser/denser grains situated at the bottom of graded layers . The top sublevel is rather thicker than the others and somewhat whitish instead of greyish , as apparent also in Localities 6 and 7 . Some lateral variability is not surprising in continental environments , which are normally affected by strong morphogenetic processes and/or lateral changes in the sedimentary environments . Consequently , lateral variability can also be expected within the sequence of the Footprint Tuff , even if the involved volcanic depositional processes were rather uniform over a wide area around Laetoli and gave the whole sequence a remarkably homogeneous aspect throughout its outcrops . The correlation between Site G and Site S cannot be absolutely undisputable , at least for the time being , because the original profile could not be examined directly . However , the geological and morphological setting of the area , as well as the characteristics of the newly exposed sequence , indicate with a very good margin of confidence that the newly discovered tracks belong to the Footprint Tuff . To provide a more accurate correlation within the Footprint Tuff , we observe that the Site S tracks were printed on the uppermost level of the Finely Layered Grey and White Tuff ( unit 4 in the description provided in this paper ) , which corresponds to the lower subunit of the Footprint Tuff . The lithological change to the overlying subunit is very evident and marked by a sharp surface , often underlined by a thin crack . However , because of the aforementioned dissimilarities , it is not possible to assess with reasonable confidence whether this stratigraphic position also corresponds to the top of level 14 in the standard sequence ( Hay , 1987 , p . 35 , fig . 2 . 6 ) , i . e . to the same stratigraphic position as the Site G trackways . Our results show that no matter which method is employed to estimate stature and body mass ( see Material and methods ) , the two individuals S1 and S2 were taller and had a larger body mass than the G individuals . The estimated about 165 cm stature of S1 is quite remarkable , exceeding G2 by more than 20 cm ( Table 3 ) . In order to contextualise the australopithecine and early Homo stature estimates and range of variability obtained from the footprints within a broader picture ( Figure 12 ) , and to compare them with a larger sample , we extended our analysis to consistent data based on skeletal elements , namely femurs ( see Materials and methods for details ) . Figure 12 shows the estimated stature of australopithecine and early Homo individuals by species between 4 . 0 and 1 . 0 Ma . The predicted stature of S1 exceeds any australopithecine: a mean value of 158 cm was estimated for the large Au . afarensis individual from Woranso-Mille ( Haile-Selassie et al . , 2010; Lovejoy et al . , 2016 ) , while the Hadar individuals range from 109 to 143 cm ( McHenry , 1991; Ward et al . , 2012 ) ( Figure 12 ) . The stature of S1 falls within the range of modern Homo sapiens maximum values; it also fits the available Homo erectus sensu lato estimates based on fossil remains ( Ruff and Walker , 1993 ) and on footprints ( Bennett et al . , 2009 ) ( Figure 12 ) . At the same time , the 41 to 48 kg body mass range estimated for S1 ( Table 3 ) falls easily within the range of male Au . afarensis ( 40 . 2–61 . 0 kg ) ( Grabowski et al . , 2015 ) . These results extend the dimensional range of the Laetoli track-makers and identify S1 as a large-size individual , probably a male ( Plavcan , 1994; Grabowski et al . , 2015 ) . 10 . 7554/eLife . 19568 . 021Figure 12 . Estimates of predicted stature of fossil hominin individuals by species over time for the interval 4–1 Ma . Solid symbols ( or crosses in bold ) refer to stature estimates based on actual femur length; open symbols refer to stature estimates based on estimated femur length , in turn based on femur head diameter . For Laetoli and Ileret , stature estimates are based on footprint length ( see Materials and methods ) . For Laetoli , Ileret and Woranso-Mille , the average value and range of predicted stature are shown . Colours are associated to the geographical location of each fossil/footprint site on the map . See Supplementary file 5 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 19568 . 021 These findings provide independent evidence for large body-size individuals among hominins as ancient as 3 . 66 Ma . Consequently , we may emphasise the conclusions by Grabowski et al . ( 2015 ) and Jungers et al . ( 2016 ) , who reported that the body sizes of the australopithecines and of the early Homo representatives were similar , but also that certain australopithecine individuals ( at least of Au . afarensis ) were comparable with later Homo species , including H . erectus s . l . and H . sapiens . Thus , our results support a nonlinear evolutionary trend in hominin body size ( Di Vincenzo et al . , 2015; Jungers et al . , 2016 ) and contrast with the idea that the emergence of the genus Homo and/or the first dispersal out of Africa was related to an abrupt increase in body size ( McHenry and Coffing , 2000; Antón et al . , 2014; Maslin et al . , 2015 ) . The identification of large-size individuals among the australopithecines – i . e . hominins commonly presumed to be small-bodied on average – shows also that the available fossil record can be misleading , resulting in an underestimate of the hominin phenotypic diversity in any given period . Moreover , ascribing the S1 tracks to a possible male requires that we reconsider the sex and age of the other Laetoli individuals , who have been object of a plethora of interpretations ( and associated illustrations largely disseminated to the public ) since Mary Leakey’s work ( Leakey , 1981 ) . The most parsimonious option is that sex and age of the hominins represented at Site G cannot be determined , as subadult individuals could possibly be present among them . However , the body-mass estimates suggest some observations as G1 and G3 fall within the range of putative Au . afarensis females ( 25 . 5–38 . 1 kg , according to Grabowski et al . [2015] ) , whereas G2 and S2 span across the upper female and the lower male ranges ( 40 . 2–61 . 0 kg , according to Grabowski et al . [2015] ) . All of these individuals are definitively smaller than the body mass calculated from the S1 tracks . A possible tentative conclusion is that the various individuals represented at Laetoli are: S1 , a male; G2 and S2 , females; G1 and G3 , smaller females or juvenile individuals . Evidence for either marked or moderate body-size variation in Au . afarensis , based on data collected in a single site , was limited until now to the fossil assemblage from the Hadar 333 locality , dated to 3 . 2 Ma ( with body masses ranging from 24 . 5 to 63 . 6 kg ) . The new estimates for the Laetoli individuals indicate an even more marked variation in body size within the same hominin population , at 3 . 66 Ma . Consequently , the combined records from Laetoli and Hadar suggest that large-bodied hominins existed in the African Pliocene for over 400 , 000 years , between 3 . 66 and 3 . 2 Ma . At the same time , these data contrast with the hypothesis of a temporal trend of body-size increase among Au . afarensis between the more ancient Laetoli and the more recent Hadar fossil samples ( Lockwood et al . , 2000 ) . The impressive record of bipedal tracks from Laetoli Locality 8 ( Site G and the new Site S ) may open a window on the behaviour of a group of remote human ancestors , envisaging a scenario in which at least five individuals ( G1 , G2 , G3 , S1 and S2 ) were walking in the same time frame , in the same direction and at a similar moderate speed . This aspect must be evaluated in association with the pronounced body-size variation within the sample , which implies marked differences between age ranges and a considerable degree of sexual dimorphism in Au . afarensis . Significant implications about the social structure of this stem hominin species derive from these physical and behavioural characteristics , suggesting that reproductive strategies and social structure among at least some of the early bipedal hominins were closer to a gorilla-like model than to chimpanzees or modern humans . Finally , the discovery reported here opens up the intriguing possibility that additional hominin trails may also occur in the area between Site G and Site S . Extended geological observations were carried out in the Laetoli area , mostly in the nearby historical Localities 6 and 7 ( Leakey , 1987b ) , in order to compare the sequences exposed there with the new Site S sequence and to assess its stratigraphic position . Unfortunately , correlation with the stratigraphic sequence of Site G ( Locality 8 ) is impossible because this historical site is completely covered by protection features and cannot be used for direct comparison . In Site S , field observation and detailed sequence descriptions were carried out on excavation profiles following the standard formalized by Catt ( 1990 ) . Basic observations on grain size , shape and mineralogy were carried out in the field using a 10x magnification hand lens . Higher-detail analyses were carried out in the laboratory , using a standard Leica stereomicroscope . The survey of the new tracks at Site S in September 2015 was focused on obtaining 3D models for documentation and morphometric analysis . The survey method is the Structure from Motion technique , an image-based process supported by in situ topographic measurements . This technique was chosen because of its technical advantages ( relatively short time of data acquisition and processing; light and handy equipment; reduced costs ) and excellent results in terms of resolution . The equipment used in the fieldwork is a DSLR camera with 15 . 3 ( 4853 × 3198 ) megapixels and two different lenses: EF 24 mm f/2 . 8 for general shots of the excavations and EF 50 mm f/1 . 4 USM for details of the tracks . When necessary , the camera was mounted on a 4 m-long telescopic rod . A measuring tape and a water level were used for the measurement of the control points ( i . e . , circular targets with 35 mm diameter ) . Considering the small size of the surfaces to be detected , this measuring technique provided very high accuracy results . The Ngorongoro Conservation Area Authority ( NCAA ) , in whose jurisdiction the site is , provided the permit for the fieldwork as per letter with Ref . No . NCAA/D/157/Vol . IV of June 5 , 2015 . Hominin and non-hominin tracks were recognised in four test-pits at Site S , namely L8 , M9 , TP2 and M10 . The original 2 × 2 m square shape of L8 – the first test-pit where bipedal tracks were discovered – was modified during the study in order to follow the trail , and consequently took the complex shape in Figure 2 ( southern side: 2 m; western oblique side: 4 m ) . M9 was excavated some 14 m to the SSE of L8 and kept the planned size of 2 × 2 m . Following the interpolated alignment of the bipedal trackway , a third smaller test-pit , TP2 ( 1 × 1 . 2 m ) ( Figure 6 ) was excavated at some 8 m to the SSE of M9 . Finally , a fourth test-pit , M9 ( 2 × 3 m ) was excavated about 15 m to the east of M9 ( Figure 2 ) . After the excavation , the 52 targets of the control point system were immediately positioned: 14 in L8 , 10 in M9 , 14 in TP2 and 14 in M10 . Each test-pit was entirely surveyed at lower resolution and then detailed 3D models of some inner portions ( single prints or groups of close prints ) were acquired ( Figures 4–6 ) . We positioned four perimeter targets on the ground at the corners of each test-pit , and four inner targets around each sub-area surveyed in detail . The following list shows the target IDs in relation to the four test-pits and selected areas ( AF: animal footprints ) : In order to optimize the timing of the fieldwork , we decided not to model in detail some of the hominin tracks , i . e . L8/S1-5 ( visible only in its posterior portion ) , L8/S1-6 ( virtually invisible due to the poor state of preservation of the Footprint Tuff ) , L8/S1-7 ( damaged and excessively deep due to the lowering of the tuff cropping out on the scarp of the terrace ) , M9/S1-1 and M9/S1-4 ( both filled by compact matrix ) . In the second step , the perimeter target positions were measured . We placed two rods equipped with a spherical level on successive pairs of targets and we marked points at the same height on the rods for each pair by using the water level device . The vertical distance between these points and the targets , as well as their mutual distance , were recorded . Repeating this process for all pairs of targets , the relative plan position and the height of the control points were determined respectively by trilateration and by levelling . A preliminary accuracy check was carried out during fieldwork , by using trilateration graphic rules in plan and by the method of successive levelling for heights . By assigning a z-coordinate to the first control point , all subsequent coordinates were derived from addition and subtraction of heights between two successive points . The check was performed by computing the algebraic sum of all height differences , and by verifying that the obtained value was close to zero . Finally , the error obtained in each test-pit was distributed to every z-coordinate of the points , in order to reduce it ( Supplementary file 1 ) . The photographic survey was carried out by three shooting modes: ( i ) using the camera with the 24 mm lens , mounted on a telescopic rod at 4 m above the test-pits , in order to record each test-pit , as well as the spatial connection between test-pits; ( ii ) using the camera freehand with the 24 mm lens , in order to acquire additional shots of each test-pit; and ( iii ) using the camera close to the ground with the 50 mm lens , in order to acquire detailed sub-areas . More than 2 , 000 photos were taken , for a total of about 50 GB . Data processing started by checking measurements in plan and height . This step is preliminary to the definition of the control point coordinates . The trilateration method was used to obtain x , y coordinates of the control points in plan . For each test-pit , six measurements were taken at the same height: the length of the four sides of the perimeter and the length of the two diagonals . Redundant measurements were used to compute the errors . In addition to a preliminary graphical control by CAD software ( Autodesk AutoCAD ) , the automatic calculation software MicroSurvey STAR*NET was used to adjust , by least squares technique , a new set of x , y coordinates and heights of the control points ( Supplementary file 2 ) . The report provided by the software shows that the residues of adjustments never exceeded 10 mm ( Supplementary file 2 ) , which is a fully acceptable figure considering the size of the test-pits . Once the adjusted x , y , z coordinate of all the control points ( Supplementary file 3 ) were computed , we used them to scale and locate in the 3D space the 3D models built by the Structure from Motion technique ( see below ) . The pictures were first calibrated to reduce lens geometric distortion , and tone adjustment was applied in order to homogenize them and to reduce the effects of different lighting conditions during shooting . Subsequently , the software Agisoft Photoscan was used to generate 3D spatial data starting from the pictures , through the following phases: ( i ) alignment of the images; ( ii ) creation of the dense point cloud; ( iii ) transformation of the dense point cloud into a surface ( mesh ) ; ( iv ) application of the texture to the mesh ( Supplementary file 4 ) . A series of orthophotos ( with and without textures ) were extracted from the 3D models ( Figure 2—figure supplements 1 , 2 and 3 and Figure 11—figure supplement 1 ) . A check on dense point cloud density was also carried out by CloudCompare , software for 3D point cloud and triangular mesh processing ( Figure 2—figure supplements 1 , 2 and 3 and Figure 11—figure supplement 1 ) . At the end of the September 2015 field season , we also surveyed a first-generation fiberglass cast of the southern portion of the Site G trackway ( about 4 . 7 m in length ) ( Figure 11 ) kept at the Leakey Camp at Olduvai Gorge . The cast includes the following tracks in the direction of walking: G1–39 , 38 , 37 , 36 , 35 , 34 , 33 , 27 , 26 , 25 on the western side and G2/G3–31 , 30 , 29 , 28 , 27 , 26 , 25 , 24 , 20 , 19 and 18 on the eastern side . Data acquisition and processing ( Supplementary file 4 ) were performed following the workflow described above for the Site S test-pits . We positioned four perimeter control points and 11 inner targets . The latter were used to model in detail six selected tracks ( G2/G3–29 , G1–35 , G1–34 , G2/3–26 , G2/3–25 and G2/3–18 , listed in the direction of walking ) ( Figure 11—figure supplement 1 ) . We used footprint size to estimate the stature of the Laetoli track-makers by means of different approaches . The easiest method follows Tuttle ( 1987 ) and consists of estimating the stature starting from the footprint length considering the ratio between foot length and stature in modern humans . Given that the foot length in H . sapiens is generally about 14% to 16% of stature ( Tuttle [1987] , and references therein ) , we computed two estimates for the Laetoli hominins assuming that their feet were , respectively , 14% and 16% of their body height ( Tables 2–3 ) . This method , however , is not fully reliable because it is based on the body proportions of modern humans , and because it does not take into account that the footprint length does not accurately reflect the foot length . For this last reason , we also estimated stature using the method of Dingwall et al . ( 2013 ) , who published some equations based on regressions of stature by footprint length in modern Daasanach people ( from the Lake Turkana area , Kenya ) . In particular , given the probable low walking speed of the Laetoli hominins ( see below ) , we used the 'walk only' equation ( Standard Error of Estimate , SEE = 5 . 4 ) ( Dingwall et al . , 2013 ) . The obtained results ( Tables 2–3 ) fall within the range of statures estimated with the first method ( except for G1 and G3 , for which slightly taller statures were calculated ) . Finally , to assess how the results were influenced by considering modern human data , we also computed some estimates using the foot:stature ratio known for Au . afarensis ( Dingwall et al . , 2013 ) . This ratio is 0 . 155–0 . 162 ( Dingwall et al . , 2013 ) , so we obtained stature estimates ( Tables 2–3 ) predictably close to or slightly lower than the lower limit of the estimates given by the Tuttle ( 1987 ) method . Similarly , we estimated the body mass of the Laetoli track-makers using the 'walk only' regression equation that relates footprint area ( i . e . , footprint length x max . width ) to body mass ( SEE = 3 . 7 ) ( Dingwall et al . , 2013 ) . For S2 only , we used the relationship between the footprint length and body mass ( SEE = 3 . 8 ) ( Dingwall et al . , 2013 ) because of the enlarged morphology of TP2/S2-1 . As for the stature , we re-calculated the mass using the known ratio between foot length and body mass in Au . afarensis ( 0 . 543–0 . 632 ) ( Dingwall et al . [2013] , and references therein ) . The latter method resulted in estimates significantly lower than those computed by the aforementioned regression equation based on modern human data ( Tables 2 and 3 ) . For both of the described methods , mean estimates of stature and body mass for S1 were computed by averaging the estimates obtained from individual tracks ( Tables 2 and 3 ) . The average footprint length values were considered more reliable than minimum values ( which from a theoretical point of view could be regarded as more representative of the foot length ) for the following reasons . It must be pointed out that the stature and body-mass estimates for S2 must be considered with caution because they are based on a single preserved footprint . The same goes for G2 , given the very low number of tracks for which the length can be measured ( Leakey , 1981 ) . We also drew some inferences about the walking speed ( Table 3 ) , which is closely related to the stride length: in two individuals of the same body size , the one walking faster shows longer stride length . Nevertheless , the body proportions ( i . e . , stature , h ) of the track-maker must be considered , because they influence the stride length ( L ) and consequently the velocity ( v ) . We followed the power law computed by Alexander ( 1976 ) : ( 1 ) v=0 . 25g0 . 5L1 . 67h−1 . 17 where g is the gravitational acceleration ( 9 . 81 m/s2 ) . Equation ( 1 ) is widely used to estimate walking speed in humans and other animals ( Bennett and Morse [2014] , and references therein ) . Speed was further estimated following the method of Dingwall et al . ( 2013 ) . We used the regression equation that relates the speed to the ratio between stride length and average footprint length for each trail , obtaining values ( Table 3 ) about twice those calculated with the equation ( 1 ) . The transitional speed from walk to run is around 2 . 2 m/s ( Dingwall et al . , 2013 ) . As the speed of the Laetoli track-makers is significantly lower than 2 . 2 m/s , we used the 'walk only' regression equation ( Dingwall et al . , 2013 ) for our speed estimates . After computing the walking speed of S1 and G1–G3 with the aforementioned two methods , we obtained the relative speed ( i . e . , walking speed/estimated stature ) ( Table 3 ) , which is a good indicator with which to compare the gait of different individuals regardless of their body proportions . Figure 12 was designed in order to compare graphically the stature estimates of the Laetoli individuals with those obtained for other hominin specimens . With the exception of the other footprint locality taken into account , Ileret in Kenya ( Bennett et al . , 2009; Dingwall et al . , 2013 ) , all other stature data are based on skeletal elements , namely femurs . Early hominin stature reconstructions are notoriously difficult to assess: the limited number of intact long bones available in the fossil record often requires reconstruction of the long bone length from fragmentary remains , before different methods can be used to estimate the stature; the eventual results can differ according to the method employed . Thus , in an attempt to provide a synthetic picture of stature among australopithecines and early Homo , and to ensure that the results are comparable , we relied on a limited number of different datasets . Data are presented in Supplementary file 5 . For the geological age of the considered specimens and for their taxonomic attributions , we followed Grabowski et al . ( 2015 ) , unless otherwise indicated . Two kinds of femur lengths were used for stature reconstruction: ( i ) the femur lengths of intact bones or femur length estimates based on reconstructions of incomplete bones; ( ii ) femur length estimates based on femur head diameters ( FHD ) , according to the method given in McHenry ( 1991 ) . Morphometric data about complete or reconstructed femurs derive from McHenry ( 1991 ) , unless otherwise indicated ( mostly fossils discovered after 1991 ) . FHD values are from Grabowski et al . ( 2015 ) . The two different equations cited in McHenry ( 1991 ) and in Jungers et al . ( 2016 ) were employed in stature reconstructions . As put into evidence in Supplementary file 5 , the results are largely equivalent , with minor differences not relevant for the purpose of this analysis . Consequently , we used stature estimates obtained using the equation published by Jungers et al . ( 2016 ) to compile Figure 12 . Three-dimensional research-quality data are available from the MorphoSource digital repository ( http://morphosource . org ) without restrictions .	Fossil footprints are extremely useful tools in the palaeontological record . Their physical features can help to identify their makers , but can also be used to infer biological information . How did the track-maker move ? How large was it ? How fast was it going ? Footprints of hominins ( namely the group to which humans and our ancestors belong ) are pretty rare . Nearly all of the hominin footprints discovered so far are attributed to species of the genus Homo , to which modern humans belong . The only exceptions are the footprints that were discovered in the 1970s at Laetoli ( in Tanzania ) on a cemented ash layer produced by a volcanic eruption . These are thought to have been made by three members of the hominin species Australopithecus afarensis – the same species as the famous “Lucy” from Ethiopia – around 3 . 66 million years ago . The extent to which body shape and size varied between different members of Au . afarensis – for example , between males and females – has been the subject of a long debate among researchers . Based on the skeletal remains found so far in East Africa , some scholars believe that individuals only varied moderately , as in modern humans , while others state that it was pronounced , as in some modern apes like gorillas . Masao et al . have now unearthed new bipedal footprints from two individuals who were moving on the same surface and in the same direction as the three individuals who made the footprints documented in the 1970s . The estimated height of one of the new individuals ( about 1 . 65 metres ) greatly exceeds those previously published for Au . afarensis . This evidence supports the theory that body size varied considerably amongst individuals within the species . Masao et al . tentatively suggest that the new footprints can be considered as a whole with the 1970s ones . The tall individual may have been the dominant male of a larger group , the others smaller females and juveniles . Thus , considerable differences may have existed between males and females in these remote human ancestors , similar to modern gorillas . The newly discovered tracks are only 150 metres away from the previously discovered sets of footprints . This leaves open the possibility that additional tracks may be unearthed nearby that will further our knowledge about the variability and behaviour of our extinct ancestors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2016	New footprints from Laetoli (Tanzania) provide evidence for marked body size variation in early hominins
Arbuscular mycorrhiza ( AM ) symbioses contribute to global carbon cycles as plant hosts divert up to 20% of photosynthate to the obligate biotrophic fungi . Previous studies suggested carbohydrates as the only form of carbon transferred to the fungi . However , de novo fatty acid ( FA ) synthesis has not been observed in AM fungi in absence of the plant . In a forward genetic approach , we identified two Lotus japonicus mutants defective in AM-specific paralogs of lipid biosynthesis genes ( KASI and GPAT6 ) . These mutants perturb fungal development and accumulation of emblematic fungal 16:1ω5 FAs . Using isotopolog profiling we demonstrate that 13C patterns of fungal FAs recapitulate those of wild-type hosts , indicating cross-kingdom lipid transfer from plants to fungi . This transfer of labelled FAs was not observed for the AM-specific lipid biosynthesis mutants . Thus , growth and development of beneficial AM fungi is not only fueled by sugars but depends on lipid transfer from plant hosts . Arbuscular mycorrhiza ( AM ) is a widespread symbiosis between most land plants and fungi of the Glomeromycota ( Smith and Read , 2008 ) . The fungi provide mineral nutrients to the plant . These nutrients are taken up from the soil and released inside root cortex cells at highly branched hyphal structures , the arbuscules ( Javot et al . , 2007 ) . For efficient soil exploration , arbuscular mycorrhiza fungi ( AMF ) develop extended extraradical hyphal networks . Their growth requires a large amount of energy and carbon building blocks , which are transported mostly as lipid droplets and glycogen to the growing hyphal tips ( Bago et al . , 2002 , 2003 ) . AMF are obligate biotrophs , as they depend on carbon supply by their host ( Smith and Read , 2008 ) . In the past , detailed 13C-labeled tracer-based NMR studies demonstrated that hexose sugars are a major vehicle for carbon transfer from plants to fungi ( Shachar-Hill et al . , 1995 ) . In addition , a fungal hexose transporter , with high transport activity for glucose is required for arbuscule development and quantitative root colonization as shown by host induced gene silencing ( Helber et al . , 2011 ) , indicating the importance of hexose transfer for intra-radical fungal development . AMF store carbon mainly in the form of lipids ( Trépanier et al . , 2005 ) . The predominant storage form is triacylglycerol ( TAG ) and the major proportion of FAs found in AMF is composed of 16:0 ( palmitic acid ) , and of 16:1ω5 ( palmitvaccenic acid ) . The latter is specific to AM fungi and certain bacteria and is frequently used as marker for the detection of AM fungi in soil ( Graham et al . , 1995; Bentivenga and Morton , 1996; Madan et al . , 2002; Trépanier et al . , 2005 ) . Fungus-specific 16:1ω5 FAs are not exclusive to glycerolipids but also incorporated into membrane phospholipids ( van Aarle and Olsson , 2003 ) . Furthermore , 18:1ω7 and 20:1∂11 are considered specific for AMF but do not occur in all AMF species ( Madan et al . , 2002; Stumpe et al . , 2005 ) . It has long been assumed that AMF use sugars as precursors for lipid biosynthesis ( Pfeffer et al . , 1999 ) . However , de novo biosynthesis of fungal fatty acids ( FAs ) was only observed inside colonized roots and not in extraradical mycelia or spores ( Pfeffer et al . , 1999; Trépanier et al . , 2005 ) . The authors concluded that AM fungi can produce FAs only inside the host . The hypothesis that plants directly provide lipids to the fungus could not be supported at that time ( Trépanier et al . , 2005 ) , due to experimental limitations and the lack of appropriate plant mutants . However , recently available whole genome sequences of AMF have revealed that genes encoding multi-domain cytosolic FA synthase subunits , typically responsible for most of the de novo 16:0 FA synthesis in animals and fungi , are absent from the genomes of the model fungi Rhizophagus irregularis , Gigaspora margarita and Gigaspora rosea ( Wewer et al . , 2014; Ropars et al . , 2016; Salvioli et al . , 2016; Tang et al . , 2016 ) . Hence , AMF appear to be unable to synthesize sufficient amounts of 16:0 FAs , but their genomes do encode the enzymatic machinery for 16:0 FA elongation to higher chain length and for FA desaturation ( Trépanier et al . , 2005; Wewer et al . , 2014 ) . Development of fungal arbuscules is accompanied by activation of a cohort of lipid biosynthesis genes in arbuscocytes ( arbuscule-containing plant cells ) ( Gaude et al . , 2012a , 2012b ) . Furthermore , lipid producing plastids increase in numbers and together with other organelles such as the endoplasmic reticulum change their position and gather in the vicinity of the arbuscule ( Lohse et al . , 2005; Ivanov and Harrison , 2014 ) , symptomatic of high metabolic activity to satisfy the high demands of arbscocytes for metabolites including lipids . The importance of plant lipid biosynthesis for arbuscule development has been demonstrated by Medicago truncatula mutants in AM-specific paralogs of two lipid biosynthesis genes FatM and REDUCED ARBUSCULAR MYCORRHIZA2 ( RAM2 ) ( Wang et al . , 2012; Bravo et al . , 2017 ) . FatM encodes an ACP-thioesterase , which terminates fatty acid chain elongation in the plastid by cleaving the ACP off the acyl group releasing free FAs and soluble ACP ( Jones et al . , 1995 ) . RAM2 encodes a glycerol 3-phosphate acyl transferase ( GPAT ) and is most similar to Arabidopsis GPAT6 . In Arabidopsis , GPAT6 acetylates the sn-2 position of glycerol-3-phosphate with an FA and cleaves the phosphate from lysophosphatidic acid , thereby producing sn-2-monoacylglycerol ( ßMAG , Yang et al . , 2010 ) . Mutations in both FatM and RAM2 impair arbuscule branching ( Wang et al . , 2012; Bravo et al . , 2017 ) . In addition , arbuscule branching requires a complex of two half ABC transporters STR and STR2 ( Zhang et al . , 2010; Gutjahr et al . , 2012 ) . The substrate of STR/STR2 is unknown but other members of the ABCG transporter family are implicated in lipid transport ( Wittenburg and Carey , 2002; Wang et al . , 2011; Fabre et al . , 2016; Hwang et al . , 2016; Lee et al . , 2016 ) . Therefore , and due to its localization in the peri-arbuscular membrane ( Zhang et al . , 2010 ) it was speculated that the STR/STR2 complex may transport lipids towards arbuscules ( Gutjahr et al . , 2012; Bravo et al . , 2017 ) . Transcriptional activation of RAM2 and STR is controlled by the GRAS transcription factor REDUCED ARBUSCULAR MYCORRHIZA1 ( RAM1 ) ( Gobbato et al . , 2012; Park et al . , 2015; Pimprikar et al . , 2016 ) and also in ram1 mutants , arbuscule branching is impaired ( Park et al . , 2015; Xue et al . , 2015; Pimprikar et al . , 2016 ) . Thus , RAM1 , FatM , RAM2 and STR/STR2 appear to form an AM-specific operational unit for lipid biosynthesis and transport in arbuscocytes . Consistently , they were found to be absent from genomes of plants that have lost the ability to form AM ( Delaux et al . , 2014; Favre et al . , 2014; Bravo et al . , 2016 ) . Here , we analyzed two Lotus japonicus mutants identified in a forward genetic screen , which are impaired in arbuscule branching ( Groth et al . , 2013 ) . Positional cloning combined with genome resequencing revealed mutations in a novel AM-specific β-keto-acyl ACP synthase I ( KASI ) gene and in the L . japonicus ortholog of M . truncatula RAM2 . KASI likely acts upstream of RAM2 in producing 16:0 FAs . The identity of the genes and the phenotypes led us to hypothesize that AMF may depend on delivery of 16:0 FAs from the plant host . Using a combination of microscopic mutant characterization , lipidomics and isotopolog profiling of 16:0 and 16:1ω5 FAs in roots and extraradical fungal mycelium , we provide strong evidence for requirement of both genes for AM-specific lipid biosynthesis and cross-kingdom lipid transfer from plants to AMF . We previously identified two L . japonicus mutants disorganized arbuscules ( dis-1 , SL0154-N ) and SL0181-N ( red ) deficient in arbuscule branching ( Groth et al . , 2013 ) ( Figure 1A–B ) . Both mutants also suffered from a reduction in root length colonization and blocked the formation of lipid-containing vesicles of the fungus Rhizophagus irregularis ( Figure 1C–D ) . We identified the causative mutations with a combination of classical mapping and next generation sequencing ( see Materials and methods ) . DIS encodes a β-keto-acyl ACP synthase I ( KASI , Figure 1—figure supplements 1A–C and 2 ) . KASI enzymes catalyze successive condensation reactions during fatty acyl chain elongation from C4:0-ACP to C16:0-ACP ( Li-Beisson et al . , 2010 ) . SL0181-N carries one mutation ( ram2-1 ) in the L . japonicus orthologue of the previously identified Medicago truncatula REDUCED ARBUSCULAR MYCORRHIZA2 ( RAM2 , Figure 1—figure supplements 3 and 4 ) . Arabidopsis GPAT6 has been shown to produce ß-MAG with a preference for 16:0 FAs ( Yang et al . , 2012 ) . Therefore , we hypothesized that DIS and RAM2 act in the same biosynthetic pathway . 10 . 7554/eLife . 29107 . 003Figure 1 . DIS and RAM2 are required for arbuscule branching and vesicle formation . Arbuscule phenotype and complementation of dis ( A ) and ram2 ( B ) mutants . The fungus was stained with wheat-germ agglutinin ( WGA ) -AlexaFluor488 . ( C-D ) Percent root length colonization of dis ( C ) and ram2 ( D ) mutants as compared to wild-type . Different letters indicate significant differences among treatments ( ANOVA; posthoc Tukey ) . ( C ) : n = 13; p≤0 . 1 , F2 , 10 = 8 . 068 ( total & int . hyphae ) ; p≤0 . 001 F2 , 10 = 124 . 5 ( arbuscules ) ; p≤0 . 001 , F2 , 10 = 299 . 1 ( vesicles ) ( D ) : n = 15; p≤0 . 1 , F2 , 12 = 10 . 18 ( total & int . hyphae ) ; p≤0 . 001 F2 , 12 = 57 . 86 ( arbuscules ) ; p≤0 . 001 , F2 , 12 = 72 . 37 ( vesicles ) . ( A-D ) Plants were inoculated with R . irregularis and harvested at 5 weeks post inoculation ( wpi ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00310 . 7554/eLife . 29107 . 004Figure 1—figure supplement 1 . Identification of the dis mutation . ( A–B ) Genetic map of the DIS locus on chromosome 4 . Numbers next to marker positions refer to the proportion of recombinant individuals among the number of analyzed F2 mutant plants . Rough mapping had previously identified the position of the dis mutation on the south arm of chromosome 4 ( Groth et al . , 2013 ) . ( A ) In the first fine-mapping round , the interval narrowed down by recombinants comprised 19 EMS-induced SNPs ( red stars ) , that could be confirmed by re-sequencing the mutant genome using next generation sequencing . ( B ) Further fine mapping resulted in an interval with 3 of these confirmed SNPs . ( C ) Physical map of the DIS locus . LjT followed by a number refers to TAC clones . CM followed by a number refers to contigs . One of the three SNPs causes a G to A transition in exon 3 of chr . 4 . CM004 . 1640 . r2 . a resulting in an amino acid change from glycine to arginine at position 190 of the protein product , which shares 79% sequence identity with a β-keto-acyl ACP synthase I ( KASI ) from Arabidopsis thaliana . Black boxes indicate exons separated by introns . ( D ) The DIS gene is duplicated in tandem . ( E ) Gene structure of DIS , DIS-LIKE and KASI . Black boxes display exons separated by introns ( black lines ) . Grey boxes indicate determined un-translated regions . ( F ) DIS , DIS-LIKE and KASI are predicted to contain a plastid transit peptide ( green ) . The catalytic triad is shown in blue and the location of mutations identified by TILLING in the DIS gene are shown in red . We chose the dis-4 mutant for further analysis because the mutation resulted in a glycine replacement , which likely affects the functionality of the protein . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00410 . 7554/eLife . 29107 . 005Figure 1—figure supplement 2 . Protein sequence alignment of L . japonicus DIS with other KASI proteins . ( A ) Sequence alignment of LjDIS , LjDIS-LIKE , LjKASI , AtKASI and E . coli KASI and KASII . ( B ) Identity matrix of LjDIS , LjDIS-LIKE , LjKASI and AtKASI . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00510 . 7554/eLife . 29107 . 006Figure 1—figure supplement 3 . Identification of mutation in the RAM2 gene . ( A ) Genetic map of the red locus on chromosome 6 . Numbers next to the marker position refer to the proportion of recombinant individuals among the number of analysed F3 ( black ) and F4 ( grey ) segregating and mutant plants respectively . Fine mapping narrowed down the interval between TM0553 and TM0302 . Red arrows indicate the genomic interval that contains the causative mutation . ( B ) Gene structure of L . japonicus RAM2 with locations of the identified EMS-induced mutation at position 1663 ( star , ram2-1 ) leading to an amino acid exchange from glycine to glutamic acid at position 555 of the RAM2 protein and LORE1 insertion ( triangle , ram2-2 ) . Black boxes indicate exons separated by intron ( thin black line ) . Grey boxes indicate untranslated regions ( UTRs ) comprising 77 bp ( 5’UTR ) and 151 bp ( 3’UTR ) . ( C ) Co-segregation analysis of arbuscule phenotype and mutation in the RAM2 gene in a number of F3 and F4 plants from segregating populations containing only the mutation on chromosome 6 . The number of plants analysed per generation , arbuscule phenotype , genotype at markers TM0053 and TM0302 and the nucleotide observed at position 1663 in the RAM2 gene are indicated . The ram2 mutation at position 1663 clearly co-segregates with the stunted arbuscule phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00610 . 7554/eLife . 29107 . 007Figure 1—figure supplement 4 . Protein sequence alignment of L . japonicus RAM2 with M . truncatula RAM2 . Sequence alignment ( A ) and identity matrix ( B ) of LjRAM2 , Lj1g3v2301880 . 1 , MtRAM2 and Medtr7g067380 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 007 We identified additional allelic dis mutants by TILLING ( Figure 1—figure supplement 1E , Supplementary file 1 ) ( Perry et al . , 2003 ) and a ram2 mutant caused by a LORE1 insertion in the RAM2 gene ( Figure 1—figure supplement 3B ) ( Małolepszy et al . , 2016 ) . Among the allelic dis mutants we chose dis-4 for further investigation because it suffers from a glycine replacement at the border of a conserved ß-sheet ( Figure 1—figure supplement 2 ) , which likely affects protein folding ( Perry et al . , 2009 ) . Both allelic mutants dis-4 and ram2-2 phenocopied dis-1 and ram2-1 , respectively . Furthermore , transgenic complementation of both dis-1 and ram2-1 with the wild-type versions of the mutated genes restored arbuscule-branching and wild-type-like levels of root length colonization and vesicle formation ( Figure 1A-B ) . Taken together this confirmed identification of both causal mutations . Transcript levels of both DIS and RAM2 increased in colonized roots ( Figure 3—figure supplement 1A ) . To analyze the spatial activity pattern of the DIS and RAM2 promoters during colonization we fused 1 . 5 kb for DIS and 2 . 275 kb for RAM2 upstream of the translational start site to the uidA gene . Consistent with a role of both genes in arbuscule development GUS activity was predominantly detected in arbuscocytes ( arbuscule-containing cells ) in both wild-type and the corresponding mutant roots ( Figure 2—figure supplement 1A–B ) . To correlate promoter activity with the precise stage of arbuscule development we used nuclear localized YFP as a reporter . To visualize the fungus , the promoter:reporter cassette was co-transformed with a second expression cassette containing secreted mCherry fused to the SbtM1 promoter . This promoter drives expression in colonized cells , in cells neighboring apoplastically growing hyphae and in cells forming pre-penetration apparatuus ( PPAs , cytoplasmic aggregations that assemble in cortex cells prior to arbuscule development ) ( Genre et al . , 2008; Takeda et al . , 2009 , 2012 ) . When expressed under the control of the SbtM1 promoter , secreted mCherry accumulates in the apoplast surrounding fungal structures and PPAs , thereby revealing the silhouette of these structures ( Figure 2A–B , Videos 1–2 ) . Nuclear localized YFP fluorescence indicated activity of both promoters in cells containing PPAs ( c , Videos 1–2 ) and containing sparsely branched ( d ) or mature ( e ) arbuscules . Furthermore , we rarely detected YFP fluorescence in non-colonized cells in direct neighborhood of arbuscocytes , which were possibly preparing for PPA formation ( a ) . However , YFP signal was absent from cells containing collapsed arbuscules ( f ) , indicating that the promoters were active during arbuscule development and growth but inactive during arbuscule degeneration ( Figure 2A–B ) . RAM2 promoter activity was strictly correlated with arbuscocytes , while the DIS promoter showed additional activity in cortical cells of non-colonized root segments ( Figure 2A–B , Figure 2—figure supplement 1C–D , Videos 3–6 ) . 10 . 7554/eLife . 29107 . 008Figure 2 . Arbuscocyte-specific expression of DIS and RAM2 is sufficient for arbuscule branching . Promoter activity indicated by nuclear localized yellow fluorescence in colonized transgenic L . japonicus wild-type roots transformed with constructs containing a 1 . 5 kb promoter fragment of DIS ( A ) or a 2 . 275 kb promoter fragment of RAM2 ( B ) fused to NLS-YFP . ( A-B ) Red fluorescence resulting from expression of pSbtM1:SP-mCherry labels the apoplastic space surrounding pre-penetration apparatuus ( PPAs ) and fungal structures , thereby evidencing the silhouette of these structures . a Colonized root , b non-colonized part of colonized root , c PPAs , ( white arrow heads indicate the silhouette of fungal intraradical hyphae ) d small arbuscules , e fully developed arbuscules f collapsed arbuscules . Merged confocal and bright field images of whole mount roots are shown . ( C-D ) Transgenic complementation of dis-1 ( C ) and ram2-1 ( D ) hairy roots with the respective wild-type gene driven by the PT4 promoter . The mutant gene was used as negative control . White arrowheads indicate arbuscules . ( E-F ) Quantification of AM colonization in transgenic roots shown in ( C-D ) . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 15; p≤0 . 001 ) among genotypes for each fungal structure separately . Int . hyphae , intraradical hyphae . ( E ) : F2 , 12 = 26 . 53 ( total ) , F2 , 12 = 46 . 97 ( arbuscules ) , F2 , 12 = 27 . 42 ( vesicles ) . ( F ) F2 , 12 = 341 . 5 ( total ) , F2 , 12 = 146 . 3 ( arbuscules ) , F2 , 12 = 35 . 86 ( vesicles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00810 . 7554/eLife . 29107 . 009Figure 2—figure supplement 1 . DIS and RAM2 promoter activity in wild type and dis and ram2 mutants . GUS activity in colonized transgenic L . japonicus wild-type and mutant roots transformed with constructs containing a 1 . 5 kb promoter fragment of DIS ( A ) or a 2 . 275 kb promoter fragment of RAM2 ( B ) fused to the uidA gene . Left micrographs: bright field channel to detect GUS-staining , middle micrographs: GFP-channel to detect ( WGA ) -AlexaFluor488 stained fungal structures . Right micrographs: Merge . ( C-D ) Single optical section of z-stack shown in Figure 2Aa ( C ) and Figure 2Ba ( D ) showing that DIS and RAM2 promoter activity is detected exclusively in the cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 00910 . 7554/eLife . 29107 . 010Video 1 . 3D animation of Figure 2Ac illustrating that the silhouette of the fungal intraradical hyphae ( red fluorescent vertical line ) aligns with the silhouette of pre-penetration apparatuus ( red fluorescent bag-like structure ) . Yellow fluorescence in nuclei indicates activation of pDIS:YFP . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01010 . 7554/eLife . 29107 . 011Video 2 . 3D animation of Figure 2Bc illustrating that the silhouette of the fungal intraradical hyphae ( red fluorescent vertical line ) aligns with the silhouette of pre-penetration apparatuus ( red fluorescent bag-like structure ) . Yellow fluorescence in nuclei indicates activation of pRAM2:YFP . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01110 . 7554/eLife . 29107 . 012Video 3 . Scan through confocal z-stack of Figure 2Aa illustrating correlation of DIS promoter activity with arbuscocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01210 . 7554/eLife . 29107 . 013Video 4 . Scan through confocal z-stack of Figure 2Ab illustrating DIS promoter activity exclusively in the cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01310 . 7554/eLife . 29107 . 014Video 5 . Scan through confocal z-stack of Figure 2Ba illustrating correlation of RAM2 promoter activity with arbuscocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01410 . 7554/eLife . 29107 . 015Video 6 . Scan through confocal z-stack of Figure 2Bb illustrating absence of RAM2 promoter activity from non-colonized cells . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 015 To examine , whether arbuscocyte-specific expression of DIS and RAM2 is sufficient for fungal development we complemented the dis-1 and ram2-1 mutants with the corresponding wild-type genes fused to the arbuscocyte-specific PT4 promoter ( Volpe et al . , 2013 ) . This restored arbuscule-branching , vesicle formation as well as root length colonization in the mutants ( Figure 2C–F ) , showing that arbuscocyte-specific expression of DIS and RAM2 suffices to support AM development . Thus , expression of lipid biosynthesis genes in arbuscocytes is not only important for arbuscule branching but also for vesicle formation and quantitative colonization . Growth and development of dis and ram2 mutants are not visibly affected ( Figure 3—figure supplement 2 ) , although they carry defects in important lipid biosynthesis genes . RAM2 is specific to AM-competent plants ( Wang et al . , 2012; Delaux et al . , 2014; Favre et al . , 2014; Bravo et al . , 2016 ) and activated in an AM-dependent manner ( Figure 2 , Figure 3—figure supplement 1A ) ( Gobbato et al . , 2012 , 2013 ) . Plants contain an additional GPAT6 paralog , which likely fulfills the housekeeping function ( Figure 1—figure supplement 4 , Yang et al . , 2012; Delaux et al . , 2015 ) . To understand whether the same applies to DIS we searched the L . japonicus genome for additional KASI genes . We detected three paralogs KASI , DIS and DIS-LIKE ( Figure 1—figure supplement 1D–E and Figure 1—figure supplement 2 ) , of which only DIS was transcriptionally activated in AM roots ( Figure 3—figure supplement 1A ) . Phylogenetic analysis revealed a split of seed plant KASI proteins into two different clades , called KASI and DIS ( Figure 3 ) . Members of the KASI clade , are presumably involved in housekeeping functions as this clade contains the product of the KASI single copy gene in Arabidopsis ( Wu and Xue , 2010 ) . Members of the DIS clade are found specifically in AM-host dicotyledons and in a gymnosperm ( Figure 3 ) . As confirmed by synteny analysis ( Figure 3—figure supplement 3 ) , DIS is absent from all eight analyzed non-host dicotyledon genomes , a phylogenetic pattern similar to other symbiosis genes ( Delaux et al . , 2014; Favre et al . , 2014; Bravo et al . , 2016 ) . The occurrence of DIS in Lupinus species , which lost AM competence but still form root nodule symbiosis , may be a relic from the AM competent ancestor . An apparently , Lotus-specific , and thus recent duplication of the DIS gene resulted in an 87% identical copy ( DIS-LIKE ) located directly adjacent to DIS in a tail-to-tail orientation ( Figure 1—figure supplements 1B–C , 2 ) . DIS-LIKE was expressed at very low levels and not induced upon AM ( Figure 3—figure supplement 1A ) . Nevertheless , because of its sequence similarity to DIS , we examined whether DIS-LIKE is also required for arbuscule formation using the dis-like-5 mutant , which suffers from a glycine replacement at position 180 at the border of a highly conserved β-sheet that likely affects protein function ( Perry et al . , 2009 ) ( Supplementary file 1 , Figure 1—figure supplement 2 ) . However , in roots of dis-like-5 AM and arbuscule development was indistinguishable from wild type ( Figure 3—figure supplement 1B ) . Therefore , DIS-LIKE might have lost its major role in arbuscule development after the duplication . 10 . 7554/eLife . 29107 . 016Figure 3 . Phylogenetic tree of KASI proteins in land plants . Protein sequences were aligned using MAFFT . Phylogenetic trees were generated by neighbor-joining implemented in MEGA5 ( Tamura et al . , 2011 ) . Partial gap deletion ( 95% ) was used together with the JTT substitution model . Bootstrap values were calculated using 500 replicates . DIS likely originated before the angiosperm divergence ( red star ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01610 . 7554/eLife . 29107 . 017Figure 3—source data 1 . Accession numbers for protein sequences used in the phyologenic tree . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01710 . 7554/eLife . 29107 . 018Figure 3—figure supplement 1 . Transcript accumulation of KASI and RAM2 genes . ( A ) Transcript accumulation of DIS , DIS-LIKE , KASI and RAM2 in control ( mock ) and R . irregularis colonized ( AM ) roots and in different organs of L . japonicus assessed by qRT-PCR . Expression values were normalized to those of the constitutively expressed gene EF1α ( DIS , DIS-LIKE , KASI ) and Ubiquitin10 ( RAM2 ) . Black circles represent three biological replicates . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 15; p≤0 . 05 , F4 , 14 ( KASI ) = 1 . 191 , F4 , 14 ( DIS ) = 8 . 412 , F4 , 14 ( DIS-LIKE ) = 4 . 563; p≤0 . 001 , F4 , 14 = 67 . 41 ( RAM2 ) ) . AM plants were inoculated with R . irregularis . Control and AM plants were harvested 5 wpi . ( B ) Arbuscule phenotype in wild type and dis-like-5 mutant roots after 5 wpi with R . irregularis as indicated by acid ink staining . White arrow heads indicate arbuscules . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01810 . 7554/eLife . 29107 . 019Figure 3—figure supplement 2 . Shoot phenotypes of dis and ram2 mutants . dis and ram2 mutants do not show growth differences in shoot growth as compared to Gifu wild-type . The image has been taken 17 weeks post planting ( size bar , 5 cm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 01910 . 7554/eLife . 29107 . 020Figure 3—figure supplement 3 . Genomic comparison of the DIS locus in host and non-host species . Synteny analysis of a ~ 200 kb sized region in the Lotus japonicus , Medicago truncatula ( green ) , Populus trichocarpa ( orange ) , Phaseolus vulgaris ( pink ) , Solanum lycopersicum ( blue ) and Carica papaya ( yellow ) genomes containing the DIS locus . The genomic block is well conserved in these host species . By contrast , no DIS homolog was detected in the corresponding genomic block of Arabidopsis thaliana ( red ) . The red rectangle indicates the DIS and DIS-LIKE locus , DIS is indicated in yellow . The sequences above Lotus correspond to the forward strand and those below Lotus to the reverse strand . The orange strip on the left side corresponds to a non-assembled region of the L . japonicus genome . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 020 We examined whether DIS can substitute the phylogenetically related housekeeping KASI . To this end we transgenically complemented an Arabidopsis kasI mutant ( Wu and Xue , 2010 ) with Lotus DIS driven by the Arabidopsis KASI promoter . Arabidopsis kasI exhibits an altered FA profile and reduced rosette growth ( Wu and Xue , 2010 ) . Complementation with DIS restored both wild-type-like rosette growth and FA accumulation . The kasI phenotypes persisted when the dis-1 mutant allele was transformed as a negative control ( Figure 4C–E ) . In the reverse cross-species complementation AtKASI driven by the DIS promoter restored colonization , arbuscule branching and vesicle formation in dis-1 roots ( Figure 4A–B ) . Furthermore , DIS contains a KASI-typical plastid transit peptide and - as predicted - localizes to plastids in Nicotiana benthamiana leaves and L . japonicus roots ( Figure 1—figure supplement 1F Figure 4F–G ) . Thus , the enzymatic function of DIS is equivalent to the housekeeping KASI of Arabidopsis and the AM-specific function must result from its AM-dependent expression pattern . 10 . 7554/eLife . 29107 . 021Figure 4 . DIS function is equivalent to a canonical KASI . ( A ) Microscopic AM phenotype of transgenic dis-1 mutant and wild-type hairy roots transformed with either an empty vector ( EV ) or the Arabidopsis KASI gene fused to the L . japonicus DIS promoter . White arrowheads indicate arbuscules . ( B ) Quantification of AM colonization in transgenic roots of dis-1 transformed with EV ( open circles ) , dis-1 transformed with pDIS-AtKASI ( grey circles ) and wild-type transformed with EV ( black squares ) . int . hyphae , intraradical hyphae . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 15; p≤0 . 001 ) among genotypes for each fungal structure separately . F2 , 12 = 0 . 809 ( total and intraradical hyphae ) , F2 , 12 = 43 . 65 ( arbuscules ) , F2 , 12 = 0 . 0568 ( vesicles ) . ( C ) Rosettes of Arabidopsis , kasI mutant , Col-0 wild-type plants and kasI mutant plants transformed either with the native AtKASI gene , the dis-1 mutant or the DIS wild-type gene driven by the Arabidopsis KASI promoter at 31 days post planting . ( D ) Rosette fresh weight of kasI mutant , Col-0 wild-type plants , one transgenic pAtKASI:AtKASI complementation line ( Wu and Xue , 2010 ) and two independent transgenic lines each of kasI mutant plants transformed either with the dis-1 mutant or the DIS wild-type gene driven by the Arabidopsis KASI promoter at 31 days post planting . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 70; p≤0 . 001; F6 , 63 = 34 . 06 ) among genotypes . ( E ) Q-TOF MS/MS analysis of absolute amount of digalactosyldiacylglycerols ( DGDG ) containing acyl chains of 16:x + 18:x ( 34:x DGDG ) or di18:x ( 36:x DGDG ) derived from total leaf lipids of the different Arabidopsis lines . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 32; ( p≤0 . 05 , F 6 , 25 = 14 . 48 ( 36:6 ) ) . ( F ) Subcellular localization of DIS in transiently transformed Nicotiana benthamiana leaves . Free RFP localizes to the nucleus and cytoplasm ( upper panel ) . RFP fused to DIS co-localizes with the Arabidopsis light harvesting complex protein AtLHCB1 . 3-GFP in chloroplasts ( lower panel ) . ( G ) Subcellular localization in plastids of DIS-YFP expressed under the control of the L . japonicus Ubiquitin promoter in R . irregularis colonized ( upper panel ) and non-colonized ( lower panel ) L . japonicus root cortex cells . BF , bright field; IH , intercellular hypha; A , arbuscule . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 021 To characterize the role of DIS in determining the lipid composition of non-colonized and colonized roots we quantified triacylglycerols ( TAGs ) , diacylglycerols ( DAGs ) , galactolipids and phospholipids in wild-type and dis-1 . The lipid profile of colonized roots contains both plant and fungal lipids , however using the fungal marker FA 16:1ω5 and previous data on fungus-specific lipids ( Wewer et al . , 2014 ) , many fungal lipids can be clearly distinguished from plant lipids . The lipid profile of non-colonized roots was not affected by the dis-1 mutation . However , the strong and significant increase of 16:0 and 16:1 ( most probably fungus-specific 16:1ω5 ) containing TAGs , which is characteristic for colonization of wild-type roots ( Wewer et al . , 2014 ) was abolished in dis-1 ( Figure 5A–D , Figure 5—figure supplement 1B ) . Also , AM- and fungus-specific DAG and phospholipid molecular species were enhanced in colonized wild-type roots but not in colonized dis-1 roots ( Figure 5—figure supplements 1A and 2 ) . In contrast , galactolipids were not affected by root colonization or genotype ( Figure 5—figure supplement 3 ) . In summary , DIS affects the glycerolipid and phospholipid profile of colonized L . japonicus roots and does not interfere with lipid accumulation in the non-colonized state . Most lipids affected by the DIS mutation are fungus-specific and therefore reflect the amount of root colonization and of fungal lipid-containing vesicles . However , since the root lipid profile is hardly affected , absence of FA elongation by DIS was the cause of reduced lipid accumulation and root colonization . 10 . 7554/eLife . 29107 . 022Figure 5 . Lack of characteristic accumulation of triacylglycerols in AM-defective mutants . ( A-D ) Quantitative accumulation of ( A ) total triacylglycerols , ( B ) tri16:0-triacylglycerol ( C ) tri16:x-triacylglycerols and ( D ) of triacylglycerols harbouring 16:x and 18:x FA-chains in non-colonized and R . irregularis colonized wild-type and dis-1 roots . Different letters indicate significant differences ( ANOVA; posthoc Tukey ) ( A ) : n = 18; p≤0 . 001; F3 , 14 = 68 . 16 . ( B ) : n = 18; p≤0 . 001; F3 , 14 = 68 . 48 . ( C ) : n = 19; p≤0 . 01 , F3 , 15 = 7 . 851 ( 16:1-16:1-16:1 ) ; p≤0 . 001 , F3 , 15 = 14 . 52 ( 16:0-16:1-16:1 ) ; p≤0 . 001 , F3 , 15 = 39 . 22 ( 16:0-16:0-16:1 ) . ( D ) : n = 19; p≤0 . 001 , F3 , 15 = 12 . 15 ( 48:x ) , F3 , 15 = 15 . 56 ( 50:x ) ; p≤0 . 01 , F3 , 15 = 22 . 93 ( 54:x ) . ( E-G ) Quantitative accumulation of ( E ) total triacylglycerols , ( F ) tri16:0-triacylglycerols , ( G ) tri16:x-triacylglycerols and ( H ) of triacylglycerols harbouring 16:x and 18:x FA-chains in colonized roots of L . japonicus wild-type Gifu , wild-type MG-20 and arbuscule-defective mutants . Different letters indicate significant differences ( ANOVA; posthoc Tukey ) . ( E ) : n = 40; p≤0 . 001; F8 , 31 = 38 . 42 . ( F ) Left: absolute tri16:0 TAG content: n = 40; p≤0 . 001; F8 , 31 = 19 . 05 . Right: tri16:0 TAG proportion among all TAGs , n = 40; p≤0 . 001; F8 , 31 = 14 . 21 . ( G ) : p≤0 . 001; n = 41 , F8 , 32 = 86 . 16 ( 16:1-16:1-16:1 ) ; n = 39 , F8 , 30 = 24 . 16 ( 16:0-16:1-16:1 ) ; n = 40 , F8 , 31 = 17 . 67 ( 16:0-16:0-16:1 ) . ( H ) : n = 40; p≤0 . 001 , F8 , 31 = 39 . 26 ( 48:x ) , F8 , 31 = 28 . 93 ( 50:x ) ; p≤0 . 01 , F8 , 31 = 19 . 78 ( 52:x ) ; p≤0 . 05 , F8 , 31 = 13 . 77 ( 54:x ) . ( A-H ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02210 . 7554/eLife . 29107 . 023Figure 5—source data 1 . Raw data for lipid profiles in Figure 5 and Figure 5—figure supplements 1–3 and 5–11 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02310 . 7554/eLife . 29107 . 024Figure 5—figure supplement 1 . Diacylglycerol ( DAG ) and triacylglycerol ( TAG ) profiles of L . japonicus WT and dis-1 control and AM roots . ( A ) Profile of diacylglycerols in control and AM-colonized L . japonicus WT and dis-1 roots . ( B ) Profile of triacylglycerols in control and AM-colonized L . japonicus WT and dis-1 roots . ( A–B ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . ‘L . japonicus and R . irregularis’ marks lipids which are found in both organisms according to ( Wewer et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02410 . 7554/eLife . 29107 . 025Figure 5—figure supplement 2 . Profiles of phospholipids in non-colonized and colonized L . japonicus WT Gifu and dis-1 roots . ( A ) Absolute amounts of phosphatidic acid ( PA ) species . ( B ) Absolute amounts of phosphatidylinositol ( PI ) species . ( C ) Absolute amounts of phosphatidylcholine ( PC ) species . ( D ) Absolute amounts of phosphatidylethanolamine ( PE ) species . ( E ) Absolute amounts of phosphatidylserine ( PS ) species . ( A–D ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . ‘L . japonicus and R . irregularis’ marks lipids which are found in both organisms according to Wewer et al . ( 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02510 . 7554/eLife . 29107 . 026Figure 5—figure supplement 3 . MGDG and DGDG profiles do not differ among L . japonicus wild-type and mutant roots . ( A ) Relative amounts of monogalactosyldiacylglycerol ( MGDG ) in control and colonized roots of Gifu WT and dis-1 . ( B ) Relative amount of digalactosyldiacylglycerol ( DGDG ) in control and colonized roots of Gifu WT and dis-1 . ( C ) Relative amounts of monogalactosyldiacylglycerols ( MGDG ) containing acyl chains of 16:x + 18:x ( 34:x MGDG ) , di18:x ( 36:x MGDG ) or 18:x + 20:x ( 38:x MGDG ) in the different colonized genotypes . ( D ) Relative amount of digalactosyldiacylglycerols ( DGDG ) containing acyl chains of 16:x + 18:x ( 34:x DGDG ) , di18:x ( 36:x DGDG ) or 18:x + 20:x ( 38:x DGDG ) of the different colonized genotypes . ( A–D ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02610 . 7554/eLife . 29107 . 027Figure 5—figure supplement 4 . All arbuscule-deficient mutants show reduced root length colonization . Quantitative AM colonization in root samples employed for lipidomics ( Figure 3D–F , Figure 5E–H , Figure 7 , Figure 5—figure supplements 1–3 and 5–11 ) as determined by modified grid-line intersect methods after acid-ink staining . WT Gifu , WT MG-20 and all AM-deficient mutants in the Gifu background ( ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 and ram2-2 ) and the str mutant in the MG-20 background . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 45 ) among genotypes for each fungal structure separately . p≤0 . 05 , F8 , 36 = 21 . 69 ( total and intraradical hyphae ) ; p≤0 . 001 , F8 , 36 = 62 . 1 ( arbuscules ) , F8 , 36 = 176 . 5 ( vesicles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02710 . 7554/eLife . 29107 . 028Figure 5—figure supplement 5 . Total fatty acid and free fatty acid profiles of colonized L . japonicus WT and AM-defective mutant roots . ( A ) Total amounts of fatty acids ( FAME ) in colonized L . japonicus roots of the different genotypes . Fatty acid methyl esters ( FAME ) were prepared from total root lipids and analysed by GC . Different letters indicate significant differences ( ANOVA; posthoc Tukey; p≤0 . 01; ( n = 42 , F8 , 33 = 29 . 91 ( 16:1 ) ; n = 43 , F8 , 34 = 20 . 25 ( 16:0 ) ; n = 43 , F8 , 34 = 11 . 34 ( 18:3 ) ; F8 , 34 = 13 . 14 ( 18:2 ) ) . ( B ) Free fatty acid composition in colonized L . japonicus roots from Gifu WT , MG-20 WT , ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Free fatty acids were isolated from total root lipids and converted into fatty acid methyl esters for quantification by GC Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 44; ( p≤0 . 001 , F8 , 35 = 230 . 6 ( 16:0 ) ; p≤0 . 001 , F8 , 35 = 257 . 7 ( 16:1 ) ; F8 , 35 = 222 . 5 ( 18:1 ) ; F8 , 35 = 15 . 48 ( 18:2 ) ; F8 , 35 = 8 . 225 ( 18:3 ) ) . ( A–B ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02810 . 7554/eLife . 29107 . 029Figure 5—figure supplement 6 . Triacylglycerol ( TAG ) profiles of colonized L . japonicus WT and AM-defective mutant roots . Absolute amounts of triacylglycerol molecular species in colonized L . japonicus roots of WT Gifu , WT MG-20 ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Black arrow indicates accumulation of tri 16:0 TAG in ram2-1 and ram2-2 . Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 02910 . 7554/eLife . 29107 . 030Figure 5—figure supplement 7 . Phosphatidic acid ( PA ) profiles in L . japonicus WT and AM-defective mutants . Absolute amounts of phosphatidic acid molecular species in colonized L . japonicus roots of WT Gifu , WT MG-20 ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Black arrow indicates accumulation of 32:0 ( di16:0 ) PA in ram2-1 and ram2-2 . Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03010 . 7554/eLife . 29107 . 031Figure 5—figure supplement 8 . Profile of phosphatidylcholines ( PC ) in L . japonicus WT and AM-defective mutants . Absolute amounts of phosphatidylcholine molecular species in colonized L . japonicus roots of WT Gifu , WT MG-20 , ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . ‘L . japonicus and R . irregularis’ marks lipids which are found in both organisms according to Wewer et al . ( 2014 ) . Arrow highlights the exclusive accumulation of unusual 32:0 ( di16:0 ) PC in ram2-1 and ram2-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03110 . 7554/eLife . 29107 . 032Figure 5—figure supplement 9 . Phosphatidylethanolamine ( PE ) profile in L . japonicus WT and AM-defective mutants . Absolute amounts of phosphatidylethanolamine molecular species in colonized L . japonicus roots of WT Gifu , WT MG-20 , ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . ‘L . japonicus and R . irregularis’ marks lipids which are found in both organisms according to Wewer et al . ( 2014 ) . Arrow highlights the exclusive accumulation of unusual 32:0 ( di16:0 ) PE in ram2-1 and ram2-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03210 . 7554/eLife . 29107 . 033Figure 5—figure supplement 10 . Phosphatidylinositol ( PI ) profile in L . japonicus WT and AM-defective mutants . Absolute amounts of phosphatidylinositol molecular species in colonized L . japonicus roots of WT Gifu , WT MG-20 , ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Bars represent means ± standard deviation ( SD ) of 3–5 biological replicates . ‘L . japonicus and R . irregularis’ marks lipids which are found in in both organisms according to Wewer et al . ( 2014 ) . Arrow highlights the exclusive accumulation of unusual 32:0 PI in ram2-1 and ram2-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03310 . 7554/eLife . 29107 . 034Figure 5—figure supplement 11 . Phosphatidylserine ( PS ) profile in L . japonicus WT and AM-defective mutants . Absolute amounts of phosphatidylserine molecular species in colonized L . japonicus roots of WT Gifu WT , WT MG-20 , ram1-3 , ram1-4 , dis-1 , dis-4 , ram2-1 , ram2-2 and str . Bars represent means ± standard deviation ( SD ) of 3–5 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 034 Similar to dis and ram2 L . japonicus mutants in the ABCG half-transporter STR and the GRAS protein RAM1 are affected in arbuscule branching ( Kojima et al . , 2014; Pimprikar et al . , 2016; Xue et al . , 2015 ) , quantitative root colonization and formation of lipid-containing fungal vesicles ( Figure 5—figure supplement 4 ) . Moreover , the AM-dependent transcriptional activation of DIS and KASIII , the latter of which is a single copy gene in L . japonicus and produces precursors for DIS-activity by catalyzing FA chain elongation from C2 to C4 , was absent from ram1 mutants ( Figure 6 ) . In contrast , induction of the single copy gene KASII , which elongates fatty acyl chains from C16 to C18 was not hampered by RAM1 deficiency . Thus , RAM1 may play an important role in the regulation of lipid biosynthesis in arbuscocytes , since it also mediates expression of RAM2 and STR ( Gobbato et al . , 2012; Park et al . , 2015; Pimprikar et al . , 2016; Luginbuehl et al . , 2017 ) . 10 . 7554/eLife . 29107 . 035Figure 6 . Loss of RAM1 affects AM-dependent induction of KASIII and DIS . ( A ) RAM1 effects on AM-dependent induction of KASIII and DIS , which catalyze 16:0 FA biosynthesis , and absence of effects on KASII . According to BLAST analysis via Kazusa ( http://www . kazusa . or . jp/lotus/ ) and NCBI ( http://www . ncbi . nlm . nih . gov/ ) KASIII and KASII are single copy genes in L . japonicus . Transcript accumulation of KASIII , DIS and KASII in non-colonized ( open circles ) and colonized ( black circles ) roots of Gifu WT , ram1-3 and ram1-4 . Different letters indicate different statistical groups ( ANOVA; posthoc Tukey; p≤0 . 001; n = 23 F5 , 12 = 65 . 04 ( KASIII ) ; n = 24 F5 , 18 = 54 . 42 ( DIS ) ; n = 18 F5 , 12 = 33 . 11 ( KASII ) ) . Transcript accumulation was determined by qRT-PCR and the housekeeping gene Ubiquitin10 was used for normalization . AM plants were inoculated with R . irregularis and harvested 5 wpi . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 035 We hypothesized that RAM1 , DIS , RAM2 and STR form a specific operational unit for lipid biosynthesis and transport in arbuscocytes . Therefore , we directly compared their impact on the AM-specific root lipid profile and measured galactolipids , phospholipids , TAGs and also total and free fatty acids in colonized roots of ram1 , dis , ram2 , str mutants and wild-type in parallel . Consistent with our previous observation in dis-1 , galactolipid accumulation was similar in colonized roots of wild-type and all mutants ( Figure 5—figure supplement 3C–D ) . In contrast , total 16:0 FAs ( FAMEs ) as well as 16:1 and 18:1 ( likely 18:1ω7 FA of fungal origin ) FAs were strongly reduced in all colonized mutants compared to the corresponding wild-type . Free FAs showed a similar pattern except for 18:1 FAs ( Figure 5—figure supplement 5 ) . Also for TAGs and phospholipids , AMF-specific molecular species and 16:0 FA containing molecular species were strongly reduced in all mutants ( Figure 5E–H , Figure 5—figure supplements 6–11 ) . However , the two allelic ram2 mutants formed an exception . They specifically over-accumulated 16:0-16:0 FA-containing phospholipids in particular 32:0 PA and 32:0-PC but also to a smaller extend 32:0-PE and 32:0-PI ( Figure 5—figure supplements 6–10 ) . A similar pattern was observed for tri-16:0 TAGs ( Figure 5F ) . This suggests that RAM2 acts downstream of DIS in a biosynthetic pathway and uses the 16:0 FAs synthesized by DIS in arbuscocytes as substrates . In the absence of functional RAM2 the FA products of DIS , are probably redirected into phospholipid biosynthesis and storage lipid biosynthesis via PA and PC ( Li-Beisson et al . , 2010 ) leading to the observed higher accumulation of 16:0 FA containing lipid species in ram2 mutants . This higher accumulation of specific lipids did not correlate with colonization levels in ram2 mutants ( Figure 5—figure supplement 4 ) confirming that reduced colonization levels are not the primary cause for altered lipid profiles in the colonized mutant roots . Instead , defective AM-specific lipid biosynthesis in the mutants more likely impairs fungal development . The first step in TAG and phospholipid production after FA biosynthesis is the esterification of FAs with glycerol by GPATs in the plastid or endoplasmic reticulum to produce α-MAGs ( sn1/3-MAGs , Li-Beisson et al . , 2010 ) . RAM2 is predicted to produce a different type of glycerolipid ß-MAG ( sn2-MAG ) with a preference towards 16:0 and 18:1 FAs ( Yang et al . , 2010; Wang et al . , 2012; Yang et al . , 2012 ) . To examine the role of RAM2 in MAG biosynthesis , we quantified α-MAG and ß-MAG species in colonized roots of wild-type and all mutants . The abundance of ß-MAGs was generally lower than that of α-MAGs ( Figure 7 ) . The amount of most α-MAG species did not differ among the genotypes . Only the fungus-specific 16:1 and 18:1ω7 α-MAGs were reduced in all mutants reflecting the lower fungal biomass ( Figure 7A ) . Fungus-specific ß-MAGs with 16:1 and 18:1ω7 acyl groups were not detected and most ß-MAG molecular species accumulated to similar levels in all genotypes . Exclusively the levels of 16:0 ß-MAGs were significantly lower in all mutants as compared to the corresponding wild-type roots ( Figure 7B ) . This supports a role of RAM2 in 16:0 ß-MAG synthesis during AM and a role of DIS in providing 16:0 FA precursors for RAM2 activity . A low accumulation , of 16:0 ß-MAGs in ram1 mutants is consistent with RAM1’s role in regulating the FA and lipid biosynthesis genes ( Figure 6 ) ( Gobbato et al . , 2012; Pimprikar et al . , 2016 ) . In str 16:0 ß-MAGs likely did not accumulate because of reduced RAM2 expression in str roots due to low root length colonization and/or a regulatory feedback loop ( Bravo et al . , 2017 ) . 10 . 7554/eLife . 29107 . 036Figure 7 . sn-1 monoacylglycerol ( α-MAG ) and sn-2 monoacylglycerol ( β-MAG ) profiles of colonized L . japonicus wild-type and AM-defective mutant roots . ( A ) Total amounts of α-MAG molecular species in the different genotypes . ( B ) Total amounts of β-MAG molecular species in the different genotypes . 16:0 β-MAG levels are significantly reduced in all mutant lines compared to the respective wild-type . ( A–B ) Bars represent means ±standard deviation ( SD ) of 3–5 biological replicates . Black asterisk indicates significant difference of mutants vs . wild-type according to Student’s t-test , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 036 In plants , ß-MAGs serve as precursors for cutin polymers at the surface of aerial organs ( Yang et al . , 2012; Yeats et al . , 2012 ) . For their use in membrane or storage lipid biosynthesis they first need to be isomerized to α-MAGs ( Li-Beisson et al . , 2010 ) . The recruitment of a GPAT6 ( RAM2 ) instead of a α-MAG-producing GPAT for AM-specific lipid synthesis supports the idea that RAM2-products are destined for something else than membrane biosynthesis of the host cell . Since AM fungal genomes lack genes encoding cytosolic FA synthase subunits ( Wewer et al . , 2014; Ropars et al . , 2016; Tang et al . , 2016 ) we hypothesized that 16:0 ß-MAGs synthesized by DIS- and RAM2 are predominantly delivered to the fungus . To test this hypothesis , we examined lipid transfer by FA isotopolog profiling . Isotopologs are molecules that differ only in their isotopic composition . For isotopolog profiling an organism is fed with a heavy isotope labelled precursor metabolite . Subsequently the labelled isotopolog composition of metabolic products is analyzed . The resulting characteristic isotopolog pattern yields information about metabolic pathways and fluxes ( Ahmed et al . , 2014 ) . We could not detect fungus-specific 16:1ω5 ß-MAGs in colonized roots ( Figure 7B ) . Therefore , we reasoned that either a downstream metabolite of ß-MAG is transported to the fungus , or alternatively , ß-MAG is rapidly metabolized in the fungus prior to desaturation of the 16:0 acyl residue . Since the transported FA groups can be used by the fungus for synthesizing a number of different lipids , we focused on total 16:0 FA methyl esters ( FAMEs , subsequently called FAs for simplicity ) and 16:1ω1 FAMEs as markers for lipid transfer . We fed L . japonicus wild-type , dis-1 , ram2-1 and str with [U-13C6]glucose and then measured the isotopolog composition of 16:0 FAs and 16:1ω5 FAs in L . japonicus roots and in associated extraradical fungal mycelium with spores . To generate sufficient hyphal material for our measurements the fungus was pre-grown on split Petri dishes in presence of a carrot hairy root system as nurse plant ( Figure 8—figure supplement 1 ) . Once the fungal mycelium had covered the plate , L . japonicus seedlings were added to the plate on the side opposing the carrot root . During the whole experiment , the fungus was simultaneously supported by the carrot hairy root and the L . japonicus seedling . Once the L . japonicus roots had been colonized , labelled glucose was added to the side containing L . japonicus . After an additional week , FAs were esterified and extracted from colonized L . japonicus roots and from the associated extraradical mycelium and the total amount of 13C labelled 16:0 and 16:1ω5 FAs as well as their isotopolog composition was determined . In L . japonicus wild-type 13C-labeled 16:0 and 16:1ω5 FAs were detected in colonized roots as well as in the extraradical fungal mycelium ( Figure 8A–B , Figure 8—figure supplement 2A–B ) , indicating that 13C-labelled organic compounds were transferred from the root to the fungus . No labelled FAs were detected in the fungal mycelium when the fungus was supplied with [U-13C6]glucose in absence of a plant host ( Figure 8A–B , Figure 8—figure supplements 2A–B , 3 ) , indicating that the fungus itself could not metabolize labelled glucose to synthesize FAs . The three mutants incorporated 13C into 16:0 FAs at similar amounts as the wild-type but hardly any 13C was transferred to the fungus ( Figure 8A–B , Figure 8—figure supplement 2A–B ) . 10 . 7554/eLife . 29107 . 037Figure 8 . Isotopolog profiling indicates lipid transfer from plant to fungus . ( A–B ) Overall excess ( o . e . ) 13C over air concentration in 16:0 FAs ( A ) and in 16:1ω5 FAs ( B ) detected in non-colonized ( only 16:0 FAs ) and colonized carrot , L . japonicus wild-type , dis-1 , ram2-1 roots and in the extraradical mycelium of R . irregularis . P values were generated by ANOVA using the Dunnett Test for multiple comparisons to L . japonicus wild-type ( n = 29 ( 16:0 control roots ) ; n = 33 ( 16:0 root AM ) ; n = 39 ( 16:0 extraradical mycelium ) ; n = 33 ( 16:1ω5 root AM ) ; n = 39 ( 16:1ω5 extraradical mycelium ) , *p<0 . 001 , p<0 . 01 , p<0 . 05 ) . ( C ) Relative fraction of 13C isotopologs for 16:0 FAs of three replicates of carrot , L . japonicus WT Gifu , dis-1 , ram2-1 in control roots ( upper panel ) and AM roots and each of the associated R . irregularis extraradical mycelia with spores ( middle panel ) and 16:1ω5 FAs in AM roots and extraradical mycelia with spores ( lower panel ) . Individual bars and double bars indicate individual samples . Values from roots are indicated by ‘R’ and from fungal extraradical mycelia with spores by ‘M’ . For carrot and L . japonicus WT the 13C labelling pattern of 16:0 and 16:1ω5 FAs in the plant is recapitulated in the fungal extraradical mycelium . Extraradical mycelium associated with dis-1 and ram2-1 does not mirror these patterns . Compare bars for AM roots and extraradical mycelium side by side . Black numbers indicate 13C o . e . for individual samples . Colors indicate 13C-isotopologs carrying one , two , three , etc . 13C-atoms ( M + 1 , M + 2 , M + 3 , etc . ) . ( D ) Schematic and simplified illustration of carbon flow and 12C vs . 13C-carbon contribution to plant lipid metabolism and transport to the fungus in the two-compartment cultivation setup used for isotope labelling . Carbohydrate metabolism and transport is omitted for simplicity . ERM , extraradical mycelium . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03710 . 7554/eLife . 29107 . 038Figure 8—source data 1 . Raw data for isotopolog profiles in Figure 8 and Figure 8—figure supplements 2 , 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03810 . 7554/eLife . 29107 . 039Figure 8—figure supplement 1 . Two-compartment cultivation setup used for labelling experiments . ( A ) Schematic representation of cultivation setup which was used for 13C-glucose labelling experiments ( Figure 8 , Figure 8—figure supplement 2–4 ) . [U-13C6]Glucose as substrate was either applied to the carrot compartment or the Lotus compartment . Colonized roots and extraradical mycelia populating the plate were harvested separately . ( B ) Photo of the 2-compartment setup . 2 week old Lotus seedlings were cultivated for 4 weeks on this setup . 100 mg of [U-13C6]Glucose was applied one week before harvest . ( C ) Quantitative AM colonization as determined by the modified grid-line intersect method after acid-ink staining in roots of genotypes indicated in the figure from plants grown in the Petri dish system ( A and B ) in parallel with the plants used for isotopolog profiling . Different letters indicate significant differences ( ANOVA; posthoc Tukey; n = 25 ) among genotypes for each fungal structure separately . p≤0 . 01 , F4 , 20 = 32 . 49 ( total and intraradical hyphae ) ; F4 , 20 = 110 . 1 ( arbuscules ) , F4 , 20 = 112 . 6 ( vesicles ) . ( D ) Arbuscule area and ( E ) frequency distribution of arbuscule area in the root samples used in ( C ) . 10 arbuscules were analysed per root system . For wild-type Gifu , MG20 and ram2-1 five , for str three and for dis-1 two root systems were available . Different letters in ( D ) indicate significant differences ( ANOVA; posthoc Tukey; n = 196 ) in arbuscule area among genotypes . p≤0 . 001 , F4 , 191 = 127 . 4 . ( E ) Representative bright-field images of arbuscules in roots of the samples analyzed in C-D . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 03910 . 7554/eLife . 29107 . 040Figure 8—figure supplement 2 . Isotopolog profiles of wild-type MG20 and str . ( A–B ) Overall excess ( o . e . ) of 13C over air concentration in 16:0 FAs ( A ) and in AMF specific 16:1ω5 FAs ( B ) detected in non-colonized ( only 16:0 FAs ) and colonized , L . japonicus wild-type and str roots and in the extraradical mycelium of R . irregularis . Five biological replicates of each genotype and treatment are shown . Black asterisks indicate statistically significant differences between mutant lines and wild-type according to Student’s t-test p<0 . 05; **p<0 . 01 . ( C ) Relative fraction of 13C isotopologs for 16:0 fatty acids of five replicates ( individual bars and double bars ) of L . japonicus WT MG-20 and str in control roots ( upper panel ) and AM roots and each of the associated R . irregularis extraradical mycelia ( middle panel ) . The same is shown for fungus-specific 16:1ω5 FAs in AM roots and extraradical mycelia ( lower panel ) . Values from roots are indicated by ‘R’ and from fungal extraradical mycelia by ‘M’ . For L . japonicus wild-type the 13C labelling pattern of 16:0 and 16:1ω5 FAs in the plant is recapitulated in the fungal extraradical mycelium . Extraradical hyphae associated with str do not mirror these patterns . Compare bars for AM roots and extraradical hyphae side by side . Black numbers indicate 13C overall excess for individual samples . Colors indicate 13C-isotopologues carrying one , two , three , etc . 13C-atoms ( M + 1 , M + 2 , M + 3 , etc . ) . ( n . d . = not detected ) . ( D ) Schematic and simplified illustration of carbon flow and 12C vs . 13C contribution to plant lipid metabolism and transport to the fungus in the two-compartment cultivation setup used for isotope labelling . Carbohydrate metabolism and transport is omitted for simplicity . ERM , extraradical mycelium . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 04010 . 7554/eLife . 29107 . 041Figure 8—figure supplement 3 . Proportion of 16:0 and 16:1ω5 FA containing only non-labelled 12C in plant and fungal tissue . Proportion of 12C 16:0 fatty acids ( M + 0 ) in non-colonized and colonized carrot , L . japonicus Gifu wild-type , dis-1 , ram2-1 roots and in the extraradical mycelium of R . irregularis ( A ) as well as in L . japonicus MG-20 wild-type , str roots and in the extraradical mycelium of R . irregularis ( C ) . Proportion of non-labeled 12C AMF specific 16:1ω5 fatty acids ( M + 0 ) in colonized carrot , L . japonicus Gifu wild-type , dis-1 , ram2-1 roots and in the extraradical mycelium of R . irregularis ( B ) as well as in L . japonicus MG-20 wild-type , str and in the extraradical mycelium of R . irregularis ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 04110 . 7554/eLife . 29107 . 042Figure 8—figure supplement 4 . Isotopolog profiles of additional samples . Relative fraction of 13C isotopologs for 16:0 fatty acids ( individual bars and double bars ) of D . carota , L . japonicus WT Gifu , dis-1 , ram2-1 in control roots ( upper panel ) and AM roots and each of the associated R . irregularis extraradical mycelia ( middle panel ) and 16:1ω5 FAs in AM roots and extraradical mycelia ( lower panel ) . Values from roots are indicated by ‘R’ and from fungal extraradical mycelia by ‘M’ . Compare bars for AM roots and extraradical hyphae side by side . Isotopolog profiles shown here and in Figure 8C were generated from 3 independent experiments for L . japonicus wild-type and , 2 independent experiments for L . japonicus mutants and carrot roots . Transfer of 13C-label from plant to fungus is higher for carrot than for L . japonicus wild-type . This is possibly caused by the fungus being exclusively dependent on carrot when carrot is labelled , while lipid transfer from L . japonicus competes with un-labeled transfer from carrot from the other side of the petri dish . Whatever the isotopolog pattern of wild-type roots , it is mirrored in the extrarical fungal mycelium , indicating lipid transfer . However , the isotopolog pattern is for most cases not mirrored in extraradical mycelium associated with lipid biosynthesis mutants . ( n . d . = not detected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 042 Remarkably , the isotopolog profile of 16:0 FAs was close to identical between colonized L . japonicus roots and the connected extraradical mycelium , for 11 independent samples of wild-type Gifu ( Figure 8C–D , Figure 8—figure supplement 4 ) and for 5 independent samples of wild-type MG20 ( Figure 8—figure supplement 2C–D ) . Moreover , the isotopolog profile of fungus-specific 16:1ω5 FAs mirrored the profile of 16:0 FAs ( Figure 8C , Figure 8—figure supplements 2 , 4 ) . Pattern conservation between root and associated extraradical mycelium occurred independently of pattern variation among individual samples . Since the fungus does not incorporate 13C into the analyzed FAs in the absence of the plant ( Figure 8A–B , Figure 8—figure supplement 2A–B ) this conserved pattern demonstrates transfer of 16:0 FA-containing lipids from the host plant to the fungus because the plant determines the isotopolog pattern of fungal 16:0 and 16:1ω5 FAs . The 16:0 FA isotopolog pattern of colonized dis-1 , ram2 and str mutant roots resembled the wild-type profile , indicating intact uptake and metabolism of labelled glucose . However , the 16:0 FA isotopolog pattern of the extraradical mycelium associated with mutant roots and the fungal 16:1ω5 FA profile inside and outside the roots differed strongly from the 16:0 FA profile of the mutant host roots ( Figure 8C , Figure 8—figure supplements 2C , 4 ) , consistent with very low FA transfer from the mutant plants to the fungus . The losses in isotopolog profile conservation between plant and fungal FAs in the mutants likely result from dilution of labelled FAs by unlabeled FAs from the carrot hairy root ( Figure 8D , Figure 8—figure supplements 1 and 2D ) and/or from biases due to quantification of FAs at the detection limit . To confirm that the plant determines the fungal FA isotopolog pattern we switched plant system and profiled isotopologs after labelling carrot root organ culture ( ROC ) in the absence of L . japonicus seedlings ( Figure 8D , Figure 8—figure supplement 1 ) . In these root organ cultures , sugar uptake from the medium does not compete with photosynthesis , as in whole seedlings . Additionally , the carrot roots explore a larger surface of the Petri dish , increasing access to substances in the nutrient medium . Consequently , and likely because of increased uptake of labelled glucose from the medium , the isotopolog pattern of carrot ROCs differed from Lotus and was shifted towards more highly labeled 16:0 FA isotopologs . This fingerprint was again recapitulated in the extraradical fungal mycelium as well as in fungus-specific 16:1ω5 FAs inside and outside the root for 10 independent samples ( Figure 8C , Figure 8—figure supplement 4 ) . These data provide strong support for direct transfer of a 16:0 FA containing lipid from plants to AMF ( Figure 9 ) . 10 . 7554/eLife . 29107 . 043Figure 9 . Schematic representation of plant fatty acid and lipid biosynthesis in a non-colonized root cell and a root cell colonized by an arbuscule . In non-colonized cells FAs are synthesized in the plastid , bound via esterification to glycerol to produce LPA in the ER , where further lipid synthesis and modification take place . Upon arbuscule formation AM-specific FA and lipid biosynthesis genes encoding DIS , FatM and RAM2 are activated to synthesize specifically high amounts of 16:0 FAs and 16:0-ß-MAGs or further modified lipids ( this work and Bravo et al . , 2017 ) . These are transported from the plant cell to the fungus . The PAM-localized ABCG transporter STR/STR2 is a hypothetical candidate for lipid transport across the PAM . Desaturation of 16:0 FAs by fungal enzymes ( Wewer et al . , 2014 ) leads to accumulation of lipids containing specific 16:1ω5 FAs . Mal-CoA , Malonyl-Coenzyme A; FA , fatty acid; KAS , β-keto-acyl ACP synthase; GPAT , Glycerol-3-phosphate acyl transferase; PAM , periarbuscular membrane; LPA , lysophosphatic acid; MAG , monoacylglycerol; DAG , diacylglycerol; TAG , triacylglycerol; PA , phosphatidic acid; PC , phosphatidylcholine; PE , phosphatidylethanolamine; PS , phosphatidylserine; CDP-DAG , cytidine diphosphate diacylglycerol; PG , phosphatidylglycerol; PI , phosphatidylinositol . DOI: http://dx . doi . org/10 . 7554/eLife . 29107 . 043 Here we identified DIS and RAM2 , two AM-specific paralogs of the lipid biosynthesis genes KASI and GPAT6 using forward genetics in Lotus japonicus . The dis and ram2 mutants enabled us to demonstrate lipid transfer from plants to AMF using isotopolog profiling . During AM symbiosis , an array of lipid biosynthesis genes is induced in arbuscocytes ( Gaude et al . , 2012a , 2012b ) , indicating a large demand for lipids in these cells . Indeed , two genes encoding lipid biosynthesis enzymes , the thioesterase FatM and the GPAT6 RAM2 , have previously been shown to be required for arbuscule branching in M . truncatula ( Wang et al . , 2012; Bravo et al . , 2017; Jiang et al . , 2017 ) . Both enzymes have a substrate preference for 16:0 FAs ( Salas and Ohlrogge , 2002; Yang et al . , 2012; Bravo et al . , 2017 ) and , consistent with this , we and others observed that colonized ram2 mutant roots over-accumulate 16:0 FA containing phospholipids and TAGs ( Figure 7 , [Bravo et al . , 2017] ) , indicating re-channeling of superfluous 16:0 FAs in the absence of RAM2 function and placing RAM2 downstream of FatM ( Figure 9 ) . Our discovery of DIS , a novel and AM-specific KASI gene , now provides evidence for the enzyme which synthesizes these 16:0 FAs in arbuscocytes . The arbuscule phenotype , as well as the lipid profile of colonized dis mutants is very similar to fatm and ram2 mutants except for the accumulation of 16:0 FA-containing lipids in ram2 ( Figure 1 , Figure 5 and all figure supplements ) , consistent with the predicted function . Together , this strongly suggests that DIS , FatM and RAM2 act in the same lipid biosynthesis pathway , which is specifically and cell-autonomously induced when a resting root cortex cell differentiates into an arbuscocyte ( Figure 2A–B , Figure 9 , [Bravo et al . , 2017] ) . Interestingly , DIS was exclusively found in genomes of AM-competent dicotyledons and a gymnosperm ( Figure 3 ) . This implies that DIS has been lost at the split of the mono- from dicotyledons . Despite the phylogenetic divergence , DIS and the single copy housekeeping KASI gene of Arabidopsis are interchangeable ( Figure 5 ) . Therefore , the specificity of DIS to function in AM symbiosis is probably encoded in its promoter ( Figure 2 ) . In monocotyledons , the promoter of the housekeeping KASI gene may have acquired additional regulatory elements , sufficient for arbuscocyte-specific activation , thus making DIS dispensable . We provide several pieces of complementary evidence that lipids synthesized by DIS and RAM2 in the arbuscocyte are transferred from plants to AMF and are required for fungal development . We fed host plants with [U-13C6]glucose and subsequently determined the isotopolog profile of freshly synthetized 16:0 and 16:1ω5 FAs in roots and associated fungal extraradical mycelia ( Figure 8 ) . This showed that: ( 1 ) AMF were unable to incorporate 13C into FAs when fed with [U-13C6]glucose in absence of the host plant . ( 2 ) When associated with a wild-type host , the fungal extraradical mycelium accumulated 13C labelled 16:0 FAs and the isotopolog profile of these 16:0 FAs was almost identical with the host profile . ( 3 ) The 16:0 FA isotopolog fingerprint differed strongly between two different wild-type plant systems ( Lotus seedling and carrot hairy root ) but for each of them the fungal mycelium recapitulated the isotopolog profile . Therefore clearly , the plant dominates the profile of the fungus , because it is impossible that the fungus by itself generates the same FA isotopolog pattern as the plant – especially in the absence of cytosolic FA synthase . Therefore , this result provides compelling evidence for interkingdom transfer of 16:0 FAs from plants to AMF . ( 4 ) In agreement , the isotopolog profile of fungus-specific 16:1 ω5 FAs inside and outside the root also resembled the plant 16:0 FA profile . ( 5 ) Colonized dis and ram2 mutant roots resembled the 16:0 FA isotopolog profile of L . japonicus wild-type roots . However , the 16:0 FA profile of the fungal extraradical mycelium and the 16:1ω5 FA profile inside the roots showed a very different pattern , consistent with very low transport of labelled FAs to the fungus when associated with the mutants . ( 6 ) DIS and RAM2 are specifically required for the synthesis of 16:0 ß-MAG ( Figure 7 ) and the predominant FA chain length found in AM fungi is precisely 16 . ( 7 ) dis and ram2 roots do not allow the formation of lipid-containing fungal vesicles and accumulate very low levels of fungal signature lipids ( Figure 5 and figure supplements ) . Together this strongly supports the idea that DIS and RAM2 are required to provide lipids for transfer to the fungus . Consequently , in the mutants , the fungus is deprived of lipids . The L . japonicus mutants were originally identified due to their defective arbuscule branching ( Groth et al . , 2013 ) . The promoters of DIS and RAM2 are active in arbusocytes and already during PPA formation , the earliest visible stage of arbuscocyte development . Together with the stunted arbuscule phenotype of dis , ram2 and fatm mutants ( Figure 1 [Bravo et al . , 2017] ) this suggests that plant lipids are needed for arbuscule growth , probably to provide material for the extensive plasma-membrane of the highly branched fungal structure . It also indicates that the arbuscule dictates development of the AMF as a whole , since lipid uptake at the arbuscule is required for vesicle formation , full exploration of the root and development of extraradical mycelia and spores . Defective arbuscule development was also observed for the different and phylogenetically distantly related AMF Gigaspora rosea ( Groth et al . , 2013 ) , which similar to R . irregularis lacks genes encoding cytosolic FA synthase from their genomes ( Wewer et al . , 2014; Tang et al . , 2016 ) . Hence the dependence on plant lipids delivered at the arbuscule is likely a common phenomenon among AMF and a hallmark of AMF obligate biotrophy . Despite the obvious central importance of lipid uptake by the arbuscule , the fungus can initially colonize the mutant roots with a low amount of intraradical hyphae and stunted arbuscules ( Figure 1 , Figure 5—figure supplement 4 ) . The construction of membranes for this initial colonization may be supported by the large amounts of lipids stored in AMF spores . This would be consistent with the frequent observation that in wild-type roots , at initial stages of root colonization , AMF form arbuscules immediately after reaching the inner cortex and before colonizing longer distances , possibly as a strategy to accquire lipids quickly after the reserves in the spore have been depleted . Alternatively , it is possible that plant housekeeping enzymes provide lipids to intraradical hyphae before arbuscule formation . Activity of the housekeeping KASI may also be responsible for slightly higher colonization levels observed for dis in some experiments as compared to other mutants . It has recently been reported that photosynthetic wild-type nurse plants can restore arbuscule-branching in Medicago ram2 and str mutants ( Jiang et al . , 2017; Luginbuehl et al . , 2017 ) , suggesting that lipids can be supplied to arbuscules via the extraradical hyphal network and intraradical hyphae through this route support arbuscule fine-branching . Based on four observations , we favor an alternative szenario , in which lipids need to be provided cell-autonomously by the arbuscocyte to support arbuscule fine-branching . However , we cannot exclude that our observations differ from the reported observations due to growth conditions or plant species . ( 1 ) Presence of nurse carrot hairy roots did not restore arbuscule branching in dis , ram2 and str ( Figure 8—figure supplement 1C–F ) . ( 2 ) dis and ram2 were found in a forward genetics screen based on their stunted arbuscule phenotype . In this screen , the fungal inoculum was provided via chive nurse plants ( Groth et al . , 2013 ) . ( 3 ) Map-based cloning of Lotus dis , ram2 and str ( Kojima et al . , 2014 ) was performed with segregating mutant populations grown in the same pot , in which the wild-type and heterozygeous siblings acted as nurse plants on the homozygeous mutants . In this system , the stunted arbuscule phenotype was easily observable . ( 4 ) Arbuscule branching in a rice str mutant was not restored by wild type nurse plants ( Gutjahr et al . , 2012 ) . It still remains to be shown , which types of lipids are transported from the plant arbuscocyte to the fungal arbuscule and how . RAM2 is the most downstream acting enzyme in arbuscocyte-specific lipid biosynthesis known to date ( Figure 9 ) . It is predicted to synthesize ß-MAG and we and others have shown that 16:0 ß-MAGs are indeed reduced in colonized roots of dis , fatm and ram2 mutants , providing evidence that this is likely the case ( Figure 7 , [Bravo et al . , 2017] ) . Although , we cannot exclude that a downstream metabolite of 16:0 ß-MAG is transported to the fungus , 16:0 ß-MAG as transport vehicle for 16:0 FAs to the fungus is a good candidate because conceptually this molecule may bear certain advantages . It has been shown in Arabidopsis that ß-MAGs are not used for plant storage or membrane lipid biosynthesis but rather as pre-cursors for cuticle formation ( Li et al . , 2007 ) . The production of ß-MAGs could therefore , be a way , to withdraw FAs from the plants own metabolism to make them available to the fungus . In addition , ß-MAGs are small and amphiphilic and could diffuse across the short distance of the hydrophilic apoplastic space between plant and fungal membrane . At the newly growing arbuscule branches the distance between the plant and fungal membrane is indeed very small and has been measured to be 80–100 nm on TEM images of high-pressure freeze-substituted samples ( Bonfante , 2001 ) . However , we could not detect fungus-specific 16:1ω5 ß-MAGs in colonized roots . This could mean that the fungus metabolizes them before desaturation of the 16:0 FAs to synthesize membrane and storage lipids . Alternatively , ß-MAGs may not be taken up by the fungus . ß-MAGs are known to isomerize to α-MAGs in acid or basic conditions ( Iqbal and Hussain , 2009 ) . It is therefore , possible that they isomerize in the acidic periarbuscular space ( Guttenberger , 2000 ) before being taken up by the arbuscule . How are MAGs transported across the peri-arbuscular membrane ? Good candidates for MAG transporters are the ABCG half transporters STR and STR2 . Similar ABCG transporters have been implicated genetically in cuticle formation , which also requires ß-MAGs ( Pighin et al . , 2004; Panikashvili et al . , 2011; Yeats et al . , 2012 ) . The half ABCG transporters STR and STR2 are both independently required for arbuscule branching and they need to interact to form a full transporter ( Zhang et al . , 2010 ) . We found that colonized roots of a L . japonicus str mutant , did not allow the formation of fungal vesicles and had the same lipid profile as dis and ram2 ( Figure 5 and figure supplements ) . Furthermore , our 13C labelling experiment demonstrated that str mutants do not transfer lipids to the fungus ( Figure 8—figure supplement 2 ) . Although these are encouraging indications , strong evidence for the role of STR in lipid transport across the periarbuscular membrane is still lacking and the substrate of STR remains to be determined . Therefore , currently , it cannot be excluded that mutation of str has an indirect effect on lipid transport and alternative mechanisms for example lipid translocation via vesicle fission and fusion are possible . Nevertheless , also in AMF , several ABC transporter genes are expressed in planta ( Tisserant et al . , 2012; Tang et al . , 2016 ) . They are not characterized , but if lipid transport via ABC transporters instead of other mechanisms would play a role , some of them could be involved in uptake of lipids into the fungal cytoplasm . We found that mutants in the GRAS gene RAM1 are impaired in AM-specific lipid accumulation in colonized roots and in AM-mediated activation of DIS and the single copy gene KASIII ( Figure 6 ) , in addition to FatM , RAM2 and STR ( Wang et al . , 2012; Park et al . , 2015; Pimprikar et al . , 2016; Luginbuehl et al . , 2017 ) . This suggests that plants have evolved an AM-specific regulatory module for lipid production in arbuscocytes and delivery to the fungus . It remains to be shown , whether RAM1 regulates lipid biosynthesis genes directly and how this occurs mechanistically . Our finding that plants transfer lipids to AMF completely changes the previous view that the fungus receives only sugars from the plant ( Pfeffer et al . , 1999; Trépanier et al . , 2005 ) . It will now be interesting to determine the relative contributions of sugar and lipid transfer to AMF , and whether this may be a determinant of variation in root length colonization and extraradical mycelium formation depending on the plant-fungal genotype combination ( Sawers et al . , 2017 ) . An interesting question refers to why AMF have lost the genes encoding cytosolic FA synthase to depend on the lipid biosynthesis machinery of the host . FA biosynthesis consumes more energy than biosynthesis of carbohydrates and organic carbon provided by the plant needs to be transported in fungal hyphae over long distances from the inside of the root to the extremities of the extraradical mycelium . Therefore , it is conceivable that supply of plant lipids to the fungus plus fungal lipid transport is more energy efficient for the symbiosis as a whole than fungal carbohydrate transport plus fungal lipid biosynthesis . Hence , inter-organismic lipid transfer followed by loss of fungal FA biosynthesis genes may have been selected for during evolution because it likely optimized the symbiosis for most rapid proliferation of extraradical mycelium , thus ensuring efficient mineral nutrient acquisition from the soil for supporting the plant host . Lipid transfer across kingdoms has also been observed in human parasites or symbiotic bacteria of insects ( Caffaro and Boothroyd , 2011; Elwell et al . , 2011; Herren et al . , 2014 ) . It will be interesting to learn whether this is a more widespread phenomenon among biotrophic inter-organismic interactions . Lotus japonicus ecotype Gifu wild-type , ram1-3 , ram1-4 , dis-1 , dis-4 , dis-like-5 , ram2-1 , ram2-2 and ecotype MG-20 wild-type and str mutant ( kindly provided by Tomoko Kojima ( NARO , Tochigi , Japan ) seeds were scarified and surface sterilized with 1% NaClO . Imbibed seeds were germinated on 0 . 8% Bacto Agar ( Difco ) at 24°C for 10–14 days . Seedlings were cultivated in pots containing sand/vermiculite ( 2/1 vol . ) as substrate . For colonization with Rhizophagus irregularis roots were inoculated with 500 spores ( SYMPLANTA , Munich , Germany or Agronutrition , Toulouse , France ) per plant . Plants were harvested 5 weeks post inoculation ( wpi ) ; except for dis-1 complementation in Figure 1A , which was harvested at 4 wpi . Arabidopsis thaliana seeds of Col-0 wild-type , kasI mutant in the Col-0 background and the transgenically complemented kasI mutant were surface sterilized with 70% EtOH +0 . 05% Tween 20% and 100% EtOH , germinated on MS-Medium for 48 hr at 4°C in the dark followed by 5–6 days at 22°C ( 8 hr light/dark ) . The L . japonicus dis mutant ( line SL0154 , [Groth et al . , 2013] ) resulting from an EMS mutagenesis program ( Perry et al . , 2003 , 2009 ) was backcrossed to ecotype Gifu wild-type and outcrossed to the polymorphic mapping parent ecotype MG-20 . The dis locus segregated as a recessive monogenic trait and was previously found to be linked to marker TM2249 on chromosome 4 ( Groth et al . , 2013 ) . We confirmed the monogenic segregation as 26 of 110 individuals originating from the cross to MG-20 ( χ2: P ( 3:1 ) =0 . 74 ) and 32 of 119 individuals originating from the cross to Gifu ( χ2: P ( 3:1 ) =0 . 63 ) exhibited the mutant phenotype . To identify SL0154-specific mutations linked to the dis locus , we employed a genome re-sequencing strategy . Nuclear DNA of Gifu wild-type and the SL0154 mutant was subjected to paired end sequencing ( 2 × 100 bp ) of a 300–500 bp insert library , on an Illumina Hi-Seq 2000 instrument resulting in between 16 . 7 and 19 . 5 Gigabases per sample , equivalent to roughly 35–41 fold coverage assuming a genome size of 470 Megabases . Reads were mapped to the reference genome of MG-20 v2 . 5 ( Sato et al . , 2008 ) and single nucleotide polymorphisms identified using CLC genomics workbench ( CLC bio , Aarhus , Denmark ) . SL0154-specific SNPs were identified by subtracting Gifu/MG-20 from SL0154/MG-20 polymorphisms . 19 potentially EMS induced ( 11x G->A , 8x C->T ) SNPs called consistently in all mapped reads from SL0514 but not in Gifu were identified between the markers TM0046/TM1545 , the initial dis target region ( Figure 1—figure supplement 1A . In a screen for recombination events flanking the dis locus , 63 mutants out of 254 total F2 individuals of a cross MG-20 x SL0154 were genotyped with markers flanking the dis locus ( Figure 1—figure supplement 1B ) . Interrogating recombinant individuals with additional markers in the region narrowed down the target interval between TM2249 and BM2170 ( 2 cM according to markers; ca . 650 kb ) . In this interval , 3 SL0154-specific SNPs with typical EMS signature ( G to A transition ) remained , of which one was predicted to be located in exon 3 of CM0004 . 1640 . r2 ( reference position 40381558 in L . japonicus genome version 2 . 5; http://www . kazusa . or . jp/lotus/ ) , a gene annotated as ketoacyl- ( acyl carrier protein ) synthase . This co-segregation together with phenotyping of one additional mutant allele obtained through TILLING ( Supplementary file 1 , [Perry et al . , 2003 , 2009] ) as well as transgenic complementation ( Figure 1A ) ) confirmed the identification of the mutation causing the dis phenotype of the SL0514 line . The two remaining mutations in the target region were located in a predicted intron of chr4 . CM0004 . 1570 . r2 . a , a cyclin-like F-box protein ( reference position: 40356684 ) and in a predicted intergenic region ( reference position: 40364479 ) . Untranslated regions of DIS and DIS-LIKE were determined using the Ambion FirstChoice RLM RACE kit according to manufacturer`s instructions ( http://www . ambion . de/ ) . DIS sequence information can be found under the NCBI accession number KX880396 . The L . japonicus Gifu mutant reduced and degenerate arbuscules ( red , line SL0181-N ) resulting from an EMS mutagenesis ( Perry et al . , 2003 , 2009 ) was outcrossed to the ecotype MG-20 and previously reported to segregate for two mutations , one on chromosome 1 and one on chromosome 6 ( Groth et al . , 2013 ) . They were separated by segregation and the mutation on chromosome 1 was previously found in the GRAS transcription factor gene REDUCED ARBUSCULAR MYCORRHIZA 1 ( RAM1 ) ( Pimprikar et al . , 2016 ) . A plant from the F2 population , which showed wild-type phenotype but was heterozygous for the candidate interval on chromosome 6 and homozygous Gifu for the candidate interval on chromosome 1 was selfed for producing an F3 . The F3 generation segregated for only one mutation as 38 out of 132 individuals exhibited the mutant phenotype ( χ2: P ( 3:1 ) =0 . 68 ) . A plant from the F3 population , which displayed wild-type phenotype but was heterozygous for the candidate interval on chromosome 6 was selfed for producing an F4 . The F4 generation also segregated for only one mutation as 17 out of 87 individuals exhibited the mutant phenotype ( χ2: P ( 3:1 ) =0 . 76 ) . To identify the mutation on chromosome 6 linked to the previously identified interval ( Groth et al . , 2013 ) , we employed additional markers for fine mapping in F3 segregating and F4 mutant populations . This positioned the causative mutation between TM0082 and TM0302 ( Figure 1—figure supplement 3A ) . Due to a suppression of recombination in this interval we could not get closer to the mutation and also next generation sequencing ( see [Pimprikar et al . , 2016] for the methodology ) failed to identify a causative mutation . The Medicago truncatula ram2 mutant displays stunted arbuscules similar to our mutant ( Wang et al . , 2012 ) . L . japonicus RAM2 had not been linked to any chromosome but was placed on chromosome 0 , which prevented identification of a RAM2 mutation in the target interval on chromosome 6 . Therefore , we sequenced the RAM2 gene by Sanger sequencing . Indeed , mutants with stunted arbuscule phenotype in the F3 and F4 generation carried an EMS mutation at base 1663 from G to A leading to amino acid change from Glycine to Glutamic acid , which co-segregated with the mutant phenotype ( Figure 1—figure supplement 3B-C ) . An additional allelic mutant ram2-2 ( Figure 1—figure supplement 3B ) caused by a LORE1 retrotransposon insertion ( Małolepszy et al . , 2016 ) and transgenic complementation with the wild-type RAM2 gene confirmed that the causative mutation affects RAM2 ( Figure 1B ) . Untranslated regions of RAM2 were determined using the Ambion FirstChoice ( R ) RLM RACE kit according to manufacturer’s instructions ( http://www . ambion . de/ ) . A 1345 bp long sequence upstream of ATG was available from the http://www . kazusa . or . jp/lotus/blast . html . To enable cloning a 2275 bp promoter fragment upstream of ATG of RAM2 the remaining upstream sequence of 1047 bp was determined by primer walking on TAC Lj T46c08 . L . japonicus RAM2 sequence information can be found under the NCBI accession number KX823334 and the promoter sequence under the number KX823335 . Genes and promoter regions were amplified using Phusion PCR according to standard protocols and using primers indicated in Supplementary file 2 . Plasmids were constructed as indicated in Supplementary file 3 . For localization of DIS in L . japonicus hairy roots the LIII tricolor plasmid ( Binder et al . , 2014 ) was used . The plasmid containing 35S:RFP for localization of free RFP in Nicotiana benthamiana leaves was taken from Yano et al . ( 2008 ) . Hypocotyls of L . japonicus were transformed with plasmids shown in Supplementary file 3 for hairy root induction using transgenic Agrobacterium rhizogenes AR1193 as described ( Takeda et al . , 2009 ) . Five plants per pot were sown . One week before transformation the primary bolt was cut off to induce growth of secondary floral bolts . 5 ml LB culture of A . tumefaciens transformed with a binary vector was incubated at 28°C , 300 rpm over night . 500 μl of the preculture was added to 250 μl LB medium with appropriate antibiotics . This culture was incubated again at 28°C , 300 rpm over night until an OD600 of 1 . 5 was reached . Plants were watered and covered by plastic bags the day before the dipping to ensure high humidity . The cells were harvested by centrifugation ( 10 min , 5000 rpm ) and resuspended in infiltration medium ( 0 . 5 x MS medium , 5% sucrose ) . The resuspended cell culture was transferred to a box and Silwet L-77 was added ( 75 μl to 250 ml medium ) . The floral bolts of the plants were dipped into the medium for 5 s and put back into plastic bags and left in horizontal position for one night . After that , plants were turned upright , bags were opened and mature siliques were harvested . For rosette growth assays T3 plants were used . 31 days post sowing the rosettes were photographed and then cut and dried in an oven at 65°C for the determination of rosette dry weight . For promoter:GUS analysis L . japonicus hairy roots transformed with plasmids containing the DIS and RAM2 promoter fused to the uidA gene and colonized by R . irregularis were subjected to GUS staining as described ( Takeda et al . , 2009 ) . To correlate DIS and RAM2 promoter activity precisely with the stage of arbuscule development two expression cassettes were combined in the same golden gate plasmid for simultaneous visualization of arbuscule stages and promoter activity . The fungal silhouette including all stages of arbuscule development and pre-penetration apparatuus were made visible by expressing secretion peptide coupled mCherry under the control of the SbtM1 promoter region comprising 704 bp upstream of the SbtM1 gene ( Takeda et al . , 2009 ) . Promoter activity was visualized using a YFP reporter fused to a nuclear localization signal ( NLS ) . N . benthamiana leaves were transiently transformed by infiltration of transgenic A . tumefaciens AGL1 as described ( Yano et al . , 2008 ) . For analysis of transcript levels , plant tissues were rapidly shock frozen in liquid nitrogen . RNA was extracted using the Spectrum Plant Total RNA Kit ( www . sigmaaldrich . com ) . The RNA was treated with Invitrogen DNAse I amp . grade ( www . invitrogen . com ) and tested for purity by PCR . cDNA synthesis was performed with 500 ng RNA using the Superscript III kit ( www . invitrogen . com ) . qRT-PCR was performed with GoTaq G2 DNA polymerase ( Promega ) , 5 x colorless GoTaq Buffer ( Promega ) and SYBR Green I ( Invitrogen S7563 , 10 . 000x concentrated , 500 µl ) - diluted to 100x in DMSO . Primers ( Supplementary file 2 ) were designed with primer3 ( 58 ) . The qPCR reaction was run on an iCycler ( Biorad , www . bio-rad . com/ ) according to manufacturer’s instructions . Thermal cycler conditions were: 95°C 2 min , 45 cycles of 95°C 30 s , 60°C/62°C 30 s and 72°C 20 s followed by dissociation curve analysis . Expression levels were calculated according to the ΔΔCt method ( Rozen and Skaletsky , 2000 ) . For each genotype and treatment three to four biological replicates were tested and each sample was represented by two to three technical replicates . L . japonicus KASI , DIS , DIS-LIKE , RAM2 , Lj1g3v2301880 . 1 ( GPAT6 ) protein sequences were retrieved from Lotus genome V2 . 5 and V3 . 0 respectively ( http://www . kazusa . or . jp/lotus/ ) and A . thaliana KASI , E . coli KASI , E . coli KASII , M . truncatula RAM2 and Medtr7g067380 ( GPAT6 ) were obtained from NCBI ( http://www . ncbi . nlm . nih . gov ) . The sequences from L . japonicus were confirmed with a genome generated by next generation sequencing in house . Protein alignment for DIS was performed by CLC Main Workbench ( CLC bio , Aarhus , Denmark ) . The Target Peptide was predicted using TargetP 1 . 0 Server ( www . cbs . dtu . dk/services/TargetP-1 . 0/ ) . RAM2 Protein alignment was performed by MEGA7 using ClustalW . The percentage identity matrix was obtained by Clustal Omega ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) . To collect sequences for phylogeny construction corresponding to potential DIS orthologs , Lotus DIS and KASI ( outgroup ) protein sequences were searched in genome and transcriptome datasets using BLASTp and tBLASTn respectively . The list of species and the databases used are indicated in Figure 3—source data 1 . Hits with an e-value >10−50 were selected for the phylogenetic analysis . Collected sequences were aligned using MAFFT ( http://mafft . cbrc . jp/alignment/server/ ) and the alignment manually checked with Bioedit . Phylogenetic trees were generated by Neighbor-joining implemented in MEGA5 ( Tamura et al . , 2011 ) . Partial gap deletion ( 95% ) was used together with the JTT substitution model . Bootstrap values were calculated using 500 replicates . A ~200 kb sized region in the L . japonicus genome containing the DIS locus ( CM00041640 . r2 . a ) was compared to the syntenic region in A . thaliana ( Col-0 ) using CoGe Gevo ( https://genomevolution . org/CoGe/GEvo . pl - ( Lyons et al . , 2008 ) as described in Delaux et al . ( 2014 ) . Loci encompassing DIS orthologs from Medicago truncatula , Populus trichocarpa , Carica papaya , Phaseolus vulgaris and Solanum lycopersicum were added as controls . Rhizophagus irregularis in colonized L . japonicus roots was stained with acid ink ( Vierheilig et al . , 1998 ) . Root length colonization was quantified using a modified gridline intersect method ( McGonigle et al . , 1990 ) . For confocal laser scanning microscopy ( CLSM ) fungal structures were stained with 1 μg WGA Alexa Fluor 488 ( Molecular Probes , http://www . lifetechnologies . com/ ) ( Panchuk-Voloshina et al . , 1999 ) . For quantification of AM colonization in L . japonicus roots a light microscope ( Leica ) with a 20x magnification was used . For observation of GUS-staining in L . japonicus hairy roots an inverted microscope ( Leica DMI6000 B ) was used with 10x and 20x magnification . Transformed roots were screened by stereomicroscope ( Leica MZ16 FA ) using an mCherry fluorescent transformation marker or the pSbtM1:mCherry marker for fungal colonization ( for Figure 2A and B ) . Confocal microscopy ( Leica SP5 ) for WGA-AlexaFluor488 detection using 20x and 63x magnification was performed as described ( Groth et al . , 2010 ) . Transgenic roots showing mCherry fluorescence signal due to SbtM1 promoter activity linked with fungal colonization were cut into pieces immediately after harvesting . The living root pieces were placed on a glass slide with a drop of water , covered by a cover slip and immediately subjected to imaging . Sequential scanning for the YFP and RFP signal was carried out simultaneously with bright field image acquisition . YFP was excited with the argon ion laser 514 nm and the emitted fluorescence was detected from 525 to 575 nm; RFP was excited with the Diode-Pumped Solid State laser at 561 nm and the emitted fluorescence was detected from 580 to 623 nm . Images were acquired using LAS AF software . Several z-optical sections were made per area of interest and assembled to a z-stack using Fiji . The z-stack movies and 3D projections were produced using the 3D viewer function in Fiji ( Schindelin et al . , 2012 ) . Approximately 50–100 mg of root or leaf material was harvested , weighed and immediately frozen in liquid nitrogen to avoid lipid degradation . The frozen samples were ground to a fine powder before extraction with organic solvents . Total lipids were extracted as described previously ( Wewer et al . , 2011 , 2014 ) . Briefly , 1 mL chloroform/methanol/formic acid ( 1:1:0 . 1 , v/v/v ) was added and the sample was shaken vigorously . At this point the internal standards for TAG and fatty acid analysis were added . Phase separation was achieved after addition of 0 . 5 mL 1M KCl/0 . 2 M H3PO4 and subsequent centrifugation at 4000 rpm for 5 min . The lipid-containing chloroform phase was transferred to a fresh glass tube and the sample was re-extracted twice with chloroform . The combined chloroform phases were dried under a stream of air and lipids were re-dissolved in 1 mL chloroform to yield the total lipid extract . For phospho- and glycerolipid analysis 20 µl of the total lipid extract were mixed with 20 µl of the internal standard mix and 160 µl of methanol/chloroform/300 mM ammonium acetate ( 665:300:35 , v/v/v ) ( Welti et al . , 2002 ) . For triacylglycerol analysis 500 µl of the total lipid extract were purified by solid phase extraction on Strata silica columns ( 1 ml bed volume; Phenomenex ) as described ( Wewer et al . , 2011 ) . TAGs were eluted from the silica material with chloroform , dried under a stream of air and re-dissolved in 1 mL methanol/chloroform/300 mM ammonium acetate ( 665:300:35 , v/v/v ) . Total lipids were extracted into chloroform and dried as described above . 15–0 FA and a mixture of 15–0 α-MAG and β-MAG were added as internal standard before the extraction . Dried extracts were resuspended in 1 ml n-hexane and applied to silica columns for solid-phase extraction with a n-hexane:diethylether gradient . Free fatty acids were eluted with a mixture of 92:8 ( v/v ) n-hexane:diethylether as described bevore ( Gasulla et al . , 2013 ) and pure diethylether were used for elution of MAG . For measurement of total fatty acids , 100 μl of the total lipid extract were used . For measurement of free fatty acids , the SPE-fraction containing free fatty acids was used . Fatty acid methyl esters ( FAMEs ) were generated from acyl groups of total lipids and free fatty acids by addition of 1 mL 1N methanolic HCL ( Sigma ) to dried extracts and incubation at 80°C for 30 min ( Browse et al . , 1986 ) . Subsequently , FAMEs were extracted by addition of 1 mL n-hexane and 1 mL of 0 . 9% ( w/v ) NaCl and analyzed on a gas chromatograph with flame-ionization detector ( GC-FID , Agilent 7890A PlusGC ) . FAMEs were separated on an SP 2380 fused silica GC column ( Supelco , 30 mx 0 . 53 mm , 0 . 20 μm film ) as described ( Wewer et al . , 2013 ) , with a temperature -gradient starting at 100°C , increased to 160°C with 25°C/min , then to 220°C with10°C/min and reduced to 100°C with 25 °C/min . FAMEs were quantified in relation to the internal standard pentadecanoic acid ( 15:0 ) . For MAG measurement , dried diethylether fractions were resuspended in 4:1 ( v/v % ) pyridine:N-Methyl-N- ( trimethylsilyl ) trifluoroacetamide ( MSTFA ) , incubated at 80°C for 30 min , dried and re-suspended in hexane prior to application on an Agilent 7890A Plus gas chromatography-mass spectrometer . MAGs were quantified by extracted ion monitoring , using [M+ - 103] for α-MAGs and [M+ - 161] for β-MAGs as previously reported for 16:0 MAG ( Destaillats et al . , 2010 ) and 24:0 MAG ( Li et al . , 2007 ) . Phosphoglycerolipids ( PC , PE , PG , PI , PS ) , glycoglycerolipids ( MGDG , DGDG , SQDG ) and triacylglycerol ( TAG ) were analyzed in positive mode by direct infusion nanospray Q-TOF MS/MS on an Agilent 6530 Q-TOF instrument as described previously ( Lippold et al . , 2012; Gasulla et al . , 2013 ) . A continuous flow of 1 µl/min methanol/chloroform/300 mM ammonium acetate ( 665:300:35 , v/v/v ) ( Welti et al . , 2002 ) was achieved using a nanospray infusion ion source ( HPLC/chip MS 1200 with infusion chip ) . Data are displayed as X:Y , where X gives the number of C atoms of the fatty acid chain and Y the amount of desaturated carbo-carbon bonds inside that fatty acid chain . Internal standards for phospho- and glycoglycerolipid analysis were prepared as described previously ( Gasulla et al . , 2013; Wewer et al . , 2014 ) . The following standards were dissolved in 20 µl of chloroform/methanol ( 2:1 , v/v ) : 0 . 2 nmol of each di14:0-PC , di20:0-PC , di14:0-PE , di20:0-PE , di14:0 PG , di20:0 PG , di14:0 PA and di20:0 PA; 0 . 03 nmol of di14:0-PS and di20:0-PS; 0 . 3 nmol of 34:0-PI; 0 . 15 nmol of 34:0-MGDG , 0 . 10 nmol of 36:0-MGDG; 0 . 2 nmol of 34:0-DGDG , 0 . 39 nmol of 36:0 DGDG and 0 . 4 nmol of 34:0 SQDG . 1 nmol each of tridecanoin ( tri-10:0 ) and triundecenoin ( tri-11:1 ) , and 2 nmol each of triarachidin ( tri-20:0 ) and trierucin ( tri22:1 ) were used as internal standards for TAG quantification ( Lippold et al . , 2012 ) . For quantification of total fatty acids and free fatty acids 5 µg of pentadecanoic acid ( FA 15:0 ) was added to the samples ( Wewer et al . , 2013 ) . The method for cultivation and stable isotope labelling of Lotus japonicus and Daucus carota hairy roots as well as for isotopolog profiling are described in more detail at Bio-protocol ( Keymer et al . , 2018 ) . To determine lipid transfer from L . japonicus to the fungus we used the carrot root organ culture system ( Bécard et al . , 1988 ) to obtain sufficient amounts of fungal material for isotopolog profiling . ( On petri dishes this was not possible with L . japonicus and in particular the lipid mutants alone ) . One compartment ( carrot compartment ) of the 2- compartmented petri dish system ( Trépanier et al . , 2005 ) was filled with MSR-medium ( 3% gelrite ) containing 10% sucrose to support the shoot-less carrot root , and the other compartment ( Lotus compartment ) was filled with MSR-medium ( 3% gelrite ) without sucrose . Ri T-DNA transformed Daucus carota hairy roots were placed in the carrot compartment . 1 week later , roots were inoculated with R . irregularis . Petri dishes were incubated at constant darkness and 30°C . Within 5 weeks R . irregularis colonized the carrot roots and its extraradical mycelium spread over both compartments of the petri dish and formed spores . At this stage two 2 week old L . japonicus seedlings ( WT , dis-1 , ram2-1 ) were placed into the Lotus compartment ( Figure 8—figure supplement 1 ) . The plates were incubated at 24°C ( 16 hr light/8 hr dark ) . To keep the fungus and root in the dark the petri dishes were covered with black paper . 3 weeks after Lotus seedlings were placed into the petri dish [U-13C6]glucose ( 100 mg diluted in 2 ml MSR-medium ) ( Sigma-Aldrich ) was added to the Lotus compartment . Therefore , only Lotus roots but not the carrot roots took up label . For transfer experiments with carrot roots no Lotus plant was placed into the Lotus compartment and the [U-13C6]glucose was added to the carrot compartment . 1 week after addition of [U-13C6]glucose the roots were harvested . The extraradical mycelium was extracted from the agar using citrate buffer pH 6 and subsequent filtration , after which it was immediately shock-frozen in liquid nitrogen . Root and fungal samples were freeze dried and subsequently derivatised with 500 µl MeOH containing 3 M HCl ( Sigma-Aldrich ) at 80°C for 20 hr . MeOH/HCL was removed under a gentle stream of nitrogen and the methyl esters of the fatty acids were solved in 100 µl dry hexane . Gas chromatography mass spectrometry was performed on a GC-QP 2010 plus ( Shimadzu , Duisburg , Germany ) equipped with a fused silica capillary column ( equity TM-5; 30 m by 0 . 25 mm , 0 . 25-µm film thickness; Supelco , Bellafonte , PA ) . The mass detector worked in electron ionization ( EI ) mode at 70 eV . An aliquot of the solution was injected in split mode ( 1:5 ) at an injector and interface temperature of 260°C . The column was held at 170°C for 3 min and then developed with a temperature gradient of 2 °C/min to a temperature of 192°C followed by a temperature gradient of 30°C/min to a final temperature of 300°C . Samples were analyzed in SIM mode ( m/z values 267 to 288 ) at least three times . Retention times for fatty acids 16:1ω5 ( unlabeled m/z 268 ) and 16:0 ( unlabeled m/z 270 ) are 12 . 87 min and 13 . 20 min , respectively . Data were collected with LabSolution software ( Shimadzu , Duisburg , Germany ) . The overall 13C enrichment and the isotopolog compositions were calculated according to ( Lee et al . , 1991 ) and ( Ahmed et al . , 2014 ) . The software package is open source and can be downloaded by the following link: http://www . tr34 . uni-wuerzburg . de/software_developments/isotopo/ . Four independent labeling experiments were performed . Overall excess ( o . e . ) is an average value of 13C atoms incorporated into 16:0/16:1ω5 fatty acids . Lunularia cruciata: For this species , the raw RNAseq reads have been previously deposited to NCBI under the accession number SRR1027885 . It is annotated with Rhizophagus irregularis ( 10% of sequences ) as the transcriptome was partly prepared from Lunularia plant tissue colonized by the fungus Rhizophagus irregularis . The corresponding Lunularia transcriptomic assembly is available at www . polebio . lrsv . ups-tlse . fr/Luc_v1/ All statistical analyses ( Source code 1 ) were performed and all boxplots were generated in R ( www . r-project . org ) .	Most land plants are able to form partnerships with certain fungi – known as arbuscular mycorrhiza fungi – that live in the soil . These fungi supply the plant with mineral nutrients , especially phosphate and nitrogen , in return for receiving carbon-based food from the plant . To exchange nutrients , the fungi grow into the roots of the plant and form highly branched structures known as arbuscules inside plant cells . Due to the difficulties of studying this partnership , it has long been believed that plants only provide sugars to the fungus . However , it has recently been discovered that these fungi lack important genes required to make molecules known as fatty acids . Fatty acids are needed to make larger fat molecules that , among other things , store energy for the organism and form the membranes that surround each of its cells . Therefore , these results raise the possibility that the plant may provide the fungus with some of the fatty acids the fungus needs to grow . Keymer , Pimprikar et al . studied how arbuscules form in a plant known as Lotus japonicus , a close relative of peas and beans . The experiments identified a set of mutant L . japonicus plants that had problems forming arbuscules . These plants had mutations in several genes involved in fat production that are only active in plant cells containing arbuscules . Further experiments revealed that certain fat molecules that are found in fungi , but not plants , were present at much lower levels in samples from mutant plants colonized with the fungus , compared to samples from normal plants colonized with the fungus . This suggests that the fungi colonizing the mutant plants may be starved of fat molecules . Using a technique called stable isotope labelling it was possible to show that fatty acids made in normal plants can move into the colonizing fungus . The findings of Keymer , Pimprikar et al . provide evidence that the plant feeds the fungus not only with sugars but also with fat molecules . The next challenge will be to find out exactly how the fat molecules are transferred from the plant cell to the fungus . Many crop plants are able to form partnerships with arbuscular mycorrhizal fungi . Therefore , a better understanding of the role of fat molecules in these relationships may help to breed crop plants that , by providing more support to their fungal partner , may grow better in the field .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2017	Lipid transfer from plants to arbuscular mycorrhiza fungi
Morphogenesis and physiology of tissues and organs requires planar cell polarity ( PCP ) systems that orient and coordinate cells and their behaviors , but the relationship between PCP systems has been controversial . We have characterized how the Frizzled and Dachsous-Fat PCP systems are connected through the Spiny-legs isoform of the Prickle-Spiny-legs locus . Two different components of the Dachsous-Fat system , Dachsous and Dachs , can each independently interact with Spiny-legs and direct its localization in vivo . Through characterization of the contributions of Prickle , Spiny-legs , Dachsous , Fat , and Dachs to PCP in the Drosophila wing , eye , and abdomen , we define where Dachs-Spiny-legs and Dachsous-Spiny-legs interactions contribute to PCP , and provide a new understanding of the orientation of polarity and the basis of PCP phenotypes . Our results support the direct linkage of PCP systems through Sple in specific locales , while emphasizing that cells can be subject to and must ultimately resolve distinct , competing PCP signals . Planar cell polarity ( PCP ) is the coordinated orientation of cell structures and behaviors within the plane of a tissue . Manifestations of PCP include the orientation of hairs , bristles , stereocilia , and ommatidia , as well as orientated cell divisions and cell movements . Two conserved molecular systems play key roles in the establishment and maintenance of PCP: the Frizzled ( Fz ) PCP pathway and the Dachsous ( Ds ) -Fat PCP pathway ( Goodrich and Strutt , 2011; Matis and Axelrod , 2013 ) . These involve distinct components , but share common features , including the polarized localization of key components within cells , and the existence of asymmetric intercellular interactions that enable this polarity to be propagated from cell to cell . Many processes are influenced by both PCP systems , but the relationship between them has been controversial . The Drosophila Ds-Fat PCP pathway includes the cadherin family proteins Ds and Fat , which interact between neighboring cells ( Ma et al . , 2003; Matakatsu and Blair , 2004; Strutt and Strutt , 2002 ) . Binding between Ds and Fat is modulated by Four-jointed ( Fj ) , a Golgi-localized kinase that phosphorylates their cadherin domains ( Brittle et al . , 2010; Ishikawa et al . , 2008; Simon et al . , 2010 ) . Ds and Fj are expressed in opposing gradients , which orient Ds-Fat PCP ( Casal et al . , 2002; Mao et al . , 2006; Strutt and Strutt , 2002; Yang et al . , 2002; Zeidler et al . , 1999; 2000 ) . Fat protein in a cell within a Ds-Fj gradient preferentially accumulates along the side where it contacts cells with higher Ds and lower Fj; Ds protein localizes in a complementary orientation ( Figure 1A ) ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . How polarization of Ds and Fat proteins establishes PCP is incompletely understood , but it is achieved in part through the unconventional myosin Dachs , whose membrane localization is regulated by Fat ( Mao et al . , 2006 ) . Mammalian homologues of Ds and Fat , Dchs1 and Fat4 , also influence PCP ( Mao et al . , 2011a; Saburi et al . , 2008; Zakaria et al . , 2014 ) , and human FAT4 can rescue PCP phenotypes of Drosophila fat mutants ( Pan et al . , 2013 ) . 10 . 7554/eLife . 09946 . 003Figure 1 . Localization of Pk and Sple in wing discs , and their interaction with Dachs and Ds . ( A ) Schematic diagram illustrating the general direction of PCP protein polarity ( arrows ) , expression gradient of Ds ( magenta ) and organization of Ds-Fat and Fz PCP pathway components in the Drosophila wing disc . ( B ) Western blots , using antibodies indicated on the right , showing the results of co-immunoprecipitation experiments between V5-tagged Dachs ( lanes 1–3 , 7 ) , Ds-ICD ( lanes 4–6 , 8 ) or GFP ( lanes 9–11 ) of Flag-tagged Sple ( lanes 1 , 4 , 9 ) , Sple-N ( lanes 2 , 5 , 10 ) , Pk ( lanes 3 , 6 , 11 ) or GFP ( lanes 7 , 8 ) . Upper panels ( Input ) show blots on lysates of S2 cells , lower panels ( IP-V5 ) show blots on proteins precipitated from these lysates by anti-V5 beads . Similar results were obtained in three independent biological replicates of this experiment . ( C–G ) Portions of wing imaginal discs with clones of cells expressing GFP:Sple ( C , E , F ) or GFP:Pk ( D , G ) ( green ) , stained for expression of E-cadherin ( blue ) , and showing either anti-Wg ( C–E ) or hh-Gal4 UAS-mCD8-RFP ( F , G ) ( red ) . White arrows indicate direction of polarization of Sple or Pk . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 00310 . 7554/eLife . 09946 . 004Figure 1—figure supplement 1 . Proteins used in co-immunoprecipitation assays . Schematics of tagged isoforms of full length Dachs , Ds intracellular domain , full length Sple , Sple N-terminal domain and full length Pk . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 00410 . 7554/eLife . 09946 . 005Figure 1—figure supplement 2 . Ds and Fj gradients in wing discs . ( A , B ) Expression of Ds ( magenta ) and fj-lacZ ( cyan ) in a wild-type wing disc . ( C ) Portion of a wing disc with clones of cells expressing Dachs:Cit ( green ) , stained for Wg ( red ) and E-cadherin ( blue ) . White arrow indicates the direction of Dachs polarization . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 005 The Drosophila Fz PCP pathway includes the asymmetrically distributed transmembrane proteins Fz and Van Gogh ( Vang , also known as Strabismus ) , which act together with the cadherin family protein Starry night ( Stan , also known as Flamingo ) ( Figure 1A ) ( Chae et al . , 1999; Goodrich and Strutt , 2011; Park et al . , 1994; Shimada et al . , 2001; Strutt , 2001; Usui et al . , 1999; Vinson and Adler , 1987 ) . Stan interacts with Stan-Fz heterodimers in neighboring cells ( Chen et al . , 2008; Struhl et al . , 2012; Strutt and Strutt , 2008 ) ; interactions between Vang and Fz have also been reported ( Wu and Mlodzik , 2008 ) . Each of these transmembrane complexes is associated with distinct cytoplasmic proteins , Fz-Stan associates with Dishevelled ( Dsh ) and Diego ( Dgo ) ( Axelrod , 2001; Feiguin et al . , 2001; Shimada et al . , 2001; Strutt , 2001 ) . Vang-Stan associates with Prickle-Spiny legs ( Pk-Sple ) ( Bastock et al . , 2003; Jenny et al . , 2003 ) . Polarization of these protein complexes can propagate from cell to cell through heterophilic intercellular interaction between Stan and Fz-Stan complexes ( Chen et al . , 2008; Goodrich and Strutt , 2011; Struhl et al . , 2012 ) . Polarization of complexes within a cell is reinforced by inhibitory intracellular interactions between asymmetrically localized components ( Goodrich and Strutt , 2011 ) . Mutation of any of the core members of the Fz PCP system will impair the polarization of all of the others , emphasizing their mutual dependency for polarization ( Strutt and Strutt , 2009 ) . Genetic interactions between Ds-Fat pathway genes and Fz pathway genes , together with observations of altered Fz pathway protein localization in Ds-Fat pathway mutants , led to suggestions that the Ds-Fat pathway acts upstream of the Fz pathway ( Ma et al . , 2003; Yang et al . , 2002 ) . According to this hypothesis , the Ds-Fat pathway acts as a ‘global module’ that provides long-range directional information through tissue-wide Fj and Ds gradients , whereas the Fz pathway acts as a ‘core module’ that establishes robust polarization that can propagate locally , and effects cellular polarity . This suggestion was challenged by observations that clones of cells mutant for or over-expressing ds , fj or fat in the abdomen can affect PCP non-autonomously even in the absence of Fz pathway components ( Casal et al . , 2006 ) . Additionally , in the abdomen , combining mutations in both Ds-Fat and Fz pathway genes can have more severe effects on PCP than single mutants , suggesting that these pathways can act in parallel ( Casal et al . , 2006; Donoughe and DiNardo , 2011; Repiso et al . , 2010 ) . There are also some manifestations of PCP , such as oriented cell divisions in the developing wing , which are influenced by the Ds-Fat pathway and not the Fz pathway ( Baena-Lopez et al . , 2005 ) . Nonetheless , other studies have provided evidence of cross-talk between PCP systems , and implicated the Pk-Sple locus in helping to mediate this cross-talk . The Pk-Sple locus produces two functional isoforms: Prickle ( Pk ) and Spiny-legs ( Sple ) , which share a common , LIM-domain containing C-terminus , but unique N-termini ( Figure 1—figure supplement 1 ) ( Gubb et al . , 1999 ) . These isoforms have distinct roles: for example , mutations that only affect pk disrupt PCP in the wing and notum , but not in the eye and leg , whereas mutations that only affect sple exhibit a complementary specificity . The observations that mutations that affect both isoforms ( pk-sple ) have milder effects on PCP than isoform-specific alleles in the wing , notum , and leg , and that over-expression of Sple or Pk leads to PCP phenotypes reminiscent of loss-of-function of pk , or sple , respectively , led to the suggestion that a balance between Pk and Sple isoforms is necessary for normal PCP ( Gubb et al . , 1999 ) . Studies of PCP establishment in the pupal wing revealed that it occurs in distinct phases , and suggested that influences of Sple are correlated with influences of the Ds-Fat pathway ( Hogan et al . , 2011; Merkel et al . , 2014 ) . Moreover , examination of PCP protein localization revealed that a coupling between the polarization of components of the Ds-Fat and Fz systems could be induced by expression of Sple ( Merkel et al . , 2014 ) . Additionally , in the abdomen , the Fj and Ds gradients are oriented oppositely within anterior ( A ) versus posterior ( P ) compartments of each segment ( Casal et al . , 2002 ) . Since hairs always point posteriorly , this led to the suggestion that there could exist a ‘rectification’ mechanism , which would reverse the influence of these gradients on hair polarity . The observations that Sple over-expression reverses polarity in P compartments , and that pk-sple mutants reverse polarity in part of the A compartment , led to the suggestion that Pk and Sple might be involved in this rectification ( Lawrence et al . , 2004 ) . Two potential mechanisms by which Pk-Sple might influence the relationship between PCP pathways have recently been suggested . It was reported that Dachs could directly interact with Pk and Sple , and that Ds and Dachs could influence Sple localization in wing discs ( Ayukawa et al . , 2014 ) . It has also been proposed that Pk-Sple could connect PCP pathways through an influence on microtubule orientation ( Olofsson et al . , 2014 ) . Vesicles containing Fz and Dsh have been observed to move along apical non-centrosomal microtubules towards the distal side of wing cells , with the proximal-distal alignment of microtubules and consequent directional transport of Fz pathway components dependent upon the Ds-Fat pathway ( Harumoto et al . , 2010; Matis et al . , 2014; Olofsson et al . , 2014; Shimada et al . , 2006 ) . Pk and Sple also influence the orientation of apical microtubules , such that the plus ends of microtubules are preferentially found at either the high end or the low end of the Ds gradient , depending on whether Pk or Sple , respectively , is the predominant isoform ( Matis et al . , 2014; Olofsson et al . , 2014 ) . Relative differences in expression of isoforms consistent with their distinct requirements have also been reported: Pk at higher levels than Sple in larval wing discs , and Sple at higher levels than Pk in eye discs ( Ayukawa et al . , 2014; Merkel et al . , 2014; Olofsson et al . , 2014 ) . While these studies are suggestive of a key role for Pk-Sple in linking PCP pathways , the extent to which these or other mechanisms link PCP pathways , and their contribution to orienting PCP , remain unclear . Here , we demonstrate that Dachs and Ds can each physically interact with Sple , and control its localization in the wing , eye and abdomen . Our studies complement observations of Ayukawa et al . ( 2014 ) in identifying requirements for Dachs and Ds in Sple localization , but differ regarding the nature of these requirements . We also extend understanding of the relationship between Ds-Fat and Fz PCP pathways by identifying organ and region-specific differences in their interactions , and illustrate how this relationship between pathways can explain poorly understood features of PCP mutant phenotypes . Our results establish control of Sple localization as a key mechanism by which the Ds-Fat pathway coordinates with Fz to influence PCP , and enhance our understanding of how PCP is coordinated in developing tissues . Components of each of the two major PCP pathways are polarized along the proximal-distal axis of the larval wing imaginal disc and pupal wing ( Figure 1A ) ( Goodrich and Strutt , 2011; Matis and Axelrod , 2013 ) . Components of both pathways are required for the normal distal orientation of wing hairs . Motivated by observations implicating the pk-sple locus in modulating the influence of the Ds-Fat pathway on wing hair and ridge polarity ( Hogan et al . , 2011 ) , we initiated experiments to examine the localization of the distinct Pk and Sple isoforms and their potential regulation by Ds-Fat PCP . This was achieved by expressing GFP-tagged isoforms in clones of cells . Consistent with recent studies examining isoform-specific localization ( Ayukawa et al . , 2014; Sagner et al . , 2012; Strutt et al . , 2013 ) , we observed that GFP:Pk was polarized towards the proximal sides of cells , except just anterior to the anterior-posterior compartment boundary , where GFP:Pk was instead polarized towards the anterior sides of cells ( Figure 1D , G , 2I ) . By contrast , GFP:Sple was polarized towards the distal sides of cells throughout the wing disc ( Figure 1C , E , F , 2I ) . The distinct localization of Pk and Sple expressed in wing discs indicates that they can respond to distinct spatial cues . 10 . 7554/eLife . 09946 . 006Figure 2 . Localization of Sple in Ds-Fat pathway mutants . ( A–H ) Portions of wing imaginal discs with clones of cells expressing GFP:Sple ( green ) in dGC13/ d210 ( A and B ) , ft8/ ftG-rv ( C ) , ds36D/ dsUA071 ( D ) , dGC13 ds36D/ dGC13 dsUA071 ( E and F ) and dGC13 ft8/ dGC13 ftG-rv ( G and H ) mutants . Discs were stained for E-cad ( blue ) and Wg ( red ) . Wg is expressed along the D-V boundary and in proximal rings , the locations of Wg expression shown are indicated . White arrows indicate general direction of Sple polarization . ( I ) Rose plots summarizing orientation of Sple or Pk in the indicated genotypes , with proximal at left , distal at right , and in the central row , anterior at top . Orientations were scored separately in the three regions depicted in the cartoon , the number of cells scored is indicated by N . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 006 The localization of Sple to the distal side of wing disc cells is similar to that of Dachs and Ds ( Figure 1—figure supplement 2 ) ( Ambegaonkar et al . , 2012; Brittle et al . , 2012; Mao et al . , 2006; Rogulja et al . , 2008 ) . To investigate whether this shared localization could reflect physical association , we assayed for co-immunoprecipitation of epitope-tagged proteins expressed in cultured Drosophila S2 cells ( Figure 1—figure supplement 1 ) . Indeed , V5-tagged Dachs could co-immunoprecipitate Flag-tagged Sple ( Figure 1B , lane 1 ) . Dachs and Sple interact through the unique N-terminus of Sple , because Dachs also co-precipitated a construct comprising only the Sple N-terminus ( Sple-N ) , but did not co-precipitate a full length Pk construct ( Figure 1B , lanes 2 and 3 ) . Interaction with Ds was investigated by expressing a construct comprising the entire intracellular domain of Ds ( Ds-ICD ) . Both Sple and Sple-N also interacted with Ds-ICD , whereas Pk did not ( Figure 1B , lanes 4–6 ) . We note that Ayukawa et al . ( 2014 ) similarly reported an ability of Dachs to interact with Sple , based on co-immunoprecipitation of proteins expressed in human HEK293 cells . However , our results differ in that they reported that Dachs could also interact with Pk , whereas we could not detect any interaction between Pk and Dachs above non-specific background ( defined by precipitation observed using GFP:V5 instead of Dachs:V5 , Figure 1B , lane 11 ) . Also , Ayukawa et al . ( 2014 ) reported that they could detect an interaction between Sple-N and Ds-ICD , but could not detect an interaction between Ds-ICD and full length Sple , leading them to suggest a requirement for other components such as Dachs , whereas we did detect this interaction ( Figure 1B , lane 4 ) . Altogether , our results establish that Dachs and Ds can each independently interact with Sple , and that they do so through its unique N-terminal region . To determine whether the shared distal localization and physical interaction between Dachs and Sple are reflective of a functional role for Dachs in localizing Sple , we examined GFP:Sple in dachs mutant wing discs . Indeed , GFP:Sple localization was altered , as throughout most of the developing wing disc its localization became similar to that of Pk: on the proximal side of cells , and in fewer , more discrete puncta ( Figure 2A , I ) . Along the A-P compartment boundary , GFP:Sple was instead localized towards anterior side of cells , as is GFP:Pk ( Figures 2I , 3A ) ( Sagner et al . , 2012 ) . Intriguingly , however , in the most proximal part of the wing pouch , GFP:Sple generally maintained a distal localization ( Figure 2B , I ) . Thus , Dachs is required for the distal localization of GFP:Sple throughout most of the wing pouch , but not in the proximal wing . 10 . 7554/eLife . 09946 . 007Figure 3 . Additional characterization of Pk and Sple localization in mutants . Portions of wing discs with clones of cells expressing GFP:Sple ( A , F , G ) or GFP:Pk ( B–E ) ( green ) in dGC13/ d210 ( A–C ) , ft8/ ftG-rv ( D ) , ds36D/ dsUA071 ( E ) , dGC13 pk30 ( F ) , and vangstbm6 ( G ) mutants . Discs were stained for E-cad ( blue ) and Wg ( red ) ( B , D , E , F and G ) or hh-Gal4 UAS-mCD8-RFP ( red ) ( A , C ) . The white arrows indicate direction of polarization . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 007 In ds mutant wing discs , GFP:Sple was in most cases unpolarized ( localized to cell membrane on all sides ) , but sometimes partially polarized ( on multiple cell sides but with a clear bias ) , or polarized in random directions ( Figure 2D , I ) . The observation of unpolarized GFP:Sple is consistent with the inference that Dachs can localize GFP:Sple , because Dachs is localized to all membranes in an unpolarized fashion in ds mutants ( Ambegaonkar et al . , 2012 ) . Dachs is similarly localized to the membrane in an unpolarized fashion in fat mutants ( Mao et al . , 2006 ) , and we always observed unpolarized membrane localization of GFP:Sple in fat mutant wing discs ( Figure 2C ) . To confirm that mis-localization of Dachs is responsible for the mis-localization of GFP:Sple in ds or fat mutant wing discs , we examined GFP:Sple in ds dachs and fat dachs double mutants . In both cases , GFP:Sple reverted to a Pk-like localization , including a proximal , punctate orientation throughout most of the wing pouch , and an anterior orientation near the A-P boundary ( Figure 2E–I ) . One remarkable feature of ds dachs or fat dachs mutant discs is that GFP:Sple localization is preferentially proximal even in proximal regions of the wing pouch , where GFP:Sple localization was preferentially distal in dachs mutants ( Figure 2B , F , H , I ) . This implies that the distal localization of GFP:Sple here in dachs mutants is Ds-dependent , and hence that Dachs and Ds each have the ability to independently localize GFP:Sple . Thus , in proximal cells , where Ds expression is higher , Dachs and Ds could provide redundant localization cues for GFP:Sple . In distal cells , by contrast , we suggest that Dachs could be required for GFP:Sple localization because Ds expression is too low . In ds or fat mutants , Dachs is mis-localized , and Ds is either absent ( in ds mutants ) or unpolarized with reduced junctional accumulation ( in fat mutants ) ( Ma et al . , 2003; Mao et al . , 2009; Strutt and Strutt , 2002 ) , and consequently GFP:Sple becomes mis-localized . Finally , in the absence of both Dachs and Ds localization cues , as in fat dachs or ds dachs mutants , GFP:Sple follows Pk localization cues . In principle this could occur either because Sple is able to respond directly to the same cues as Pk through shared motifs , or because Sple can bind to Pk . The observation that GFP:Sple localized proximally in dachs pk mutant wing discs ( Figure 3F ) indicates that Sple can respond to Pk localization cues even in the absence of Pk . Localization of Pk was not visibly altered within larval wing discs by dachs , ds , or fat mutations ( Figure 3 ) . Fz-PCP is not required for distal localization of Sple , as GFP:Sple remains preferentially distal in vang mutant wing discs ( Figure 3G ) . Cytoplasmic levels of Sple were also visibly increased in vang mutants , consistent with studies of the influence of Vang on Pk levels ( Strutt et al . , 2013 ) . We note that our studies agree with Ayukawa et al . ( 2014 ) in reporting an influence of dachs and ds on Sple localization , but differ in that they reported that in ds mutants Sple polarity was reversed , resembling Pk , whereas we observe either a complete absence , or a randomization , of Sple polarization in ds mutants , consistent with Sple being regulated by Dachs . Also , they did not report observing the difference in localization of Sple between distal and proximal regions of dachs mutants , which we determined reflects a Dachs-independent regulation of Sple by Ds . Nonetheless , our studies agree that a direct connection between the Fz and Ds-Fat PCP pathways can be mediated through Dachs and Ds-dependent control of Sple . The link between PCP pathways mediated through Sple has important implications for how PCP is oriented , and suggests explanations for the basis of both pk and fat mutant polarity phenotypes . sple mutation does not result in a hair polarity phenotype in the wing , whereas pk mutants have a characteristic wing polarity phenotype , in which hairs in much of the wing are mis-oriented away from the wing margin , and wing margin bristles can point proximally ( Figure 4D , H; Figure 4—figure supplement 1 ) ( Gubb et al . , 1999 ) . The observation that Pk and Sple can localize in opposite directions , whereas wing hairs normally point in a single direction , implies that cells must ultimately choose which of these two distinct localization cues to follow . Normally , they choose the Pk cue ( Figure 4A ) , presumably because the Pk isoform is more abundant than the Sple isoform in the wing ( Ayukawa et al . , 2014; Merkel et al . , 2014; Olofsson et al . , 2014 ) , and hence it dictates polarization . Indeed , if Sple is over-expressed , then hair polarity is reversed even more strongly than in pk mutants , and can align with Ds-Fat PCP ( Ayukawa et al . , 2014; Doyle et al . , 2008; Gubb et al . , 1999; Merkel et al . , 2014; Olofsson et al . , 2014 ) , and the expression of GFP:Sple in clones is sufficient to alter wing hair polarity ( Figure 4—figure supplement 1 ) . These observations suggest that wing hair polarity in pk mutants could be directed by the Ds-Fat pathway dependent polarization of Sple ( Figure 4A ) . As this linkage in most of the wing depends upon dachs ( Figure 2I ) , this hypothesis predicts that the pk wing hair polarity phenotype should be suppressed by dachs mutation ( Figure 4A ) . Indeed , when we tested this by comparing wing hair and bristle orientation in pk versus dachs pk mutants , this suppression was observed ( Figure 4C–J ) . This result can also explain the observation that over-expression of Fat could suppress the pk hair polarity phenotype ( Hogan et al . , 2011 ) , because over-expression of Fat removes Dachs from the membrane ( Mao et al . , 2006 ) , which , as Dachs functions at membranes ( Pan et al . , 2013; Rauskolb et al . , 2011 ) , is functionally equivalent to dachs mutation . 10 . 7554/eLife . 09946 . 008Figure 4 . Contribution of Dachs and Sple to PCP mutant wing phenotypes . ( A ) Cartoons depicting inferred protein localization and hair orientation ( brown ) in wing cells of the indicated genotypes to explain rescue of pk by dachs , and rescue of fat by sple . Faint Sple and Ds indicate lower levels . ( B ) Schematic adult wing to show approximate location of panels shown in close-up , as indicate by letters . ( C–N ) Close-ups of portions of wings ( as indicated in panel B ) to show hair and bristle orientation in the indicated genotypes . Arrows indicate general direction of polarity . ( C–F ) Show wing margin bristles , ( G–N ) show wing hairs , in wild type ( C , G ) , pk30 ( D , H ) , dGC13/ d210 ( E , I ) , dGC13 pk30 ( F , J ) , UAS-RNAi-fat/+; C765-Gal4/UAS-dcr2 ( K , M ) and sple1 UAS-RNAi-fat/ sple1; C765-Gal4/UAS-dcr2 ( L , N ) . Suppression of pk polarity phenotypes by dachs was 100% penetrant . For suppression of fat phenotypes by sple , near the proximal anterior wing margin ( region K , L ) in 10/10 fat RNAi wings scored hairs point predominantly towards the wing margin , whereas in 7/8 fat sple wings scored hairs point predominantly distally , and in 1/8 wings scored a substantial fraction of hairs ( ∼1/4 ) point towards the wing margin . Near the anterior cross-vein ( region M-N ) , in 8/9 fat RNAi wings scored hairs point predominantly proximally , and in 1/9 wings scored hairs point predominantly towards the L3 vein , whereas in 5/8 fat sple wings scored hairs point distally , in 2/8 they point towards the L3 vein , and in 1/8 they point proximally . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 00810 . 7554/eLife . 09946 . 009Figure 4—figure supplement 1 . Hair polarity in wing is not affected by loss of sple or dachs . ( A ) Schematic adult wing to show approximate location of panels shown in close-up , as indicate by letters . ( B ) Pupal wing with clones of cells expressing GFP:Sple ( green ) and F-actin in hairs stained by phalloidin ( red ) . ( C–J ) Close-ups of portions of wings ( as indicated in panel A ) to show hair orientation in the indicated genotypes . Arrows indicate general direction of polarity . Hair polarity in wild type ( C , E ) , sple1 mutant ( D , F ) , UAS-dcr2; UAS-RNAi-fat/nub-Gal4 ( G ) , UAS-dcr2; nub-Gal4; UAS-RNAi-dachs ( H ) , UAS-dcr2; UAS-RNAi-fat/nub-Gal4; UAS-RNAi-dachs ( I ) , and fat8/fatG-rv; UAS-wts/tub-Gal4 ( J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 009 The determination that Ds-Fat and Fz pathways are molecularly linked by physical interaction between Dachs and Sple also provides a new perspective on polarity phenotypes of fat and ds . The altered wing hair polarity in fat or ds mutants has been interpreted as indicating that Fat and Ds have a normal role in directing hair polarity in regions of the wing . Indeed , recent studies have inferred that Ds-Fat PCP influences core protein polarization in the wing by orienting microtubules ( Harumoto et al . , 2010; Matis et al . , 2014; Olofsson et al . , 2014 ) . However , as hair polarity in the wing is normally Pk-dependent rather than Sple-dependent , and as we found that Ds-Fat PCP in the wing influences Sple localization but not Pk localization , we considered an alternative model: rather than reflecting a normal role in directing hair polarity , these phenotypes of fat and ds could stem from the inappropriate accumulation of Dachs , leading to inappropriate localization of Sple , which in some contexts could interfere with the normal Pk-dependent polarization cues ( Figure 4A ) . Consistent with this hypothesis , ds hair polarity phenotypes are suppressed by dachs ( Brittle et al . , 2012 ) , and we confirmed that fat wing hair polarity phenotypes ( generated using wing-specific RNAi ) are also suppressed by dachs ( Figure 4—figure supplement 1 ) . This hypothesis further predicts that fat PCP phenotypes could be suppressed by sple ( Figure 4A ) , and while we did not observe completely normal hair polarity in fat sple wings , we did observe a partial suppression , including restoration of normal , distally oriented polarity in two regions affected by loss of Fat: near the proximal anterior wing margin , and near the anterior cross-vein ( Figures 4B , K–N; Figure 4—figure supplement 1 ) . By contrast , when Hippo pathway phenotypes of fat are rescued by Warts over-expression , wing polarity remains abnormal in the proximal wing ( Feng and Irvine , 2007 ) ( Figure 4—figure supplement 1 ) . Thus , while Sple is not required for normal wing hair polarity , it mediates a connection between Ds-Fat and Fz pathways that contributes to abnormal hair polarity in the absence of fat . PCP in the eye has been studied for its influence on the organization and orientation of ommatidia ( Jenny , 2010 ) . The eight photoreceptor cells within each ommatidia are arranged in a characteristic pattern that comes in two chiral forms . This chirality is determined by which of two neighboring photoreceptors becomes the R3 cell and which becomes R4 . This decision is dependent upon Notch signaling , which is biased by Fz PCP such that the cell at the R3-R4 interface with higher Fz becomes R3 ( Cooper and Bray , 1999; Fanto and Mlodzik , 1999; Strutt et al . , 2002; Tomlinson and Struhl , 1999 ) . The two chiral forms are established in mirror symmetry with respect to the dorsal-ventral compartment boundary , termed the equator ( Figure 5—figure supplement 1 ) . In sple mutants , ommatidial chirality is randomized , whereas in pk mutant eyes ommatidial chirality is normal ( Gubb et al . , 1999 ) . Ds and Fj are expressed in complementary gradients in the eye ( Figure 5—figure supplement 1 ) , and experiments manipulating Ds and Fj expression have revealed that these gradients instruct normal polarity ( Simon , 2004; Strutt and Strutt , 2002; Yang et al . , 2002; Zeidler et al . , 1999 ) . However , the relationship between Fz and Ds-Fat PCP pathways in the eye and how this influences polarity has remained unclear . Also , in contrast to the wing and abdomen , where dachs mutation suppresses fat PCP phenotypes , dachs mutation has little effect on fat PCP phenotypes in the eye ( Brittle et al . , 2012; Mao et al . , 2006; Sharma and McNeill , 2013 ) . We hypothesized that the influence of Ds-Fat PCP on ommatidial polarity might be accounted for by an ability of Ds to polarize Sple independently of dachs , as in the proximal wing . Ommatidia form progressively in a wave of differentiation that sweeps across the eye disc , initiated within a line of cells that form the morphogenetic furrow . We analyzed GFP:Sple localization at the 5-cell precluster stage of ommatidial formation , when R3-R4 specification occurs . GFP:Sple localized to the equatorial side of cells within both R3 and R4 , which places it within R4 at the R3-R4 interface ( Figure 5A ) . Equatorial polarization of Sple co-localizes it with Vang ( Strutt , 2001 ) , and is consistent with the observation that it interacts with Ds , since Ds is also polarized to the equatorial side of cells in the eye disc ( Brittle et al . , 2012 ) . This equatorial polarization of Sple was disrupted in ds or fat mutants ( Figure 5C , E ) , but was not affected by mutation of dachs ( Figure 5B ) , nor could dachs mutation prevent the mis-localization of Sple in ds or fat ( Figure 5D , F ) . In ds or fat mutants , Sple localization was partially randomized within R3 and R4 , and also partially unpolarized , in that it was often detected on multiple cell junctions . However , it was never detected along the cell junction with the more anterior cells within the ommatidial cluster ( R2 and R5 ) ( Figure 5 ) . 10 . 7554/eLife . 09946 . 010Figure 5 . Sple and Pk localization in photoreceptor cells . Localization of GFP:Sple ( A–F ) or GFP:Pk ( G , H ) in cells with expression in R4 or R3 photoreceptor cells in wild type ( A , G ) , dGC13/ d210 ( B ) , ds36D/ dsUA071 ( C ) , dGC13 ds36D/ dGC13 dsUA071 ( D ) , ft8/ ftG-rv ( E ) , dGC13 ft8/ dGC13 ftG-rv ( F ) , and sple1 ( H ) mutants . Rose plots summarize localization based on the indicated number ( N ) of examples , with equatorial to the right , polar to the left , and anterior ( towards the morphogenetic furrow ) at top . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 01010 . 7554/eLife . 09946 . 011Figure 5—figure supplement 1 . Polarity and gradients in eye discs . ( A ) Cartoon illustrating the arrangement and orientation of photoreceptor cells in eye discs at the 5-cell pre-cluster stage . ( B ) Schematic illustrating the subcellular localization of Ds-Fat and Fz PCP pathway components at the R4 and R3 interface . ( C , D ) Expression of Ds ( magenta ) and fj-lacZ ( cyan ) in a wild-type eye disc . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 01110 . 7554/eLife . 09946 . 012Figure 5—figure supplement 2 . Sple and Pk polarity in front of the morphogenetic furrow , and influence of GFP:Pk on PCP . ( A , B ) Portions of eye imaginal discs in front of the morphogenetic furrow with clones of cells expressing GFP:Sple ( A ) or GFP:Pk ( B ) ( green ) , stained for expression of E-cadherin ( blue ) . The white arrows indicates general direction of Sple or Pk polarization . ( C , D ) Portions of eye imaginal discs stained for Elav ( marks photoreceptor cells ) and Prospero ( Pros , overlap with Elav marks R7 ) and expressing GFP:Pk ( green ) . ( C ) control disc with no GFP:Pk clones . ( D ) Disc with 1-day-old GFP:Pk clones . ( E ) Disc with 3-day-old GFP:Pk clones , yellow arrowheads highlight ommatidia with abnormal polarity . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 012 We also examined localization of GFP:Pk within R3 and R4 , and found that it too localized to the equatorial side of both cells ( Figure 5G ) . How might Pk localize equatorially if it cannot interact with Ds ? We hypothesized that this might arise from an ability of Pk-Sple proteins to multimerize , or from the interactions that lead core components of the Fz pathway to adopt a shared , discrete , polarized localization . In such cases , equatorial polarization of Pk would not come about because it directly responds to an equatorial-polar signal like Ds , but rather because it can interact with Sple and/or Vang , which recruit it to equatorial cell junctions . In support of this , we found that in sple mutants , Pk localization is altered such that it becomes partially randomized , resembling Sple localization in fat or ds mutants ( Figure 5H ) . Thus , equatorial Pk localization in R3 and R4 depends upon Sple . The posterior bias in Pk localization in sple mutants , and the similar posterior bias in Sple localization within ds or fat mutants , tend to place Pk or Sple towards the side of the cell nearest the morphogenetic furrow ( Figure 5 ) . The morphogenetic furrow is a source of local signals for multiple pathways , including Notch , Hedgehog and Decapentaplegic , which might , indirectly at least , influence Fz PCP orientation in these mutant eye discs , as they have been implicated in influencing PCP in other contexts ( Sagner et al . , 2012; Struhl et al . , 1997 ) . Finally , we observed that in front of the morphogenetic furrow , GFP:Sple and GFP:Pk exhibit distinct localization profiles , with GFP:Sple accumulating on the equatorial sides of cells , as it does behind the furrow , but Pk:GFP accumulating on the anterior sides of cells , which in this region is the side closest to the morphogenetic furrow ( Figure 5—figure supplement 2 ) . Thus , behind the morphogenetic furrow , Pk and Sple co-localize in an Sple-dependent process , whereas in front of the morphogenetic furrow , Sple and Pk can localize differently . Our observation that GFP:Pk localizes , like Sple , on the equatorial sides of R3 and R4 behind the furrow was initially unexpected , because abnormal PCP has been reported in adult eyes with clones of cells expressing GFP:Pk ( Strutt et al . , 2013 ) . However , the observation of a distinct localization for GFP:Pk in front of the furrow suggested that the timing of GFP:Pk expression might be important . Polarization of GFP:Pk or GFP:Sple is examined in clones induced 1 day before dissection , as it is scored in ommatidia in which these transgenes are expressed only in R3 or only in R4 . These clones are induced when cells are in or behind the furrow . When we examined PCP in eye discs , using Prospero and Elav to stain ommatidia , PCP appeared unaffected by clones expressing GFP:Pk for one day ( Figure 5—figure supplement 2 ) . By contrast , within 3-day-old clones , which would have been initiated in front of the morphogenetic furrow , PCP was disturbed ( Figure 5—figure supplement 2 ) . These observations indicate that Pk over-expression can disturb PCP in the eye , but PCP ultimately becomes refractory to Pk over-expression . Hairs in the Drosophila abdomen point posteriorly; this orientation is influenced by components of both the Fz and Ds-Fat PCP pathways ( Casal et al . , 2002; 2006; Lawrence et al . , 2004 ) . In analyzing the relationship between PCP pathways in the abdomen , we focused on the pleural cells , which form in lateral and ventral regions , but have also examined polarity in tergites , which form on the dorsal side of the abdomen . As the subcellular localizations of Dachs , Sple and Pk within pupal abdominal cells have not been described , we first characterized their distributions in pleural cells of wild-type animals at pupal stages , with posterior compartments marked using hh-Gal4 and UAS-RFP transgenes . Dachs:GFP and Sple:GFP were polarized towards the anterior sides of cells within A compartments , and towards the posterior sides of cells within P compartments ( Figure 6A , B , H ) . This is consistent with their being polarized in response to the Ds and Fj gradients , as the Fj and Ds gradients are oriented oppositely within anterior ( A ) versus posterior ( P ) compartments of each segment ( Figure 6—figure supplement 1 ) ( Casal et al . , 2002 ) , and Dachs and Sple accumulate on the sides of cells that face towards lower Ds levels and higher Fj levels . Pk:GFP , by contrast , was polarized towards the anterior sides of cells within both A and P compartments ( Figure 6C , H ) . Thus , in A compartments Pk:GFP and Sple:GFP polarize in the same direction , whereas in P compartments they polarize in opposite directions . 10 . 7554/eLife . 09946 . 013Figure 6 . Localization of Dachs , Sple and Pk in abdominal pleura . ( A–G ) Pleura of wildtype ( A–C ) , ft8/ ftG-rv ( D , E ) , dGC13 ft8/ dGC13 ftG-rv ( F ) and sple1/sple1 ( G ) pupae with clones of cells expressing GFP:Dachs ( A , D ) , GFP:Sple ( B , E , F ) and GFP:Pk ( C , G ) ( green ) . Posterior compartments are marked by hh-Gal4 UAS-mCD8-RFP ( red ) . Anterior-posterior body axis is indicated at top . ( H ) Rose plots depicting polarization of GFP:Dachs , GFP:Sple or GFP:Pk in pleural cells of the indicated genotypes; anterior polarization is to the left and posterior polarization is to the right . For wild type and dachs mutants cells were scored separately in A and P compartments . For fat , ds , fat dachs , and ds dachs the anterior compartment was further subdivided into a front region ( A1 , anterior-most 8 cells ) , and the remainder of the A compartment ( A* ) . For sple , the A compartment was subdivided into a front region of 5 cells ( Af ) , a back region of 10 cells ( Ab ) , and a middle region comprising the rest of the compartment ( Am ) ; P compartment localization is summarized in Figure 6—figure supplement 1 . Scoring of sub-regions of the A compartment in wild-type is shown in Figure 6—figure supplement 1 . Apparent variations in cell size are mostly due to the flexibility of the pleura , which is easily stretched or compressed . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 01310 . 7554/eLife . 09946 . 014Figure 6—figure supplement 1 . Gradients influencing PCP in the abdomen . ( A ) Schematic illustrating the orientation of hairs and approximate gradients of Ds , Fj and Hh expression in the abdomen ( Casal et al . , 2002; Struhl et al . , 1997 ) . ( B ) Rose plots depicting polarization of GFP:Sple or GFP:Pk in pleural cells of the indicated genotypes . Additional wild-type analysis is for comparison to mutants shown in Figure 6; anterior polarization is to the left and posterior polarization is to the right . For GFP:Sple the anterior compartment was subdivided into a front region ( A1 , anterior-most 8 cells ) , and the remainder of the A compartment ( A* ) . For GFP:Pk the A compartment was subdivided into a front region of 5 cells ( Af ) , a back region of 10 cells ( Ab ) , and a middle region comprising the rest of the compartment ( Am ) . ( C ) Pleura of dGC13 sple1/ dGC13 sple1 pupa with clones of cells expressing GFP:Pk ( green ) . Posterior compartments are marked by hh-Gal4 UAS-mCD8-RFP ( red ) . Anterior-posterior body axis is indicated at top . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 014 Consistent with prior studies ( Lawrence et al . , 2004 ) , we observed that pk-sple mutants reverse hair polarity within the center of the A compartment , while the P compartment , and the edges of the A compartment , exhibit normal hair polarity ( Figure 7B , F ) . The A compartment in tergites encompasses all of the hairs and bristles in the anterior , pigmented part of each abdominal segment , plus approximately two rows of hairs in the unpigmented region posterior to this ( Struhl et al . , 1997 ) . The P compartment in tergites encompasses the remaining hairs posterior to the A compartment , plus a region of naked cuticle . In pleura we estimated the A and P compartment regions based on the neighboring tergites , but can not make precise assignments of compartment identity for hairs near compartment boundaries . We extended analysis of pk-sple by examining isoform-specific alleles . pk mutant alleles have normal polarity in A compartments , but mostly reversed polarity in the P compartment within pleura ( Figure 7C ) , although not in tergites ( Figure 7G ) . sple mutant alleles have normal polarity in the P compartment , and abnormal polarity , including hair reversal but also sideways or swirling hair orientations , within the center of the A compartment ( Figure 7D , H ) . 10 . 7554/eLife . 09946 . 015Figure 7 . Influence of Pk and Sple on hair polarity in the abdomen . ( A–D ) Hair polarity in pleura revealed by F-actin ( phalloidin staining ) in wild type ( A ) , and in pk-sple14 ( B ) , pk30 ( C ) and sple1 ( D ) mutants . Yellow asterisk indicates the position of the spiracle , which forms near the center of the anterior compartment . Yellow arrows indicate the region where hair orientation is normal , and red arrows indicate the region where hair orientation is disrupted . Dashed yellow lines mark approximate boundaries between regions with normal and abnormal polarity . ( E–H ) Hair polarity in tergites of wild-type ( E ) , pk-sple14 ( F ) , pk30 ( G ) and sple1 ( H ) mutant animals . Black arrows indicate regions where hair orientation is normal , and blue arrows indicate the region where hair orientation is abnormal . Dashed blue line mark approximate boundaries between regions with normal and abnormal polarity . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 015 The influence of pk-sple on polarity in the P compartment of the pleura thus appears reminiscent of the situation in the wing: Pk and Sple can respond to opposing localization cues . In the absence of Pk , cells respond instead to Sple-dependent cues , leading to a reversal of polarity ( Sple localized normally in pk mutants , Figure 8I , K ) . The essential contribution of Sple to the pk phenotype is confirmed by its suppression in pk-sple mutants ( Figure 7 ) . The influence of Pk-Sple on polarity in A compartments is reminiscent of the situation in the eye . Sple , not Pk plays the key role in establishing polarity here , and Pk was mis-localized within the central region of A compartments of sple mutants ( Figure 6G , H ) , hence Sple contributes to localization of Pk here . 10 . 7554/eLife . 09946 . 016Figure 8 . Localization of Dachs , Sple and Pk in abdominal pleura of additional genotypes . ( A–J ) Pleura of ds36D/dsUA071 ( A , B , F ) , dachsGC13/dachs210 ( C , G ) , ds36D dachsGC13/dsUA071 dachsGC13 ( D , H ) , ft8/ftG-rv ( E ) , pk30 ( I ) and ft8 dachsGC13/ftG-rv dGC13 ( D , J ) mutant pupae with clones of cells expressing of GFP:Dachs ( A ) , GFP:Sple ( B , C , D , I ) and GFP:Pk ( E-H , J ) ( green ) . Posterior compartments are marked by hh-Gal4 UAS-mCD8-RFP ( red ) . ( K ) Rose plots depicting polarization of GFP:Dachs , GFP:Sple or GFP:Pk in pleural clones of the indicated genotypes; anterior polarization is to the left and posterior polarization is to the right . Clones were scored separately in A and P compartmentsDOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 016 In fat or ds mutants , hair polarity is disturbed in much of the A and P compartments , although a small region at the front of the A compartment exhibits normal hair polarity ( Figures 9A , E , 10A , E ) ( Casal et al . , 2002 ) . To determine whether the abnormal polarity could be explained by mis-localization of Dachs and/or Ds , and a consequent mis-localization of Sple and/or Pk , we assessed both genetic interactions and protein localization . The disruption of polarity within A compartments in fat or ds mutants was correlated with mis-localization of Dachs throughout the A compartment ( mostly uniform Dachs in fat mutants , and randomized Dachs in ds mutants , Figures 6D , 8A , K ) , and mis-localization of Sple everywhere except the most anterior region of the A compartment ( Figure 6E , H , 8B ) . Moreover , mutation of dachs suppressed the hair polarity phenotypes of fat and ds in A compartments ( Figures 9C , F , 10C , G ) ( Mao et al . , 2006 ) , and also suppressed the mis-localization of Sple ( Figures 6F , H , 8D ) . These observations suggest that ds and fat polarity phenotypes in the anterior abdomen can be accounted for by a Dachs-dependent mis-localization of Sple , as we had observed in the wing . Mutation of dachs alone does not disrupt polarity in A compartments ( Figures 9B , 10B ) . It could be that in the absence of Dachs and Ds , Sple is localized by the same cues that localize Pk , as Pk localization remained normal within A compartments of ft or ds mutants ( Figure 8E , F , K ) . 10 . 7554/eLife . 09946 . 017Figure 9 . Influence of Ds-Fat PCP on hair polarity in abdominal pleura . Hair polarity in pleura revealed by F-actin ( phalloidin staining ) in ft8/ftG-rv ( A ) , dGC13/d210 ( B ) , dGC13 ft8/dGC13 ftG-rv ( C ) , ft8 sple1/ftG-rv sple1 ( D ) , ds36D/dsUA071 ( E ) , dGC13 ds36D/dGC13 dsUA071 ( F ) , ft8 pk30/ftG-rv pk30 ( G ) , ft8 pk-sple14/ftG-rv pk-sple14 ( H ) , dGC13 pk30 ( I ) and dGC13 sple1 ( J ) mutant animals . Yellow asterisk indicates the position of the spiracle . Yellow arrows indicate the region where hair orientation is normal , and red arrows indicate the region where hair orientation is disrupted . Dashed yellow lines mark approximate boundaries between regions with normal and abnormal polarity . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 01710 . 7554/eLife . 09946 . 018Figure 10 . Influence of Ds-Fat PCP on hair polarity in abdominal tergites . Hair polarity in tergites of ft8/ftG-rv ( A ) , dachsGC13/dachs210 ( B ) , ft8 dachsGC13/ftG-rv dachsGC13 ( C ) , ft8 sple1/ftG-rv sple1 ( D ) , ds36D/dsUA071 ( E ) , dachsGC13 sple1/dachsGC13 sple1 ( F ) , ds36D dachsGC13/dsUA071 dachsGC13 ( G ) , and dachsGC13 pk30/dachsGC13 pk30 ( H ) mutant animals . Black arrows indicate the region where hair orientation is normal , and blue arrows indicate regions where hair orientation is disrupted . Dashed blue line mark approximate boundaries between regions with normal and abnormal polarity . DOI: http://dx . doi . org/10 . 7554/eLife . 09946 . 018 In P compartments , Dachs is mis-localized in fat or ds mutants , and there is also a partial mis-localization of Sple ( , Figures 6 , 8 ) . However , mutation of dachs alone causes a reversal of hair polarity in P compartments ( Figures 9B , 10B ) ( Matakatsu and Blair , 2008 ) . This reversal of polarity is associated with a reversal of Pk localization ( Figure 8G , K ) . The P compartment of the abdomen thus differs from other regions we have examined both in that there is a strong PCP phenotype associated with mutation of dachs , and in the mis-localization of Pk in dachs mutants . One potential explanation for this could be that in the absence of Dachs , Pk localization becomes governed by Sple , which retains its normal posterior localization in dachs mutants ( Figures 6G , 8C ) . In dachs sple mutants , Pk was partially randomized , but an overall posterior bias in localization was still observed ( Figure 6—figure supplement 1 ) . In dachs sple or dachs pk mutant abdomens , there is still some reversal of hair polarity in P compartments , although the region of reversal appears narrow than in dachs mutants ( Figure 10F , H ) . Pk localization is also disturbed in P compartments of fat or ds mutants ( Figure 8E , F , K ) , as well as in fat dachs or ds dachs double mutants ( Figure 8H , J , K ) . Thus , while there are some similarities in control of PCP between abdominal P compartments and wings , there are also differences , hence distinct mechanisms contribute to the control PCP in each of these body regions . Our results implicate interactions between Dachs and Sple , and between Ds and Sple , as a connection point between PCP pathways coordinating polarity in multiple Drosophila organs . We found that both Dachs and Ds can each independently bind to Sple , but not Pk , through the unique N-terminus of Sple . Dachs and Ds also each have the ability to influence Sple localization , and we identified some places where Dachs is necessary and sufficient to localize Sple ( i . e . distal wing ) , and others where Dachs is dispensable but Ds is required ( e . g . proximal wing , eye ) . The influence of Ds on Sple localization is essential for some manifestations of PCP , such as ommatidial polarity in the eye . Indeed , because Fz PCP can propagate from cell to cell , we propose that localization of Sple by Ds could account for the long-range influence of Ds-Fat borders on PCP in the eye ( Sharma and McNeill , 2013; Strutt and Strutt , 2002; Yang et al . , 2002 ) . The influence of dachs on PCP is generally mild , likely due to partial redundancy with Ds in localizing Sple . However , a strong influence of Dachs is revealed in the context of additional PCP mutations , as it contributes to the disturbed Sple localization and hair polarity in the wing and abdomen of fat and ds mutants , as well as the reversal of wing hair polarity in pk mutants . Several mechanisms by which Pk-Sple could influence the polarization of components of the Fz-PCP system have been described . These include physically associating with Vang , and promoting clustering , endocytosis , and/or degradation of Vang or other PCP proteins ( Bastock et al . , 2003; Cho et al . , 2015; Jenny et al . , 2005; Strutt et al . , 2013; Tree et al . , 2002 ) . They have also been identified as influencing the orientation of apical non-centrosomal microtubules that can traffic components of the Fz-PCP system , including Fz and Dsh ( Matis et al . , 2014; Olofsson et al . , 2014 ) . The various activities ascribed to Pk-Sple are not mutually exclusive , and it could play multiple roles . However , the only unique functions clearly attributed to Sple as opposed to Pk are its interactions with Ds and Dachs , and distinct localization . Fz and Fat localize to opposite sides of cells in wings , but to the same side in eyes ( Matis and Axelrod , 2013 ) ( Figure 1A , Figure 5—figure supplement 1 ) . Because Sple expression is relatively high in eye and low in wing ( Ayukawa et al . , 2014; Merkel et al . , 2014; Olofsson et al . , 2014 ) , our studies are consistent with a molecular explanation for ‘rectification’ of this relationship between PCP pathways in which interaction of Sple with Dachs and Ds links PCP pathways in eyes but not in wings; coupling between pathways that depends upon physical interactions with Sple and not Pk could similarly explain the opposite relationships between hair polarity and Ds and Fj gradients in A versus P abdominal compartments . Thus , the solution our results support for the controversy over the relationship between the two PCP pathways is that in some contexts they operate in sequence , with directional information passed from Ds-Fat PCP to Fz PCP via Sple , whereas in other contexts they are uncoupled . While vertebrates have a Ds homologue that is required for PCP , Dchs1 ( Mao et al . , 2011a ) , Dchs1 must influence PCP in mammals through a distinct mechanism , as Pk is conserved in vertebrates , but the Sple isoform is not . Even in flies , the linkage of Dachs and Ds to Sple cannot be the sole mechanism by which Ds-Fat signaling influences PCP , as some manifestations of cell polarity controlled by Ds-Fat , e . g . oriented cell divisions , do not require Sple ( Baena-Lopez et al . , 2005; Gubb et al . , 1999 ) . Moreover , when mis-expressed , Ds and Fat can alter PCP even in flies lacking a functional Fz PCP system ( e . g . fz- stan- flies ) ( Casal et al . , 2006 ) . It has also been proposed that PCP in the wing is influenced by shear forces generated by contraction of the wing hinge ( Aigouy et al . , 2010 ) , and disruption of these shear forces in fat or ds mutants might occur through a mechanism that depends upon dachs but not sple , due either to effects of Dachs on Hippo signaling ( Cho et al . , 2006 ) or on cytoskeletal tension ( Bosveld et al . , 2012; Mao et al . , 2011b ) . Such additional influences of Dachs might explain why fat wing hair polarity phenotypes are more strongly suppressed by loss of dachs than by loss of sple . One revelation from analysis of Pk and Sple localization is that not only can PCP be oriented differently in different places or at different times ( Hogan et al . , 2011; Sagner et al . , 2012 ) , even at one place , cells can be subject to simultaneous , competing , polarity cues . For example , in the wing , where GFP:Sple localizes differently from GFP:Pk , cells must thus choose between competing polarity cues . Normally , they choose Pk localization cues , because Sple expression is low ( Ayukawa et al . , 2014; Merkel et al . , 2014; Olofsson et al . , 2014 ) . Nonetheless , the Sple expressed in the wing is functional and able to direct PCP , as evidenced by the dachs- and sple-dependent reversals of hair polarity in pk mutants . Based on observations that pk-sple alleles could have weaker phenotypes than isoform-specific alleles , and that over-expression of Pk or Sple could result in phenotypes reminiscent of sple or pk alleles , respectively , it was proposed that PCP requires a balance between Pk and Sple ( Gubb et al . , 1999 ) . However , we suggest that their relationship is better described as a competition . In the wing disc , Pk expression is more abundant than Sple expression , hence Pk ‘wins’ , and cells orient in response to cues that are unrelated to Ds-Fat PCP . When Pk is removed , then Sple can direct PCP , and hair polarity becomes governed by Ds-Fat PCP . Wild-type PCP requires Sple in some places , and Pk in others , but we know of no results that would require a balance between these two isoforms at any one place and time . We further propose that the competition between Sple and Pk is carried out by feedback mechanisms that promote polarization . Positive feedback mechanisms , which reinforce the accumulation of co-localized proteins , together with negative feedback mechanisms , which inhibit the accumulation of oppositely localized proteins , are a staple of PCP systems , and have been widely viewed as a means of amplifying , maintaining , and propagating polarization in response to weak polarity signals . The observation that cells sometimes need to choose between competing polarity signals leads us to emphasize that feedback mechanisms could also have a distinct , fundamentally important role in PCP that has not previously been considered – they enable cells to make a discrete choice between competing polarity signals . The observation that the relative level of Pk versus Sple influences how cells respond to competing polarity signals , with that choice then amplified by feedback , also has implications for the interpretation of GFP:Pk and GFP:Sple localization profiles . We take the localization of these proteins as indicators of the polarity signals that cells ‘see’ when that isoform predominates . This is not necessarily the same as their localization under endogenous expression conditions . For example , endogenous Pk localization might normally match Sple in the eye even in front of the furrow , because it is recruited to equatorial sides of cells by interactions with Sple and Vang . Analysis of wing hair polarity played a central role in development of the hypothesis that Ds-Fat functions as a ‘global’ PCP module and Fz as a ‘core’ PCP module , with polarity guided by the vectors of Fj and Ds expression ( Ma et al . , 2003 ) . However , since Ds-Fat signaling modulates Sple , but not Pk , localization , and Pk , but not Sple , is normally important for wing hair polarity , we infer that Ds-Fat PCP does not normally play a significant role in directing wing hair polarity . Instead , we propose , as also suggested by ( Blair , 2014 ) , that the hair polarity phenotypes of ds or fat mutants are better understood as a de facto gain-of-function phenotype , resulting from inappropriate accumulation of Dachs on cell membranes , which then leads to inappropriate localization of Sple , and abnormal polarity . This would also explain how Ds-Fat signaling , stripped of polarizing information , could nonetheless rescue PCP phenotypes: for example , how uniform Ds expression can rescue hair polarity in ds fj mutants ( Matakatsu and Blair , 2004; Simon , 2004 ) , and how expression of the intracellular domain of Fat can rescue hair polarity in fat mutants ( Matakatsu and Blair , 2006 ) , as these manipulations suppress the membrane accumulation of Dachs that would otherwise occur in mutant animals . More recently , it has been proposed that Ds-Fat PCP provides directional information to orient Fz PCP in the wing by aligning and polarizing apical non-centrosomal microtubules that can traffic Fz and Dsh ( Harumoto et al . , 2010; Matis et al . , 2014; Olofsson et al . , 2014 ) . While disorganization of these microtubules is observed in fat or ds mutants , we suggest that the inference that Ds-Fat thus orients PCP in wing via these microtubules is incorrect . There is evidence both in imaginal discs and in axons that Pk-Sple can orient microtubules ( Ehaideb et al . , 2014; Olofsson et al . , 2014 ) . Sple is mis-localized in fat or ds mutant wing discs . Thus , we propose that the effects of ds and fat mutants on microtubules within the wing are likely a consequence of abnormal Sple localization , which disrupts microtubule orientation , but need not be interpreted as evidence for a normal role of Ds-Fat signaling in orienting microtubules or Fz PCP in the wing . The A compartment of the abdomen can be compared to the eye ( Sple-dependent ) , and the P compartment of the abdomen can be compared to the wing ( Pk-dependent ) . However , we also observed striking differences in the apparent influence of Ds-Fat pathway mutations on PCP between the abdomen and other organs . One possible contribution to these differences is the metamerism of the abdomen , and the local propagation of PCP . For example , the disturbance of Pk localization in P compartments even in fat dachs , ds dachs or dachs sple mutants would not be predicted based on the lack of detected physical interaction of Ds or Dachs with Pk , and lack of influence of these mutations on Pk in the wing . However , the P compartment is adjacent to two A compartments , and disturbances of polarity in these neighboring A compartments could potentially spread into the P compartment . The relatively normal polarity within A compartments of fat dachs or ds dachs mutants would also not be predicted if Sple is normally establishing polarity in response to the polarization of Ds and Dachs , and this result is not easily explained by hypothesizing propagation of normal polarity from neighboring compartments , since polarity is partially reversed in P compartments of these genotypes . Instead , we propose that in the absence of the primary PCP signal ( in this case provided by the Ds-Fat pathway ) , cells use information from secondary signals , which appear to provide polarizing information in the A compartment that parallels that provided by the Ds-Fat pathway . For investigation of Dachs , Sple or Pk localization we used act> y+>EGFP-dachs /TM6b , act> y+>EGFP-sple /TM6b , or act> y+>EGFP-pk /TM6b ( Bosveld et al . , 2012; Strutt et al . , 2013 ) . Clones with posterior compartments marked were made by crossing to y w hs-FLP[122]; If/CyOGFP; hh-Gal4 UAS-mCD8-RFP /TM6b . Mutant backgrounds examined in clones were ftG-rv/ ft8 , d210/ dGC13 , dsUA071/ ds36D , dGC13 ftG-rv/ dGC13 ft8 , dGC13 dsUA071/dGC13 ds36D , sple1/ sple1 , dGC13 pk30/ dGC13 pk30 , vangstbm6 . Flip-out clones were induced by heat shock at 33°C either 24 hr ( wing disc , eye disc ) or 12 hr ( abdomen ) before dissection . Additional mutant backgrounds were pk30/ pk30 , and pk-sple14/ pk-sple14 . UAS-RNAi-fat ( vdrc 9396 ) or UAS-RNAi-dachs ( vdrc 12555 ) was expressed using nub-Gal4 or C765-Gal4 along with UAS-dcr2 . Tissues were fixed in 4% paraformaldehyde in PBS followed by permeabilization in PBS with 1% BSA and 0 . 1% Triton X-100 . Primary antibodies used include rat anti-E-cad ( 1:200 , DSHB ) , anti-Pros ( 1:200 , DSHB ) , anti-Elav ( 1:200 , DSHB ) and mouse anti-Wg ( 1:400 , DSHB ) . Secondary antibodies used were labeled with DyLight405 , Cy3 or Alexa647 ( 1:100 , Jackson Immuno Research , West Grove , PA ) ; GFP and RFP were detected by autofluorescence . Alexa488-phalloidin ( 1:10 , Life Technologies ) was used to stain hairs in pleura . Protein localization in pleura was determined at ∼48 hr after pupal formation . Images were captured on a Leica TCS-SP5 confocal microscope or PerkinElmer Volocity spinning disc confocal microscope . Sple was amplified by PCR from pUAST-pksple ( Gubb et al . , 1999 ) using forward primer ( 5’- CTCGAACCACGGCGGCCGCCAACATGAGCAGCCTGTCAACCGGTGGAG -3’ ) and reverse primer ( 5’- GTGGTTCGAGGGTACCCGAGATGATGCAGTTCTTGTCCTTG -3’ ) and cloned using NotI and KpnI sites into pUAST-TM-EGFP:3XFlag ( Mao et al . , 2009 ) after removal of TM-EGFP by NotI/KpnI digestion to generate pUAST-sple:3XFLAG . Sple ( N ) ( first 349 amino acids ) was amplified by PCR from pUAST-pksple using forward primer ( 5’- ACTCTGAATAGGGAATTGGGAATTCCAACATGAGCAGCCTGTCAACCGGTG -3’ ) and reverse primer ( 5’- GTAGTCGCCTCGAGCCGCGGCCAGCTCATTTGACTCCTGCTGGGCG -3’ ) and inserted using InFusion cloning kit into pUAST-app:3XFlag2XStop ( gift of B . Staley ) after removal of TM-EGFP by EcoRI/NotI digestion , to generate pUAST-sple:3XFLAG . Pk was amplified by PCR from pUAST-pkpk ( Gubb et al . , 1999 ) using forward primer ( 5’- ACTCTGAATAGGGAATTGGGAATTCCAACATGGATACCCCAAATCAAATGC -3’ ) and reverse primer ( 5’- GTAGTCGCCTCGAGCCGCGGCCAGCGAGATGATGCAGTTCTTGTCC -3’ ) and inserted using InFusion cloning kit into pUAST-app:3XFlag2XStop after by EcoRI / NotI digestion , to generate pUAST-pk:3XFLAG2XStop . Tagged isoforms of Dachs , Ds-ICD , Sple , Sple-N , Pk and GFP ( control ) were expressed in S2 cells by transient transfection using Effectene of plasmids pUAST-attB-d:V5 , His ( Mao et al . , 2006 ) , pUAST-TM-DS-ICD:FLAG , V5 , His ( Mao et al . , 2009 ) , pUAST-attB-sple:3xFLAG , pUAST-attB-sple ( N ) :3xFLAG-2xStop , pUAST-attB-pk ( N ) :3xFLAG-2xStop , pAc-3XFLAG:GFP ( Oh et al . , 2009 ) , pAc-GFP:V5 ( Feng and Irvine , 2009 ) . Cells were harvested 48 hr after transfection and lysed in RIPA ( 50 mM Tris-HCl , pH 8 . 0; 150 mM NaCl; 1% NP-40; 0 . 5% Sodium deoxycholate; 0 . 1% SDS; 1 mM EDTA; 1 mM DTT and 10% glycerol , supplemented with protease inhibitor cocktail ( Roche ) and phosphatase inhibitor cocktail ( CalBiochem ) ) . Cell lysates were precleared by incubation in 10 μl Protein-A beads for 2 hr at 4°C and later incubated with 10 μl anti-V5 beads at 4°C overnight for co-immunoprecipitation . Anti-V5 beads were washed four times with RIPA , boiled in Laemmli sample buffer at 100°C for 5 min , and run on SDS-PAGE gels . Western blotting was performed using rabbit anti-V5 ( 1:5000 , Bethyl Labs ) and mouse anti-Flag ( 1:3000 Sigma ) , and fluorescentconjugated secondary antibodies ( Odyssey ) . Polarity vectors were determined manually by estimating the vector of Sple , Pk , or Dachs polarization within a cell , and comparing it to a reference vector . The images analyzed were projections through several confocal sections . Wing discs were examined at mid- to late third instar , eye discs were examined at late third instar , and abdomens were examined at 48 h after puparium formation . Vectors of polarization were determined within scoreable cells . Scoreable cells included single cell clones , and cells within small clones , or irregular portions of larger clones , with unlabeled cells on two or three sides , and for which any GFP observed could be clearly assigned to a single cell . Vectors of polarization were drawn from the center of the cell to the strongest visible accumulation of GFP-tagged proteins . In cases where a broad region of similar intensity was observed , vectors were drawn toward the center of membrane GFP accumulation . Counterstaining with E-cadherin was employed to ensure that all junctions of scored cells were visible in the images . In wing discs , polarity was determined separately in distal , proximal and A-P boundary regions , using Wg and Hh expression as references . The proximal region was defined as cells within 5 cells of the fold at the edge of the wing pouch . The reference vector for the proximal region was a line perpendicular to the tangent of the proximal Wg ring at the point closest to the cell being scored . The A-P boundary region was defined as cells within 5 cell diameters anterior to the edge of Hh expression . The reference vector for the A-P boundary region was a line drawn parallel to the A-P boundary . The distal region was defined as cells within the wing pouch , excluding the proximal and A-P boundary regions , and also limited to the central three quarters of the D-V Wg stripe ( Figure 2I ) . Cells overlapping the D-V boundary were excluded from analysis . The reference vector for the distal region was a line perpendicular the D-V boundary Wg stripe . In eye discs , the reference vector was a line parallel to the tangent of the morphogenetic furrow ( the poles of the eye disc , where the morphogenetic furrow is not perpendicular to the equator , were not analyzed ) . In the abdomen , the reference vector was a line perpendicular to the A-P compartment boundaries . Cells in different regions were scored separately as indicated in the figure legends , based on observed regional differences in polarity in certain genotypes . In all tissues , the angle between the vector of polarization and the reference vector was calculated using ImageJ , and rose plots summarizing the distribution of angles were generated using Matlab .	Animals have many asymmetric organs . Wings , for example , are aerodynamically shaped and have a clear front , back , top and bottom , and even additions to these organs , such as feathers on the wing , often need to be oriented in a specific manner . This kind of orientation arises when cells divide and grow asymmetrically in a flat plane . The asymmetry is established at the level of single cells when proteins are not equally spread throughout a cell , but rather asymmetrically distributed . Such cells are said to be ‘planar polarized’; and many experiments addressing this so-called planar cell polarity have been conducted in fruit flies , because they can be genetically altered easily . Previous studies have shown that two signaling pathways—called Frizzled and Dachsous-Fat—regulate how individual cells orient themselves within a flat sheet of cells that forms fruit fly’s wing . The two pathways are not independent , but it is unclear how they are linked . In particular , there has been conflicting evidence as to whether the Dachsous-Fat pathway controls the Frizzled pathway or whether the two act in parallel . Now , Ambegaonkar and Irvine have discovered new roles for a protein that is involved in both pathways , called 'Spiny-legs' . This protein was known to be important in the Frizzled pathway , but , when it was tracked with a fluorescent tag in developing wing cells it also accumulated in areas where two proteins that make up part of the Dachsous-Fat pathway were located . Biochemical experiments showed that both of these proteins ( which are called Dachs or Dachsous ) could physically interact with Spiny-legs . Ambegaonkar and Irvine therefore deleted the genes for Dachs or Dachsous in fruit flies and observed that Spiny-legs no longer organized itself in the proper way , implying that Dachs and Dachsous control where Spiny-legs goes within cells . When this analysis was extended to other fruit fly organs , such as the eyes , Ambegaonkar and Irvine found that Dachsous was more important than Dachs for the correct localization of Spiny-legs . Additionally , the Frizzled and Dachsous-Fat pathways seemed to compete for interactions with Spiny-legs . This connection between the two pathways helps to explain how cells behave when several different signals reach them . It also shows how different organs can reuse conserved components of the pathways to make different end products . Future studies should aim to work out the number of systems that polarize cells and how they are connected in different tissues .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	Coordination of planar cell polarity pathways through Spiny-legs
In response to proteasome dysfunction , mammalian cells upregulate proteasome gene expression by activating Nrf1 . Nrf1 is an endoplasmic reticulum-resident transcription factor that is continually retrotranslocated and degraded by the proteasome . Upon proteasome inhibition , Nrf1 escapes degradation and is cleaved to become active . However , the processing enzyme for Nrf1 remains obscure . Here we show that the aspartyl protease DNA-damage inducible 1 homolog 2 ( DDI2 ) is required to cleave and activate Nrf1 . Deletion of DDI2 reduced the cleaved form of Nrf1 and increased the full-length cytosolic form of Nrf1 , resulting in poor upregulation of proteasomes in response to proteasome inhibition . These defects were restored by adding back wild-type DDI2 but not protease-defective DDI2 . Our results provide a clue for blocking compensatory proteasome synthesis to improve cancer therapies targeting proteasomes . Proteasome inhibition elicits a response to restore proteasome activity , or a 'bounce-back response , ' where Nrf1 is the responsible transcription factor that upregulates expression of all proteasome subunit genes in a concerted manner in human cells ( Radhakrishnan et al . , 2010; Steffen et al . , 2010 ) . Proteasome inhibitors such as bortezomib and carfilzomib have been in clinical use for treatment of cancers , especially multiple myeloma , but this bounce-back response attenuates the ability of proteasome inhibitors to kill cancer cells ( Radhakrishnan et al . , 2010 ) . Therefore , genes regulating Nrf1 activation could be useful drug targets for increasing efficacy of proteasome inhibition in cancer treatment . When Nrf1 is produced , the bulk of the polypeptide is inserted into the ER lumen and glycosylated , with a short cytosolic N-terminus followed by a single transmembrane domain ( Radhakrishnan et al . , 2014; Zhang et al . , 2007 ) . The luminal region of Nrf1 is continually retrotranslocated to the cytosol by the p97/VCP ATPase complex , accompanied by deglycosylation and ubiquitination . Under normal circumstances , Nrf1 is promptly degraded by the proteasome . In contrast , when proteasome activity is compromised , Nrf1 escapes degradation and is proteolytically cleaved to the active form which enters the nucleus and enhances expression of target genes including proteasome subunits ( Radhakrishnan et al . , 2014 , 2010; Sha and Goldberg , 2014; Steffen et al . , 2010 ) . However , the processing enzyme for Nrf1 remains obscure . To identify genes important for Nrf1 activation , we performed a genome-wide small interfering RNA ( siRNA ) screen ( Figure 1—figure supplement 1A ) . Our approach used the well-characterized subcellular localization of Nrf1 accumulation in the nucleus in response to proteasome inhibition . HEK293A cells were transfected with pooled siRNA ( a pool of 4 unique siRNAs per gene ) and then treated with the proteasome inhibitor bortezomib to induce accumulation and nuclear translocation of Nrf1 ( Steffen et al . , 2010 ) . Cells were then fixed and stained with anti-Nrf1 antibody . The ratio of the nuclear to cytoplasmic fluorescent intensities was assessed by high-content microscopy and automated image analysis ( Figure 1—figure supplement 1B ) . p97 siRNA treatment served as a positive control , which abolished Nrf1 translocation following bortezomib treatment while increasing cytoplasmic Nrf1 ( Figure 1A ) ( Radhakrishnan et al . , 2014 ) . We observed a high degree of assay robustness ( Z’-factor > 0 . 5 , Figure 1—figure supplement 1C ) in the primary screen . The initial candidate genes with B score < –3 . 2 ( Figure 1B ) or which were picked up by visual inspection of the raw image data were further tested using four individual siRNAs in two different cell lines ( HEK293A and HT1080 cells ) ( Figure 1C ) . The subsequent candidates that had more than two hits in either cell line were finally examined whether the siRNAs mitigated upregulation of PSMA3 , a proteasome subunit gene . Consequently , we obtained 14 candidate genes that may impair activation of Nrf1 in response to bortezomib treatment ( Figure 1D ) . These hits included SEL1L , a co-factor of the ubiquitin ligase HRD1 , which catalyzes ER-associated degradation ( ERAD ) of Nrf1 ( Iida et al . , 2011; Steffen et al . , 2010; Tsuchiya et al . , 2011 ) and FAF2/UBXD8 , a p97-recruiting molecule in ERAD ( Meyer et al . , 2012 ) , validating our screening approach ( Figure 1A , B , and D ) . 10 . 7554/eLife . 18357 . 003Figure 1 . A genome-wide siRNA screen for regulators of Nrf1 translocation to the nucleus in response to proteasome inhibition . ( A ) Representative images of Nrf1 localization in control cells ( no siRNA ) and cells transfected with siRNA targeting p97 or SEL1L in the primary screen . Yellow-boxed regions are magnified and displayed in the right panels . ( B ) B score of all samples in the primary screen . Data are ordered from lowest to highest . Dashed blue line represents a cutoff value for positive hits . Some of the representative final hits are shown as red dots . The list of B scores for all samples in the primary screen are available in the Figure 1—source data 1 . ( C ) Workflow and summary of the genome-wide siRNA screen . ( D ) List of the 14 final hit genes and the score in each assay throughout the screening process . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 00310 . 7554/eLife . 18357 . 004Figure 1—source data 1 . List of B-score in the primary screen . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 00410 . 7554/eLife . 18357 . 005Figure 1—figure supplement 1 . Methods for the genome-wide screen . ( A ) A schematic view of the workflow in the primary screen using HEK293A cells . ( B ) Definition of nucleus and cytoplasm in the screen . Region of interest ( ROI ) of the nucleus was defined as a circle drawn two pixels inside of the outermost DAPI signal . ROI of the cytoplasm was defined as a three-pixel wide ring around DAPI signal . No gap between the rings and DAPI signals was set in the analysis using HEK293A cells , whereas a one-pixel gap was set in the analysis using HT1080 cells . ( C ) The well score of negative ( no siRNA ) and positive ( si p97 ) controls . The well score is the fluorescence intensity ratio of ROI of nucleus to cytoplasm as defined in ( B ) . The Z’ factor was calculated from 12 individual wells on each plate . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 005 Among the final hit genes , we focused on DDI2 , because it has a typical retroviral aspartyl protease domain , and therefore is a candidate Nrf1 processing enzyme ( Krylov and Koonin , 2001 ) . In negative and positive ( p97 siRNA ) control cells treated with bortezomib , the majority of Nrf1 is localized in the nucleus and the cytoplasm , respectively ( Figure 2A ) . DDI2 knockdown partially inhibited nuclear translocation of Nrf1 , accompanied by an increase in cytoplasmic Nrf1 . To determine if there is a defect in Nrf1 processing by DDI2 knockdown , we examined which Nrf1 species were observed in each knockdown . In negative control cells , Nrf1 was hardly detected in the absence of bortezomib , but bortezomib treatment increased the processed , active form of Nrf1 as well as the full-length , cytosolic form that is retrotranslocated into the cytosol by p97 ( Figure 2B ) ( Radhakrishnan et al . , 2014 ) . In p97 knockdown cells , the luminal ER form of Nrf1 that is N-glycosylated accumulated while the processed form almost disappeared both in the presence and absence of bortezomib . In DDI2 knockdown cells , Nrf1 was not detected in the absence of bortezomib , similar to control cells . However , in the presence of bortezomib , the full-length , cytosolic form of Nrf1 markedly accumulated ( Figure 2B ) . These results indicate that DDI2 is involved in conversion of full-length , cytosolic Nrf1 to the processed , active form . 10 . 7554/eLife . 18357 . 006Figure 2 . DDI2 is involved in Nrf1 processing and translocation to the nucleus . ( A ) Representative images of Nrf1 localization . HEK293A cells were transfected with a non-targeting control ( negative control ) , DDI2 , or p97 siRNA and then treated with 50 nM bortezomib for 14 hr before fixation . ( B ) Immunoblotting of whole-cell lysates of cells in ( A ) treated with or without bortezomib . Nrf1 is detected as three different forms; a glycosylated form ( G ) , full-length form ( FL ) , and processed form ( P ) . ( C ) Immunoblotting of Nrf1 after deglycosylation treatment . HEK293A cells were transfected with DDI2 or p97 siRNA , followed by transfection with Nrf1-3×Flag , and then treated with or without 50 nM bortezomib . The cell lysates were treated with or without Endo H . DeG denotes deglycosylated Nrf1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 00610 . 7554/eLife . 18357 . 007Figure 2—figure supplement 1 . The expression and localization of DDI2 were not affected by bortezomib treatment . ( A ) Relative mRNA expression of DDI2 in HCT116 cells treated with or without bortezomib ( 50 nM , 14 hr ) . The data represent mean + standard error of the mean ( SEM ) ( n = 3 , biological replicates ) . ( B ) Subcellular localization of DDI2 . HEK293A cells stably expressing Venus-Sec61β ( ER marker ) were transfected with Flag-DDI2 and treated with or without bortezomib . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 007 We further examined the N-glycosylation status of Nrf1 using cells transfected with Nrf1 tagged with 3×Flag at the C-terminus . N-glycosylated Nrf1 accumulated in p97 knockdown cells was sensitive to endoglycosidase H ( Endo H ) treatment and the deglycosylated form migrated faster in SDS-PAGE , consistent with a previous report ( Figure 2C ) ( Radhakrishnan et al . , 2014 ) . In contrast , the full-length form of Nrf1 that is significantly accumulated in DDI2 knockdown cells was not sensitive to Endo H treatment ( Figure 2C ) . Note that bortezomib treatment alone causes some accumulation of Endo H-sensitive N-glycosylated Nrf1 and that the deglycosylated species was detected at almost the same molecular weight as the processed , active form of Nrf1 observed in cells treated with bortezomib alone , similar to previously reported observations ( Figure 2C ) ( Radhakrishnan et al . , 2014 ) . These results demonstrate that the form of Nrf1 accumulated in DDI2-depleted cells is not N-glycosylated , further supporting the role of DDI2 in the processing of Nrf1 rather than in deglycosylation or retrotranslocation . The X-ray crystal structure analysis of the retroviral aspartyl protease ( RVP ) domain of budding yeast Ddi1p has revealed that it is a dimer with a similar fold to that of the human immunodeficiency virus type 1 ( HIV-1 ) protease , with identical geometry of the double D[S/T]GA motif of the active site ( Sirkis et al . , 2006 ) . The HIV-1 protease typically cleaves substrates between two hydrophobic residues ( Konvalinka et al . , 2015 ) . Nrf1 has been shown to be cleaved between Trp103 and Leu104 to become active ( Radhakrishnan et al . , 2014 ) , which conforms with the cleavage motif by retroviral aspartyl proteases . Accordingly , we asked whether the protease activity of DDI2 is required for Nrf1 processing . DDI2 has a ubiquitin-like domain ( UBL ) at the N-terminus and a RVP domain near the C-terminus ( Figure 3A ) . Bortezomib treatment increased the processed form of Nrf1 ( Figure 3B ) . Knockdown of DDI2 reduced the processed form and increased full-length Nrf1 ( Figure 3B ) . This effect was rescued by introducing siRNA-resistant wild-type DDI2 but not a protease-dead DDI2 in which the active site aspartic acid 252 was replaced with asparagine ( D252N ) . We also found that a DDI2 mutant lacking the UBL domain only partially restored the effect of DDI2 knockdown ( Figure 3B ) . These results suggest that the protease activity of DDI2 is required for cleavage of Nrf1 and that the UBL domain plays some role in the cleavage . 10 . 7554/eLife . 18357 . 008Figure 3 . The protease activity of DDI2 is necessary for Nrf1 processing and its transcriptional activity . ( A ) Schematic diagram of wild-type ( WT ) and each mutant of DDI2 . Ubiquitin-like ( UBL ) domain and retroviral protease-like ( RVP ) domain are represented as filled rectangles . The putative aspartyl protease active site amino acid sequence is shown . ( B ) HEK293A cells were transfected with DDI2 siRNA and after 24 hr were transfected with a plasmid encoding WT or mutant DDI2 shown in ( A ) , followed by 50 nM bortezomib treatment for 14 hr before harvest . The signal intensity ratio of Nrf1 full-length form ( FL ) to the processed form ( P ) was calculated , where the ratio for bortezomib treatment alone was set as 1 . ( C ) Immunoblotting of whole cell lysates of DDI2 WT knock-in ( KI ) , DDI2 knockout ( KO ) , and DDI2 D252N KI HCT116 cells . The cells transfected with Nrf1-3×Flag were treated with or without 50 nM borteaomib . ( D ) Relative mRNA expression of the proteasome genes PSMA3 and PSMB5 in WT , DDI2 KO , DDI2 WT KI , and DDI2 D252N KI HCT116 cells . mRNA levels of target genes were normalized by GUSB mRNA levels . The data represent mean + standard error of the mean ( SEM ) ( n = 3 , biological replicates ) . Statistical comparison was made by Tukey’s test ( p<0 . 05 ) . ( E ) Proteasome peptidase activity of cell lysates of the indicated cell lines . The data represent mean + SEM ( n = 3 , biological replicates ) . Statistical comparison was made by Tukey’s test ( *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 00810 . 7554/eLife . 18357 . 009Figure 3—figure supplement 1 . Genome editing of DDI2 locus by CRISPR-Cas9 system . ( A ) Schematic diagram of DDI2 wild-type ( WT ) , knockout ( KO ) , and knock-in ( KI ) alleles . Red arrowhead indicates sgRNA targeting site , and blue arrows indicate PCR primers for confirmation of homologous recombination . puroR indicates a puromycin resistant cassette . ( B ) Confirmation of successful genome editing at the DDI2 locus . The genome DNA of each cell line was extracted and amplified by PCR using primer pairs shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18357 . 009 To confirm the necessity of the protease activity of DDI2 , we generated DDI2 knockout ( KO ) and protease-dead DDI2 ( D252N ) knock-in ( KI ) cells as well as wild-type DDI2 knock-in cells ( Figure 3 , Figure 3—figure supplement 1A , B ) . In DDI2 knockout cells and DDI2 D252N knock-in cells , the full-length form of Nrf1 was accumulated upon bortezomib treatment , whereas the processed form was accumulated in wild-type DDI2 knock-in cells ( Figure 3C ) . These results further support the requirement of DDI2 protease activity for Nrf1 activation . We then examined whether a lack of the catalytic activity of DDI2 abolishes the 'bounce-back' response after proteasome inhibition . In parental HCT116 cells , bortezomib treatment caused an increase in mRNA levels of the proteasome subunit genes PSMA3 and PSMB5 ( Figure 3D ) . Knockout of DDI2 strongly suppressed this response , further supporting the importance of DDI2 in Nrf1 activation ( Figure 3D ) . Interestingly , the basal expression of proteasome subunits was also decreased in DDI2-deficient cells . In wild-type DDI2 knock-in cells , mRNA levels of the proteasome subunits were upregulated in response to bortezomib , similar to the parental cells . In contrast , DDI2 D252N knock-in cells did not undergo such a response , similar to DDI2 knockout cells ( Figure 3D ) . These results suggest that the processing of Nrf1 by the aspartyl protease activity of DDI2 is required for upregulation of proteasome gene expression mediated by Nrf1 in response to proteasome inhibition . Nrf1 has also been found to regulate basal expression of proteasome subunits , the extent of which varies between cell types ( Lee et al . , 2013 , 2011 ) . We observed that knockout and D252N DDI2 knock-in cells had significantly lower proteasome activity compared to wild-type DDI2 knock-in cells , suggesting that DDI2 is also involved in basal expression of proteasomes through its catalytic activity ( Figure 3E ) . In conclusion , we identified DDI2 as a protease that is required for Nrf1 processing and the bounce-back response induced by proteasome inhibition . However , there remain several questions to be answered . How can the involvement of DDI2 be reconciled with a previous report that demonstrated a defect in Nrf1 processing by strong inhibition of the proteasome , leading to the conclusion that the proteasome is the processing enzyme for Nrf1 ( Sha and Goldberg , 2014 ) ? In terms of substrate specificity , the cleavage site of Nrf1 ( P1: W , P1’: L ) does not seem to be a sequence preferred by the proteasome ( Toes et al . , 2001 ) ; rather it conforms to a cleavage motif of RVP ( Konvalinka et al . , 2015 ) . It could be that the proteasome activity is required for function of DDI2 or some other factors that is involved in Nrf1 processing . Related to this , the mechanism by which DDI2 acts as a Nrf1 processing protease remains unclear . DDI2 is not induced by bortezomib at either protein or mRNA level ( Figure 2B and Figure 2—figure supplement 1A ) . Furthermore , the subcellular localization of DDI2 seems to be unaffected by bortezomib treatment ( Figure 2—figure supplement 1B ) . Since DDI2 is suggested to be active even when the proteasome activity is not compromised ( Figure 3D and E ) , a specific activation mechanism under proteasome impairment may not exist . An intriguing finding is that the UBL domain of DDI2 plays some role in Nrf1 processing ( Figure 3B ) . It has been shown that the UBL domain of Ddi1p is an atypical UBL that binds ubiquitin ( Nowicka et al . , 2015 ) . Binding of DDI2 with ubiquitinated proteins , possibly Nrf1 itself , would be promoted by proteasome inhibition and may facilitate Nrf1 processing by DDI2 . Lastly , whether DDI2 directly cleaves Nrf1 remains unknown . We have tested a recombinant fragment of Nrf1 encompassing the processing site as a substrate for recombinant DDI2 , but failed to detect its cleavage . Other factors might be required for in vitro reconstitution of Nrf1 processing by DDI2 , such as substrate unfolding , co-activators of DDI2 , and a set of specific experimental conditions . Understanding the mechanism by which DDI2 cleaves Nrf1 and establishing an in vitro assay for the enzymatic activity of DDI2 should provide useful information for developing a DDI2 inhibitor that would block compensatory proteasome synthesis to improve cancer therapies targeting proteasomes . In the primary screen , Dharmacon siGENOME SMARTpool siRNA library ( GE Dharmacon , Lafayette , CO ) was used . To prepare screening plates , the siRNAs in each well were suspended in 1 × siRNA buffer ( Thermo Fisher Scientific , Waltham , MA ) and 2 . 5 pmol siRNA ( 2 . 5 μL/well ) was dispensed into black , clear bottom , 384-well plates ( Greiner , Kremsmünster , Austria ) . For each well , a mixture of 10 μL DMEM and 0 . 1 μL Lipofectamine RNAiMAX ( Invitrogen , Carlsbad , CA ) was added . After 40 min incubation , 2000 cells/well of HEK293A cells were seeded . After 48 hr culture , bortezomib was added into each well to a final concentration of 10 nM . Cells were fixed with 4% PFA after 12 hr bortezomib treatment . Cells were then stained with Nrf1 antibody ( sc-13031; Santa Cruz Biotechnology , Dallas , TX ) and DAPI , and the fluorescent images were acquired and analyzed by CellInsight High Content Screening Platform ( Thermo Fisher Scientific ) . The fluorescence signal ratio of the nucleus to the cytoplasm was used as a raw measured value . The value was fitted in a two-way median polish method to exclude positional effects in the 384-well plate , and then the B score was calculated on a per-plate basis using the following formula . B score = ( Xi − Median ) / MAD ( Xi: measured value , MAD: median of absolute deviation ) In the secondary screen , four individual siRNAs contained in the library were purchased from Dharmacon and used . HEK293A and HT1080 cells transfected with each siRNA were analyzed by the same method as in the primary screening . In the third screen , HEK293A cells were treated with each hit siRNA and the expression level of the proteasome gene PSMA3 was measured by quantitative RT-PCR . Human DDI2 cDNA was synthesized from total RNA extracted from HT1080 cells using the indicated primers . Forward: 5’-ATGCTGCTCACCGTGTACTGTGTGC-3’ , Reverse: 5’-TCATGGCTTCTGACGCTCTGCATCC-3’ . DDI2 UBL deletion mutant was synthesized using 5’-AACTTACCCCGAATAGATTTCAG-3’ for a forward primer . siRNA resistant mutations were introduced without changing amino acid sequence using the following primers . Forward: 5’- TAATGTTGTATATTAACTGCAAAGTGAATGGACATCCTG-3’ , Reverse: 5’- CGACCTGTCCAAAACTTTCCGGAGCCTCTTCCATAGC-3’ . Human Nrf1 cDNA was synthesized from total RNA extracted from HEK293A cells using the indicated primers . Forward: 5’-ATGCTTTCTCTGAAGAAATACTTAACG-3’ , Reverse: 5’-TCACTTTCTCCGGTCCTTTGG-3’ . PCR was performed using PrimeSTAR Max DNA polymerase ( Takara Bio , Shiga , Japan ) . Amplified fragments were subcloned into pIRES vector ( Clontech Laboratories , Mountain View , CA ) and all constructs were confirmed by sequencing . HEK293A cells were purchased from Thermo Fisher Scientific . HCT116 cells were obtained from RIKEN BRC . The cell lines were tested negative for mycoplasma contamination by DAPI staining . The authors performed no further authentication of the cell lines . HEK293A cells and HCT116 cells were cultured under standard conditions . cDNAs were transfected into cells using PEI-MAX ( Mw: 40 , 000 ) . siRNAs targeting DDI2 or p97 and siGENOME Non-Targeting siRNA #2 were purchased from GE Dharmacon . siRNA was transfected into cells with Lipofectamine RNAi MAX ( Invitrogen ) . The sequences of siRNAs targeting DDI2 and p97 were as follows: DDI2 , 5’-GCCAAGUAGUGAUGCUUUA-3’; p97 , 5’-GUAAUCUCUUCGAGGUAUA-3’ . The cell lines were established using the CRISPR/Cas9 system . Single guide RNAs ( sgRNA ) were designed using CRISPR direct ( http://crispr . dbcls . jp/ ) and cloned into a pX330 vector . The sgRNA sequence for DDI2 was 5’-ACTCGAGCTCGCACAGCGCG-3’ . Targeting constructs for gene knockout were designed to insert a puromycin resistance cassette at the locus of the start codon . Targeting constructs for DDI2 knock-in were designed to insert DDI2 wild type or D252N cDNA in-frame downstream of DDI2 exon 1 . A puromycin resistance cassette was also inserted into this region . The sgRNA vector and targeting vector were transfected in HCT116 cells . After 48 hr transfection , cells were cultured in medium supplemented with 4 μg/mL puromycin . After two weeks drug selection , colonies were picked up and successful homologous recombination was confirmed by PCR method . PCR was performed using EmeraldAmp PCR Master Mix ( Takara Bio ) . The following primers were used for confirmation of genome editing . DDI2 Forward: 5’-ATGCTGCTCACCGTGTACTGTGTGC-3’ , DDI2 intron 1 Reverse: 5’-GCAAGCTGAGTAGGGAAATGAAACCACCAA-3’ , Puro forward: 5’-GTCACCGAGCTGCAAGAACTCTTCC-3’ . Cells were harvested 12 hr after 20 nM bortezomib treatment . Total RNA of cells was isolated using High Pure RNA isolation kit ( Roche , Basel , Switzerland ) and were reverse-transcribed using ReverTra Ace qPCR RT kit ( Toyobo , Osaka , Japan ) . Quantitative RT-PCR was performed using THUNDERBIRD Probe qPCR Mix ( Toyobo ) , Universal ProbeLibrary Probe ( Roche ) , and LightCycler 480 ( Roche ) . The sequences of primers used were as follows: PSMA3 , 5’-GAAGAAGCAGAGAAATATGCTAAGG-3’ and 5’-GGCTAAATAGTTACATTGGACTGGAG-3’; PSMB5 , 5’-CATGGGCACCATGATCTGT-3’ and 5’-GAAATCCGGTTCCCTTCACT-3’; GUSB , 5’-CGCCCTGCCTATCTGTATTC-3’ and 5’-TCCCCACAGGGAGTGTGTAG-3’ . 24 hr after transfection of siRNA , cells were transfected with cDNA and cultured for a further 48 hr . 50 nM bortezomib was added 14 hr prior to cell lysis . Cells were lysed in buffer containing 42 mM Tris-HCl ( pH 6 . 8 ) , 1 . 72% SDS , 5 . 6% glycerol , 5% 2-mercaptoethanol , and 0 . 01% bromophenol blue ( SDS sample buffer ) for whole-cell lysate . The samples were subjected to SDS-PAGE , transferred to polyvinylidene fluoride membrane , and analyzed by immunoblotting . All images were taken using Fusion SL4 ( M&S Instruments ) . Rabbit polyclonal antibody against DDI2 was raised by immunizing keyhole limpet hemocyanin ( KLH ) conjugated synthetic DDI2 C-terminal ( residues 385–399 ) peptides . The following antibodies were purchased: Nrf1 ( sc-13031; Santa Cruz ) , p97 ( MA3-004; Invitrogen ) , GAPDH ( sc-32233; Santa Cruz ) , Flag ( F1804; Sigma Aldrich , St . Louis , MO ) . Cells were fixed in 4% paraformaldehyde 72 hr after transfection of siRNA and 16 hr after 10 nM bortezomib treatments . The cells were incubated with primary antibodies , and then incubated with DAPI ( Nacalai Tesque , Kyoto , Japan ) and secondary antibodies , either Goat anti-rabbit or anti-mouse IgG secondary antibody Alexa Fluor 488 or Alexa Fluor 647 conjugate ( Invitrogen ) . All images were acquired by TCS SP5 or TCS SP8 ( Leica Microsystems , Wetzlar , Germany ) . Cells were lysed in ice-cold buffer containing 25 mM Tris-HCl ( pH 7 . 5 ) , 0 . 2% Nonidet P-40 , 1 mM dithiothreitol , 2 mM ATP , and 5 mM MgCl2 . The hydrolysis of the fluorogenic peptide , succinyl-Leu-Leu-Val-Tyr-7-amino-4-methylcoumarin ( Suc-LLVY-MCA ) ( Peptide Institute , Osaka , Japan ) was measured in 50 mM Tris-HCl ( pH 8 . 0 ) at 37°C by ARVO MX 1420 ( PerkinElmer , Waltham , MA ) . Cells were lysed in ice-cold phosphate buffered saline ( PBS ) containing 0 . 5% Triton X-100 . After centrifugation ( 20 , 000 g , 10 min ) , the cell lysates were subjected to deglycosylation reactions with Endo Hf ( New England BioLabs , Ipswich , MA ) following the manufacturer’s protocol . A biological replicate was considered as each independent experiment . Each different clone of the same genotypes was also considered as a biological replicate in the experiments using mutant cell lines obtained by the CRISPR-Cas9 system . Technical replicates were multiple analyses of the same sample in an experiment . The results are expressed as mean + standard error of the mean ( SEM ) of three biological replicates ( n = 3 ) . Significant differences were considered as probabilities less than 5% ( p<0 . 05 ) .	The proteasome is a machine that destroys unnecessary or damaged proteins inside cells . This role of the proteasome is essential for cell survival , and so when the proteasome is inhibited , cells produce new proteasomes to compensate . Upon proteasome inhibition , a protein called Nrf1 is activated and executes this “bounce-back” response . Some cancer treatments aim to kill cancer cells by inhibiting proteasomes , but these treatments may be unsuccessful if the bounce-back response is not also prevented . Therefore , understanding how Nrf1 is activated is an important issue . Nrf1 is produced at a structure called the endoplasmic reticulum in cells and is continually destroyed by the proteasome . On the other hand , when proteasomes are inhibited , Nrf1 accumulates and is cleaved into an active form , which moves to the cell nucleus to start producing proteasomes . However , it was not known which molecule cleaves Nrf1 . Koizumi et al . set out to discover this molecule by screening the genetic material of human cells , and identified a gene that encodes a protease ( an enzyme that cleaves other proteins ) called DDI2 . The loss of DDI2 from cells prevented Nrf1 from being cleaved and entering the nucleus , resulting in low levels of proteasome production . Further experiments showed that a mutant form of DDI2 that lacked protease activity was unable to cleave Nrf1 , confirming DDI2’s role in activating Nrf1 . Deleting DDI2 from cells does not completely prevent the cleavage of Nrf1 , and so some other cleaving enzyme might exist; the identity of this enzyme remains to be discovered . Future work is also needed to establish exactly how DDI2 cleaves Nrf1 . This could help to develop a DDI2 inhibitor for cancer treatment that could be used in combination with existing proteasome inhibitors .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "short", "report" ]	2016	The aspartyl protease DDI2 activates Nrf1 to compensate for proteasome dysfunction
Intelligence and education are predictive of better physical and mental health , socioeconomic position ( SEP ) , and longevity . However , these associations are insufficient to prove that intelligence and/or education cause these outcomes . Intelligence and education are phenotypically and genetically correlated , which makes it difficult to elucidate causal relationships . We used univariate and multivariable Mendelian randomization to estimate the total and direct effects of intelligence and educational attainment on mental and physical health , measures of socioeconomic position , and longevity . Both intelligence and education had beneficial total effects . Higher intelligence had positive direct effects on income and alcohol consumption , and negative direct effects on moderate and vigorous physical activity . Higher educational attainment had positive direct effects on income , alcohol consumption , and vigorous physical activity , and negative direct effects on smoking , BMI and sedentary behaviour . If the Mendelian randomization assumptions hold , these findings suggest that both intelligence and education affect health . Intelligence and educational attainment are associated with many socioeconomic and health outcomes ( Cutler and Lleras-Muney , 2006; Clark and Royer , 2013; Deary and Johnson , 2010; Deary , 2012; Hill et al . , 2016a ) . However , the causal relationships between intelligence , education , and health outcomes are unclear , in part , because intelligence and education are strongly correlated . On average , children who score more highly in intelligence tests tend to remain in school for longer , ( Deary et al . , 2007 ) and people who remained in school for longer tend to have higher intelligence later in life ( Ritchie and Tucker-Drob , 2018 ) . Intelligence and educational attainment are partially heritable , and strongly genetically correlated ( rg=0 . 70 ) ( Lee et al . , 2018; Hill et al . , 2019; Hill et al . , 2016b; Hill et al . , 2018 ) . A review of the effects of educational attainment on socio-economic and health outcomes in later life found some evidence that increases in education led to lower mortality , especially for older cohorts ( Galama et al . , 2018 ) . However , there was little consistent evidence of effects on other outcomes such as obesity ( Galama et al . , 2018 ) . A systematic review of 28 studies provided evidence using quasi-experiential designs that each year of education causes measured IQ to increase by on average 1 to 5 points ( Ritchie and Tucker-Drob , 2018 ) . Large prospective studies have found that intelligence test scores taken at age 11 strongly associate with educational attainment later in life , a link that is , in part , explained by shared genetic effects ( Deary et al . , 2007; Calvin et al . , 2012 ) . However , there are no quasi-experimental studies of the effects of intelligence on education because intelligence is less subject to perturbation by natural experiments such as policy reforms . In order to design successful interventions aimed at the amelioration of health conditions , it is necessary to determine the extent to which intelligence , education , or both , are causal factors for health outcomes . If the health differences are mainly due to differences in intelligence that are independent from education , then changes to the length of schooling are unlikely to affect population health . If , however , educational attainment has a causal effect on health and socioeconomic outcomes later in life , and these effects are independent of intelligence , the health of the population may be able to be raised by implementing a policy change aimed at increasing the duration of education ( Deary and Johnson , 2010 ) . Mendelian randomization is an approach that can provide evidence about the relative causal effects of intelligence and education on social and health outcomes under specific assumptions ( Davey Smith and Ebrahim , 2003; Davies et al . , 2018a ) . Mendelian randomization generally uses single nucleotide polymorphisms ( SNPs ) that associate with traits of interest , in this case intelligence and educational attainment , as proxy variables for a trait . At each SNP , offspring inherit at random one of their mother’s two possible alleles and one of their father’s two possible alleles . With the exception of somatic mutation , SNPs are invariant post-conception , so it is not possible for the environment or developing disease to affect inherited DNA . Thus , SNPs are not affected by reverse causation , which can distort causal interpretations of observational epidemiological associations . Segregation of alleles at germ cell formation ( Mendel’s first law ) and the independent assortment of alleles with respect to the rest of the genome excepting proximal DNA segments ( Mendel’s second law ) , and the lack of environmentally influenced effects on survival from conception through to live birth and study entry , leads to genetic variants being largely unrelated to factors that would confound conventional observational studies ( Davey Smith , 2011 ) . Instrumental variable interpretations of Mendelian randomization analysis depend on three assumptions , ( 1 ) the genetic variants associate with the risk factors of interest , ( 2 ) the genetic variants-outcome associations are not confounded by potentially unmeasured factors , and ( 3 ) the genetic variants only affect the outcomes via their effect on the exposures of interest ( in this case intelligence or education ) ( Davies et al . , 2018a ) . One approach would be to estimate the effects of intelligence and education separately using SNPs identified in GWAS for intelligence or education ( Hill et al . , 2019; Okbay et al . , 2016 ) . However , this could be misleading , as SNPs that affect intelligence are likely to also affect education and vice versa . Thus , the SNPs associated with intelligence and education are pleiotropic – and most affect both traits . However , it is unclear whether any effects of the SNPs on the health and socioeconomic outcomes later in life via education and intelligence are vertically or horizontally pleiotropic ( Davey Smith and Hemani , 2014 ) . Consider a Mendelian randomization study with one exposure - education . Vertically pleiotropic effects would occur if the SNPs being used as proxies for education affect the outcomes first via their effects on intelligence , which then has a downstream effect on educational attainment ( Figure 1a ) . Horizontally pleiotropic effects could occur if a SNP being used as a proxy for education directly affected an outcome via intelligence without being mediated via education ( Figure 1b ) . In a Mendelian randomization analysis using education as a single exposure , this horizontal pleiotropy would violate the third Mendelian randomization assumption . Alternatively , the SNPs being used as proxies for education could affect intelligence , and consequently educational attainment , but all of the effects of the SNPs on the outcome could be mediated via intelligence ( Figure 1c ) . Pleiotropy would have similar consequences for a Mendelian randomization study with intelligence as a single exposure . One method which can provide evidence about which of these scenarios is most likely is multivariable Mendelian randomization ( Sanderson et al . , 2018; Burgess et al . , 2015 ) . This approach simultaneously estimates the effects of two or more risk factors using a potentially overlapping set of SNPs . Multivariable Mendelian randomization estimates the direct effect of each risk factor – i . e . the direct effect of intelligence on outcomes that is not mediated via the effect of intelligence on education , and the direct effect of education that is not mediated via the effect of education on intelligence ( Figure 2 ) . It is important to note that multivariable Mendelian randomization does not overcome bias due to horizontally pleiotropic effects via other mechanisms , for example , effects via character or personality . Here , we used a large sample from the UK Biobank and SNPs associated with intelligence and education to estimate the direct effects of intelligence and education on a range of health and socioeconomic outcomes . Our primary analysis uses two-sample multivariable Mendelian randomization of the effects of intelligence and education on a range of socioeconomic and health outcomes . This approach makes most efficient use of available data . We present single sample Mendelian randomization analysis using polygenic risk scores in the UK Biobank as sensitivity analyses . We investigated the direction of causation between intelligence and education using bidirectional two-sample Mendelian randomization . We restricted analysis to SNPs that were available in the Hill et al . ( 2019 ) intelligence genome-wide association study ( GWAS ) , the discovery sample of Okbay et al . ( 2016 ) educational attainment GWAS , and the UK Biobank Haplotype Reference Consortium panel ( HRC ) . We selected all SNPs that were associated with intelligence at p<5 × 10−08 . Of these SNPs , we then selected the lead SNPs within each 10 , 000 kb genomic region by selecting the SNP with the lowest p-value . We then repeated this process to select all SNPs that associated with education at p<5 × 10−08 , and selected lead SNPs within each region . This resulted in 270 SNPs that were associated with either intelligence or education , or both . Some of these SNPs represented the same signal , so we clumped a combined list of SNPs by selecting the SNP with the lowest p-value for education . This resulted in 219 SNPs , of which 144 were associated with intelligence , but not education at p<5 × 10−08 , 38 were associated with education , but not intelligence at p<5 × 10−08 , and 37 were associated with both intelligence and education at p<5 × 10−08 . All 219 SNPs associated with either trait were included in the multivariable Mendelian randomization analysis for both intelligence and education ( for a flow chart of selection of SNPs see Supplementary file 1 - Figure 2 ) . We estimated the effects of intelligence on education and vice versa using non-overlapping data from the UK Biobank . We used two-sample summary data Mendelian randomization methods including the inverse variance weighted ( IVW ) , MR-Egger , weighted median and modal estimators , see methods for details ( Bowden et al . , 2016a; Bowden et al . , 2016a; Hartwig et al . , 2017 ) . The inverse variance weighted ( IVW ) estimates implied that a one standard deviation ( SD ) increase in intelligence increased years of education by 0 . 52 SD ( 95% CI: 0 . 48 to 0 . 56 , p-value=2 . 2 × 10−145 ) . The IVW estimates imply that a one SD increase in education increased intelligence by 0 . 77 SD ( 95% CI: 0 . 68 to 0 . 86 , p-value=3 . 0 × 10−62 ) . There was more evidence of heterogeneity in the estimated effects of education across the SNPs ( I2 = 0 . 60 , 95% CI: 0 . 48 to 0 . 69 ) than in the estimated effects of intelligence ( I2 = 0 . 48 , 95% CI: 0 . 38 to 0 . 56 ) . The heterogeneity statistic indicates the variability of the estimated effects between SNPs; a value of zero indicates no heterogeneity and one indicates high heterogeneity . One explanation for this heterogeneity across SNPs is if the SNPs have horizontally pleiotropic effects . The effects of each of the SNPs on intelligence and education , along with the IVW , MR-Egger , weighted median and weighted mode estimates are presented in Figure 3 and Supplementary file 1 - Table 1 and 2 . The estimated effect of intelligence was consistent and robust across four different estimators ( IVW , MR-Egger , weighted median and weighted mode ) . The estimated effect of education was robust across IVW , weighted median and weighted mode , but the MR-Egger estimate was attenuated by over half , one explanation for this is if the education SNPs have unbalanced horizontally pleiotropic effects on intelligence or the INSiDE assumption is violated . There was little statistical evidence that the estimated effects of intelligence on education were biased by horizontal pleiotropy ( MR-Egger intercept = 0 . 001 , 95% CI: −0 . 005 to 0 . 003 , p-value=0 . 68 ) . However , these tests for pleiotropy are likely to have low power . There was modest evidence that the estimated effects of education on intelligence were affected by horizontal pleiotropy ( MR-Egger intercept = 0 . 009 , 95% CI: 0 . 001 to 0 . 016 , p-value=0 . 03 ) . We estimated the total effect of intelligence and education on each of the social and health outcomes using two-sample summary data Mendelian randomization ( Supplementary file 1 - Figure 3 ) . This approach estimates the total effect of a one SD change in intelligence or education including any effects mediated through the other ( or any other , for example character ) trait . The estimates of the total effects for each outcome for intelligence and education were generally in a consistent direction . The inverse-variance weighted estimates imply that a one standard deviation increase in intelligence reduced risk of: high blood pressure by 4 . 0 percentage point ( pp ) ( 95% CI: 0 . 02 to 0 . 06 , p-value=1 . 2×10−05 ) ; diabetes by 0 . 8pp ( 95% CI: 0 . 1 to 1 . 6 , p-value=0 . 04 ) ; having had a heart attack by 1 . 2pp ( 95% CI: 0 . 7 to 1 . 7 , p-value=5 . 5 × 10−07 ) ; of reporting having seen a GP for nerves , anxiety , tension or depression by 0 . 5pp ( 95% CI: 0 . 3 to 0 . 5 , p-value=8 . 1×10−08 ) . The results suggested that a one standard deviation increase in intelligence was unlikely to have a large effect on risk of mortality or cancer ( risk difference = −0 . 001 , 95% CI: −0 . 006 to 0 . 003 , p-value=0 . 57 , and = −0 . 01 , 95% CI: −0 . 02 to 0 . 003 , p-value=0 . 12 ) . A one standard deviation reduced risk of being an ever or current smoker by 4 . 9pp ( 95% CI: 2 . 8 to 6 . 9 , p-value=4 . 5 × 10−06 ) and 3 . 5pp ( 95% CI: 2 . 4 to 4 . 7 , p-value=1 . 2×10−09 ) . Intelligence also had substantial total effects on household income , increasing the probability of reporting a household income higher than £18 , 000 , £31 , 000 , £52 , 000 and £100 , 000 by 11 . 4pp ( 95% CI: 9 . 8 to 12 . 9 , p-value<1 . 4×10−45 ) , 13 . 4pp ( 95% CI: 11 . 5 to 15 . 3 , p-value=2 . 1×10−44 ) , 11 . 9pp ( 95% CI: 10 . 3 to 13 . 6 , p-value=3 . 4×10−47 ) , and 4 . 1pp ( 95% CI: 3 . 3 to 4 . 9 , p-value=7 . 7×10−24 ) respectively . Higher intelligence generally had beneficial total effects on all of the continuous outcomes . A one standard deviation increase in intelligence was estimated to increase: grip strength by 0 . 32 kg ( 95% CI: 0 . 03 to 0 . 67 , p-value=0 . 07 ) ; height by 1 . 44 cm ( 95% CI: 0 . 95 to 1 . 94 , p-value=1 . 4×10−08 ) ; alcohol consumption by 0 . 29 ( 95%CI: 0 . 23 to 0 . 35 ) ; days of vigorous physical activity per week by 0 . 17 ( 95%CI: 0 . 10 to 0 . 24 , p-value=6 . 7×10−06 ) ; and moderate physical activity by 0 . 32 ( 95%CI: 0 . 23 to 0 . 40 , p-value=1 . 1×10−13 ) and reduce: BMI by 0 . 91 kg/m2 ( 95% CI: 0 . 65 to 1 . 18 , p-value=7 . 7×10−12 ) ; and systolic and diastolic blood pressure by 1 . 09 ( 95%CI: 0 . 65 to 1 . 54 , p-value=1 . 5 × 10−06 ) and 1 . 86 ( 95%CI: 1 . 12 to 2 . 61 , p-value=9 . 1×10−07 ) mmHg respectively; and hours watching television per day by 0 . 49 ( 95%CI: 0 . 42 to 0 . 57 , p-value=4 . 3×10−20 ) . The effects of a one SD increase in education were very similar in direction and magnitude to those of intelligence . An exception was that there was little evidence that education affected frequency of vigorous physical activity ( mean difference = 0 . 00 95% CI: −0 . 15 to 0 . 15 , p-value=0 . 997 ) . Next , we estimated the direct effect of each exposure using multivariable Mendelian randomization ( Sanderson et al . , 2018 ) . The direct effect of intelligence ( or , mutatis mutandis , education ) is the effect of intelligence that is not mediated via education ( or intelligence ) . Multivariable Mendelian randomization estimates the effects of two exposures using the two sets of ( overlapping ) SNPs as instruments . We restricted the analysis to SNPs in linkage equilibrium which were identified in the intelligence and/or education GWAS at p<5 × 10−08 clumped on r2 = 0 . 01 within 10 , 000 kb using the 1000 genomes reference panel ( Hemani et al . , 2018 ) . Some SNPs that were selected from the intelligence and education GWAS were closely positioned in the genome and were correlated . For these pairs of SNPs , we selected the SNP that most strongly associated with education in the GWAS . Sensitivity analysis which clumped SNPs using their association in the intelligence GWAS is presented in the supplementary materials; however , the results were virtually identical . The instruments strongly predicted both education and intelligence in the single sample analysis ( the minimum Sanderson-Windmeijer multivariable F-statistic was 21 . 6 ) ( Sanderson et al . , 2018; Sanderson and Windmeijer , 2015 ) . Sanderson-Windmeijer F-statistics tests the strength of the SNP-exposure conditional on the other exposure ( intelligence or education ) . Because the effects of the SNPs on intelligence and education are similar ( but not identical ) , the Sanderson-Windmeijer multivariable F-statistics are smaller than standard univariable F-statistics . The estimates of the direct effects presented in Figure 4 are less precise than the estimates of the total effects . It is not possible to test the strength of the instruments to jointly predict both of the exposures for the two-sample analysis . However , the Sanderson-Windmeijer tests of the strength of the instruments in the single sample analysis is likely to provide a lower bound of their strength . In the two-sample analysis the estimates of the direct effects of intelligence were attenuated compared to the total effects . As seen ( Figure 4 and Supplementary file 1 - Figure 3 ) , a one SD increase in intelligence score increased the probability of a household income above £18 , 000 and £52 , 000 by 5 . 2pp ( 95% CI: 1 . 5 to 8 . 9 , p-value=0 . 007 ) and 4 . 7pp ( 95% CI: 0 . 8 to 8 . 6 , p-value=0 . 02 ) respectively . A SD increase in intelligence score increased alcohol consumption by 0 . 19 ( 95%CI: 0 . 06 to 0 . 32 , p-value=0 . 005 on a five-unit Likert scale ) . A SD increase in intelligence score decreased rates of vigorous and moderate physical activity by 0 . 34 ( 95%CI: 19 . 6 to 49 . 1 , p-value=8 . 2 × 10−06 ) and 0 . 31 ( 95%CI: 0 . 12 to 0 . 50 , p-value=1 . 9 × 10−03 ) days per week . The direct effects of education that were attenuated compared to the total effects . A one SD increase in education increased the probability of having higher income across the entire income distribution . The direct effects of education on household income on the risk difference scale were between 2 . 0 and 6 . 6 times as large as the direct effects of intelligence . A SD increase in education resulted in a 1 . 00 kg/m2 ( 95% CI: 0 . 06 to 1 . 93 , p-value=0 . 04 ) decrease in BMI , which was larger than the direct effects of intelligence . The direct effects of education on alcohol consumption were similar to intelligence: an increase of 0 . 21 ( 95%CI: 0 . 01 to 0 . 41 , p-value=0 . 04 ) . Each SD increase in education reduced television consumption by 46 . 7 min per day ( 95% CI: 33 . 6 to 59 . 8 , p-value=1 . 8 × 10−11 ) and increased vigorous physical activity by 0 . 31 days per week ( 95% CI: 0 . 09 to 0 . 54 , p-value=0 . 007 ) . A potential source of bias in Mendelian randomization studies is sample overlap . If the same sample is used to detect the SNPs used as instruments as is used in Mendelian randomization the results can be biased towards the observed exposure-outcome association ( Burgess et al . , 2016 ) . We minimised the possibility of this bias by restricting our main results to two entirely non-overlapping samples ( see methods and supplementary materials ) . We used a restricted set of SNPs for intelligence detected without using UK Biobank data as a sensitivity analysis using individual-level data . The results from both approaches were consistent . Pleiotropy could explain our results if the SNPs we used as instruments directly affect the outcome through mechanisms other than intelligence and education , e . g . personality . However , a systematic review of studies investigating the effects of non-cognitive skills on academic and health outcomes later in childhood found evidence of modest effects on academic outcomes , and very few estimates of their associations with health outcomes later in life ( Smithers et al . , 2018 ) . Furthermore , the published estimates were consistent with substantial small study and publication bias . The measure of intelligence in UK Biobank is relatively crude: a 13 item verbal-numeric reasoning test . In a multivariable adjusted phenotypic analysis , this would cause measurement error on the exposure and attenuation of the coefficient on intelligence . However , Mendelian randomization is implemented here as a form of instrumental variable analysis , and therefore is less likely to be affected by measurement error on the exposures than conventional analyses ( Sargan , 1958 ) . Both intelligence and educational attainment had similar values of the Sanderson-Windmeijer F-statistic . The intelligence GWAS we used to identify SNPs associated with intelligence was conducted in older adults . Intelligence is relatively stable across the life-course . For example , Deary et al . ( 2004 ) found that scores on the Moray House Test ( a mental ability test ) taken at age 11 and around age 77 were correlated ( r = 0 . 66 ) ( Deary et al . , 2004 ) . On average , SNPs associated with intelligence in adults had similar effects on intelligence in children ( rg = 0 . 71 ) ( Hill et al . , 2016b ) . As a result our estimates of the effects of intelligence will partially reflect the effects of adult intelligence on the outcomes . Non-genetic quasi-experimental evidence suggests that length of schooling affects adult intelligence ( Ritchie and Tucker-Drob , 2018 ) . If adult intelligence affects the outcomes , then the estimated direct effect of education would be attenuated by any direct effects of education on the outcomes mediated via adult intelligence . Thus , our estimates of the direct effects of intelligence may be overestimates because they also include effects of adult intelligence . The effects of childhood intelligence could be estimated using SNPs identified in a GWAS of intelligence in children . However , currently available GWAS of childhood intelligence are considerably smaller than those available for adult intelligence . Mendelian randomization studies using samples of unrelated individuals can be biased by bias due to population stratification and difference in ancestry across variants and selection and participation bias ( Haworth et al . , 2019; Taylor et al . , 2018 ) . We investigated this by adjusting for additional covariates , and bias component plots ( Davies et al . , 2017 ) . These analyses suggested that while the genetic scores for intelligence and education associate with some measures of early life experience and place of birth , the bias induced in our estimates may be limited . Another potential explanation for these results is dynastic effects which occur if parents’ intelligence or education directly affects their offspring’s outcomes . SNPs associated with education in GWAS also associate with parental education . Non-inherited parental alleles at these loci also associate with offspring’s outcomes via their expression in parental phenotypes ( Kong et al . , 2018 ) . Furthermore , substantial fractions of the GREML-SNP estimates of the heritability of education may be due to indirect effects of parents ( Young et al . , 2018 ) . Thus , our estimates of the effect of intelligence and education are likely to attribute these parental effects to the offspring’s characteristics . Similarly , parents do not mate randomly and assort on educational attainment . The associations of non-inherited alleles of SNPs known to associate with education provide evidence about the size of these effects . The association of offspring outcomes and the non-transmitted polygenic score for education was 31% and 29% of the size of the association of the transmitted scores from fathers and mothers respectively . Similarly , the size of the association of the non-transmitted education polygenic score with a broad measure of health outcomes was 37% and 42% of the transmitted score for fathers and mothers respectively . These results suggest that some of the effects we report could be due to assortative mating , dynastic ( genetic nurture ) effects , which can cause bias in Mendelian randomization studies ( Hartwig et al . , 2018 ) . We investigated this using the siblings in UK Biobank , but our results were underpowered . Future studies should investigate this further using within-family studies with larger samples ( Lawlor et al . , 2017; DiPrete et al . , 2018; Warrington et al . , 2018; Brumpton et al . , 2019 ) . The effects we report may be specific to the time-period that the UK Biobank participants have lived through . For example , we find evidence of effects on smoking rates , particularly in single sample analysis . However , smoking rates have declined since the 1960s , 1970s , and 1980s , when the UK Biobank participants left school . Changes to education policies today , such as recent changes in the United Kingdom to mandate education , or part-time training and apprenticeships up to the age of 18 , may not have the same impact as we report ( UK Government , 2018 ) . The UK Biobank is more educated than the general population , which could cause selection bias . However , we have previously shown that reweighting the sample to account for the under sampling of less educated people did little to affect the Mendelian randomization results ( Davies et al . , 2018b ) . We have only investigated the effect of intelligence and education on a limited number of outcomes reported in UK Biobank . There are differences in morbidity and mortality by intelligence and education for a wide range of disease outcomes , and the conclusions we report here , that the direct effects of education are bigger than intelligence may not hold for other outcomes . Future studies should apply multivariable Mendelian randomization to summary data on outcomes to investigate this hypothesis as efficiently as possible ( Inoue and Solon , 2010; Angrist and Krueger , 1995; Pierce and Burgess , 2013 ) . Multivariable Mendelian randomization is a flexible approach for estimating and evaluating possible pleiotropic pathways . Future studies could exploit these methods to elucidate the mechanisms that mediate the effects of intelligence and education on outcomes later in life . In summary , we found evidence from genetic association studies that both intelligence and education might affect health and social outcomes later in life . The direct effects for education are larger than for intelligence , suggesting that much of the effect of intelligence on outcomes later in life may be mediated via the effect of intelligence on education . The UK Biobank is a cohort study that recruited 503 , 317 people aged between 38 and 73 years old between 2006 and 2010 in 21 study centres across the UK . See Supplementary file 1 - Figure 1 for an illustration of the inclusions and exclusion of samples into the study . UK Biobank received ethical approval from the Research Ethics Committee ( REC reference for UK Biobank is 11/NW/0382 ) . Mendelian randomization estimates can be affected by weak instrument bias if overlapping samples are used to select SNPs associated with the exposures ( intelligence and education ) and the outcomes ( Burgess et al . , 2016 ) . To minimise risk of this bias we estimated the SNP-outcome associations using a sample that excluded any participants who were included in the GWAS used to select the SNPs . The bivariate analysis estimated the SNP-outcome associations using participants that were not included in the intelligence GWAS reported by Hill et al . ( 2019 ) or the education GWAS reported by Okbay et al . ( 2016 ) . This excludes participants who took the verbal-numeric reasoning test or were in the UK Biobank interim release . Therefore , we used the remaining 124 , 661 participants to estimate the association between the SNPs and the outcomes of interest in the two-sample analysis . For the single sample analysis ( more details below ) , we constructed weighted allele scores for intelligence and education using the Sniekers GWAS which only included the UK Biobank interim release and the Okbay discovery sample ( Sniekers et al . , 2017 ) . These GWAS are smaller than Hill GWAS , so detected fewer SNPs associated with intelligence at p-value<5 × 10−08 . This fact means that the analyses using these SNPs are less precise than the two-sample analysis used in the primary analysis . We used 77 , 882 participants who took the verbal-numeric reasoning test but were not included in the interim release . However , because the single sample analysis was restricted to individuals with the verbal-numeric reasoning scores , we can test whether the intelligence and education genetic scores sufficiently associate with intelligence and education . We used the Sanderson-Windmeijer F-test to test whether the polygenic scores explained sufficient variation in intelligence ( education ) conditional on education ( intelligence ) . We used a broad measure of depression that the participant had seen a GP for nerves , anxiety , tension , or depression ( ID:2090 ) at either the initial assessment centre visit or any repeat assessment centre visit , or a HES record indicating depression as a primary or secondary reason for admission ( ID: 41202 , 41204 and ICD-10 = F32 , F33 , F34 , F38 , F39 ) ( Howard et al . , 2018 ) . The only participants who had measures of arterial stiffness were those who took the verbal-numeric reasoning test , therefore we excluded this outcome . More details of the phenotype definitions are provided elsewhere ( Davies et al . , 2018c ) . We defined educational attainment using the same algorithm as the educational attainment GWAS ( Okbay et al . , 2016 ) . Educational attainment was coded using the answer to touch-screen questions about qualifications . We assigned participants to their highest level of education reported at either assessment centre visit . Those with degrees were assigned to 20 years of education; NVQs , HND or HNC qualification were assigned to 19 years of education; other professional qualifications were assigned to 15 years; A-levels or AS levels were assigned to 13 years; GSCEs , O-levels or CSEs were assigned 10 years of education; and none of the above were assigned to 7 years . Previous studies have found that using self-reported age had little impact on estimates of the effect of education ( Sanderson et al . , 2019 ) . We dropped individuals who stated prefer not to say or did not have a value for this question from the analysis . This measure of educational attainment was standardised to mean zero and variance one . We defined intelligence using the ‘verbal-numeric reasoning’ score from the test taken at the baseline assessment centre visits . We replaced missing values of this test for the initial assessment centre visit with values taken at first repeat assessment visit ( N = 15 , 404 ) . The participants answered 13 logic questions within two minutes . We standardised this variable to mean zero and variance one . All analyses included 40 principal components of genetic variation , sex , age , year and month of birth , and an interaction of month and year of birth and sex . Full details of our genotype quality control pipeline are described elsewhere ( Mitchell et al . , 2017 ) . In brief , we excluded participants who had mismatching sex , those with non XX or XY sex chromosomes , extreme heterozygosity or missingness . We limited the analysis to 11 , 554 , 957 SNPs on the HRC panel , of which we further limited to the 7 , 303 , 122 SNPs which were available in both the Hill and Okbay GWAS . We selected independent SNPs ( r2 <0 . 01 within 10 , 000 kb ) that were associated with either intelligence or education at p<5 × 10−08 . Where there was more than one SNP in a region that associated with the trait at p<5 × 10−08 we selected the SNP with the lowest p-value . We then took the combined list of SNPs associated with either intelligence or education and repeated this process of clumping and selecting the SNP with the lowest p-value in the Okbay GWAS to create a list of SNPs for both traits that were in linkage equilibrium . See Supplementary file 1 - Figure 2 for flowchart . We used bi-directional Mendelian randomization to investigate the direction of causation between intelligence and education . We estimated the effect of intelligence on education using the 181 lead SNPs from Hill et al . and data on educational attainment from participants of the UK Biobank who did not take the verbal-numeric reasoning test and were not in the interim release ( Hill et al . , 2019 ) . These individuals were not included in the Hill et al . GWAS . We estimated the effects of education on intelligence using the 75 SNPs reported in the Okbay et al . ( 2016 ) discovery sample and samples with intelligence measures from UK Biobank . These samples do not overlap . We estimated the effects using two-sample summary data Mendelian randomization . The primary analysis used inverse variance weighted estimators . As sensitivity analyses , we used MR-Egger , weighted median , and weighted mode estimators to investigate whether pleiotropy biased the IVW estimates ( Bowden et al . , 2016a; Hartwig et al . , 2017; Bowden et al . , 2016b ) . We adjusted all the summary estimates for month and year of birth , sex , interactions of month and year of birth and sex and 40 principal components of genetic variation . We report estimates of the instrument strength and heterogeneity of the instrument-exposure association and the estimated effect of each of the exposures on the social and health outcomes across different SNPs . We estimated the total effect of intelligence and education on each of the health and social outcomes using univariable Mendelian randomization . We used 181 SNPs from Hill et al . ( 2018 ) and 75 SNPs from the discovery sample of Okbay et al . ( 2016 ) as instruments for intelligence and education respectively . Our primary analysis used IVW estimators which assume no directional pleiotropy . These estimates ignore possible pleiotropic or mediated effects via the other exposure: education and intelligence . We estimated the effect of each of the phenotypes using methods that were robust to other forms of pleiotropy using MR-Egger , weighted median and weighted mode estimators ( Bowden et al . , 2016a; Hartwig et al . , 2017; Bowden et al . , 2016b ) . These can obtain consistent estimates of the causal effect if the pleiotropic effects are independent of the effects on the exposure , or if the majority or most frequent variants are not pleiotropic . We used multivariable Mendelian randomization to estimate the direct effects of intelligence and education on each of the health and social outcomes . This method has been described in detail elsewhere ( Sanderson et al . , 2018 ) . In brief , this method is based on standard instrumental variable methods which allow for multiple exposures . Each exposure must be sufficiently explained by the set of instruments have an instrument that explains a sufficient proportion of the variation in the exposure , conditional on the other exposure . In the single sample , the SNPs can be correlated with each other ( i . e . in linkage disequilibrium ) and in the single or two sample case can also correlate with more than one exposure . The strength of association of the proposed instruments and the exposure can be tested using Sanderson-Windmeijer tests ( Sanderson and Windmeijer , 2015 ) . As described elsewhere it is possible to use these methods with summary data from two separate samples to estimate the SNP-exposure and SNP-outcome associations . The two-sample approach allows us to efficiently combine information from multiple studies , not all of which have measured intelligence , education and the outcomes ( Pierce and Burgess , 2013 ) . Hence we can integrate more data and have more precise estimates . We estimated the multivariable effects of intelligence and education using linear regression weighted for by one over the standard error of the SNP-outcome association . We conducted a series of sensitivity analyses to investigate how sensitive our results were to the methods we used . The cleaned analysis dataset will be uploaded to the UK Biobank archive . Please contact access@ukbiobank . ac . uk for further information . The analytic scripts used to clean the data and produce the results are available on GitHub ( Davies , 2019; copy archived at https://github . com/elifesciences-publications/ukbiobank-intell-vs-ea ) .	Highly educated people tend to be healthier and have higher incomes than those with less schooling . This might be because education helps people adopt a healthier lifestyle , as well as qualifying them for better-paid jobs . But , on average , highly educated people also score more highly on cognitive tests . This may explain why they tend to adopt healthier behaviours , such as being less likely to smoke . Because education and intelligence are so closely related , it is difficult to tease apart their roles in people’s health . Davies et al . have now turned to genetics to explore this question , focusing on genetic variation associated with intelligence and education levels . Analysing genetic and lifestyle data from almost 140 , 000 healthy middle-aged volunteers from the UK Biobank study suggested that together , intelligence and education influence many life outcomes , but also that they have independent effects . For instance , there is evidence that more intelligent people tend to earn more , irrespective of their education . However , more educated people also tend to earn more , even after accounting for their intelligence . They also tend to have lower BMIs , be less likely to smoke , and engage in less sedentary behaviour and more frequent vigorous exercise in midlife . For each of these outcomes , the effects of education are all in addition to the effects of intelligence . Education and intelligence thus affect life outcomes together and independently . Overall , the results of Davies et al . suggest that extending education , for example by increasing school-leaving age , could make the population as a whole healthier . However , the individuals in the current study grew up when smoking was far more common than it is today . Some of the observed effects on health may thus be due to differences in smoking rates between groups with different levels of education . If so , increasing education may not have as much impact today as it did in the past . It is also possible that these findings reflect the effects of the family environment , for example how parents influence their offspring . Larger studies are needed to investigate this hypothesis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2019	Multivariable two-sample Mendelian randomization estimates of the effects of intelligence and education on health
Emerging evidence suggests that dopamine may modulate learning and memory with important implications for understanding the neurobiology of memory and future therapeutic targeting . An influential hypothesis posits that dopamine biases reinforcement learning . More recent data also suggest an influence during both consolidation and retrieval . Eighteen Parkinson’s disease patients learned through feedback ON or OFF medication , with memory tested 24 hr later ON or OFF medication ( 4 conditions , within-subjects design with matched healthy control group ) . Patients OFF medication during learning decreased in memory accuracy over the following 24 hr . In contrast to previous studies , however , dopaminergic medication during learning and testing did not affect expression of positive or negative reinforcement . Two further experiments were run without the 24 hr delay , but they too failed to reproduce effects of dopaminergic medication on reinforcement learning . While supportive of a dopaminergic role in consolidation , this study failed to replicate previous findings on reinforcement learning . Phasic changes in dopamine level are believed to encode the reward prediction error ( RPE ) , which measures the difference between the reward expected after an action , and the reward actually received ( Schultz et al . , 1993 , Schultz et al . , 1997 ) . In turn the RPE guides reinforcement learning ( RL ) such that behaviour is adapted to changing surroundings . Several studies have taken advantage of the dopaminergic depletion in Parkinson’s disease ( PD ) in the substantia nigra pars compacta and ventral tegmental area ( Agid et al . , 1989; Shulman et al . , 2011 ) . PD patients are frequently treated with dopamine replacement therapy ( levodopa and dopamine agonists ) , and thus by comparing patients in ON and OFF medication states the effects of dopamine depletion can be investigated . One influential study using such a procedure found that dopaminergic state modulated RL from positive and negative feedback ( Frank et al . , 2004 ) . This study used the Probabilistic Selection Task ( PST ) , in which participants see two Japanese Hiragana symbols on the screen at the same time , and must pick one , receiving either ‘Correct’ or ‘Incorrect’ feedback ( see Figure 1 ) . This feedback is determined probabilistically , so that card A in pair AB is given positive feedback on 80% of trials , and negative feedback on 20% , and vice versa for card B . Pairs CD and EF have probabilities 70–30% and 60–40% , respectively . During the learning trials , if a participant chooses card A over card B , this could be because they have learned that card A is more often rewarded and should be chosen , or that card B is more often punished and should be avoided – one cannot tell these apart from this choice . So , a novel pairs test is given where all the cards are shown in all possible combinations ( e . g . AB , AC , AD , AE… ) , without feedback , and from this the percentage of times that card A is chosen , and card B avoided , can be used as benchmarks for positive and negative reinforcement , respectively . 10 . 7554/eLife . 26801 . 003Figure 1 . Diagram of the learning trials of the Probabilistic Selection Task . In each pair one card is more likely to be rewarded ( shown ‘Correct’ feedback ) than the other , with card A in pair AB rewarded 80% of trials , and card B on 20% of trials . For pairs CD and EF , the probabilities are 70–30% and 60–40% . DOI: http://dx . doi . org/10 . 7554/eLife . 26801 . 003 PD patients chose A more and avoided B less when ON medication , and vice versa for OFF medication ( Frank et al . , 2004 ) . This suggested better learning from positive reinforcement and poorer negative reinforcement ON medication , while patients OFF showed the opposite pattern . Importantly , patients OFF medication were better at negative reinforcement than healthy age-matched controls ( HC ) , suggesting that PD actually improved some aspect of RL . The explanation for these effects was provided with a model of the basal ganglia . In the Go-NoGo model , the two main pathways from the striatum are proposed to underlie positive and negative reinforcement . The direct pathway , which is mainly activated by striatal neurons containing dopamine D1 receptors and therefore is activated by a dopamine increase during a positive RPE , underlies positive reinforcement . The indirect pathway which is inhibited by D2 receptors , and therefore activated when a dopamine decrease signals a negative RPE , allows negative reinforcement . When PD patients are ON medication , the higher dopamine levels activate D1 receptors and inhibit D2 receptors , thus biasing towards the direct pathway , improving positive reinforcement . When OFF medication , the lower dopamine levels mean less D1 activation , and less D2 inhibition , thus increasing indirect pathway activity , and improving negative reinforcement . While this view is persuasive , some more recent studies have cast doubts on the extent to which dopamine is involved in the learning part of RL , and how much is the expression of that learning . In Frank et al . ( 2004 ) study , the RL test was given immediately after learning , and it was only in this test that the effects were seen . However , as patients were ON for both learning and testing , or OFF for both , it meant that the dopaminergic effects could have occurred at either point . One study attempted to correct for this by adding a one hour delay between learning and testing , allowing PD patients to learn OFF medication , then be tested ON medication ( along with ON-ON and OFF-OFF conditions ) ( Shiner et al . , 2012 ) . It was found that being ON medication at the time of testing produced greater expression of positive reinforcement than being OFF , while it had no effect during learning . Additionally , RPE signals in the striatum during learning were not affected by medication state , but ventromedial PFC and nucleus accumbens signals in the test phase only tracked the value of the stimuli when patients were ON medication . This suggests that dopamine may not play as much of a role in learning , but instead influence the choices made after the expected rewards have been learnt . Other studies have also shown dopaminergic D2 receptor effects mainly on the expression rather than learning ( Eisenegger et al . , 2014; Pessiglione et al . , 2006 ) . The direct link between dopamine and RL , as opposed to expression of information , was further questioned by another study , which found that dopaminergic effects could be shown even when rewards were not actually given during learning ( Smittenaar et al . , 2012 ) . Participants received one of two shapes , probabilistically , after selecting a stimulus , and only after the learning trials were they told that one shape corresponded to winning money , and one to losing money . Thus , the reward/monetary associations were created separately to the stimulus-outcome associations . However , they still found that PD patients ON medication during testing ( after finding out about the money ) , showed higher accuracy on the most rewarded stimulus , and lower accuracy on the most punished stimulus . This shows that it is possible to generate this effect without any reinforcement learning actually taking place , suggesting dopamine influences value-based decision making . A recent extension to standard RL models offers a mechanism by which dopamine may influence expression of learning . The OpAL model ( Collins and Frank , 2014 ) has separate learning rates and choice parameters for the direct and indirect pathways , which learn from positive and negative reinforcement , respectively . By allowing dopamine to affect the choice parameter , it can bias towards choosing the stimulus that learned mainly from the direct pathway , or from the indirect pathway , thus lending more weight to the positive or negative reinforcement the stimulus received . In addition to evidence of dopamine affecting expression of learned values , there is also evidence of it affecting consolidation . PD patients ON medication showed an increase in accuracy on an RL task after a 20 min delay , while those OFF medication showed a large decrease ( Coulthard et al . , 2012 ) . This was despite all PD patients showing the same behaviour during the learning trials . It is still possible that this is a retrieval effect , and that it was simply not seen during the learning trials as the values were still being updated , but it is also possible that the dopaminergic medication preserved the synaptic weight changes induced during learning , thus improving memory for the learned items . This explanation ties in nicely with models of synaptic consolidation based on the synaptic tagging and capture hypothesis ( Clopath et al . , 2008; Frey and Morris , 1998; Redondo and Morris , 2011 ) . In these models , early synaptic changes are induced during learning , but decay away unless actively prolonged , which is achieved by the changes setting a ‘tag’ which plasticity-related proteins ( PRPs ) must act on to make the changes permanent . Dopamine is hypothesised to set the threshold for the synthesis of these PRPs , so that higher levels of dopamine mean a lower threshold . Thus , PD patients OFF medication would have a higher PRP threshold , and therefore lower consolidation of early synaptic changes , leading to a delayed impairment of memory , as seen in Coulthard et al . ( 2012 ) . Here , we sought to investigate the hypotheses that: These hypotheses were tested in experiment 1 where surprisingly we did not show the expected effects of dopamine on novel pairs choices . We undertook a further two experiments to investigate the effects of dopamine and delays on RL , and the effects of procedural changes to the PST . As experiment 2 did not replicate the findings of Frank et al . ( 2004 ) , an exact replication was run to ensure we observed the well-described effect of dopamine enhancing positive reinforcement or impairing negative reinforcement . Dopaminergic medication seemed to affect memory performance on the PST . When patients were OFF medication on day 2 , being ON medication the day before ( during learning ) prevented a decrease in their memory over the 24 delay . Interestingly , both day 1 ON conditions had a pattern of a slight increase in memory scores ( albeit non-significantly different to zero ) , while both day 1 OFF conditions ( and HC ) showed a decrease on average . As this score was the difference between 30 min and 24 hr delay tests , it suggests that day 1 dopamine improved consolidation of the learned values sometime after 30 min , preventing a decay in the memory . This is in line with a previous study showing a benefit of dopamine at the time of learning on memory testing 20 min later ( Coulthard et al . , 2012 ) , although it was only seen at longer delays here . All PD patients went back ON medication immediately after the day 1 session regardless of day 2 condition , so all patients were in an ON state for the hours after learning ( see Figure 6 for diagram ) . This means the day 1 ON and OFF groups differed in dopaminergic activity until about 1 . 5 hr after learning , when the medication would have reached peak concentration . This gives a short time window for day 1 medication to affect consolidation of RL . 10 . 7554/eLife . 26801 . 020Figure 6 . A diagram of the timing of PD medication withdrawal for all four conditions in experiment 1 . Blue is when patients were ON medication , red hatched bars when they were OFF , and yellow bars the PST phases . In order for patients to be fully OFF medication during testing , they were withdrawn from their dopaminergic medications a minimum of 15 hr prior to testing ( >24 hr for long-lasting medications ) . Note that in all conditions , patients were ON medication for a few hours after the day 1 session , to minimise the time spent OFF medication . DOI: http://dx . doi . org/10 . 7554/eLife . 26801 . 020 This finding fits with the wider literature implicating dopamine in memory and consolidation mechanisms ( Lisman et al . , 2011; Shohamy and Adcock , 2010 ) . Dopamine given before or after learning can improve consolidation ( Bernabeu et al . , 1997; de Lima et al . , 2011; Furini et al . , 2014; Péczely et al . , 2016; Rossato et al . , 2009 ) , although there is still debate on the time course of its effects . Synaptic tagging and capture models suggest dopaminergic effects would take place within a few hours of learning ( Clopath et al . , 2008; Redondo and Morris , 2011 ) , and consolidation effects on RL have been reported over shorter times before ( Coulthard et al . , 2012 ) . Synaptic tagging has mainly been studied in hippocampal circuits , and may relate to the binding of separate experiences within a time window of hours or days ( Shohamy and Adcock , 2010 ) . However , the PST is assumed to rely on basal ganglia functioning , at least when there are short delays between action and feedback as there were here ( Foerde and Shohamy , 2011 ) . Combining computational synaptic tagging and capture models with basal ganglia RL models would show whether such an explanation could explain this effect . Further behavioural work fractioning the time window after learning where dopamine impacts on consolidation of RL could also be illuminating . Consolidation has not often been the focus in RL studies , rather learning or testing effects , but a few studies have shown that RL consolidation is affected by dopamine or sleep . Three studies have found that sleep affects performance on the Weather Prediction Task ( Barsky et al . , 2015; Djonlagic et al . , 2009; Lerner et al . , 2016 ) . While this task is different to the PST , it is not unreasonable to expect sleep to also affect other RL tasks similarly . Dopamine has also been shown to affect sleep consolidation for reward-related memory ( Feld et al . , 2014 ) , and while reward-related memory is partly a declarative memory process , it relies on the same reward-processing brain regions that underlie RL ( Wittmann et al . , 2005 ) . An alternate explanation is that dopamine state during the 30 min memory block affected reconsolidation of the RL values , allowing the values to be reconsolidated properly and recalled accurately the next day . Dopamine has been implicated in reconsolidation ( Rossato et al . , 2015 ) , although it has not been investigated in the RL domain . Experiment 1 sought to separate the effects of dopamine during learning and during testing on positively and negatively reinforced information , but found no effects of either . PD patients also did not differ from HC . If our study were simply underpowered we might expect the results to at least be in the direction predicted by previous studies . Interestingly , the direction of effect was opposite to the expected effect , with the day 1 ON conditions having the highest amount of avoid-B selections . The classic view is that dopamine improves positive reinforcement , at the cost of impaired negative reinforcement , so it is unclear why the condition in which patients have the most dopamine would show greatest expression of negative reinforcement . Due to this unexpected pattern , another experiment was run without the 24 hr delay between learning and testing , to try to replicate the expected pattern of behaviour . Experiment 2 used the same modified PST as experiment 1 , but with testing immediately after learning . Again , PD patients did not differ from HC , and showed no effect of medication . This is in contradiction to previous studies which found that PD patients had greater expression of positive reinforcement when ON medication , and greater expression of negative reinforcement when OFF medication ( Frank et al . , 2004; Shiner et al . , 2012 ) . It is surprising that we were unable to replicate the findings of dopamine affecting positive and negative reinforcement , especially in experiment 3 which was designed to be an exact replication of the original study ( Frank et al . , 2004 ) . We now look at the possible differences between the studies . The main results reported here were on unfiltered data , but when the data filtering used in previous studies was applied , it made little difference to the results . The average accuracy on novel pairs in experiment 3 was much lower than reported in Frank et al . ( 2004 ) , where the patients ON medications achieved 78% accuracy in choosing A and patients OFF medications achieved 82% accuracy in avoiding B . By contrast the corresponding accuracies in our experiment were very close to chance ( 53% and 51% , respectively; 54 . 5% and 46 . 7% for filtered data ) . So , since the patients were unable to well express any learned preferences , it is not surprising that there was no difference in expression preference learned from positive and negative feedback . One important thing to consider is whether there were sample differences that could explain the disparity between our results and previous studies ( e . g . Frank et al . , 2004 ) . Our samples were very closely matched in age , gender and disease severity to the PD patients tested ON medication in Frank et al . ( 2004 ) . T-tests on the Frank et al . ( 2004 ) PD sample and our experiment 3 sample showed no significant differences in ages or Hoehn and Yahr stages , and χ2 tests showed no significant difference in gender distributions between PD and HC ( p>0 . 1 ) . The PD patients ( p<0 . 001 ) and HC ( p=0 . 046 ) in experiment 3 had fewer years of education on average than those in Frank et al . ( 2004 ) , which could have contributed to the failure to learn this task , although our experiment 1 sample also had fewer years of education ( p<0 . 001 ) and they could learn the modified PST . Given that other studies have found effects of dopamine on RL tasks when using different samples , the differences here are unlikely to be dependent on demographic characteristics . Many studies have used the PST ( or variations ) , and while the findings are not always exactly the same , the general pattern of higher dopamine levels either improving expression of positive reinforcement ( Rutledge et al . , 2009; Shiner et al . , 2012; Smittenaar et al . , 2012; Voon et al . , 2010 ) , impairing negative reinforcement ( Cools et al . , 2006; Frank et al . , 2007b; Mathar et al . , 2017 ) , or both ( Bódi et al . , 2009; Maril et al . , 2013; Palminteri et al . , 2009; Piray et al . , 2014; Wittmann et al . , 2005; Voon et al . , 2010 ) has been seen multiple times . Likewise , low dopamine conditions have shown either worse positive reinforcement ( Eisenegger et al . , 2014; Jocham et al . , 2011; Kobza et al . , 2012 ) , better negative reinforcement ( Bódi et al . , 2009; Cox et al . , 2015; Frank et al . , 2004; Voon et al . , 2010 ) , or both ( Palminteri et al . , 2009; Piray et al . , 2014 ) . However , not all PST studies have agreed with these findings . One study found that while PD patients with greater left hemisphere pathology showed dopaminergic medication modulated reward and punishment learning , patients with greater right hemisphere pathology showed no such effects ( Maril et al . , 2013 ) . They also pointed out that left-hemisphere patients are more common , so are likely to be over-represented in study samples unless specifically balanced . We found no effects of laterality of symptoms in our data . Additionally , the PST was found to have low test-retest reliability when retested 7–8 weeks later ( Baker et al . , 2013 ) ; as well as low correlation between performance at different time points , participants initially classed as ‘positive learners’ or ‘negative learners’ were labelled differently when retested . The study also failed to find any effects of several dopaminergic genes on RL , in contradiction to previous genetic studies ( Doll et al . , 2011; Frank and Hutchison , 2009; Frank et al . , 2007a ) . This study questions the idea that dopamine improves positive reinforcement and/or impairs negative reinforcement , and raises doubts about the reliability of the PST . A recent study has shown that performance on the PST depends on the discriminability of the stimuli used , with the difference in discriminability of stimuli A vs B , and stimuli C vs D affecting whether healthy participants showed choose-A or avoid-B biases ( Schutte et al . , 2017 ) . While the majority of the studies ( ours included ) using the PST counterbalanced which specific stimuli were A and B ( etc . ) , it is still possible that differences in discriminability between the stimuli within and across experiments may have affected results . There are many variations of the PST , including using different stimuli ( Waltz et al . , 2007 ) smiling and frowning faces as feedback ( Aberg et al . , 2015 , 2016; Gold et al . , 2013; Jocham et al . , 2011 ) , using money as feedback ( Kunisato et al . , 2012; Rustemeier et al . , 2012 ) , changing the number of pairs ( Doll et al . , 2014 ) , the probabilities of reward ( Doll et al . , 2014; Evans and Hampson , 2015 ) , the number of trials ( Cicero et al . , 2014; Evans and Hampson , 2015 ) , and the filtering criterion ( Evans and Hampson , 2015; Waltz et al . , 2007 ) . Small changes such as changes to the stimuli used , or changing the delay between action and feedback can have large effects on how this task works ( Foerde and Shohamy , 2011; Schutte et al . , 2017 ) . Care should be taken when comparing across such procedures , as we , and others , have shown that small modifications to RL task procedure can have large effects on behaviour . Some studies are now using simpler RL tasks that require learning to predict the outcome associated with one stimulus shown , rather than picking from two simultaneously shown stimuli ( Bódi et al . , 2009; Herzallah et al . , 2013; Mattfeld et al . , 2011; Simon and Gluck , 2013; Tomer et al . , 2014 ) . Tasks including punishments as well as rewards are frequently used , and often have a simpler probabilistic structure with the same probability for pairs used for the reward and punishment pairs ( Eisenegger et al . , 2014; Naef et al . , 2017; Palminteri et al . , 2009; Pessiglione et al . , 2006 ) . These simpler tasks may make it easier for participants to learn , and have lesser effects of stimulus discriminability . However , discriminability effects have not been tested for in these tasks , and to date only the Weather Prediction Task has published test-retest reliability data ( Aron et al . , 2006 ) . So whether other RL tasks are actually better than the PST or have the same issues remains to be seen . It would be useful if reliability and validation data were included in publications for tasks such as these in future . It is interesting to consider what the implication of our results are for the theories of basal ganglia function . In particular , how our results can be reconciled with the observations that dopaminergic neurons activate striatal D1 neurons , which are believed to be involved in activating movements , while they inhibit the D2 neurons , which are thought to be involved in movement inhibition ( Kravitz et al . , 2010 ) . It has been recently proposed that these neurons encode not only the expected value of an action in the difference of their activity , but also the variance of the reward in the sum of their activity ( Mikhael and Bogacz , 2016 ) . In PST , the stimuli with reward probability closer to 50% have a high variance of reward , because on some trials they result in positive and on some trials in negative feedback . In the simulations of the PST these stimuli strongly activated both D1 and D2 neurons ( Mikhael and Bogacz , 2016 ) . Thus on the simulated novel-pair trials with such stimuli and the stimulus A , the D1 neurons selective for both options were activated , and hence increasing the level of dopamine had little affect the accuracy in choose-A ( or even decreased it for some variants of the model ) . These simulations show that the level of dopamine may little influence the accuracy the PST even if the dopaminergic modulation differentially affects the striatal D1 and D2 neurons . Dopamine and PD did not affect expression of positive or negative reinforcement when tested immediately or 24 hr after learning . Dopamine during learning improved the consolidation of RL memories over 24 hr . The original PST had very low accuracy , and the modifications made to the PST had large effects , increasing learning and novel pairs accuracy , and increasing the amount of avoid-B selections participants made . This highlights the effects that can be induced by small changes to these types of tasks . These experiments failed to replicate the previously reported effects of dopamine and PD on RL , suggesting the effect may be weak . Ethical approval was obtained from the NHS Research Ethics Committee at Frenchay , Bristol . All participants gave written consent , in accordance with the Declaration of Helsinki . As experiment 1 failed to show predicted effects of medication or disease state , further experiments were run . As we had made some modification to the PST based on our pilot data , we designed experiment 2 to test whether performance using our modified PST was as predicted from previous work ( e . g . Frank et al . , 2004 ) with no delay between learning and testing . In all trials , selections of the cards with the highest probability of reward were taken as the ‘optimal’ choice , regardless of what feedback they produced on that specific trial . In the novel pairs block , the learning pairs ( AB and CD ) were excluded from the analysis as in previous studies ( Frank et al . , 2004 ) . Between-subjects ANOVAs were used to compare PD patients to HC , and paired sample t-tests to compare the PD medication conditions ( with Bonferroni corrections for multiple comparisons ) . Cohen’s d and ηp2 effect sizes are given for significant results from t-tests and ANOVAs , respectively . Additional analyses were conducted after data filtering; if a participant scored 50% or lower on the AB choice in the novel pairs task , they were assumed to have not learnt the task properly and that data were excluded . Each condition was checked separately , so one medication condition’s data could be excluded while all other remain . This filtering was only applied to the novel pairs data and analysis . Some data were missing from the analysis , due to experimenter error or computer errors . Two final learning blocks , two 30 min memory blocks and one novel pairs block were missing from experiment 1 . All error bars in the figures are standard error of the mean ( SEM ) . MATLAB ( RRID:SCR_001622 ) was used for data processing ( code available at https://github . com/johnPGrogan/Effects-of-dopamine-on-RL-consolidation-in-PD/releases/tag/v1 . 0; Grogan , 2017 ) , and SPSS ( RRID:SCR_002865 ) for statistical tests . A copy of the code is available at https://github . com/elifesciences-publications/Effects-of-dopamine-on-RL-consolidation-in-PD . We did not obtain consent from participants to share individual data from this study , thus only summary statistics are presented in the figures , tables and text . Individual data are not provided in the source data files , although the summary statistics are .	Brain cells release a naturally occurring chemical called dopamine . The release of this chemical affects how people respond to their ever-changing environment , including how they learn from rewards and punishments . Parkinson’s disease is a condition where the brain cells that make dopamine start to die , and so the levels of dopamine in the brain begin to drop . Parkinson’s disease patients are routinely given drugs to bring their dopamine levels back up to near-normal levels . About 13 years ago , researchers found that when patients with Parkinson’s disease were given dopamine-medication they were better at learning from rewards and worse at learning from punishments . If the patients were withdrawn from their dopamine-medications they were worse at learning from rewards but better at learning from punishments . However , it was not clear if this was because the dopamine affects the learning process , or if it affects how people remember what they learned and how they make choices later on . To better understand how dopamine is involved in learning in people with Parkinson’s disease , Grogan et al . looked at the effects of dopamine on memory over a timescale of 24 hours . People with Parkinson’s disease and healthy volunteers were shown a choice of symbols and given the chance to learn which gave a reward – a picture of a smiling face – and which gave a punishment – a frowning face . If the Parkinson’s disease patients had taken their dopamine-medication before learning the task , their memory did not worsen over the next 24 hours . This suggests that having dopamine in the brain around the time of learning helped the patients to store the memory . The patients , however , were not any better at learning from rewards when taking their medication , which contradicts some earlier studies . To explore this further , Grogan et al . copied the exact same task from the 13-year-old study , and still did not find that patients were better at learning from reward when taking dopamine . These findings could help scientists to better understand what dopamine does during learning and memory , and how the brain normally works . Finally , Parkinson’s disease causes problems with memory . A clearer picture of the types of memory problems patients have , and of how their dopamine-medication can help , might make it easier for clinicians to treat patients with Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Effects of dopamine on reinforcement learning and consolidation in Parkinson’s disease
Axis elongation is a conserved process in which the head-to-tail or anterior-posterior ( AP ) axis of an embryo extends . In Drosophila , cellular rearrangements drive axis elongation . Cells exchange neighbours by converging into transient multicellular vertices which resolve through the assembly of new cell interfaces parallel to the AP axis . We found that new interfaces elongate in pulses correlated with periodic contractions of the surrounding cells . Inhibiting actomyosin contractility globally , or specifically in the cells around multicellular vertices , disrupted the rate and directionality of new interface assembly . Laser ablation indicated that new interfaces sustained greater tension than non-elongating ones . We developed a method to apply ectopic tension and found that increasing AP tension locally increased the elongation rate of new edges by more than twofold . Increasing dorsal-ventral tension resulted in vertex resolution perpendicular to the AP direction . We propose that local , periodic contractile forces polarize vertex resolution to drive Drosophila axis elongation . Axis elongation is a conserved morphogenetic process is which the basic body plan of an animal is established . In vertebrates , axis elongation involves convergence and extension movements mediated by cell intercalation , cell migration , and oriented cell division ( Bénazéraf and Pourquié , 2013 ) . In Drosophila , axis elongation occurs in an epithelial monolayer referred to as the germband , which lengthens by more than two-fold along the anterior-posterior ( AP ) axis of the animal , while narrowing along the dorsal-ventral ( DV ) axis ( Figure 1—figure supplement 1A ) . The changes in germband architecture are largely driven by cell intercalation ( Irvine and Wieschaus , 1994 ) . Cell intercalation facilitates changes in tissue architecture through neighbour exchange events . In vertebrates , cell intercalation drives many developmental processes , including primitive streak formation in chick embryos ( Voiculescu et al . , 2007 ) ; gut organogenesis ( Chalmers and Slack , 2000 ) , neural tube closure ( Davidson and Keller , 1999 ) , and elongation of kidney tubules ( Lienkamp et al . , 2012 ) in Xenopus; epiboly in Xenopus ( Keller , 1980 ) and zebrafish ( Warga and Kimmel , 1990 ) ; convergence and extension of the mesoderm in Xenopus ( Wilson et al . , 1989; Shih and Keller , 1992 ) , zebrafish ( Yin et al . , 2008 ) , and mouse ( Yen et al . , 2009 ) ; and visceral endoderm migration ( Migeotte et al . , 2010; Trichas et al . , 2012 ) , eye lid closure ( Heller et al . , 2014 ) , neural plate elongation ( Williams et al . , 2014 ) , palate fusion ( Kim et al . , 2015 ) , and limb bud elongation ( Lau et al . , 2015 ) in mouse . During Drosophila axis elongation , cell intercalation is driven by polarized actomyosin contractility , which promotes the disassembly of interfaces separating anterior and posterior cell neighbours ( AP interfaces ) , to form multicellular vertices where four or more cells converge ( Bertet et al . , 2004; Zallen and Wieschaus , 2004; Blankenship et al . , 2006 ) . Polarized disassembly of cell contacts is also associated with cell intercalation in chick ( Rozbicki et al . , 2015 ) , Xenopus ( Shindo and Wallingford , 2014 ) , and mouse embryos ( Williams et al . , 2014; Lau et al . , 2015 ) . Following contraction of AP interfaces in the Drosophila germband , multicellular vertices are systematically resolved through the assembly of new contacts separating dorsal and ventral cell neighbours ( DV interfaces , Figure 1—figure supplement 1B , Video 1 ) . While vertex resolution and the subsequent assembly of new cell-cell interfaces drive tissue elongation , little is known about the mechanisms that regulate these processes . Myosin turnover between phosphorylated and unphosphorylated states is important for the directionality of vertex resolution ( Kasza et al . , 2014 ) . Computational modelling suggests that periodic contraction of the apical surface of germband cells , driven by pulsatile actomyosin networks , could promote the oriented assembly of new cell contacts ( Lan et al . , 2015 ) . However , the role of actomyosin contractility in vertex resolution remains unclear . In this study , we combine quantitative imaging with biophysical and pharmacological manipulations to investigate the mechanisms of vertex resolution in Drosophila axis elongation . We find that the assembly of new interfaces during vertex resolution occurs in pulses associated with the periodic contraction of the cells anterior and posterior to the multicellular vertex . Pulsed actomyosin contractility in the cells around the vertex is critical for the directionality and rate of assembly of the new cell interface . Local , ectopic AP tension is sufficient to accelerate the assembly of new interfaces , and local DV tension can reorient vertex resolution . Together , our results demonstrate that local , periodic actomyosin contractility directs the resolution of multicellular vertices and promotes the assembly of new cell contacts during polarized cell rearrangements in Drosophila germband extension . To investigate the mechanisms of vertex resolution during Drosophila axis elongation , we used quantitative image analysis to measure the dynamics of assembly of new DV junctions in embryos expressing Resille:GFP ( Morin et al . , 2001 ) to visualize cell outlines . We found that the assembly of new DV edges occurred in cycles of elongation and shortening ( Figure 1A–B , blue line ) , with a period of 126 ± 5 s ( n = 110 edges ) . On average , elongation pulses increased edge length by 772 ± 46 nm , while shortening pulses decreased edge length by a significantly smaller amount , 114 ± 19 nm ( n = 110 edges , p = 9 . 0 × 10−22 ) , thus resulting in net edge elongation . Germband cells undergo characteristic cycles of apical area contraction and relaxation with a period of 130 ± 3 s , and predominantly oriented along the AP axis of the embryo ( Fernandez-Gonzalez and Zallen , 2011; Sawyer et al . , 2011 ) . To examine whether the anisotropic oscillations of germband cells were associated with the assembly of new cell junctions during vertex resolution , we compared the changes in length of the nascent DV edge to the changes in apical area of the cells immediately anterior or posterior to that DV edge ( Figure 1A–B ) . In a majority of cases ( 143/220 cell-edge pairs , 65% ) , we observed a negative correlation between changes in length of the new DV junction and changes in area of the cell anterior or posterior to it ( Figure 1C ) . To calculate the dominant relationship between changes in anterior/posterior cell area and new DV edge length , we quantified the correlations after shifting the edge length backward or forward in time . Reaching the maximum correlation with small time shifts would indicate in-phase oscillations , while maximum anti-correlation with small time shifts would suggest oscillations in anti-phase . We found that short time shifts of the edge length signal maximized the anti-correlation , while longer time shifts were necessary to maximize the correlation ( p = 1 . 74 × 10−5 , Figure 1D–E ) , further suggesting that pulses of new DV edge assembly are associated with the contraction of the anterior and posterior cells . Similar analyses demonstrated that changes in length of the new edge were predominantly positively correlated with changes in the apical area of the dorsal and ventral cells , which share the new edge ( 156/220 cell-edge pairs , 71% , Figure 1—figure supplement 2 ) . Together , our results suggest that pulsed contractions of the cells in the immediate vicinity of a multicellular vertex may promote vertex resolution during Drosophila axis elongation . 10 . 7554/eLife . 10757 . 003Video 1 . Polarized cell rearrangements drive Drosophila axis elongation . Germband cells expressing Resille:GFP during germband extension . A stack was acquired every 10 s . Time is indicated as min:s . Anterior left , dorsal up . This video relates to Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00310 . 7554/eLife . 10757 . 004Figure 1 . Directional assembly of new interfaces during vertex resolution is associated with pulsatile apical contractions and requires contractile activity . ( A ) Vertex resolution during axis elongation in an embryo expressing Resille:GFP . Blue indicates the new DV interface , red labels the anterior and posterior cells . ( A’ ) Kymograph illustrating the elongation of the DV interface shown in ( A ) . Scale bar , 10 s . The interface is rotated by 90° with respect to ( A ) . Anterior down , dorsal left . ( B ) Rates of change for edge length ( blue , solid line ) , anterior cell area ( red , dashed line ) , and posterior cell area ( red , dotted line ) during the neighbour exchange event shown in ( A ) . Rate of change was calculated with respect to t + 60 s . ( C ) Correlation coefficients between changes in edge length and changes in anterior or posterior cell area ( n = 220 pairs in 110 neighbour exchange events in 13 embryos ) . ( D ) Changes in correlation between edge length and anterior ( dashed ) or posterior ( dotted ) cell area during the neighbour exchange event shown in ( A ) when the edge length signal was shifted in time in 10-s increments . Arrowheads indicate the correlation minima ( blue ) or maxima ( red ) closest to 0-s shift . ( E ) Distribution of time shifts ( absolute value ) required to obtain the minimum ( blue ) and maximum ( red ) correlations in all 220 signal pairs shown in ( C ) . ( F , G ) Rate of change in cell area in embryos injected with water ( F , n = 122 cells in 3 embryos ) or 100 mM Y-27632 ( G , n = 99 cells in 3 embryos ) . Each line represents a single cell . ( H ) Oscillation amplitude for changes in cell area in embryos injected with water ( blue ) or 100 mM Y-27632 ( red ) . Asterisks indicate p < 0 . 001 . ( I–J' ) Vertex resolution during axis elongation in embryos expressing E-cadherin:GFP and injected with water ( I ) or with 100 mM Y-27632 ( J , J’ ) . Arrowheads indicate nascent DV interfaces . ( K ) Distribution of vertex resolution angles relative to the AP axis in embryos injected with water ( blue , n = 28 vertices in 3 embryos ) or 100 mM Y-27632 ( red , n = 25 interfaces in 3 embryos ) . Angles were measured 150 s after the onset of vertex resolution . An angle of 90° with respect to the AP axis corresponds to the DV axis . ( L ) Length of new DV interfaces forming within 30° of the AP axis in embryos injected with water ( blue , n = 25 interfaces in 3 embryos ) or 100 mM Y-27632 ( red , n = 11 interfaces in 3 embryos ) . ( A , I–J’ ) Anterior left , dorsal up . Scale bars , 5 µm . ( B , F , G , L ) Time is with respect to the onset of vertex resolution , defined as the first time point in which the length of the nascent interface exceeded 1 µm . ( H , K , L ) Error bars , s . e . m . AP , anterior-posterior; DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00410 . 7554/eLife . 10757 . 005Figure 1—figure supplement 1 . Axis elongation in Drosophila is driven by neighbour exchange events . ( A ) Germband position at the beginning ( left ) , during ( centre ) , and at late stages of axis elongation ( right ) . Arrows indicate the direction of cell movement . White arrowheads delimit the germband . Scale bar , 100 µm . ( B ) Diagram ( top ) and germband cells ( bottom ) showing a neighbour exchange event . An AP interface contracts ( left , green ) , forming a vertex where four cells meet ( centre , magenta ) . The vertex resolves through the assembly of a new DV interface ( right , cyan ) . Scale bar , 5 µm . ( A , B ) Anterior left , dorsal up . AP , anterior-posterior; DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00510 . 7554/eLife . 10757 . 006Figure 1—figure supplement 2 . Dorsal and ventral cells oscillate with new DV interfaces . ( A ) Vertex resolution during axis elongation in an embryo expressing Resille:GFP . Blue indicates the new DV interface , red labels the dorsal and ventral cells . Anterior left , dorsal up . Scale bar , 5 µm . ( B ) Rates of change for edge length ( blue , solid line ) , dorsal cell area ( red , dashed line ) , and ventral cell area ( red , dotted line ) during the neighbour exchange event shown in ( A ) . Rate of change was calculated with respect to t + 60 s . ( C ) Correlation coefficients between changes in edge length and changes in dorsal or ventral cell area ( n = 220 pairs in 110 neighbour exchange events in 13 embryos ) . ( D ) Changes in correlation between edge length and dorsal ( dashed ) or ventral ( dotted ) cell area during the neighbour exchange event shown in ( A ) , when the edge length signal was shifted in time in 10-second increments . Arrowheads indicate the correlation minima ( blue ) or maxima ( red ) closest to 0 s shift . ( E ) Distribution of time shifts ( absolute value ) required to obtain the minimum ( blue ) and maximum ( red ) correlations in all 220 signal pairs shown in ( C ) . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00610 . 7554/eLife . 10757 . 007Figure 1—figure supplement 3 . Directional assembly of new DV interfaces during vertex resolution requires actomyosin contractility . ( A–A’’ ) Germband cells expressing E-cadherin:GFP ( green , A’ ) and myosin:mCherry ( magenta , A’’ ) , before ( pre-injection ) and at different times after injection with 100 mM Y-27632 . Anterior left , dorsal up . Scale bars , 5 µm . ( B–C' ) Vertex resolution during axis elongation in embryos expressing Resille:GFP and injected with water ( B ) or with 100 mM Y-27632 ( C , C’ ) . Arrowheads indicate nascent DV interfaces . Anterior left , dorsal up . Scale bars , 5 µm . ( D ) Distribution of vertex resolution angles relative to the AP axis in embryos injected with water ( blue , n = 26 vertices in 3 embryos ) or 100 mM Y-27632 ( red , n = 43 interfaces in 7 embryos ) . Angles were measured 150 s after the onset of vertex resolution . An angle of 90° with respect to the AP axis corresponds to the DV axis . Error bars , s . e . m . ( E ) Length of new DV interfaces forming within 30° of the AP axis in embryos injected with water ( blue , n = 21 interfaces in 3 embryos ) or 100 mM Y-27632 ( red , n = 19 interfaces in 7 embryos ) . Time is with respect to the onset of vertex resolution , defined as the first time point in which the length of the nascent interface exceeded 1 µm . Error bars , s . e . m . AP , anterior-posterior; DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00710 . 7554/eLife . 10757 . 008Figure 1—figure supplement 4 . Par complex localization is affected by Y-27632 , but not by Cytochalasin D . ( A–B' ) Germband cells expressing Par-6:GFP at endogenous levels and injected with water ( A ) , 100 mM Y-27632 in water ( A’ ) , 50% DMSO ( B ) , or 5 mM Cytochalasin D in 50% DMSO ( B’ ) . Anterior left , dorsal up . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 00810 . 7554/eLife . 10757 . 009Figure 1—figure supplement 5 . Oriented assembly of new DV interfaces requires actin-based contraction . ( A , B ) Germband cells expressing GFP:utrophin in embryos injected with DMSO ( A ) or with 5 mM Cytochalasin D ( B ) . Scale bars , 10 µm . ( C , D ) Rate of change in cell area in DMSO ( C ) or Cytochalasin D-injected embryos ( D ) . Each line represents a single cell ( n = 20 cells in 4 embryos in both C and D ) . ( E ) Oscillation amplitude for changes in cell area in embryos injected with DMSO ( blue ) or 5 mM Cytochalasin D ( red ) . Asterisk indicates p < 0 . 05 . Error bars , s . e . m . ( F–H ) Vertex resolution during axis elongation in embryos expressing E-cadherin:GFP ( green , top; grayscale , bottom ) and myosin:mCherry ( magenta , top ) and injected with DMSO ( F ) or with 5 mM Cytochalasin D ( G-H' ) . Arrowheads indicate nascent DV interfaces . ( I ) Distribution of vertex resolution angles relative to the AP axis in embryos injected with DMSO ( blue , n = 50 vertices in 5 embryos ) or 5 mM Cytochalasin D ( red , n = 15 vertices in 4 embryos ) . Angles were measured 150 s after the onset of vertex resolution . An angle of 90° with respect to the AP axis corresponds to the DV axis . ( J ) Length of new DV interfaces forming within 30° of the AP axis in embryos injected with DMSO ( blue , n = 43 interfaces in 5 embryos ) or 5 mM Cytochalasin D ( red , n = 9 interfaces in 4 embryos ) . Time is with respect to the onset of vertex resolution , defined as the first time point in which the length of the nascent interface exceeded 1 µm . ( I , J ) Error bars , s . e . m . ( K ) Non-resolving vertex in an embryo expressing E-cadherin:GFP ( green , top; greyscale , bottom ) and myosin:mCherry ( magenta , top ) , and injected with 5 mM Cytochalasin D . Arrowheads indicate the vertex . ( A , B , F–H' , K ) Anterior left , dorsal up . ( F–H' , K , K' ) Scale bars , 5 µm . AP , anterior-posterior; DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 009 The cyclical changes of apical area in germband cells are driven by pulsatile networks of medial-apical non-muscle myosin II ( Rauzi et al . , 2010; Fernandez-Gonzalez and Zallen , 2011; Sawyer et al . , 2011 ) . To investigate if actomyosin contractility is necessary for vertex resolution , we injected embryos expressing E-cadherin:GFP and myosin:mCherry with the Rho-kinase inhibitor Y-27632 at 100 mM . Rho-kinase is one of the main activators of myosin ( Amano et al . , 1996; Kimura et al . , 1996 ) , and treatment with Y-27632 abolishes the ability of germband cells to generate mechanical force ( Fernandez-Gonzalez et al . , 2009 ) . In Y-27632-injected embryos , germband cells displayed a rapid loss of myosin from their apical surface ( Figure 1—figure supplement 3A ) , resulting in a dramatic reduction in the amplitude of apical area oscillation ( p = 1 . 7 × 10−44 , Figure 1F–H ) . Inhibiting actomyosin contractility affected the directionality of vertex resolution: 9/25 vertices resolved within 30° of the DV axis in Y-27632-injected embryos , in contrast to 0/28 in water-injected controls ( p = 0 . 02 , Figure 1I , J’ , K , Video 2 ) . In addition , for vertices that resolved along the AP axis , inhibiting Rho-kinase reduced the rate of new edge elongation with respect to controls ( 0 . 001 ± 0 . 080 μm/min vs . 0 . 28 ± 0 . 06 μm/min , respectively , p = 0 . 01 , Figure 1I , J , L , Video 2 ) , suggesting that myosin activity facilitates the assembly of new DV interfaces . Similar results were obtained in embryos expressing Resille:GFP , a different cell outline marker ( Figure 1—figure supplement 3B–E ) . However , Rho-kinase activity can regulate the localization of the Par polarity complex ( Atwood and Prehoda , 2009; Simões et al . , 2010 ) ( Figure 1—figure supplement 4A ) , raising the possibility that abnormal vertex resolution upon Y-27632 injection was a consequence of defects in cell polarity , rather than reduced actomyosin contractility . 10 . 7554/eLife . 10757 . 010Video 2 . Actomyosin contractility is required for directional vertex resolution . Germband cells expressing E-cadherin:GFP in embryos injected with water ( left ) or 100 mM Y-27632 ( centre and right ) . A stack was acquired every 10 s . Time is indicated as min:s . Anterior left , dorsal up . This video relates to Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 010 To further investigate the role of mechanical forces in vertex resolution , we disrupted actomyosin contractility by injecting embryos with 5 mM of Cytochalasin D , a drug that blocks actin polymerization by binding to the elongating end of filaments and preventing the addition of new actin monomers ( Flanagan and Lin , 1980 ) . Cytochalasin D injection disrupted the actin cytoskeleton ( Figure 1—figure supplement 5A–B ) and reduced apical area oscillations ( p = 0 . 04 , Figure 1—figure supplement 5C–E ) , without affecting the localization of Par-6 , a member of the Par complex ( Figure 1—figure supplement 4B ) . Cytochalasin D treatment led to an 83% reduction in the rate of new DV edge assembly with respect to controls ( 0 . 07 ± 0 . 10 μm/min vs . 0 . 40 ± 0 . 05 μm/min , respectively , p = 0 . 01 , Figure 1—figure supplement 5F–G , J , Video 3 ) . In Cytochalasin D-injected embryos 4/15 vertices resolved along the DV axis , in contrast to 0/50 in DMSO-injected controls ( p = 4 . 0 × 10−15 , Figure 1—figure supplement 5F , H–I , Video 3 ) . Strikingly , in Cytochalasin D-injected embryos , 32/47 vertices persisted for at least 10 min and up to 40 min ( Figure 1—figure supplement 5K ) . Together , our results demonstrate that actomyosin contractility is necessary for the directional assembly of new interfaces during vertex resolution in Drosophila axis elongation . 10 . 7554/eLife . 10757 . 011Video 3 . Stabilization of actin filaments impairs directional vertex resolution . Germband cells expressing E-cadherin:GFP in embryos injected with 50% DMSO ( left ) or 5 mM Cytochalasin D ( centre and right ) . A stack was acquired every 10 s . Time is indicated as min:s . Anterior left , dorsal up . This video relates to Figure 1—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 011 If actomyosin contractility in the cells anterior and posterior to a resolving vertex drives directional interface assembly , then the nascent edge must be under tension . To quantify tension , we used an ultraviolet ( UV ) laser to locally irradiate and sever DV interfaces in embryos expressing E-cadherin:GFP , and particle-tracking velocimetry to quantify the change in position of the tricellular vertices once connected by the severed interface . The instantaneous retraction velocity of the vertices is proportional to the tension sustained by the interface prior to ablation ( Hutson et al . , 2003; Fernandez-Gonzalez et al . , 2009 ) . We compared retraction velocities after ablation of control DV junctions that were not actively elongating ( average length of 7 . 3 ± 0 . 3 μm , Figure 2A ) and newly forming DV edges ( average length of 3 . 4 ± 0 . 2 μm , Figure 2B ) . The retraction velocity after ablation of new DV junctions was 0 . 81 ± 0 . 08 μm/s , 32% greater than the retraction velocity after severing control DV edges ( 0 . 61 ± 0 . 03 μm/s , p = 0 . 05 , Figure 2C ) , indicating that – assuming uniform viscoelastic properties – new DV edges sustain increased mechanical tension with respect to non-elongating edges with similar orientation . New DV interfaces displayed smaller angles between the anterior or posterior cell junctions ( θavg = 136 . 6 ± 3 . 3° ) than control DV interfaces , ( θavg = 150 . 3 ± 3 . 4° , p = 0 . 02 , Figure 2A–B , D ) , and the retraction velocity after laser ablation was significantly anti-correlated with the angle between the anterior or posterior cell junctions ( r = −0 . 6 , p = 2 . 9 × 10−5 ) . Notably , no correlation was found between control or new DV interface length and instantaneous retraction velocity after ablation ( r = 0 . 04 and 0 . 35 , respectively , Figure 2E , F and Figure 2—figure supplement 1 ) , suggesting that differences in retraction velocity between control and new DV edges are independent from interface length , and determined by whether the edge is being assembled . Vertex retraction after laser ablation could result from actomyosin contractility at the interface or at another structure ( for example , another interface or a medial apical surface ) connected to the severed interface . New DV edges were myosin-depleted ( Blankenship et al . , 2006 ) ( p = 4 . 3 × 10−5 , Figure 2—figure supplement 2 ) , suggesting that vertex retraction after ablation of new DV edges was caused by tension generated elsewhere and exerted onto the new edge . Together , our data strongly suggest that mechanical tension parallel to the AP axis of the embryo contributes to vertex resolution . 10 . 7554/eLife . 10757 . 012Figure 2 . Resolving edges sustain increased mechanical tension during axis elongation . ( A , B ) Germband cells expressing E-cadherin:GFP before and after ablation of a control DV edge ( A ) or a newly forming DV edge ( B ) . White arrowheads point to the ablated interface . θ1 and θ2 indicate the angles between the junctions anterior and posterior to the ablated interface , respectively . Anterior left , dorsal up . Scale bars , 5 µm . ( A’ , B’ ) Kymographs showing the vertex displacement caused by laser ablation of the edges shown in ( A , B ) . Arrowheads indicate vertex position prior to ablation ( green ) or immediately after ( yellow ) . Interfaces are rotated by 90° with respect to ( A , B ) Anterior down , dorsal left . Scale bar , 3 s . ( C ) Retraction velocity after laser ablation in control ( blue , n = 28 ) and new ( red , n = 12 ) DV interfaces . Asterisk indicates p = 0 . 05 . Error bars , s . e . m . ( D ) Scatterplot showing interface length vs . average junction angle at the anterior and posterior ends ( θavg = ( θ1 +θ2 ) /2 ) . ( E , F ) Scatterplots showing interface length vs . retraction velocity after laser ablation for control ( E ) and new ( F ) DV interfaces . Solid lines are best-fit lines . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 01210 . 7554/eLife . 10757 . 013Figure 2—figure supplement 1 . The retraction velocity after ablation of new and control DV edges is not anti-correlated with their length . ( A–B' ) Kymographs showing the vertex displacements caused by laser ablation of relatively short ( A , B ) and long ( A’ , B’ ) control ( A ) or new ( B ) DV edges . Arrowheads indicate vertex position prior ( green ) or immediately after ( yellow ) ablation . Anterior down , dorsal left . Scale bars , 3 s . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 01310 . 7554/eLife . 10757 . 014Figure 2—figure supplement 2 . New DV edges do not display a significant myosin accumulation . ( A , B ) AP- ( A ) and newly forming DV- ( B ) oriented interfaces in embryos expressing E-cadherin:GFP ( green ) and myosin:mCherry ( magenta ) . Arrowheads indicate the interfaces . Scale bars , 2 μm . Anterior left , dorsal up . ( C ) Myosin:mCherry fluorescence in AP and newly forming DV interfaces . Asterisks indicate p < 0 . 001 . Error bars , s . e . m . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 014 To further investigate the relative contribution of anterior/posterior and dorsal/ventral cells to new DV junction assembly during vertex resolution , we disrupted actomyosin contractility specifically in the anterior and posterior , or the dorsal and ventral cells . To this end , we used a UV laser to irradiate and destroy myosin networks in the cells anterior/posterior or dorsal/ventral to four-cell vertices . Cells expressed E-cadherin:GFP to visualize cell outlines , and myosin:mCherry to track the assembly of contractile networks . Cells were re-irradiated upon assembly of medial actomyosin networks to prevent the generation of contractile forces . Irradiated cells were not extruded in the course of these experiments . Controls were four-cell vertices in which the anterior/posterior or dorsal/ventral cell pairs were sham-irradiated with the UV laser fully attenuated using a neutral density filter . When the contractile activity of anterior/posterior cells was disrupted , 4/7 four-cell vertices did not resolve ( their length was never greater than 1 μm for at least 1 min ) , in contrast to 0/10 vertices in sham-irradiated controls . In controls , the rate of new edge elongation calculated over 180 s was 0 . 47 ± 0 . 08 μm/min ( Figure 3A , C ) . Preventing contraction of the anterior/posterior cells resulted in a significant reduction of the rate of new edge elongation to 0 . 18 ± 0 . 05 μm/min for the vertices that resolved ( p = 0 . 03; Figure 3B–C ) . These results suggest that contractility in the cells anterior and posterior to a multicellular vertex is necessary for vertex resolution and the assembly of the new DV interface . 10 . 7554/eLife . 10757 . 015Figure 3 . Local actomyosin contractility is necessary for vertex resolution and new DV interface assembly . ( A , B , D , E ) Cells expressing E-cadherin:GFP ( green ) and myosin:mCherry ( magenta ) in sham-irradiated controls ( A , D ) or when UV irradiation was used to reduce local tension ( B , E ) . White arrowheads indicate resolving interfaces . Asterisks show the targeted cells . Time is with respect to the first laser irradiation . Anterior left , dorsal up . Scale bars , 5 µm . ( C , F ) Length of resolving DV interfaces over time in controls ( blue , n = 10 interfaces in C and F ) , under reduced AP tension ( red , n = 7 interfaces in C ) , or under reduced DV tension ( red , n = 7 interfaces in F ) . Discontinuities in the blue lines indicate times at which cells were targeted with the attenuated UV laser in all experiments . Error bars , s . e . m . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 015 To investigate the role of dorsal/ventral cells in vertex resolution , we prevented assembly and contraction of medial actomyosin networks in the dorsal and ventral cells using laser ablation . In contrast with the ablation of anterior/posterior cells , ablation of the DV cells did not prevent vertex resolution: 5/7 new DV interfaces reached a length of at least 1 μm , similar to 10/10 in controls . The initial rates of elongation were similar , with new DV interfaces elongating at a rate of 0 . 37 ± 0 . 11 µm/min over 60 s when contraction of the DV cells was disrupted , compared to rates of 0 . 50 ± 0 . 15 µm/min in sham-irradiated controls ( p = 0 . 48 , Figure 3D–F ) . However , ablation of the dorsal and ventral cells resulted in a significant reduction of the rate of new interface elongation over the subsequent 120 s of elongation , from 0 . 52 ± 0 . 11 μm/min in controls to -0 . 03 ± 0 . 08 μm/min ( p = 0 . 01 , Figure 3D–F ) . Notably , in 3/5 vertices that resolved when dorsal/ventral cells were ablated , new DV edges formed but were not sustained beyond 1 min , collapsing back into vertices . Together , our data suggest that dorsal/ventral cells are necessary to sustain the elongation of new DV interfaces , but not the resolution of multicellular vertices . To determine if mechanical tension from the anterior and posterior cells can promote the elongation of new DV interfaces during germband extension , we developed a method to apply ectopic local tension to resolving vertices based on wound healing ( Campinho et al . , 2013; Fernandez-Gonzalez and Zallen , 2013 ) . Upon wounding by irradiation with a UV laser , germband cells undergo apical constriction driven by medial-apical actomyosin networks ( Figure 4—figure supplement 1A ) . Apical constriction of germband cells generates ectopic tension on the surrounding cell interfaces ( Figure 4—figure supplement 1A , arrowheads ) . We used a UV laser to wound the cells anterior and posterior to resolving vertices by irradiating their medial-apical surfaces ( Figure 4A–B and Figure 4—figure supplement 1B , and Video 4 ) . Under sham-irradiation ( UV laser fully attenuated using a neutral density filter ) , the cell area and medial myosin of the anterior and posterior cells remained largely unaffected , and the new DV interface elongated at a rate of 0 . 79 ± 0 . 14 μm/min ( Figure 4A , C , E ) . Conversely , when the cells anterior and posterior were irradiated with UV light , myosin accumulated on the apical surface of the wounded cells and their apical areas decreased rapidly ( Figure 4B , D ) , resulting in ectopic , AP-oriented tension on the resolving vertex . Under ectopic tension parallel to the AP axis , new DV junctions elongated at a rate of 1 . 73 ± 0 . 30 μm/min , 2 . 1-fold faster than the elongation rate in controls ( p = 6 . 2 × 10−3 , Figure 4E ) . These results indicate that local mechanical tension parallel to the AP axis is sufficient to promote rapid assembly of new DV interfaces during vertex resolution in germband extension . 10 . 7554/eLife . 10757 . 016Figure 4 . Local mechanical tension is sufficient to promote and orient new interface assembly during vertex resolution . ( A–B' ) Cells expressing E-cadherin:GFP ( green ) and myosin:mCherry ( magenta ) in sham ( A ) or UV-irradiated ( B ) embryos . ( C , D ) Medial myosin intensity ( magenta ) and cell area ( green ) in sham ( C , n = 22 cells in 11 embryos ) and UV-irradiated embryos ( D , n = 16 cells in 8 embryos ) . ( E ) Length of resolving DV interfaces over time in controls ( blue , n = 11 interfaces ) and under increased tension along the AP axis ( red , n = 8 interfaces ) . ( F , G ) Cells expressing E-cadherin:GFP in sham ( F ) or UV-irradiated ( G ) embryos . Asterisks show the cells around a four-cell vertex ( white arrowheads ) that were irradiated . Yellow arrowheads indicate the formation of a four-cell vertex . ( A , B , F , G ) Anterior left , dorsal up . Scale bars , 5 µm . ( H ) Length of resolving interfaces over time in controls ( blue , n = 12 ) and under increased tension along the DV axis ( red , n = 13 ) . Turquoise indicates elongation parallel to the AP axis , pink denotes DV elongation . ( C–E , H ) Time is with respect to the time point when the nascent DV interface first exceeded 1 µm in length . Error bars , s . e . m . ( C , D ) Normalization is with respect to the value at 0 s . AP , anterior-posterior; DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 01610 . 7554/eLife . 10757 . 017Figure 4—figure supplement 1 . Wounded cells undergo apical constriction and induce ectopic tension on adjacent cell-cell junctions . ( A–A’’ ) Germband cells expressing E-cadherin:GFP ( green , A’ ) and myosin:mCherry ( magenta , A’’ ) , before ( pre-ablation ) and at different times after UV-irradiation of the cell denoted by the yellow asterisk . White arrowheads indicate neighbouring junctions under ectopic strain when the wounded cell constricts apically . Time after wounding is shown . Anterior left , dorsal up . Scale bars , 5 μm . ( B ) Schematic representation of a method to induce ectopic AP-oriented tension ( red arrows ) on a vertex by wounding ( yellow rays ) the neighbouring anterior and posterior cells . AP , anterior-posterior . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 01710 . 7554/eLife . 10757 . 018Video 4 . Mechanical tension promotes rapid elongation of new DV interfaces . Germband cells in embryos expressing E-cadherin:GFP ( green ) and myosin:mCherry ( magenta ) under sham irradiation ( left ) or upon wounding and apical constriction of the cells anterior and posterior to a multicellular vertex ( right ) . Arrows indicate resolving multicellular vertices . A stack was acquired every 3 s . Time is indicated as min:s . Anterior left , dorsal up . This video relates to Figure 4 . DV , dorsal-ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 10757 . 018 Our findings that ectopic tension can increase the rate of new edge elongation suggest that tension parallel to the DV axis may change the direction of vertex resolution . We compared the orientation and rate of new edge elongation in sham-irradiated embryos ( Figure 4F ) and in embryos in which we induced apical constriction of the cells dorsal and ventral to a four-cell vertex , increasing tension along the DV axis ( Figure 4G ) . All the four-cell vertices examined in control embryos ( n = 12 ) resolved within 30° of the AP axis , and the rate of new interface assembly was 0 . 69 ± 0 . 13 μm/min ( Figure 4F , H ) . When we applied ectopic tension along the DV axis , the rate of new edge elongation was not affected ( 0 . 74 ± 0 . 18 μm/min , p = 0 . 8 ) , but the orientation of the new edge changed and occurred within 30° of the DV axis in 13 out of 13 cases ( Figure 4G , H ) . Together , our data indicate that local tension can promote and orient the assembly of new cell-cell interfaces , suggesting a central role for mechanical forces during vertex resolution in Drosophila axis elongation . Polarized junction remodelling drives changes in tissue architecture from worms to mice ( Walck-Shannon and Hardin , 2014 ) . While junctional contraction and disassembly in the context of cell intercalation have been extensively explored ( Bertet et al . , 2004; Blankenship et al . , 2006; Rauzi et al . , 2008; Fernandez-Gonzalez et al . , 2009; Levayer et al . , 2011; Bosveld et al . , 2012; Shindo and Wallingford , 2014; Lau et al . , 2015 ) , little is known about the mechanisms that control the directional assembly of new cell contacts during neighbour exchange . We used quantitative imaging , and biophysical and pharmacological approaches to show that local mechanical forces can direct the assembly of new junctions during Drosophila germband extension . New junctions elongate in pulses anti-correlated with the periodic contractions of the cells anterior and posterior to the new contact . Inhibiting actomyosin contractility disrupts both the rate and directionality of new junction assembly . Disrupting contractility in the cells anterior and posterior to the new edge disrupts vertex resolution and slows down new edge elongation , while preventing contraction of the dorsal and ventral cells mainly affects the maintenance and lengthening of the new cell interface . Hypercontraction of the cells anterior and posterior to the new edge accelerates the rate of new edge assembly . Finally , applying ectopic tension orthogonal to the characteristic orientation of vertex resolution is sufficient to alter the direction of new edge formation , suggesting that mechanical forces associated with actomyosin contractility direct the assembly of new cell contacts during multicellular vertex resolution in germband extension . We show that vertex resolution occurs under increased mechanical tension , in a process that requires actomyosin contractility . Consistent with this , expression of inactive or constitutively active forms of myosin in embryos lacking the wild-type motor protein disrupts the directionality of vertex resolution during germband extension ( Kasza et al . , 2014 ) . Furthermore , mechanical tension is necessary for directional resolution of multicellular vertices in the mouse embryonic ectoderm during limb bud elongation ( Lau et al . , 2015 ) . In the Drosophila dorsal thorax , whose architecture is determined by neighbour exchange events , actomyosin contractility in new edges is tightly regulated to facilitate their elongation ( Bardet et al . , 2013 ) . Our data suggest that the increase in tension on the new contact may be caused locally by the pulsatile , anisotropic contraction of the cells around the resolving vertex . Interestingly , cells in the mouse limb bud ectoderm also display pulsed contractions that are disrupted in β-catenin mutants , and in these mutants the directionality of vertex resolution is lost ( Lau et al . , 2015 ) . Together , these results are consistent with a general role for pulsed contractile activity in orienting and promoting cell intercalation . We find that anterior/posterior and dorsal/ventral cells may play different roles during multicellular vertex resolution . Our data suggest that the anterior and posterior cells contribute to both vertex resolution and new edge elongation , while the dorsal and ventral cells are mainly necessary to support the elongation of the edge once the vertex has resolved . Recent mathematical modelling predicts that periodic actomyosin contractility in the medial-apical surface of anterior and posterior cells could drive the assembly of new edges during germband extension ( Lan et al . , 2015 ) . The pulsed contraction of the anterior and posterior cells could cause rapid membrane reorganization in the dorsal and ventral cells ( Pramanik et al . , 2009 ) , facilitating the assembly of an actin scaffold ( Pickering et al . , 2013 ) and the formation of junctions . Junctional and cytoskeletal remodelling require intact DV cells , and possibly , the continued stimulus from AP cell pulsing . The implementation of optogenetic approaches ( Guglielmi et al . , 2015 ) to locally inhibit membrane remodelling and junctional and cytoskeletal dynamics will reveal how these processes are coordinated across cells to promote directional cell rearrangements during epithelial morphogenesis . The mechanisms by which mechanical tension regulates the assembly of new cell interfaces during germband extension remain unclear . An accumulation of filamentous actin is the first known step of vertex resolution ( Blankenship et al . , 2006 ) , and in this study , we found that blocking actin polymerization results in multicellular vertices that do not resolve . Thus , actin polymerization may play a central role in vertex resolution . Mechanical forces can control actin dynamics in vitro , possibly by inducing conformational changes in the formin family of actin regulators to favour faster and more frequent polymerization of actin filaments ( Courtemanche et al . , 2013; Higashida et al . , 2013; Jegou et al . , 2013 ) . In addition , actin filaments are less susceptible to severing in the presence of increased tension ( Hayakawa et al . , 2011 ) , which may accelerate actin assembly at nascent cell interfaces . Understanding how mechanical forces impact the localization and dynamics of different actin regulators will contribute to elucidating the mechanisms by which tension promotes directional cell behaviours during Drosophila axis elongation . We used the following markers for live imaging: ubi-E-cadherin:GFP ( Oda and Tsukita , 2001 ) , sqh-sqh:mCherry ( Martin et al . , 2009 ) , resille:GFP ( Morin et al . , 2001 ) , sqh-GFP:utrophin ( Rauzi et al . , 2010 ) , and par-6Δ226 , par-6:GFP ( Wirtz-Peitz et al . , 2008 ) . Stage-7 embryos were dechorionated in 50% bleach for 90 s , rinsed , glued ventrolateral side down to a glass coverslip using heptane glue , and mounted in a 1:1 mix of halocarbon oil 27 and 700 ( Sigma-Aldrich , St . Louis , MO ) . Embryos were imaged using a Revolution XD spinning disk confocal microscope equipped with an iXon Ultra 897 camera ( Andor , Belfast , UK ) and a 1 . 5x coupling lens . For experiments using laser ablation , a 60x oil immersion lens ( Olympus , Shinjuku , Japan; NA 1 . 35 ) was used; for all other experiments , a 40x oil immersion lens ( Olympus , NA 1 . 35 ) was used . Sixteen-bit Z-stacks were acquired at 0 . 3-µm steps every 3–10 s ( 8–10 slices per stack ) . Ablations were induced using a pulsed Micropoint N2 laser ( Andor ) tuned to 365 nm . The laser delivers 120 µJ pulses at durations of 2–6 ns each . For ablation of cell boundaries , 10 consecutive laser pulses were delivered to a single spot along a cell interface . For single-cell wounds , 10 consecutive laser pulses were delivered to each of two spots spaced 2 µm apart on the medial-apical region of the cell of interest . In experiments where local tension was reduced , 10 laser pulses were delivered to a single spot on the medial-apical region of the cell of interest . Cells were re-ablated upon assembly of medial-apical myosin networks . In sham-irradiated controls , cells were targeted with the laser completely attenuated every 60 s to mimic the repeated ablations performed in the corresponding experiments . Embryos were dechorionated and glued to a coverslip as above , dehydrated for 10–15 min , and covered with a 1:1 mix of halocarbon oil 27 and 700 ( Sigma-Aldrich ) . Embryos were injected using a Transferman NK2 micromanipulator ( Eppendorf , Hamburg , Germany ) , and a PV820 microinjector ( WPI , Sarasota , FL ) attached to the spinning disk confocal microscope . Drugs ( Y-27632 , Tocris Bioscience , Bristol , UK ) ; ( Cytochalasin D , EMD Millipore , Darmstadt , Germany ) were injected into the perivitelline space , where they are predicted to be diluted 50-fold ( Foe and Alberts , 1983 ) . Y-27632 was injected at 100 mM in water; control embryos were injected with water . Cytochalasin D was injected at 5 mM in 50% DMSO; control embryos were injected with 50% DMSO . Embryos were imaged immediately after injection for at least 10 min . Image analysis was performed using algorithms developed with Matlab ( MathWorks , Natick , MA ) and DIPImage ( Delft University of Technology , Delft , Netherlands ) and integrated in our custom Scientific Image Segmentation and Analysis ( SIESTA ) software ( Fernandez-Gonzalez and Zallen , 2011; Leung and Fernandez-Gonzalez , 2015 ) . The onset of vertex resolution was established as the first time at which the length of a nascent interface exceeded 1 μm . New edge orientation was quantified relative to the AP axis of the embryo , defined as 0° , and was measured 150 s after the onset of vertex resolution . Edge length was measured as the distance between the two vertices defining the edge . To measure how fast new edges assemble , we defined the rate of elongation at time t as: ( 1 ) rate of elongation ( t ) =l ( t ) -l ( t0 ) t-t0 where l ( t ) represents the length of the edge at time t , and t0 is the time of onset of vertex resolution . The rate of elongation was calculated over the initial 90 s of interface elongation , unless indicated otherwise . Cell areas were quantified using an algorithm in which seeds were manually placed within each cell of interest in the first timepoint of a movie . Seeds were automatically expanded to delineate the cell boundaries using the watershed method ( Beucher , 1992 ) , a region-growing algorithm . Seeds were subsequently propagated to the next time point using particle image velocimetry to account for cellular movement ( Wang and Fernandez-Gonzalez , in preparation ) , and the process was iterated . To measure retraction velocities following laser ablation , we determined the change in distance between the two vertices delimiting the ablated interface , and divided this value by the sum of the ablation and the stack acquisition times . In time-lapse images , fluorescence was measured from maximum intensity projections of three apical slices . Fluorescence intensities were background-subtracted using the most frequent pixel value ( the mode ) of a maximum intensity projection of three basal slices cropped around the region of interest ( 10 μm × 10 μm ) . Intensity values were corrected for photobleaching by dividing by the mean image intensity in each time point . To quantify myosin levels in new DV edges with respect to AP edges , we imaged embryos expressing myosin:mCherry , and measured fluorescence in manually traced cell interfaces . We subtracted the image mode from the myosin fluorescence measurements as an estimate of the background . Oscillatory cell behaviours were characterized by the rate of change per minute of the corresponding magnitude , calculated as the difference of measurements collected 1 min apart . To calculate periods , rates of change were detrended by subtracting the line of best fit using the detrend function in Matlab ( Mathworks ) . The period was computed as the inverse of the dominant frequency in a fast Fourier transform of the detrended signal . To calculate the mean change in edge length during the elongation or shortening steps of new DV edge formation , we quantified the area under the curve for positive ( elongation ) or negative ( shortening ) rates of length change . The resulting numbers were the total elongation or shortening for a given edge , which divided by the number of pulses yielded the mean change in length per elongation or shortening pulse . The correlation between signal pairs was determined using the corrcoef function in Matlab ( Mathworks ) . To find the time shift required for minimum or maximum correlation between signal pairs , one signal was shifted forward and backward in time relative to the other , in increments of 10 s up to 240 s . With each increment , the correlation was recalculated . The resulting correlation curve was Gaussian-smoothed using a sigma of 10 s , and the time shifts required to obtain the first local minimum and maximum in the correlation values were determined . Sample means were compared using Student’s t-test ( Glantz , 2002 ) . The significance of correlation coefficients was calculated by transforming the correlation value into a t-statistic using the Matlab corrcoef function ( Mathworks ) . Sample distributions were contrasted using Kolmogorov–Smirnov’s test . Error bars indicate the standard error of the mean ( s . e . m . ) .	Tissues and organs form certain shapes that allow them to perform particular roles in the body . For example , the lungs form sacs that accommodate large volumes of air , while the skin forms a sheet to cover and protect our internal organs . One way to shape a tissue is for cells to swap places with their neighbours . During this rearrangement , the contacts between neighbouring cells break down before new contacts are formed with other cells . While the physical and molecular signals that guide the break down of cell contacts are well understood , less is known about how new contacts form . Early in development , animal embryos establish a head-to-tail 'axis' that helps to guide where each tissue and organ will form in the body . In fruit fly embryos , the cell rearrangements that drive this process involve cells exchanging places with their neighbours by gathering around a single point . These temporary cell clusters are then organised via new cell contacts that form parallel to the head-to-tail axis . Here , Yu and Fernandez-Gonzalez investigate the role of mechanical forces in forming new cell contacts as the head-tail axis elongates . The experiments show that disrupting the ability of the cells to generate mechanical forces inhibited the formation of new cell contacts and prevented cells from successfully swapping places . Conversely , when mechanical tension is applied at the rearrangement site , the assembly of new cell contacts happens faster . Furthermore , if the tension is applied in different orientations , new cell contacts form parallel to the direction of the mechanical force . Yu and Fernandez-Gonzalez thus show that local mechanical forces direct the assembly of new cell contacts as the head-to-tail axis forms . These forces are most likely generated by cell contractions that appear to create mechanical tension at sites of cell rearrangement . How such physical forces are converted into molecular signals remains a question for future work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Local mechanical forces promote polarized junctional assembly and axis elongation in Drosophila
To understand how long-range patterning gradients are interpreted at the cellular level , we investigate how a gradient of expression of the Four-jointed kinase specifies planar polarised distributions of the cadherins Fat and Dachsous in the Drosophila wing . We use computational modelling to test different scenarios for how Four-jointed might act and test the model predictions by employing fluorescence recovery after photobleaching as an in vivo assay to measure the influence of Four-jointed on Fat-Dachsous binding . We demonstrate that in vivo , Four-jointed acts both on Fat to promote its binding to Dachsous and on Dachsous to inhibit its binding to Fat , with a bias towards a stronger effect on Fat . Overall , we show that opposing gradients of Fat and Dachsous phosphorylation are sufficient to explain the observed pattern of Fat–Dachsous binding and planar polarisation across the wing , and thus demonstrate the mechanism by which a long-range gradient is interpreted . Planar polarity ( also known as planar cell polarity [PCP] ) refers to the coordinated polarisation of cells within the plane of a tissue such as an epithelium . How epithelia are planar polarised and how planar polarisation is co-ordinated across a tissue has intrigued researchers for decades . In the late 1950's , Locke presented evidence that orienting gradients could control epithelial tissue patterning ( Locke , 1959 ) , but despite years of research , the mechanisms by which graded cues might mediate the coordinated polarisation of individual cells remain incompletely understood ( for review see Strutt , 2009 ) . Of the known molecular systems regulating planar polarity , only for the Drosophila Fat-Dachsous-Four-jointed ( Ft-Ds-Fj ) pathway is there strong evidence for a primary role of graded activity in providing orienting cues ( Zeidler et al . , 1999 , 2000; Casal et al . , 2002; Strutt and Strutt , 2002; Yang et al . , 2002; Ma et al . , 2003; Matakatsu and Blair , 2004; Simon , 2004; Ambegaonkar et al . , 2012; Brittle et al . , 2012 ) . Ft and Ds are large atypical cadherins known to bind to each other heterophilically ( Ma et al . , 2003; Matakatsu and Blair , 2004 ) , and Fj is a kinase shown to be active in the Golgi ( Strutt et al . , 2004; Ishikawa et al . , 2008 ) . Complementary graded expression patterns of Ds and Fj ( Zeidler et al . , 1999 , 2000; Yang et al . , 2002; Ma et al . , 2003 ) result in the planar polarisation of Ft and Ds across cells , with ( in the developing wing ) Ds accumulating distally and Ft accumulating proximally ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . How the complementary expression patterns of Fj and Ds result in the accumulation of Ft and Ds on opposing cell edges is still under investigation , however , the data suggest that higher proximal expression of Ds and the opposing gradient of Fj expression leads to a gradient of Ft–Ds dimer formation ( Casal et al . , 2006; Lawrence et al . , 2008; Strutt , 2009 ) . Waddington ( 1943 ) first reported a genetic interaction between fj and ds , and subsequent work revealed that Fj was able to regulate the localisation and function of both Ft and Ds ( Casal et al . , 2002; Strutt and Strutt , 2002; Yang et al . , 2002; Ma et al . , 2003; Cho and Irvine , 2004; Strutt et al . , 2004; Casal et al . , 2006 ) . Fj is able to phosphorylate several cadherin repeats of both Ft and Ds ( Ishikawa et al . , 2008 ) , raising the possibility that this is a mechanism by which Fj regulates Ft–Ds binding . This led to examination of the result of Fj phosphorylation on Ft and Ds binding in vitro ( Brittle et al . , 2010; Simon et al . , 2010 ) . Using a cell aggregation assay , based on Drosophila S2 cells co-transfected with Fj , Ds , or Ft , Brittle et al . ( 2010 ) deduced that Fj was able to act on three previously identified serines in Ds to inhibit the binding of Ds to Ft ( Ishikawa et al . , 2008; Brittle et al . , 2010 ) . Using an alkaline phosphatase-based cell surface binding assay , Simon et al . ( 2010 ) similarly found that Fj inhibited the ability of Ds to bind to Ft and were also able to demonstrate an improvement in Ft binding to Ds . In addition , in vivo assays examining the propagation of polarity from over-expression clones also found that over-expressed Fj promoted Ft activity and inhibited Ds activity ( Brittle et al . , 2010 ) . So far there is no direct in vivo evidence that Fj normally does act on both Ft and Ds to modulate their binding during tissue patterning , or that any such activity is directly dependent on the mapped phosphorylation sites in the cadherin domains . Furthermore , it is unclear how the observed asymmetric subcellular distributions of Ft and Ds are related to the proposed differences in binding affinities between neighbouring cells and across the tissue . To address these issues , we have carried out in vivo studies of Ft and Ds behaviour , making use of both normal protein forms and variants mutated at the mapped phosphorylation sites , and using protein mobility as measured using fluorescence recovery after photobleaching ( FRAP ) as a novel assay for in vivo binding activity . Furthermore , to better understand the possible effects of Fj phosphorylation on Ft–Ds binding and how this might lead to their asymmetric subcellular distributions , we have combined our experimental approach with computational modelling . As an aid to understand the possible consequences for the generation of cellular asymmetry of Fj acting on either Ft or Ds or both , we generated a computational model reflecting a one-dimensional line of cells each with two compartments , to simulate Ft–Ds binding between cells according to a Fj gradient ( see ‘Materials and methods’ ) . In the model , Fj is allowed to phosphorylate either Ft or Ds in proportion to its concentration , and Ft and Ds are then permitted to freely bind at cell edges until they reach equilibrium ( see Figure 1A ) . To produce a representative model , we measured the gradient of Fj expression across Drosophila larval wing discs ( Figure 1C ) , using fj null clones to determine the appropriate background correction ( Figure 1C′′ ) . A gradient of 2 . 5–3 . 5% between adjacent cells ( Figure 1B ) was observed along the proximo-distal axis . We also allowed redistribution of Ft and Ds to occur in the model , in accordance with our experimental observations of protein mobility ( see Figure 2C , D , L ) . A hierarchy of binding affinities was used , based on the in vitro experiments ( Brittle et al . , 2010; Simon et al . , 2010 ) , with phosphorylated Ft ( FtP ) and unphosphorylated Ds ( Ds ) producing the strongest partnership , and unphosphorylated Ft ( Ft ) and phosphorylated Ds ( DsP ) producing the weakest partnership ( see ‘Materials and methods’ ) . 10 . 7554/eLife . 05789 . 003Figure 1 . Computational modelling of the effect of a Fj gradient on Ft–Ds binding . ( A ) Cartoon illustrating Ft–Ds binding and the predicted effect of a Fj gradient on the distribution of Ft–Ds heterodimers in a single cell . In the absence of any gradients ( i . e . , uniform Ft , Ds , and Fj expression across the tissue ) , Ft ( blue ) and Ds ( purple ) freely associate at junctions , with a certain proportion binding to form heterodimers ( indicated by double black bars ) . We predict that by adding a Fj gradient ( yellow bar ) an asymmetric distribution of Ft–Ds binding across a cell will occur . If Fj inhibits Ds binding and promotes Ft binding more strongly to the right , Ft molecules in each cell prefer to bind to Ds in the next left-most cell ( with which they have a stronger binding interaction ) , and so preferentially accumulate at left cell edges; similarly Ds molecules prefer to associate with Ft in the next right-most cell and in turn accumulate at right cell edges . Hence , the overall consequence of a Fj gradient and free movement of Ft and Ds within cells is the generation of cellular asymmetries of Ft and Ds distributions . ( B ) Graph illustrating the proximal-distal measurements of the Fj gradient across Drosophila third instar larval wing discs , showing a typical gradient of around 3% ( dashed lines indicate 95% confidence intervals , error bars indicate standard deviation ( SD ) , n = 5 ) . Cell number increases from 0 to 30 moving from proximal to distal . ( C ) Confocal images of wing discs from animals with EGFP knocked-in to the endogenous ds locus , stained for Fj ( red ) and observed for native GFP fluorescence ( green ) . A yellow line indicates where Fj gradient measurements were taken from proximal [P] to distal [D] . ( C′ ) An example zoomed-in image showing the distribution of Fj and Ds in this region . ( C″ ) fjd1 mutant clones ( absence of blue ß-gal staining , cell junctions also labelled in blue ) demonstrate that Fj is expressed throughout the tissue . Mutant clones were used to subtract background from the gradient measurements . ( D and E ) A computational model of Ft–Ds binding at cellular junctions . Each bar represents one cell and the scale indicates the amount of bound Ds or Ft in arbitrary units at junctions between cells ( note scale is cut off at 20 units ) , also see Figure 1—figure supplement 1 . Sloped top edges represent the difference between proximal and distal cell edges within one cell . A more graded slope within a cell indicates an increase in asymmetry . 23 cells are represented and the observed Fj gradient of 3% ( B ) has been used . Looking at both bound Ds ( D ) and bound Ft ( E ) shows Ds is preferentially localised distally and Ft proximally . If Fj acts on Ds only ( D and E ) , weak asymmetry is seen within cells ( 3 . 8% increase ) and a tissue gradient of higher to lower Ft–Ds binding is seen as Fj increases . If Fj acts on Ft only ( D′ and E′ ) similar weak asymmetry ( 3 . 6% increase ) is seen within cells , however , a tissue gradient of lower to higher Ft–Ds binding is observed as Fj increases . If Fj can act on both Ft and Ds ( D″ and E″ ) stronger asymmetry is seen within cells ( 7 . 9% increase ) and the tissue gradient is much reduced . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00310 . 7554/eLife . 05789 . 004Figure 1—figure supplement 1 . Representation of mathematical model . ( A ) one-dimensional row of cells with proximal ( P ) and distal ( D ) membranes in cells 1 to i . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00410 . 7554/eLife . 05789 . 005Figure 2 . Ft and Ds show mutually dependent stable fractions at junctions . Wing discs containing clones of ( A ) ds-EGFP/+ and ( B ) ft-EGFP/+ labelled for EGFP ( green ) and Ds or Ft , respectively ( red ) , show that addition of an EGFP tag does not affect protein localisation . Quantification of ( A′ ) Ds or ( B′ ) Ft junctional protein confirms expression levels of tagged proteins are similar to wild-type . FRAP analysis of ( C ) Ds-EGFP and ( D ) Ft-EGFP puncta ( bright regions , solid line ) and non-puncta ( dashed line ) reveals a large stable fraction of protein in puncta ( see Figure 2—figure supplement 1 for an example of a puncta and non-puncta region and Figure 2—figure supplement 2 for FRAP graphs with individual data points ) . Multiplying the intensity of FRAP regions prior to bleaching ( C′ and D′ ) by the stable fraction plateau ( C and D ) reveals a final ‘stable amount’ of protein at junctions ( C″ and D″ ) and a large significant difference between puncta and non-puncta regions ( both p = 0 . 0001 , error bars denote standard error ( SEM ) here and in all remaining figures ) . ( E ) Apical confocal sections of a V5-Ft antibody internalisation assay in pre-pupal wing , with time points of 0 , 10 , and 20 min , reveal persistent Ft puncta over time while quantification of total apical levels ( right panel ) shows protein internalisation is occurring ( p = 0 . 01 at 10 min and p = 0 . 001 at 20 min ) ( F ) Sub-apical confocal sections of pre-pupal wing show V5-Ft previously at the cell surface is seen in discrete intracellular puncta at t = 20 , where it co-localises with early endosome marker Hrs ( magenta ) ( F′ ) , confirming antibody internalisation rather than antibody drop off . FRAP analysis of Ds-EGFP in ftG-rv clones ( G ) and Ft-EGFP in a dsUA071 background ( H ) resulted in more rapid and increased protein recovery . Dashed lines denote 95% confidence intervals . As recovery did not reach 100% , a small stable population could remain . Also see Figure 2—figure supplement 4 for FRAP graphs with individual data points . Re-bleaching experiments suggest that this is not an experimental artefact ( see Figure 2—figure supplement 5 ) . Live images of Ds-EGFP ( I ) and Ds-EGFP with a ftG-rv clone ( I′; blue dots indicate first row of mutant cells ) show a loss of puncta and diffuse distribution of Ds-EGFP when Ft is not present . Similar distribution of Ft-EGFP ( J ) is seen in a dsUA071 mutant ( J′ ) . ( K ) Comparison of Ds-EGFP and Ft-EGFP protein levels taken at the same confocal settings suggests there is almost twice as much Ft-EGFP ( n = 4 wings ) as Ds-EGFP ( n = 2 wings ) in puncta regions ( p = 0 . 01 ) . Intensity measurements were taken from manually selected puncta using 1 µm2 ROIs . Around 50 ROI measurements were taken per wing and average intensities were plotted and analysed using an unpaired t-test . ( L ) Bar chart representing time taken for 50% protein recovery to occur during FRAP experiments . Both Ft-EGFP and Ds-EGFP mobile fractions exhibit slower recovery in puncta compared to non-puncta and mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00510 . 7554/eLife . 05789 . 006Figure 2—figure supplement 1 . Description of FRAP method . A cartoon flowchart describing the steps involved in a FRAP experiment . Initially , puncta and non-puncta regions are selected and their intensity measured ( pre-bleach intensity ) . The regions are bleached for a few seconds by increasing the laser power . Bleached regions are imaged over time and recovery of the unstable population is measured for around 5 min . Measurements are plotted and a one-phase exponential curve is fitted to the data . The plateau of this curve gives us the unstable fraction and to get the stable fraction we minus the unstable fraction from 1 , for example , 1–0 . 78 = 0 . 22 . This figure is multiplied by the pre-bleach intensity to give the stable amount . So if , for example , 78% unbleached protein has recovered , 22% of the bleached protein in the region of interest is then the stable fraction . We multiply this by the original pre-bleach intensity . So , a starting intensity of 500 units is multiplied by 0 . 22 to get a final stable amount of 110 units . However , if a mutant has a starting intensity of 250 and the same stable fraction of 22% we would say the stable amount of the mutant is 55 . The mutant has therefore much less stable protein at junctions despite having a similar stable fraction . An example of puncta and non-puncta regions in ds-EGFP is also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00610 . 7554/eLife . 05789 . 007Figure 2—figure supplement 2 . Individual data points for FRAP experiments . FRAP graphs containing individual data points with SEM for each experiment . ( A ) ds-EGFP homozygous puncta and non-puncta . ( B ) ft-EGFP homozygous puncta and non-puncta . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00710 . 7554/eLife . 05789 . 008Figure 2—figure supplement 3 . FRAP on Ds-EGFP in the pre-pupal wing . ( A ) FRAP on distal pre-pupal wings demonstrates a difference between puncta and non-puncta in the same wing region as the internalisation assay , confirming the presence of stable and unstable protein populations . ( A′ ) Intensity of puncta and non-puncta regions in the pre-pupal wing . ( A″ ) Stable amount of puncta and non-puncta in the pre-pupal wing confirms a large difference in stability between regions . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00810 . 7554/eLife . 05789 . 009Figure 2—figure supplement 4 . Individual data points for FRAP experiments . FRAP graphs containing individual data points with SEM for each experiment . ( A ) ds-EGFP ftG-rv clones ( no puncta observed ) . ( B ) ft-EGFP dsUA071 ( no puncta observed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 00910 . 7554/eLife . 05789 . 010Figure 2—figure supplement 5 . Control re-bleach FRAP experiments . ( A ) Re-bleach experiment performed on ft-EGFP dsUA071 . The re-bleach experiment confirms there are no obvious bleaching artefacts in our FRAP experiments . A FRAP region is selected and bleached as normal ( see ‘Materials and methods’ ) , recovery is allowed for 2 min prior to a second bleaching event which is performed in a smaller region within the first bleached region ( A′ ) . Recovery is measured and reaches around 100% . If there were any bleaching artefacts ( such as photodamage-induced stabilisation of a protein population ) , recovery after the second bleach would not reach 100% . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 010 Using these starting parameters , our modelling predicts that even if Fj acts on only one of either Ft or Ds , then a weak asymmetry of both Ft and Ds is produced across cells ( Figure 1D , D′ , E , E′ ) . However , in both cases there is also a shallow gradient of bound protein across the tissue . If , however , Fj acts on both Ft and Ds , there is improved cellular asymmetry and a negligible tissue gradient ( Figure 1D″ , E″ ) . Notably , neither situation predicts the approximate two-fold asymmetry in Ds distribution across the cell axis that is experimentally observed ( Brittle et al . , 2012 , see ‘Discussion’ ) . Thus , in our model , Fj acting on only Ft , or only Ds , or both results in asymmetric subcellular distributions of Ft and Ds . Therefore , in order to resolve the role of Fj in Ft–Ds asymmetry generation , we sought to understand its effects on Ft–Ds binding in vivo . As there is no method for directly measuring the strength of binding of two proteins in the in vivo context of a living tissue , we instead employed FRAP to assay the mobility of Ft and Ds in the developing Drosophila wing . During FRAP , flies with an EGFP-tagged protein ( e . g . , Ds-EGFP ) are used , and a fluorescent region of interest is selected and bleached . The same region is imaged over time and recovery of fluorescent protein is measured . Recovery is due to movement of unbleached protein into the bleached area from elsewhere in the cell and any protein that has been bleached remains bleached and is not reactivated . Therefore , if after bleaching and imaging , over time we see , for example , fluorescence recovering to 30% of the starting intensity we assume that 70% of the bleached protein remains and is therefore stably bound at the junction . We perform all of our experiments using the same microscope settings ( unless otherwise stated ) and since protein levels ( and thus fluorescent intensity ) can vary between genotypes , we must take this into account during our analysis . To do this , we multiply the starting intensity by the stable fraction to get a final stable amount ( see Figure 2—figure supplement 1 ) . We make the assumption that a decrease in the speed or size of the mobile population provides a readout of increased binding interactions . For instance , if the size of the mobile population ( unstable fraction ) increases in a particular genotype and , once the intensity is taken into account , a smaller stable amount of overall protein is calculated , we assume that binding affinity has decreased . Additionally , if the speed of protein recovery after bleaching increases in a particular mutant , we infer that binding is not as strong as ‘wild-type’ and the protein can therefore relocate more quickly . FRAP was performed as previously described ( Strutt et al . , 2011 see also ‘Materials and methods’ and Figure 2—figure supplement 1: individual data points and confidence intervals are also provided in the relevant figure supplements and Supplementary file 1 ) in the same proximal region of the wing disc ( unless otherwise stated ) using a strain of flies expressing the Ds protein endogenously tagged with EGFP ( Brittle et al . , 2012 ) , and a newly generated strain in which EGFP was inserted at the C-terminus of the endogenous Ft coding region ( see ‘Materials and methods’ ) . Both transgenes were found to be expressed and localised similarly to wild-type ( Figure 2A , B ) and insertion of the tag did not affect wing size or shape suggesting protein function is normal . Live imaging of both Ft-EGFP and Ds-EGFP ( Figure 2I , J ) revealed that the junctional populations exhibit a punctate distribution , as previously seen by immunolabelling in fixed tissue ( Ma et al . , 2003; Brittle et al . , 2012 ) and also for components of the ‘core’ planar polarity pathway ( Strutt et al . , 2011 ) . In our FRAP experiments , puncta and non-puncta regions were bleached and fluorescence recovery was measured over time ( Figure 2C , D , Figure 2—figure supplement 2 ) . The Ft-EGFP signal was almost twice the level of Ds-EGFP ( Figure 2K ) , therefore laser power had to be adjusted accordingly between experiments meaning the final stable amounts for Ds-EGFP cannot be compared directly to those for Ft-EGFP . For both puncta and non-puncta , there was some recovery of fluorescence demonstrating that the proteins were mobile at junctions . Recovery was not complete indicating that there was also a population of stable Ds-EGFP and Ft-EGFP that did not recover during the time period used . Measuring protein recovery revealed a significant difference in stability between puncta and non-puncta regions ( Figure 2C , D and Table 1 ) with puncta showing an increased amount of stable protein ( Figure 2C″ , D″ , both p = 0 . 0001 ) when taking pre-bleach intensity levels into account ( Figure 2C′ , D′ ) . Overall , the FRAP assays indicated that there were both stable and unstable populations of Ft-EGFP and Ds-EGFP present at junctions , with stable material concentrated into bright punctate regions . 10 . 7554/eLife . 05789 . 011Table 1 . Comparison of rate and stabilisation dataDOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 011GenotypeStable amountSEM of stable amountHalf timeConfidence intervalDs-EGFP Homozygous Puncta292 . 9±18 . 538 . 530 . 8 to 51 . 3Ds-EGFP Homozygous Non-Puncta72 . 1±21 . 827 . 922 . 13 to 37 . 6Ds-EGFP ftG-rv Homozygousn/an/a8 . 35 . 8 to 14 . 5Ds-EGFP fjd1 Homozygous Puncta167 . 2±21 . 633 . 8925 . 5 to 50 . 5Ds-EGFP fjd1 Homozygous Non-Puncta26 . 4±20 . 55 . 63 . 8 to 10 . 3Ft-EGFP Homozygous Puncta148 . 0±9 . 524 . 219 . 28 to 32 . 3Ft-EGFP Homozygous Non-Puncta52 . 0±10 . 69 . 57 . 2 to 13 . 9Ft-EGFP dsUA071 Homozygousn/an/a8 . 55 . 7 to 16 . 6Ft-EGFP fjd1 Homozygous Puncta114 . 1±1 . 416 . 813 . 3 to 22 . 8Ft-EGFP fjd1 Homozygous Non-Puncta52 . 6±19 . 011 . 48 . 7 to 16 . 6Ds-EGFP Homozygous Puncta Distal ( High Fj ) 378 . 0±17 . 693 . 059 . 1 to 217 . 6Ds-EGFP fjd1 Homozygous Puncta Distal142 . 7±2869 . 751 . 3 to 108 . 6 As an independent assay to confirm the presence of stable and unstable protein populations , we performed an antibody internalisation assay in the pre-pupal wing ( see ‘Materials and methods’ and Strutt et al . , 2011 ) to assess the endocytic turnover of Ft , using a Ft transgene tagged extracellularly with a V5 epitope ( Feng and Irvine , 2009 ) . The internalisation assay is performed in pre-pupal wings as the peripodial membrane prevents the assay from being effective in the wing disc . Live pre-pupal wings were dissected in Schneider's medium and incubated with antibody against V5 at 0°C followed by washing and chasing at room temperature before fixation . The amount of V5-Ft at the cell surface was determined via incubation with secondary antibody in the absence of detergent . When endocytosis was allowed by moving the tissue to room temperature the total amount of apical cell surface protein decreased over time suggesting protein internalisation was occurring and indicating the presence of an unstable cell surface population of V5-Ft ( Figure 2E ) . To confirm that protein loss was not due to antibody drop off , V5-Ft previously found at the surface was traced to internal vesicles ( Figure 2F ) and found to co-localise with the early endosome marker Hrs ( Figure 2F′ ) . Populations of V5-Ft appeared to be resistant to endocytic turnover with persistent puncta of cell surface V5-Ft observed after 20 min ( Figure 2E ) . This was consistent with the puncta stability observed in FRAP experiments both in wing discs ( Figure 2D″ ) and also in the pre-pupal wing ( Figure 2—figure supplement 3 ) . Note however , that although we assume that these remaining populations in each assay are the same , we cannot be certain that this is the case . During FRAP , we are unable to observe the fate of the bleached protein . Furthermore , in the internalisation assay , protein might be internalised and return to the same sites via exocytosis . To understand whether the observed protein stability at junctions was due to the presence of heterophilic binding between Ft and Ds , FRAP was performed on both Ft-EGFP and Ds-EGFP in ds and ft null mutant backgrounds , respectively ( Figure 2G–J , Figure 2—figure supplement 4 ) . Removal of the putative binding partner resulted in the loss of obvious puncta and the almost complete recovery of fluorescence after bleaching , indicating that the majority of protein had become unstable at junctions . Thus , the observed stability of each protein is dependent on the presence of the binding partner . Comparing data for the rate of recovery of fluorescence of Ds-EGFP ( Table 1 and Figure 2L ) further revealed that the mobile populations of Ds-EGFP show slower movement , either in puncta or non-puncta , in the presence of Ft , compared to that in the absence of Ft ( where no puncta are observed ) . This suggests that in addition to the immobile population of Ds bound to Ft which is concentrated in puncta as described in the previous section , there is also a mobile population of Ds-EGFP present in both puncta and non-puncta regions which is bound to Ft and shows less rapid movement than free Ds-EGFP ( i . e . , Ds-EGFP in the absence of Ft ) . The reduction in the rate of movement of these mobile Ds-Ft heterodimers in puncta as opposed to non-puncta might indicate the presence of a cis-dimerisation mechanism that promotes the clustering of heterodimers into the puncta . Alternatively , mechanisms such as physical interactions with the cytoskeleton similar to those seen for E-Cadherin clustering ( Cavey et al . , 2008 ) may be responsible for the reduced mobility . Our data on the rate of recovery of Ft-EGFP ( Table 1 and Figure 2L ) also support these conclusions . However , in this case there is a negligible difference in the rate of movement of Ft-EGFP in non-punctate regions in the presence of Ds , as compared to the rate of movement of Ft-EGFP in the absence of Ds , and also a faster rate of Ft-EGFP mobility within puncta as compared to Ds-EGFP in puncta . We surmise that these differences are due to there being an excess of Ft-EGFP present over Ds ( see Figure 2K ) , resulting in an increased proportion of the population of Ft-EGFP being unbound and thereby free to move into bleached regions . Overall , we conclude from our FRAP experiments using ‘wild-type’ Ft-EGFP and Ds-EGFP , that Ft and Ds bind to each other across cell membranes resulting in the production of immobile fractions of Ft and Ds at cell junctions , and a reduction in mobility of Ft and Ds at junctions . Having thus demonstrated that Ft–Ds binding can be measured in vivo by virtue of the effects that binding has on protein mobility and stability , we next sought to understand the effects of Fj on Ft–Ds binding . Fj is able to phosphorylate both Ft and Ds ( Ishikawa et al . , 2008 ) and modify the binding affinities between the two proteins in vitro ( Brittle et al . , 2010; Simon et al . , 2010 ) ; however , the functional in vivo result of these modifications is unknown . To investigate this , we performed FRAP on endogenously expressed Ft-EGFP and Ds-EGFP in a fj mutant background , measuring fluorescence recovery to infer the stability of Ft–Ds dimers ( Figure 3 , Figure 3—figure supplement 1 ) . As Ft–Ds binding across junctions is mutually dependent , we infer that if one protein gains or loses stability , so will the other . Moreover , as Fj is thought to act on Ft and Ds in opposite ways , if we remove Fj , we might not expect to see any difference in overall stability of Ft–Ds dimers as the positive and negative effects could cancel each other out . If a difference is observed , it suggests that phosphorylating one protein has a stronger effect than phosphorylating the other . 10 . 7554/eLife . 05789 . 012Figure 3 . Ft and Ds stability upon loss of Fj . Stable fraction of ( A ) Ds-EGFP and ( B ) Ft-EGFP in a fjd1 background ( yellow lines ) . Solid lines represent puncta and dashed lines represent non-puncta . ‘Wild-type’ values are plotted for comparison . Intensity of Ds-EGFP decreases ( A′ ) and Ft-EGFP increases ( B′ ) , in a fjd1 background , whereas stable amount of protein in puncta falls in both ( A″ ( p = 0 . 004 ) and B″ ( p = 0 . 04 ) ) . The reduction in Ds-EGFP levels at junctions in the absence of Fj is consistent with less Ds being bound to Ft in this situation and excess unbound Ds being removed from junctions . The reason for an increase in Ft levels is unknown , as presumably less Ft is bound to Ds , but suggests that unbound Ft is not removed from junctions . See Figure 3—figure supplement 1 for FRAP curves with individual data points . ( C ) Live images of wing discs taken during a FRAP experiment show the recovery of puncta over time in ds-EGFP and ds-EGFP in a fjd1 background . Protein recovery after 70 ( sec ) is increased in a fjd1 background and overall protein levels appear reduced . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 01210 . 7554/eLife . 05789 . 013Figure 3—figure supplement 1 . Individual data points for FRAP experiments . FRAP graphs containing individual data points with SEM for each experiment . ( A ) ds-EGFP fjd1 puncta and non-puncta . ( B ) ft-EGFP fjd1 puncta and non-puncta . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 013 The co-expression of Fj with Ds in a cell culture assay has been shown to reduce the ability of Ds to bind to Ft ( Brittle et al . , 2010 ) . Therefore , we hypothesise that removing Fj might increase the stability of any Ft–Ds dimer ( corresponding to a decrease in the height of the FRAP recovery plateau ) . However , we found that Ds-EGFP became less stable upon removal of Fj ( Figure 3A″ , p = 0 . 004 ) , with a reduction in both the stable fraction ( i . e . , an increase in the height of the plateau ) and the stable amount at junctions ( Figure 3A , A″ ) . Stills of live FRAP experiments demonstrate the increased recovery of Ds-EGFP in a fj mutant background after 70 s ( Figure 3C ) . Co-expression of Fj with Ft has been shown to increase the affinity of Ft for Ds ( Simon et al . , 2010 ) ; therefore , we should expect a loss of stability of Ft-EGFP in a fj background . In this case analysis of Ft-EGFP recovery in a fj background did result in a reduction in stable fraction and stable amount at junctions ( Figure 3B , B″ , p = 0 . 04 ) . The half time rate of recovery of Ft-EGFP and Ds-EGFP was also decreased in the absence of Fj , consistent with a reduction of binding affinity allowing more rapid mobility ( Table 1 ) . Thus , the overall result of removing Fj was a reduction in stability of the Ft–Ds dimer . A loss of stability , rather than a gain , suggests that , at least in the wing disc , the effect Fj has on Ft is dominant to any which it might have on Ds . Using endogenously tagged Ft and Ds proteins , we have uncovered a dominant effect of Fj on Ft; however , the overall result may mask any potential consequence of Fj action on Ds . As the in vitro evidence suggests that Fj modifies Ft–Ds binding via phosphorylation of their extracellular cadherin repeats ( Ishikawa et al . , 2008; Brittle et al . , 2010; Simon et al . , 2010 ) , we turned to mutations in the mapped phosphorylation sites to separate effects on each molecule . We have previously described fly strains expressing EGFP-tagged Ds phosphorylation mutants and mimetics uniformly under control of the Actin5C promoter ( Brittle et al . , 2010 ) . Under these expression conditions , protein levels are higher than endogenous and there is a loss of visible puncta meaning that FRAP experiments are performed solely in junctions and puncta and non-puncta are not distinguished . Microscope settings were also altered to take the increased intensity levels into account . All of the phosphomutant experiments were performed in relevant mutant backgrounds so the Actin5C construct was the only form of expression of the protein in question . For example , if we were looking at Act-Ds-EGFP , we performed the experiment in a ds background . If Fj does have an independent effect on Ds , we might expect that preventing Fj from phosphorylating Ds ( Ds phosphomutant ) whilst not affecting Ft phosphorylation ( by leaving Fj cellular activity intact ) should allow us to see this . Based on previous experiments where phosphorylating Ds reduces the binding affinity between Ft and Ds ( Brittle et al . , 2010 ) , we are able to hypothesise that mutating the phosphorylation sites in Ds , thus preventing Fj phosphorylation , should improve binding affinity and therefore increase the stable amount of bound protein . We also hypothesise that mutating the phosphorylation sites of Ft ( so Ft cannot be phosphorylated by Fj ) will result in a reduction in binding affinity and therefore a reduced stable amount . When comparing the stability of ‘wild-type’ Ds-EGFP ( expressed from the Act-ds-EGFP transgene ) and a form of Ds with the phosphorylation sites mutated to alanine to block phosphorylation ( Act-ds-S>Ax3-EGFP; Figure 4A , Figure 4—figure supplement 1 ) , there was a strong increase in the amount of stable protein present in the mutant ( Figure 4A″ , 196 . 3 ± 18 and 428 . 9 ± 35 intensity units respectively , p = 0 . 0004 ) . As Fj was still expressed and presumably able to act normally on Ft in this experiment , the increase in stable amount must be solely due to the inability of Fj to phosphorylate Ds . Additional loss of Fj , to remove potential Ft phosphorylation from the experiment , resulted in the expected decrease in stability , back to a level similar to ‘wild-type’ ( Figure 4B , E , Figure 4—figure supplement 1 ) , again consistent with the effect of Fj on Ft being dominant to that on Ds . A bar chart representing all of the different stable amounts of Act-DsEGFP is provided in Figure 4E . 10 . 7554/eLife . 05789 . 014Figure 4 . Effects of mutating the phosphorylation sites in Ds and Ft on in vivo stability . ( A ) Comparison of Act-ds-EGFP with Act-ds-EGFP phosphomutant ( Act-ds-S>Ax3-EGFP ) . FRAP is performed on junctions as no puncta are visible . Ds-S>Ax3-EGFP has a larger remaining stable fraction after bleaching ( A ) , is more highly expressed at junctions ( A′ ) and has a significantly larger stable amount at junctions ( A″ p = 0 . 0004 ) than ‘wild-type’ . ( B ) When Fj is removed from the phosphomutant background ( fj− Act-ds-S>Ax3-EGFP ) intensity levels drop ( B′ ) and the stable amount at junctions returns to below ‘wild-type’ levels ( B″ and E ) . ( C ) Ft-EGFP phospho-mutant ( Act-ft-S/T>Ax5-EGFP ) has a smaller stable fraction than Ft-EGFP ( Act-ft-EGFP ) , shows reduced levels at junctions ( C′ ) , and has a significantly reduced stable amount at junctions ( C″ , p = 0 . 006 ) . ( D , D′ , D″ ) In a fjd1 mutant background , the phosphomutant protein ( fj- Act-ft-S/T>Ax5-EGFP ) does not become any less stable suggesting any loss of stability in ( C ) is caused primarily by a loss of Ft phosphorylation . See Figure 4—figure supplement 1 for FRAP graphs with individual data points . ( E–F ) An overview of stable amounts in ( E ) Act-ds-EGFP and ( F ) Act-ft-EGFP shows that despite there being a significant increase in stable Ds when its phosphorylation sites are mutated , the removal of Fj results in a loss of stable Ds ( E; p = 0 . 02 ) and also Ft ( F; p = 0 . 02 ) . Thus the phosphorylation state of Ds only appears to significantly affect Ft–Ds binding if Ft is already phosphorylated by Fj . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 01410 . 7554/eLife . 05789 . 015Figure 4—figure supplement 1 . Individual data points for FRAP experiments . FRAP graphs containing individual data points with SEM for each experiment . ( A ) Act-ds-EGFP and Act-ds-S>Ax3-EGFP ( no puncta available ) . ( B ) Act-ds-S>Ax3-EGFP and fj-Act-ds-EGFP S>Ax3-EGFP ( no puncta observed ) . ( C ) Act-ft-EGFP and Act-ft-S/T>Ax5-EGFP ( no puncta observed ) . ( D ) Act-ft-S/T>Ax5-EGFP and fj- Act-ft-S/T>Ax5-EGFP ( no puncta observed ) . ( E ) Comparison of Act-ds-EGFP and Act-ds-EGFP phosphomimetic ( Act-ds-S>Dx3-EGFP ) . The expected result upon expression of a ds phosphomimetic might be a reduction in stable fraction and stable amount . However , the Act-ds-S>Dx3-EGFP stable fraction is the same as Act-ds-EGFP ( E ) , intensity levels are similar ( E′ ) and stable amounts are not significantly different ( E″ ) suggesting that mutating the phosphorylation sites has not been successful or does not have any additional effect ( see text ) . ( F ) Comparison of Act-ft-EGFP and Act-ft-EGFP phosphomimetic ( Act-ft-S/T>Dx4-EGFP ) . Mimicking the phosphorylation sites in Act-ft-EGFP might be expected to increase both the stable fraction and stable amount of protein when compared to ‘wild-type’ Act-ft-EGFP . However , the stable fraction remains the same ( F ) , and although intensity levels of the phosphomimetic are higher ( F″ ) , the stable amount is not significantly increased ( F″ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 015 We were unable to detect any decrease in stability using our previously described Ds phosphomimetic ( Act-ds-S>Dx3-EGFP ) fly strain ( Figure 4—figure supplement 1 ) . We suspect that this is due to the mutation of serines to aspartates having only a modest phosphomimetic effect caused by the lower negative charge of aspartic acid compared to a phosphate group . The effect of mimicking the phosphorylation of Ds is therefore no greater than the normal effect of Fj phosphorylation on Ds in the region of the wing being assayed . It is also likely that our FRAP assay is unable to detect subtle differences in binding affinity . We also constructed Ft phosphomutants and mimics ( see ‘Materials and methods’ ) and tested them in a similar manner . Consistent with previous results , a Ft phosphomutant ( Act-ft-S/T>Ax5-EGFP ) showed a decreased stable fraction ( Figure 4C , Figure 4—figure supplement 1 ) and a large reduction in stable amount when compared to ‘wild-type’ ( Figure 4C″ , 97 . 9 ± 15 and 194 . 5 ± 24 intensity units respectively , p = 0 . 006 ) . Additionally , a Ft phosphomimetic ( Act-ft-S/T>Dx4-EGFP ) showed slightly improved stability compared to ‘wild-type’ ( Figure 4—figure supplement 1 ) , however , this increase was not statistically significant . Removal of Fj , and therefore Ds phosphorylation , from the phosphomutant background ( Figure 4D , Figure 4—figure supplement 1 ) did not result in a significant change in stable amount at the junctions confirming that the observed reduction in stability of the phosphomutant ( Figure 4C″ ) was primarily due to a loss of Fj phosphorylation of Ft . A bar chart showing the stable amounts of Act-ft-EGFP in each condition is provided in Figure 4F . Importantly , the phosphomutant experiments have revealed in vivo affects of Fj phosphorylation on both Ft and Ds . Mutating the Ds phosphorylation sites resulted in a significantly improved binding ability , however , this improved ability was lost when Ft was also not phosphorylated . Additionally , the phosphomutant experiments have confirmed that there is a dominant effect of Fj phosphorylation on Ft . Previously , our model of Ft–Ds interactions produced consistent cellular asymmetry with a negligible tissue gradient when Fj acted upon both Ft and Ds ( Figure 1D , E ) . However , our data have shown that , although Fj can act on both Ft and Ds , the phosphorylation of Ft to increase Ft–Ds binding is dominant . Furthermore , in the absence of Ft phosphorylation ( in a Ft phosphomutant ) , the phosphorylation of Ds ( ‘wildtype’ vs fj- ) has no significant effect on the overall amount of bound protein . In order to understand how this might affect the establishment of cellular asymmetry and binding across a tissue , we modified our model parameters as follows: first , we reduced the degree by which phosphorylated Ds inhibits Ft–Ds binding in complexes where Ft is phosphorylated ( i . e . , complexes A and B ) ; second , we made the binding affinities for complexes formed from unphosphorylated Ft ( i . e . , complexes C and D ) the same ( see ‘Materials and methods’ ) . We then ran the model with Fj acting on both Ds and Ft ( Figure 5A ) . We consequently still see the establishment of cellular asymmetries in Ft–Ds distribution within each cell , but also see a tissue gradient of Ft–Ds binding that follows the Fj gradient . The model therefore predicts that as you move from a region of low Fj to high Fj , Ft–Ds dimer stability and thus the stable amount at cell junctions should increase . 10 . 7554/eLife . 05789 . 016Figure 5 . Ft–Ds binding varies across the wing in response to the Fj expression gradient . ( A ) Output of the revised computational model in which Fj acts more strongly on Ft than Ds , and in which unphosphorylated Ft binds with equal affinity to both phosphorylated and unphosphorylated Ds . When Fj can act on both Ft and Ds with a Ft bias , an intermediate outcome between acting on Ft only and acting on Ft and Ds without a bias is seen and a cellular asymmetry of 4 . 1% is observed . A tissue gradient of 4 . 7% in the model when acting on both Ft and Ds , compared to 10% when Fj acts on Ft alone ( Figure 1D′ , E′ ) suggests that Ds phosphorylation is required to counter the strong graded effects of Ft phosphorylation . ( B ) FRAP analysis of endogenously expressed Ds-EGFP reveals a larger stable fraction in regions of high Fj ( distal ) despite showing slightly less overall fluorescence ( B′ ) . Overall , a significantly increased stable amount is seen in high Fj regions ( B″ p = 0 . 01 , 23% difference ) . ( C ) FRAP analysis in a fj mutant shows that this difference is lost when taking pre-bleach intensity levels into account ( C′ , C″ ) . See Figure 5—figure supplement 1 for FRAP graphs with individual data points . ( D ) Images taken during a FRAP experiment in a region of high Fj demonstrate the low level of recovery over time when compared to Ds-EGFP recovery in a region of low Fj ( as seen in Figure 3C ) . ( E ) FRAP analysis of junctionally localised Ds-EGFP expressed from the Act-ds-S>Ax3-EGFP transgene show overall reduced stable fractions , however , increased distal intensity levels of Ds-EGFP ( E′ ) mean the overall stable amount is significantly increased in regions of high Fj ( E″ , p = 0 . 01 , 53% difference ) . In a fjd1 mutant ( F ) any observed difference in stable amount across the tissue is lost ( F″ ) when taking pre-bleach intensity levels into account ( F′ ) ( proximal equates to low Fj , distal equates to high Fj ) . This is again consistent with Fj acting primarily on Ft . See Figure 5—figure supplement 1 for FRAP curves with individual data points . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 01610 . 7554/eLife . 05789 . 017Figure 5—figure supplement 1 . Individual data points for FRAP experiments . FRAP graphs containing individual data points with SEM for each experiment . ( A ) ds-EGFP homozygous puncta at low ( proximal ) and high ( distal ) Fj regions . ( B ) ds-EGFP homozygous puncta at proximal and distal regions in a fj mutant . ( C ) Act-ds-S>Ax3-EGFP puncta at low ( proximal ) and high ( distal ) Fj regions . ( D ) Act-ds-S>Ax3-EGFP puncta at proximal and distal Fj regions in a fj mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 017 We next endeavoured to test in vivo the prediction of a gradient of Ft–Ds binding across the tissue . We have already shown that Fj protein levels are high in the distal wing pouch and lower proximally ( Strutt et al . , 2004; Figure 1B ) . We therefore would expect there to be an increase in stability of Ft and Ds as we move from regions of low Fj ( proximal ) to regions of high Fj ( distal ) . To test this , we used FRAP analysis on endogenously expressed Ds-EGFP ( Figure 5B , Figure 5—figure supplement 1 ) and were able to detect a statistically significant increase in stable junctional amount in more distal regions ( Figure 5B″ , 292 . 9 ± 19 proximally and 378 ± 18 distally , p = 0 . 01 , 23% difference [see also Supplementary file 1] ) . This difference in stable amount was lost when Fj was removed ( Figure 5C″ 167 . 2 ± 22 proximally and 142 . 7 ± 28 . 6 distally , NS [see also Supplementary file 1] ) , implying that the Fj gradient normally produced the difference across the tissue . We repeated this experiment using the Act-ds-S>Ax3-EGFP phosphomutant , to assess the result of Fj only acting on Ft . Despite relatively small differences between stable fractions ( Figure 5E , Figure 5—figure supplement 1 ) , we saw a greater difference in stable Ds-S>Ax3-EGFP at junctions between high and low Fj regions ( Figure 5E″ , 109 . 2 ± 21 proximally and 230 . 4 ± 37 distally p = 0 . 01 , 53% difference ) which was again lost upon removal of Fj ( Figure 5F″ , 97 . 8 ± 9 proximally and 78 . 2 ± 25 distally , NS ) . These results directly confirm the predictions of the model , showing not only that a slight gradient of binding strength exists across the tissue , but also that an opposing gradient of Ds phosphorylation usually acts to counter the effects of a gradient of Ft phosphorylation , resulting in a relatively even distribution of bound Ft–Ds complexes at junctions across the tissue axis from low to high Fj . A long-standing problem in developmental biology is understanding how long-range patterning gradients are interpreted at the cellular level . More specifically , with regard to understanding how cell polarity is coordinated across sheets of cells , a major goal is to determine the mechanism by which a gradient of transcription across a tissue ( produced for instance in response to a morphogen ) can be sensed by individual cells to result in each cell adopting a uniform polarisation . A particular challenge for such a sensing mechanism is that the difference in levels of transcription between adjacent cells may be very small ( e . g . , only a few percent or less of the peak expression levels ) , and at the high end of the gradient this difference needs to be read against the background of a high overall expression level . The Ft-Ds-Fj pathway in Drosophila represents an excellent system for addressing this problem . The Ft and Ds cadherins bind heterophilically between adjacent cells , and the visible readout of polarity is the asymmetric distribution of these dimers across the cell axis , such that approximately two-fold higher Ds is found on one cell edge , bound to Ft on the apposing edge of the neighbouring cell ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . However , the measured Ft asymmetry across the cell axis is weaker than that of Ds , most likely because there is a larger population of Ft present at junctions , including a presumably significant unbound fraction which is symmetrically distributed . Importantly , the asymmetric distribution of Ft–Ds dimers is a result of the patterns of transcription of the ds and fj genes as specified by upstream morphogens ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . In this study , we focus specifically on the mechanisms determining Ft–Ds subcellular polarity in the Drosophila third instar wing imaginal disc . At this stage , Fj is expressed in a gradient , high at the putative distal end of the wing ( i . e . , the centre of the wing pouch ) and low proximally towards the wing hinge ( Strutt et al . , 2004 ) , whereas Ds is relatively uniformly expressed in the visible region of the wing pouch , but higher in hinge regions of the wing ( Strutt and Strutt , 2002 ) . It is important to point out that much of the proximal wing blade is folded at the larval stage and most likely also has high levels of Ds expression and this region of high Ds expression could therefore also be promoting Ft–Ds asymmetry . Although we believe that both the Fj and Ds expression patterns are important cues for specifying Ft–Ds asymmetry in the wing pouch , the evidence suggests that boundaries of Ds expression may only be able to act as a patterning cue over a few cell diameters ( Ambegaonkar et al . , 2012; Brittle et al . , 2012 ) , and therefore throughout much of the wing pouch the Fj gradient is likely to be a dominant cue . In in vitro assays , Fj is able to mediate the phosphorylation of extracellular cadherin repeats of Ft and Ds , and phosphorylation of Ft has been shown to promote its binding to Ds , whereas phosphorylation of Ds appears to inhibit its binding to Ft ( Ishikawa et al . , 2008; Brittle et al . , 2010; Simon et al . , 2010 ) . Thus , the gradient of Fj in the wing pouch is predicted to produce opposing gradients of Ft–Ds binding affinities , with Ft–Ds binding favoured between cells moving down a Fj gradient , and Ds-Ft binding favoured between cells moving up a Fj gradient ( Figure 6A ) . It has been previously proposed that these opposing binding gradients might play a role in producing uniform Ft–Ds interactions across the tissue ( Simon et al . , 2010 ) . 10 . 7554/eLife . 05789 . 018Figure 6 . Models of Ft–Ds interactions . ( A ) Cartoon illustrating our model of Ft–Ds binding across a tissue according to a Fj gradient . Ft-P ( dark blue ) and Ds-P ( dark purple ) increase , and Ft ( light blue ) and Ds ( light purple ) decrease as you move up the Fj gradient . Weak binding ( no line ) between heterodimers can occur between all populations , intermediate binding ( double lines ) can occur when Ft-P is available and the strongest bond ( solid line ) occurs only between Ft-P and Ds . As there is a bias towards Ft phosphorylation increasing binding , more intermediate binding occurs in regions of high Fj even though less favourable Ds ( light purple ) is available . In regions of low Fj , although there is more favourable Ds available , Ft-P is less abundant meaning fewer intermediate bonds are able to form . As Ft-P allows for stronger binding in cells with higher Fj , Ds in the next left-most cell preferentially accumulates to the right cell edge meaning the Fj gradient produces cellular asymmetry . A gradient of tissue binding is also produced with cells in the left-most cell producing fewer intermediate bonds than those in the right-most cell . ( B ) Conformational changes caused by the phosphorylation of Ft and Ds could provide an explanation for the differences in binding affinity caused by apparently similar phosphorylation events . The ‘open’ conformation of Ft and the ‘closed’ conformation of Ds may bind preferably such that if Ft is phosphorylated and Ds is not , binding is strongest ( +++ ) . If Ft and Ds are both phosphorylated , binding strength is intermediate ( ++ ) . ‘Closed’ conformation of non-phosphorylated Ft produces the weakest binding and is not affected by Ds ( + ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05789 . 018 In this study , we use a mixture of computational and experimental methods to address the mechanism of action of Fj . Initially , we constructed a simple one-dimensional computational model , in which Fj activity either promotes Ft binding activity or inhibits Ds binding activity , and in which Ft–Ds dimers are then allowed to freely bind at either cell edge until they reach an equilibrium state . Using this we show that a Fj gradient can lead to cellular asymmetry of Ft–Ds if Fj acts on either Ft or Ds or both . Our predictions differ from those of a previous study ( Simon et al . , 2010 ) which suggested that Fj acting only on Ft would not result in a cellular asymmetry of activity: the origin of this difference is that in our model ( in accordance with our experimental observations ) the Ft and Ds populations are mobile and able to redistribute to the most favourable cell edge . However , in agreement with the same study , our model does predict that only in the situation that Fj acts on both Ft and Ds is the amount of Ft–Ds dimers bound at the junctions approximately the same across the entire axis of the Fj gradient . Using FRAP to measure the levels of stable amounts and mobility of Ft and Ds at cell junctions , combined with either removal of fj activity or mutation of the Fj phosphorylation sites in Ft and Ds , we directly demonstrate that Ft and Ds stably associate at cell junctions in vivo , that Fj modulates this binding with a net positive effect in promoting binding , and that as predicted from the in vitro assays , Fj acts independently on both Ft and Ds to promote or inhibit their binding , respectively . An interesting experimental observation is that the effects of Fj on Ft–Ds binding and the total amounts stably localised to cell junctions are relatively modest . For instance , blocking the phosphorylation of either Ft or Ds only decreases the stable junctional amounts by about twofold . Thus , all the molecular species ( i . e . , Ds and DsP and Ft and FtP ) contribute to the final population of bound Ft and Ds at cell junctions . Our experimental results reveal that the effect of Fj on Ft is stronger than the effect of Fj on Ds . A simple prediction of this observation , confirmed by our computational model , is that the ability of the gradient of phosphorylated Ds to oppose the gradient of phosphorylated Ft is reduced , and therefore a gradient of Ft–Ds binding is expected to be observed across the tissue ( Figures 5A and 6A ) . This prediction was confirmed experimentally . We cannot rule out that under normal circumstances the in vivo effects of Fj on Ds are in fact negligible compared to those on Ft , as our experiments make use of Ds transgenes and may not fully represent events occurring when wild-type Ds is expressed from its endogenous locus . Nevertheless , the simplest interpretation of both our observations and previous published work is that Ds phosphorylation contributes to normal patterning . In sum , our findings demonstrate an in vivo mechanism for how a Fj expression gradient is converted into cellular asymmetries via phosphorylation of both Ds and Ft . We note that our computational simulations result in relatively modest cellular asymmetries of Ft and Ds distributions ( around 10% of the total cellular levels ) , whereas in vivo , observed Ds asymmetry can be as high as twofold ( Ambegaonkar et al . , 2012; Brittle et al . , 2012 ) . However , our model is only intended to make simple predictions regarding the effects of binding and redistribution of Ft and Ds molecules in a simple one-dimensional system and does not capture the full complexity of a three-dimensional cell environment and changes that occur in protein production and degradation and other cell properties over time . Furthermore , it is generally believed that relatively weak asymmetries generated by expression gradients might subsequently be amplified by feedback mechanisms ( Ambegaonkar et al . , 2012; Brittle et al . , 2012 ) , and some molecular mechanisms that might contribute to amplification have recently been identified ( Bosch et al . , 2014; Rodrigues-Campos and Thompson , 2014 ) . With regard to possible amplification mechanisms , an intriguing observation is that the majority of stable Ft–Ds at junctions is concentrated in bright regions which we refer to as ‘puncta’ . Furthermore , even mobile junctional populations of Ft and Ds show reduced mobility within these brighter regions . This might suggest that in addition to trans-interactions between Ft and Ds in neighbouring cells , there may also be cis-interactions between Ft and Ds molecules in cell junctions , a view supported by previous experimental reports ( Matakatsu and Blair , 2006; Sopko et al . , 2009 ) . Such a clustering mechanism could provide the molecular basis of a positive feedback interaction . Previous observations have suggested that Ft–Ds puncta do not co-localise with ‘core’ planar polarity protein puncta at junctions ( Ma et al . , 2003 ) . We are unsure as to the reasons for these distinct puncta populations , they could correspond to specialised membrane domains that favour binding or they could be random regions of protein clustering driven by cis-dimerisation or cytoskeleton tethering . A final conundrum is the observation , first made in vitro , which we have now confirmed in vivo , that modification of analogous serine residues in cadherin repeats of Ft and Ds by Fj phosphorylation leads to opposite effects on their binding activity . The simplest model would be that phosphorylation either promoted or inhibited binding to a partner , but this is evidently not the mechanism in play here . A possible working hypothesis is that phosphorylation events lead to changes in the intramolecular conformation of the extracellular regions of Ft and Ds , by analogy to the way in which phosphorylation frequently acts to cause other classes of molecules to enter ‘open’ or ‘closed’ conformations ( Xu and Carpenter , 1999; Potter et al . , 2005; Bertocchi et al . , 2012 ) . A possible scenario is that phosphorylation of each molecule causes it to enter an ‘open’ conformation , and that ‘open’ Ft binds most favourably to Ds , but ‘closed’ Ds binds most favourably to Ft ( Figure 6B ) . A recent publication analysing the configuration of mammalian Ft4 and Ds1 revealed multiple hairpin-like bends in their C-terminal regions caused by the loss of calcium binding linkers ( Tsukasaki et al . , 2014 ) . It is possible that Fj phosphorylation results in conformational changes between cadherin repeat domains near to these calcium binding sites resulting in the regulation of binding strength , as suggested previously for E-cadherin ( Petrova et al . , 2012 ) . Further studies of the structures and mode of heterophilic interactions between Ft and Ds will be required to resolve this question . Wing discs were dissected from wandering third instar larvae , fixed in 4% paraformaldehyde and washed in PBS containing 0 . 1% Triton-X-100 prior to immunolabelling . Primary antibodies used for histology were rabbit anti-Ds ( Strutt and Strutt , 2002 ) , rabbit anti-Ft ( Brittle et al . , 2012 ) , guinea pig anti-Hrs ( Lloyd et al . , 2002 ) , mouse anti-βGAL ( Promega , Wisconsin , USA ) , and mouse anti-Armadillo ( DSHB , Iowa City , USA ) . A rabbit serum against Fj was generated using a fusion protein corresponding to amino acids 111–433 , affinity purified and used at 1/100 for immunostaining . Secondary antibodies used were anti-Rb RRX and Cy2 ( Jackson , Pennsylvania , USA ) , anti-guinea pig A568 and anti-mouse Cy5 ( Molecular Probes , Oregon , USA ) . Images are averages of three confocal microscope sections taken on an Olympus FV1000 ( Pennsylvania , USA ) , a Leica SP1 ( Solms , Germany ) , or a Nikon A1R ( Tokyo , Japan ) confocal and processed using ImageJ ( NIH , USA ) and Adobe Photoshop ( California , USA ) . ImageJ was used to measure the mean intensity of endogenously tagged ds-EGFP and ft-EGFP at junctions using at least nine wings of each genotype . For Fj gradient measurements , immunolabellings were carried out on fixed wild-type wing discs using the rabbit-Fj antibody . Images were taken as 0 . 2-μm confocal slices throughout the disc and averages of the full stack were taken . Regions of interest were hand drawn per cell using a membrane marker as a guide and average intensity per cell was measured . Discs containing null clones were used for antibody background subtraction . Distal cells containing maximum measured signal were normalised to 100% and gradient of a proximal row of cells was calculated . Antibody internalisation assays were carried out on 5 . 5 hr APF pupal wings as previously described ( Strutt et al . , 2011 ) . To detect V5-Ft , wings from flies of the genotype ftG-rv/ft8; P[acman] V5-ft ( Feng and Irvine , 2009 ) were dissected and incubated with anti-V5 antibody ( Novus Biologicals , Colorado , USA ) . For detection of extracellular V5-Ft , tissue was incubated in secondary antibody in the absence of detergent , and post-fixed before adding other antibodies with detergent . Internalised V5-Ft was co-stained with guinea pig anti-Hrs ( Lloyd et al . , 2002 ) . For quantitation of extracellular staining at least five wings at each time point were imaged from at least two experiments taking 0 . 15-µm sections and using constant confocal settings . An average intensity of the three brightest confocal slices at the level of the apical junctions was measured in ImageJ . Laser-off background was subtracted , and the readings were normalized to 1 . 0 at t0 . To visualise internalised V5-Ft , wings were imaged in apical and subapical regions . Statistical analysis was carried out using ordinary one-way ANOVA with Tukey's test for multiple comparisons . Constructs were generated using standard molecular biology techniques and mutagenised and PCR-amplified regions verified by sequencing . Full-length ft is a 22-kb genomic fragment from BACR11D14 ( BACPAC Resources , California , USA ) , containing the entire coding sequence and tagged with EGFP subcloned into pAttB-FRT-polyA-FRT ( derived from pAct-FRT-polyA-FRT [Strutt , 2001] ) . Point mutations in the cadherin domains were introduced using QuikChange Multi-Site Directed Mutagenesis kit ( Stratagene , California , USA ) . 5 serines or threonines identified as phosphorylation sites in Ft ( CAD3 ( S273 ) , CAD5 ( S497 ) , CAD10 ( T1052 ) , CAD11 ( S1156 ) and CAD13 ( S1387 ) ) ( Ishikawa et al . , 2008 ) were mutated to alanine to generate Ft−S/T>Ax5-EGFP . A phosphomimetic form of Ft ( Ft S/T>Dx4-EGFP ) contains point mutations to aspartate of the following residues ( CAD3 ( S273 ) , CAD10 ( T1052 ) , CAD11 ( S1156 ) , and CAD13 ( S1387 ) ) . Alleles used are described in FlyBase . Homologous recombination was used to target EGFP into the C-terminus of the ft gene at its endogenous locus using the pRK2 targeting vector ( Huang et al . , 2008 ) . ds-EGFP ( Brittle et al . , 2012 ) and ft-EGFP flies were recombined with fjd1 , ftG-rv , or dsUA071 . All endogenous FRAP experiments were performed using homozygous ds-EGFP or ft-EGFP . Clones were generated using the FLP/FRT system ( Xu and Rubin , 1993 ) and marked with arm-lacZ ( Vincent et al . , 1994 ) . Transgenes containing ft-EGFP and point mutants were integrated into the same landing site ( attP2 68A4 ) ( Groth et al . , 2004 ) by Genetivision ( Texas , USA ) . Act-ds-EGFP and point mutations used are described previously ( Brittle et al . , 2010 ) . Genotypes used were:y w Scer\FLP1Ubx . hs; ds38k/ dsUA071; AttB{w+ ActP-FRT-polyA-FRT-dsX-EGFP} /+y w Scer\FLP1Ubx . hs; dsUA071 fjP1/ ds38k fjP1; AttB{w+ ActP-FRT-polyA-FRT-dsX-EGFP} /+y w Scer\FLP1Ubx . hs; ft8/ ftG-rv; AttB{w+ ActP-FRT-polyA-FRT-ftX-EGFP} ftG-rv /+y w Scer\FLP1Ubx . hs; ft8 fjd1/ ftG-rv fjP1; AttB{w+ ActP-FRT-polyA-FRT-ftX-EGFP} /+where dsX and ftX refers to wild-type ds or ft or one of the phosphomutants . Prior to dissection , an imaging chamber was built using a 22 × 50 mm cover glass as a base slide ( Thermo Scientific , Massachusetts , USA ) . Sellotape was placed smoothly on to the centre of the slide and a 7-mm2 area was cut out using a razor blade . Wandering stage larvae were collected and cleaned by rinsing in PBS . Wing discs were dissected in Shields and Sang M3 media ( Sigma [#S3652] , Missouri , USA ) with 2% added foetal bovine serum ( M3FBS ) . Discs were transferred by pipette to the cut out area in the imaging chamber with around 4 µl of media and arranged with their apical surface facing towards the base cover glass . The M3FBS was spread evenly around the cut out area . A 13-mm circular cover glass ( Thermo Scientific ) was carefully placed over the top allowing extra media to spread to the edges . Quick drying nail varnish was used to seal the cover glass after leaving to settle for a few minutes . For FRAP , samples were imaged on an inverted Nikon A1R GaAsP confocal using a Nikon 60× oil objective lens at 11 . 76× zoom producing a FRAP region of 256 × 256 pixels with a pixel size of 0 . 07 µm . For pre- and post-bleach images , a 448 Argon laser was used at an output of 0 . 5% with varying gain settings depending upon phenotype . Eight 1 µm2 regions of interest ( ROIs ) were selected per wing and bleached using the 488 argon laser at 50% power , passing 1–3 times for between 0 . 5 and 1 . 5 s depending upon experiment . Two pre-bleach images were captured with no delay as well as an immediate post-bleach image , 10 images were then captured every 5 s , followed by 10 images every 10 s and 10 images every 30 s . The initial rapid imaging was done in order to capture adequate rate information . For analysis , ROIs were individually reselected in ImageJ at each time point and acquisition bleaching was measured in non-bleached regions . Data were corrected for acquisition bleaching and normalised against pre-bleach values . An XY graph was plotted for each wing in PRISM ( v . 6 GraphPad ) . A one-phase exponential association curve was fitted for each ROI and an average plateau value recorded . A one-phase exponential curve was generated as we observe a single mode of recovery that reaches a plateau . We note that should the mode of recovery be more complex than this a two-phase exponential curve may be more appropriate; however , we do not have relevant experimental data to support this . Pre-bleach values were averaged per wing and multiplied by their associated plateau giving a stable amount . Stable amounts were then averaged across wings producing a final figure for each genotype . Stable amounts were analysed using unpaired t-tests or ordinary one-way ANOVAs with Tukey's test for multiple comparisons . Final stable fraction graphs were produced using the average plot for each wing and intensity was averaged across ROIs . We developed a computational model of Ft–Ds binding at cellular junctions , described by a set of ordinary differential equations . In this framework , we define a one-dimensional row of cells , whose proximal and distal membranes contain the same initial amount of Ft and Ds ( Figure 1—figure supplement 1 ) . These molecules were then phosphorylated according to a linear gradient of Fj activity , applied to reflect in vivo measurements of Fj quantity . Phosphorylated and unphosphorylated molecules were allowed to bind across junctions resulting in formation of four possible complexes , each in two orientations , namely , FtP—Ds ( A ) , FtP—DsP ( B ) , Ft—Ds ( C ) , and Ft—DsP ( D ) . This consideration of all four possible complexes between phosphorylated and non-phosphorylated forms of Ft and Ds is a key distinguishing feature of our model , when compared to previous computational approaches which only considered a single binding species essentially equivalent to our complex A ( Abley et al . , 2013; Mani et al . , 2013; Jolly et al . , 2014 ) . Reaction rates were derived from mass action , giving the following equations for proximally oriented complexes:dAPidt=kaonFtPPiDsDi−1− kaoffAPi , dBPidt= kbonFtPPiDsPDi−1− kboffBPi , dCPidt= kconFtPiDsDi−1− kcoffCPi , dDPidt= kdonFtPiDsPDi−1− kdoffDPi . Subscripts denote the proximal ( P ) membrane in cell i neighbouring distal membrane ( D ) in cell i−1 . Equivalent equations were derived for distally oriented complexes . Each reaction is parameterised by rate constants kon and koff for binding and unbinding reactions of each complex . Equations for unbound molecules were derived in a similar fashion , with the addition of a simple term allowing redistribution within a cell , parameterised by the coefficient , Diff:dDsPidt= −kaonFtPDi−1DsPi− kconFtDi−1DsPi+ kaoffADi−1+ kcoffCDi−1+ Diff ( DsDi− DsPi ) , dDsPPidt= −kbonFtPDi−1DsPPi− kdonFtDi−1DsPPi+ kboffBDi−1+ kdoffDDi−1+ Diff ( DsPDi− DsPPi ) , dFtPidt= −kconFtPiDsDi−1− kdonFtPiDsPDi−1+ kcoffCPi+ kdoffDPi+ Diff ( FtDi− FtPi ) , dFtPPidt= −kaonFtPPiDsDi−1− kbonFtPPiDsPDi−1+ kaoffAPi+ kboffBPi+ Diff ( FtPDi− FtPPi ) . Binding rates between molecules are parameterised by the association constant ( kon/koff ) for each complex , since a higher ‘on’ rate constant will result in an increased concentration of complexed molecules . Relative binding strengths of different molecule combinations reflect findings from in vivo and in vitro studies ( Brittle et al . , 2010; Simon et al . , 2010 ) , such that phosphorylation of Ft ( FtP ) promotes its binding and conversely phosphorylation of Ds ( DsP ) inhibits its binding . The following hierarchy of relevant association constants was used initially , A > B = C > D , such that the complex containing the two favoured molecules , FtP and Ds making complex A , had a faster ‘on’ rate than other combinations . Initial values of kon/koff for A , B , C , and D were chosen as 1 , 1/4 , 1/4 and 1/16 , respectively , to reflect the relative differences in stable amounts measured experimentally . Binding was allowed to continue over time until convergence . Simulations were run using an in-built ode solver ( ode23s ) in MATLAB ( R2013a; MathWorks , Massachusetts , USA ) and final concentrations of bound molecules were plotted . We went on to adapt this model to reflect our experimental finding that Fj appears to have a dominant effect on Ft . To achieve this , we altered association constants of binding reactions to allow phosphorylation of Ds to have a less significant effect on binding strengths , giving A > B > C = D . Thus association constants for each complex A , B , C , and D , were given as 1 , 1/2 , 1/4 , and 1/4 , respectively . Further explanation can be found in the ‘Results’ .	Epithelial cells form sheets that line the body surfaces and internal cavities of animals—such as the skin and the lining of the gut . Certain structures on the surface of epithelial cell sheets—for example scales , hair , and feathers—are often all orientated in a particular direction . Epithelial cells with structures organised like this are described as being ‘planar polarised’ . Different proteins work together to set up planar polarity in a sheet of epithelial cells . Dachsous and Fat are two proteins that are found in the cell membranes of epithelial cells , including in the wings of the fruit fly Drosophila . These proteins bind to each other and link a cell to its neighbour . Dachsous and Fat accumulate on opposing sides of each cell: Fat accumulates on the side closest to the fly's body , and Dachsous builds up on the side closest to the wing tip . This pattern provides directional cues that help orientate surface structures , and the pattern is established , in part , by the activity of an enzyme called Four-jointed . Four-jointed adds phosphate groups onto Dachsous and Fat . The activity of the Four-jointed enzyme forms a gradient along a developing wing: levels are low near the fly's body , and high at the wing tip . Previous experiments performed on cells grown in the laboratory showed that when Four-jointed adds phosphate groups to Fat and Dachsous , it prevents Dachsous from binding to Fat . However , it also makes Fat more able to bind to Dachsous . These opposing effects are thought to cause the proteins to accumulate on opposing sides of each cell . However , this has yet to be demonstrated in real tissue , not least because of the technical difficulty of measuring whether Fat-Dachsous binding has occurred in living organisms . Here , Hale et al . overcome this challenge using a method called ‘fluorescence recovery after photobleaching’ ( or FRAP ) to measure Fat and Dachsous binding in the epithelial cells in the developing Drosophila wing . Combining these experimental results with a computational model confirmed the findings of previous laboratory studies: that Four-jointed makes it easier for Fat to bind to Dachsous , and makes it more difficult for Dachsous to bind to Fat . The opposing effects on the activity of Fat and Dachsous that result from the Four-jointed gradient in the developing wing are able to fully explain the observed patterns of Fat-Dachsous binding and of planar polarisation across the wing . Overall , Hale et al . demonstrate how a gradient of protein activity that spans many cells is sensed and interpreted by individual cells to establish planar polarity . However , exactly how the phosphate groups added to Dachsous and Fat by Four-jointed modifies how they bind to each other remains a question for future work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Cellular interpretation of the long-range gradient of Four-jointed activity in the Drosophila wing
The CRISPR system for prokaryotic adaptive immunity provides RNA-mediated protection from viruses and mobile genetic elements . When viral RNA transcripts are detected , type III systems adopt an activated state that licenses DNA interference and synthesis of cyclic oligoadenylate ( cOA ) . cOA activates nucleases and transcription factors that orchestrate the antiviral response . We demonstrate that cOA synthesis is subject to tight temporal control , commencing on target RNA binding , and is deactivated rapidly as target RNA is cleaved and dissociates . Mismatches in the target RNA are well tolerated and still activate the cyclase domain , except when located close to the 3’ end of the target . Phosphorothioate modification reduces target RNA cleavage and stimulates cOA production . The ‘RNA shredding’ activity originally ascribed to type III systems may thus be a reflection of an exquisite mechanism for control of the Cas10 subunit , rather than a direct antiviral defence . CRISPR systems provide adaptive immunity in prokaryotes against mobile genetic elements ( MGE ) . DNA sequences derived from MGE are incorporated into the host genome separated by short direct repeats , forming the CRISPR locus . This is transcribed and processed to generate CRISPR RNAs ( crRNA ) that are loaded into effector complexes , programming them to detect and subsequently destroy the cognate MGE when the cell next encounters one . Class two effector complexes include Cas9 , which has seen wide application in genome engineering . Class one complexes are more intricate , multisubunit systems that use a backbone built around Cas7 subunits which binds the crRNA . This backbone subunit is present in multiple copies and is conserved across the Class one effectors , which are further differentiated into type I , III and IV systems ( reviewed in [Makarova et al . , 2015] ) . The type I complexes , typified by Cascade ( type I-E ) from Escherichia coli , bind target DNA and then recruit the Cas3 enzyme via an interaction with the Cse1 subunit , resulting in Cas3 mediated DNA degradation ( Hochstrasser et al . , 2014 ) . In contrast , type III complexes bind invading RNA species , activating several different enzymatic activities present within the complex ( reviewed in [Tamulaitis et al . , 2017] ) . The Cas7-mediated ‘backbone cleavage’ activity degrades bound target RNA with a characteristic 6-nucleotide spacing ( Staals et al . , 2014; Tamulaitis et al . , 2014; Hale et al . , 2014 ) . The HD-nuclease domain of the large Cas10 subunit is responsible for target DNA degradation ( Elmore et al . , 2016; Estrella et al . , 2016; Kazlauskiene et al . , 2016; Jung et al . , 2015; Han et al . , 2017a ) . This is activated when target RNA is bound by the complex ( Samai et al . , 2015 ) , providing a mechanism for transcription-dependent DNA targeting ( Deng et al . , 2013 ) that allows lysogenic phage to persist in the host chromosome ( Goldberg et al . , 2014 ) . Type III CRISPR systems are much more tolerant of mismatches with nucleic acid targets than other types , making viral escape difficult ( Pyenson et al . , 2017; Manica et al . , 2011; Manica et al . , 2013 ) . Most recently a third activity of type III effector complexes has been described . For some time , it was known that many type III system operons encode a protein such as Csm6 or Csx1 that is not part of the effector complex . These genes have been shown to be important for CRISPR-based immunity in vivo ( Deng et al . , 2013; Jiang et al . , 2016 ) . The proteins have a CARF ( CRISPR-associated Rossmann fold ) domain linked to a HEPN ( higher eukaryotes and prokaryotes nucleotide binding ) domain and have weak ribonuclease activity in vitro ( Niewoehner and Jinek , 2016; Sheppard et al . , 2016 ) . CARF domains were predicted to bind some form of cyclic nucleotide , perhaps as an allosteric effector ( Lintner et al . , 2011 ) but the source of the effector was unknown . This puzzle was solved when it was demonstrated that the cyclase domain of Cas10 can synthesise cyclic oligoadenylate ( cOA ) molecules from ATP , when activated by target RNA binding ( Kazlauskiene et al . , 2017; Niewoehner et al . , 2017 ) . cOA species consist of 3 to 6 membered rings of AMP with 3’ and 5’ linkages and have not been observed in any other biological context . cOA in turn binds to and activates Csm6 and Csx1 , enhancing their ribonuclease activity . cOA thus represents a new type of second messenger , generated by type III CRISPR effector complexes , that sculpts the cellular response to invasion by MGE ( reviewed in [Koonin and Makarova , 2018] ) . The crenarchaeaon Sulfolobus solfataricus has been a model for studies of the CRISPR-Cas system for many years ( reviewed in [Garrett et al . , 2015] ) . S . solfataricus encodes two type III effector systems: a Cmr ( type III-B ) complex , which has a unique Cmr7 subunit ( Zhang et al . , 2012 ) , and a Csm ( type III-D ) complex ( Rouillon et al . , 2013 ) ( Figure 1 ) . Both have a Cas7 backbone and a large Cas10 subunit with HD and cyclase domains , but they differ in their structural organisation and nuclease activities ( Zhang et al . , 2016 ) . S . solfataricus Cmr has two distinct target RNA cleavage modes and has not been observed to cleave DNA targets ( Zhang et al . , 2016 ) . In contrast the Csm complex has the canonical Cas7-mediated ‘backbone’ target RNA cleavage activity and a DNA nuclease activity ( Zhang et al . , 2016 ) . The S . solfataricus CRISPR system also has six CRISPR-associated CARF-domain proteins of which one is predicted to be HEPN nucleases ( Sso1389 ) , two belong to the Csa3-like transcription factor family ( Sso1444 and 1445 ) and three have indeterminate function ( Sso1393 , 1397 and 2081 ) ( Figure 1 ) . This suggests that S . solfataricus can mount a sophisticated and wide-ranging antiviral response through cOA signalling . Large-scale changes in gene expression on viral infection in S . solfataricus have been reported ( Quax et al . , 2013 ) , which may be at least partly due to cOA signalling . However , to date , cOA synthesis has not been directly demonstrated in any type III system in the archaea . Recently , a HEPN family Csx1 nuclease from the related organism S . islandicus was shown to be activated by mRNA with a 3’ polyadenylate tail ( Han et al . , 2017b ) , which opens another possible mechanism for activation of this antiviral response , independent of cOA signalling . Furthermore , viral infection induces dormancy in S . islandicus ( Bautista et al . , 2015 ) , which could be consistent with activation of a cOA-dependent ribonuclease or cOA-dependent changes in transcription . Here , we demonstrate that the S . solfataricus Csm/III-D complex generates cOA in response to target RNA binding , activating a CARF-domain nuclease Csx1 for RNA degradation . We describe a new method to generate short oligoadenosine molecules with a 3’-cyclic phosphate moiety and use this to determine that cOA4 is the relevant activator for Csx1 . Activation of cOA synthesis is sensitive to changes at the 3’ end of RNA targets , next to Cas10 . The deactivation of cOA synthesis is shown to correlate with backbone cleavage towards the centre of bound target RNA , leading to rapid product release . cOA synthesis has so far been observed directly in two bacterial type III systems ( Kazlauskiene et al . , 2017; Niewoehner et al . , 2017 ) . To study the system in the archaeon S . solfataricus , we first purified the Csm complex and the CARF nuclease Csx1 ( Sso1389 ) ( Figure 1 ) . To determine whether the Csm complex in S . solfataricus can synthesise cOA , we incubated Csm with target RNA ( A26 ) in the presence of ATP . Using a radioactively labelled target RNA , we first confirmed that the wild-type Csm complex can bind and cleave the target RNA via ‘backbone-mediated’ cleavage by the Cas7 subunit ( Figure 2A ) , as demonstrated previously ( Zhang et al . , 2016 ) . Three cleavage sites ( B1-B3 ) are clearly visible , and a fourth ( B4 ) is very faint . Targeted mutations that abrogate the active site of the HD domain and cyclase domain ( denoted HD and GGDD , respectively ) did not affect this backbone cleavage mode . The presence of 1 mM ATP in these reactions also had little effect on activity . Next , we performed the same reactions shown in Figure 2A but with unlabelled target RNA , phenol-chloroform extracted the products of reactions shown in Figure 2A to deproteinise them and added this solution to a reaction containing purified Csx1 ( Sso1389 ) along with a radioactively labelled RNA substrate ( A1 ) ( Figure 2B ) . Cleavage of the target RNA by Csx1 at five positions ( arrowed ) was only observed with products of reactions that had included ATP as well as Csm with a wild-type cyclase domain . The cleavage sites map to C/U dinucleotides for this substrate ( see Table 1 ) . An active Csm HD domain was not required for stimulation of Csx1 . These data suggest that Csx1 is activated by a small molecule that is synthesised by the cyclase domain of Csm when it is incubated with target RNA in the presence of ATP , consistent with an active cOA-signalling pathway in S . solfataricus . By carrying out the reaction shown in Figure 2A in the presence of α-32P-ATP , followed by separation of reaction products by denaturing gel electrophoresis , we observed synthesis of small radioactive oligomers , dependent on an active Csm cyclase domain , consistent with cOA synthesis by the Csm complex ( Figure 2C ) . Liquid chromatography coupled with mass spectrometry ( LC-MS ) was used to confirm that the major product corresponded to cOA4 , with much lower amounts of other cyclic molecules present ( Figure 2—figure supplement 1 ) . Although cOA4 is the main product generated by the Csm cyclase domain , it is possible that Csx1 is stimulated by a minor product . For example , S . thermophilus Csm mainly generates cOA3 and cOA4 but very little of the cOA6 that is the activator of the cognate Csm6 ( Kazlauskiene et al . , 2017 ) . It is difficult to generate large amounts of cOA of defined ring size by enzymatic means; the type III systems are complex ribonucleoprotein machines , and each generates a mixture of cOA molecules with different ring sizes , often in non-ideal proportions . Linear oligoadenylates can activate CARF-domain nucleases , an effect stimulated by the presence of a cyclic 2’ , 3’-phosphate terminus ( Niewoehner et al . , 2017 ) . However , standard phosphoramidite synthesis is problematic for small oligonucleotides below six nucleotides in length . In order to generate large quantities of defined signalling species , we have devised a convenient method for the generation of linear polyadenylates with a 2’ , 3’ cyclic phosphate – a species that has an identical mass to the corresponding cOA . To do this , we took advantage of the specificity of the MazF toxin of E . coli , which cuts RNA 5’ to an ACA recognition sequence , leaving a 2’ , 3’-cyclic phosphate ( Zhang et al . , 2003 ) . MazF can be expressed at high levels in E . coli in complex with the antitoxin MazE and is activated by proteolysis ( Park et al . , 2012 ) . By feeding recombinant MazF with oligoribonucleotides of sequence 5’- ( A ) n-ACAUCAG , we could generate large quantities of linear analogues of cOAs A3 >P , A4 >P , A5 >P and A6 >P ( where ‘>P’ denotes a 2’ , 3’ cyclic phosphate moiety ) as desired ( Figure 3A ) . To test the utility of this approach , we incubated the Csx1 nuclease Sso1389 with a labelled target RNA substrate in the presence of MazF-derived A3 >P , A4 >P , A5 >P and A6 >P oligoadenylates ( Figure 3B ) . We observed clear activation of the Csx1 ribonuclease activity by A4 >P , comparable to the activation by the authentic cOA mixture generated by the S . solfataricus Csm effector , demonstrating that cOA4 is the relevant activator for this enzyme . To investigate the relationship between target RNA cleavage and cOA synthesis , we used a 205 nt RNA transcript containing the target site for crRNA A26 ( Figure 4 ) . The RNA transcript was labelled during in vitro transcription with α-32P-ATP , allowing detection of any product containing an adenosine . We showed previously that Csm degrades target RNA oligonucleotides by backbone cleavage at up to four positions , B1-B4 ( 28 ) . Under single turnover conditions ( Csm 5 µM , transcript 5 nM ) , Csm-mediated backbone cleavage at sites B1-B3 of the transcript target sequence could be observed both by the processing of the large RNA species from 205 down to 170 nt , and by the appearance of small RNA products corresponding to cleavage at sites B1 and B2 further down the gel ( Figure 4A ) . Since the six nucleotides between sites B2 and B3 lack an adenosine , this product was not visible . We quantified the fractional cleavage of the transcript over time and fitted it to an exponential equation , yielding a rate kc = 0 . 15 ± 0 . 01 min−1 ( Figure 4B ) . More detailed kinetics are described using synthetic oligonucleotide substrates , below . In the same reaction , we monitored cOA production by including α-32P-ATP ( 5 nM ) along with cold ATP ( 0 . 5 mM ) . cOA synthesis was observed from early time points and plateaued over the time course . No more than 10% of the total amount of ATP was consumed in the course of the reaction . The change in cOA4 production over time was calculated by quantifying the amount of cOA4 produced per minute during each time interval and is shown as a bar chart in Figure 4B . As the target site in the transcript RNA becomes fully cleaved between 20 and 40 min , cOA4 production shuts down . Induction of cOA synthesis in response to viral infection induces an anti-viral state with an altered gene expression landscape and activation of non-specific ribonucleases . It follows that the synthesis of cOA should be under tight control . Activation of the cyclase domain of the Csm complex occurs on target RNA binding ( Figure 2 ) , which presumably initiates a conformational change that activates the cyclase domain . To investigate this further , we tested the activation of cOA synthesis using a variety of oligonucleotide target RNA molecules ( Figure 5 ) . Viral-derived target RNA is naturally mismatched with the CRISPR-derived 5’ handle of the crRNA . When the target RNA is complementary to the 5’ handle of the crRNA sequence , or when the 5 nt mismatched 3’-end of target RNA is missing , backbone cleavage still occurs but the cyclase domain is not activated ( Figure 5—figure supplement 1 ) . This confirms previous observations with S . thermophilus Csm ( Kazlauskiene et al . , 2017 ) . We next tested the effect of mismatches by introducing pairs of mismatches at a variety of positions along the target RNA ( Figure 5 ) . All mismatched targets were still cleaved by Csm , in keeping with previous observations ( Staals et al . , 2014; Kazlauskiene et al . , 2016 ) . Target MM1 and MM2 , with two mismatches spanning cleavage site B1 and B2 respectively , abolished backbone cleavage at the mutated position , but cleavage at other sites was still observed . Single mismatches flanking site B1 reduced but did not abolish backbone cleavage at site B1 ( MM1a and 1b ) , whilst the double mismatch adjacent to site B1 ( MM0 . 5 ) reduced cleavage at that site very significantly , suggesting local perturbation . None of the double mismatches introduced 5’ to site B1 on the target RNA reduced cOA synthesis , showing that they behave very similarly to fully cognate target RNA . In contrast , a double mismatch positioned two nt ( MM0 . 5 ) or four nt ( MM1 ) from the 3’ end of the matching target RNA resulted in complete abolition of cOA synthesis . Together , these data suggest that Csm is particularly sensitive to disruptions of targets at the 3’ end nearest the Cas10 subunit , whether by truncation , base-pairing with the crRNA 5’-handle , or mismatch ( discussed later ) . Two possibilities for deactivation of the cyclase activity are when target RNA is cleaved , or when cleaved target RNA is released from the Csm complex . To measure the rates of target RNA cleavage and product release under pseudo-single turnover conditions we incubated an excess of Csm with an end-labelled target RNA oligonucleotide to follow target cleavage as a function of time . Three backbone cleavage sites ( B1-B3 ) were clearly visible on gels ( Figure 6A ) ( Zhang et al . , 2016 ) . We quantified the rates of cleavage at these three sites in triplicate experiments ( Figure 6B ) , normalised to aid comparison – this normalisation does not change the rates measured . Site B1 , which is the closest site to the Cas10 subunit , was cleaved with a rate constant kB1 of 0 . 19 min−1 under these conditions . The next backbone cleavage site , B2 , was cut about half as fast ( kB2 = 0 . 077 min−1 ) . The third site , distal to Cas10 , was cleaved at the slowest rate ( kB3 = 0 . 004 min−1 ) . The rate of cleavage at sites B1 and B2 thus corresponds very well with that observed for the transcript substrate ( Figure 4 ) . Faster target RNA cleavage at sites closer to the Cas10 domain has been observed qualitatively for other type III systems ( Staals et al . , 2014; Tamulaitis et al . , 2014 ) , and may be a general phenomenon . Having defined the kinetics of target RNA cleavage , we investigated the related cOA synthesis activity of Csm by feeding the reaction with α-32P-ATP to monitor the synthesis of cOA species . These experiments were carried out under the exact same conditions as those used to determine RNA cleavage ( in fact they can be run simultaneously with labelled oligonucleotide target RNA and α-32P-ATP ) . The amount of cOA4 generated per minute during each interval following the initiation of the reaction was quantified in triplicate and is shown in Figure 7A . The rate of cOA4 synthesis peaked at 10 min and rapidly fell away , suggesting that the cyclase domain is rapidly activated on target RNA binding , and quickly deactivated after target RNA cleavage . To investigate this further , we monitored cOA4 synthesis by spiking radioactive ATP into a reaction containing Csm , target RNA and 0 . 5 mM cold ATP at three different time points ( 0 , 20 and 40 min ) after initiation of the reaction and followed cOA4 synthesis for a further 40 min ( Figure 7B ) . Under single turnover conditions with 25 nM target RNA , we observed very low amounts of cOA4 production after 20 min of reaction . When the target RNA was increased ten-fold , this allowed cOA production to continue for a longer period . Thus , cOA synthesis is closely linked to target RNA availability , and is rapidly deactivated once target RNA is cleaved . Oligonucleotides with phosphorothioate ( P-thioate ) bonds , where a non-bridging oxygen of the phosphate linkage is replaced by sulfur , show enhanced resistance to a variety of nucleases . We therefore explored the effect of introduction of three consecutive phosphorothioate linkages centred on cleavage sites B1 and B2 ( Table 1 ) . The P-thioate target RNA was still cleaved by Csm , but the rate was significantly slower ( Figure 8 ) . In particular , a defect in cleavage at site B2 was apparent . By adding radioactive α-ATP to the reaction , we were able to monitor cOA4 simultaneously . This revealed that cOA4 production was strongly enhanced in the reaction with P-thioate target RNA , consistent with a requirement for cleavage and/or dissociation of cleaved target RNA to switch off the activity of the Cyclase domain . In the cell , target RNA cleavage by Csm may be rate-limited by the rate of product release , rather than the chemical step of a cleavage reaction ( Kazlauskiene et al . , 2016 ) . Conceivably , cleaved target RNA that remained bound to Csm could continue to activate the cyclase domain . We therefore quantified the rate of dissociation of cleaved target RNA by carrying out a cleavage reaction over time and running the products in a native gel . RNA remaining bound to Csm was held up in the wells whilst released RNA migrated through the gel . Dissociated , radioactively labelled target RNA cleavage products were trapped by the addition of a cold DNA oligonucleotide of complementary sequence to generate a RNA:DNA heteroduplex that migrated at a defined position in the gel ( Figure 9A ) . The DNA trap is complementary to the target RNA from the 5’ end to roughly the site of B3 cleavage ( Table 1 ) . The dissociated RNA product built up over the time course of the reaction and was quantified in triplicate experiments . The rate of cleaved target RNA release was determined as koff = 0 . 066 ± 0 . 003 min−1 ( Figure 9B ) , similar to the rate of cleavage of site B2 in the centre of the target RNA ( kB2 = 0 . 077 min−1 ) , significantly faster than the cleavage at site B3 . The site of cleavage of the main dissociated product was confirmed as B2 using a synthetic RNA marker of that size ( Figure 9—figure supplement 1 ) . Thus , the clearance of target RNA from the Csm complex may be governed by the rate of backbone cleavage rather than product dissociation . Furthermore , cleavage up to the centre of the target RNA ( site B2 ) appears sufficient to allow product dissociation . We also analysed the rate of synthesis of cOA4 over time ( as shown in Figure 7 ) , setting the rate at 10 min equal to one and normalising the data . The data fit a simple exponential with a rate constant kobs = 0 . 047 ± 0 . 008 min−1 . In conclusion , the rates of target RNA cleavage at site B2 , RNA product dissociation and inactivation of cOA synthesis were very similar , within experimental error . When cleavage at site B2 is perturbed by phosphorothioate modification , cOA production is enhanced significantly . These observations are consistent with activation of the cyclase domain quickly upon target RNA binding , followed by deactivation as target RNA is cleaved and dissociates from the Csm complex . We have demonstrated that the type III-D Csm complex from S . solfataricus generates cOA on binding target RNA . Given the preponderance of CARF-family proteins associated with type III systems and the conservation of the cyclase domain , this is likely to be a general phenomenon . The major product of cOA synthesis by this complex is cOA4 , with much smaller amounts of other cyclic and linear oligoadenylates observed . For the S . solfataricus system , the generation of cOA4 fits with the binding pocket modelled for the CARF domain of the transcription factor Csa3 ( Lintner et al . , 2011 ) , and with the observation that OA4 activates the CARF nuclease Csx1 in the related organism S . islandicus ( Han et al . , 2017b ) . Enterococcus italicus Csm generates mostly cOA6 , and this molecule is the effector for the corresponding Csm6 enzyme ( Niewoehner et al . , 2017 ) . In contrast , the major product generated by the S . thermophilus enzyme is cOA3 , despite the fact that the corresponding Csm6 protein is activated by cOA6 ( Kazlauskiene et al . , 2017 ) . Given that CARF domains are formed by dimeric subunits , the corresponding activators are likely to have a 2-fold axis of symmetry , which fits with observations that cOA4 and cOA6 are both activators of CARF nucleases . In contrast , cOA3 and cOA5 lack this symmetry and are unlikely to bind to CARF domains . They may be irrelevant in the context of signalling by type III systems , or alternatively may be activators for families of effector proteins that utilise a different domain for ligand binding . To circumvent the difficulty in generating defined activators of CARF domain proteins , we have developed a facile technique that takes advantage of the specificity of the E . coli toxin MazF , which cleaves RNA 5’ to an ACA sequence , generating a 3’ cyclic phosphate group . Linear analogues of cOA can be generated with the same mass and charge as the cyclic equivalents in a cost effective and scalable manner and be radioactively labelled if required . This approach could prove particularly useful for smaller molecules such as OA4 , where phosphoramidite synthesis is problematic . Target RNA binding has been shown to activate both the HD nuclease and cOA synthesis activities of diverse type III systems . This is likely to result from conformational changes in the complex , and the Cas10 subunit in particular . A 5° rotation of the Cas10 subunit on target RNA binding has been observed for T . thermophilus Cmr ( Taylor et al . , 2015 ) , which may reflect this activation . We observe that the activation of cOA synthesis is rapid , with activity detected within the first minute of incubation of Csm with target RNA transcripts . As for S . thermophilus Csm , activation requires a 3’ extension to the target RNA that does not base-pair with the 5’-handle of the crRNA ( Figure 5 and [Kazlauskiene et al . , 2017] ) . In our standard oligonucleotide substrates there is a 5 nt 3’- overhang , which fits the minimum requirement predicted from the biochemical and modelling study of the Siksnys group ( Kazlauskiene et al . , 2016 ) . This short stretch of RNA presumably interacts with the Cas10 and/or the Csm4 subunit in such a way that a conformational change occurs in this region of the complex , resulting in the activation of both the HD nuclease and GGDD cyclase active sites , paving the way for DNA degradation and , through cOA signalling , the activation of HEPN family ribonucleases . As observed for other type III systems , there is a high degree of tolerance for mismatches in target RNAs ( Pyenson et al . , 2017; Manica et al . , 2011 , 2013; Goldberg et al . , 2018 ) , which all tend to be cleaved , and thus removed from the enzyme . By extending this analysis to examine cOA synthesis , we observe that 2 nt mismatches 5’ to site B1 do activate the cyclase domain and would thus elicit a full anti-viral response . However , the target RNAs with a double mismatch at or 3’ to site B1 ( corresponding to positions within 4–5 nt of the 3’ end of target RNA base-paired with the crRNA ) , whilst still cleaved , did not activate cOA production ( Figure 5 ) . This suggests an important role for the 3’ end of the bound target RNA close to the Cas10 subunit where sequence requirements are more stringent . Presumably , destabilization of the RNA duplex in this region can lead to changes in the conformation of the single-stranded 3’ end of the target RNA , resulting in failure to activate the cyclase domain . Viral targets that manage to mutate in this key area may thus have more chance of escaping from cOA-mediated HEPN nuclease activation , which is a key determinant of type III CRISPR immunity . In contrast , substantial blocks of mismatches in this general area do not prevent phage targeting by the S . epidermidis type III-A system ( Pyenson et al . , 2017 ) . This could be due to differences between the activation of the cyclase and HD nuclease domains of Cas10 , or to differences between type III-A and III-D systems and deserves further analysis . The degree of specificity of the HEPN family nucleases against viral targets is still a matter for conjecture . The Marraffini group report that Csm6 selectively targets phage transcripts ( Jiang et al . , 2016 ) , and the spatial coupling of the HD nuclease activity of Cas10 to actively transcribing viral RNA targets could ensure at least partial selectivity for foreign nucleic acids . Nonetheless , it seems likely that host nucleic acids would be caught in the cross-fire of this ‘scorched earth’ attack , leading to a shutdown in host gene expression or even cell death – which , as pointed out previously , could be a sensible option if viral infection is established in the cell ( Elmore et al . , 2016 ) . Indeed , viral infection has been shown to result in dormancy and cell death in S . islandicus , which harbours a type III CRISPR system ( Bautista et al . , 2015 ) . The activation of the HD nuclease and cyclase domains of Cas10 as a result of target RNA binding is not an irreversible event . Logically , it could be predicted that these domains are deactivated either when target RNA is cleaved , or when it dissociates from the complex , with a time lag possible in either circumstance . Backbone cleavage of RNA was the first activity detected in type III CRISPR systems ( Hale et al . , 2009 ) , and appears to be ubiquitous . Nonetheless , this activity is not essential for type III-mediated immunity against phage infection in vivo ( Samai et al . , 2015 ) . Terns and colleagues suggested that target RNA cleavage may represent a means for deactivation of the DNA nuclease activity of type III systems , as cleavage of the RNA transcript would cause the type III effector to ‘drift away’ ( Elmore et al . , 2016 ) . Rather than spatial separation though , it seems more likely that target RNA cleavage and subsequent dissociation allows a collapse back to an inactive ground state of Cas10 , switching off the highly potent HD nuclease and cOA signalling activities . The Csm ( III-A ) complex from S . thermophilus and Thermus thermophilus also appear to cut target RNA first at the 3’ end nearest the Cas10 subunit ( Staals et al . , 2014; Tamulaitis et al . , 2014 ) , and cleavage by S . thermophilus Csm is rapid – of the order of 5 s under single turnover conditions ( Kazlauskiene et al . , 2016 ) . Thermatoga maritima Cmr ( III-B ) cleaves target RNA with six nt spacing at four sites , with cleavage fastest at site 2 , which is 11 nt from the 3’ end of the target RNA . Cleavage is rapid under single turnover conditions , reaching completion within 1 min for site 2 and 5 min for site 4 , which is 23 nt from the 3’ end ( Estrella et al . , 2016 ) . The rate of product release has not previously been quantified in any system . For S . thermophilus Csm , multiple turnover kinetic experiments suggested that product dissociation , whilst rate limiting , was moderately fast ( k = 3 min−1 ) . However , activation of the DNase activity of the HD domain persisted over a much longer timescale , 30–60 min . The authors suggested that the activated state of S . thermophilus Csm persists after target RNA dissociation ( Kazlauskiene et al . , 2016 ) , but an alternative is that there is heterogeneity in the population of Csm enzymes , with some remaining bound to RNA targets and hence preserved in an active state for much longer periods than others . For T . maritima Cmr , cleaved target RNA dissociation also correlated with the deactivation of the HD nuclease domain , although the rates were not measured ( Estrella et al . , 2016 ) . Many of these uncertainties arise from the difficulty inherent in quantifying nuclease activity in these systems . To delineate the roles of backbone cleavage and RNA dissociation in the control of cOA synthesis , we carried out a detailed kinetic analysis , quantifying the rates of target RNA cleavage and subsequent dissociation ( Figure 10 ) . We observed comparatively rapid cleavage at sites B1 ( 0 . 19 min−1 ) and B2 ( 0 . 08 min−1 ) , nearest to the 3’ end of the target RNA and proximal to the Cas10 subunit , and much slower cleavage of more distal sites . The trend for faster cleavage at the 3’ end is clear from studies of other type III systems . The overall cleavage rate of a transcript RNA , which is a more physiologically relevant target , was very similar ( Figure 4 ) . The introduction of phosphorothioate linkages spanning sites B1 and B2 reduced but did not abolish target RNA cleavage ( Figure 8 ) , and we noted a highly significant increase in cOA production with these substrates , consistent with a link between target RNA cleavage and cOA synthesis . By trapping the dissociated RNA we could quantify the rate of product release directly for the first time for a type III system . We determined a rate of 0 . 07 min−1 for dissociation of the 5’ end of the target RNA – a very close match to the rate of cleavage at site B2 in the centre of the RNA . Furthermore , the decrease in the rate of cOA synthesis by activated Csm over time also fitted an exponential with a rate of 0 . 05 min−1 , ten times faster than the rate of cleavage at site B3 . Thus , it appears that full target RNA digestion may not be required for product release . Furthermore , RNA dissociation is likely not rate limiting . Rather , the chemical step of RNA cleavage determines the rate of product release . As our assays were carried out at 70°C , close to the growth temperature of S . solfataricus , RNA duplexes of around 20 bp would be only marginally stable in the absence of stabilising protein interactions . This fits the situation in S . islandicus , where target RNA forming less than 20 bp with the crRNA was observed to dissociate ( Han et al . , 2017a ) . It is also consistent with the general observation for type III systems that target RNA cleavage products tend to accumulate under single turnover conditions rather than undergoing processive cleavage to the smallest possible products ( Staals et al . , 2014; Tamulaitis et al . , 2014; Hale et al . , 2014; Han et al . , 2017a; Zhang et al . , 2016 ) . Finally , RNA cleavage and dissociation results in the simultaneous deactivation of the cyclase domain , presumably due to rapid conformational change to the ground state . These observations are consistent with the hypothesis that target RNA cleavage ( and therefore dissociation ) in type III systems functions as an ‘off switch’ for the cyclase and HD nuclease active sites ( Estrella et al . , 2016 ) . This , rather than the protective effect of direct degradation of viral transcripts , may be the primary role of target RNA cleavage . After all , a single target RNA may trigger synthesis of many molecules of cOA , which in turn would activate multiple HEPN ribonucleases for active RNA degradation – a potent amplification of antiviral defences . Under multiple turnover conditions that may be more reflective of a high viral load in the cell , the cyclase domain is activated for longer periods ( Figure 7 ) . RNA substrates that are not bona fide targets are still cleaved and thus cleared from the binding site , but do not activate cOA synthesis , allowing the Csm complex to seek other targets . Several studies have reported that the HD nuclease activity of type III systems persists up to 1 hr , despite target RNA cleavage reaching completion within a few minutes ( Tamulaitis et al . , 2014; Estrella et al . , 2016; Kazlauskiene et al . , 2016 ) . This raises the possibility that the HD nuclease and cyclase domains experience different half-lives in the activated state , with the former persisting for much longer . It could make sense to switch off the cyclase activity quickly , as the resultant cOA would provide a much more persistent activation of CARF-family nucleases and transcription factors . A systematic study of the temporal control of HD nuclease activity is required to resolve this question . Ultimately , the removal of the cOA second messenger by an as-yet unidentified phosphodiesterase is likely to play an important part in the control of type III systems . This should be a priority for future study . The Csm ( III-D ) complex ( wild-type , HD and GGDD variants ) was expressed and purified from S . solfataricus as described previously ( Zhang et al . , 2016 ) . This involves expression of a his-tagged subunit from an arabinose-inducible vector in S . solfataricus , followed by purification of the native complex with the tagged subunit incorporated . Variants such as the HD and GGDD targeted changes can thus be introduced on the tagged subunit . The resultant protein has a mixture of all the crRNAs loaded into the Csm complex , with abundant crRNAs such as A26 present at 1–2% of the total crRNA ( Rouillon et al . , 2013 ) . A target RNA ( target RNA A26 and derivatives ) complementary to crRNA A26 is used in the assays described . Sso1389 ( encoding Csx1 ) was purchased as a synthetic gene from Integrated DNA Technologies , Coralville , IA , United States ( IDT ) and cloned into the pEHisTEV vector ( Liu and Naismith , 2009 ) . The inactive Csx1 variant H345N was generated using the QuikChange Site-Directed Mutagenesis kit as per manufacturer’s instructions ( Agilent technologies ) . The pEHisTEVCsx1 and pEHisTEVCsx1H345N constructs were transformed into C43 ( DE3 ) E . coli cells . Protein expression was induced with 0 . 4 mM isopropyl-β-D-1-thiogalactoside ( IPTG ) at an OD600 of ~0 . 6 and grown for 4 hr at 25°C . Cells were harvested and resuspended in lysis buffer containing 50 mM Tris-HCl pH 7 . 5 , 0 . 5 M NaCl , 10 mM Imidazole and 10% glycerol , and lysed by sonicating six times 2 min on ice with 2 min rest intervals . Csx1 was purified with a 5 ml HisTrapFF column ( GE Healthcare ) , washing with five column volumes ( CV ) of buffer containing 50 mM Tris-HCl pH 7 . 5 , 0 . 5 M NaCl , 30 mM Imidazole and 10% glycerol , and eluting with a linear gradient of buffer containing 50 mM Tris-HCl pH 7 . 5 , 0 . 5 M NaCl , 0 . 5 M imidazole and 10% glycerol across 15 CV . Size exclusion chromatography was used to further purify Csx1 , eluting protein with buffer containing 20 mM Tris-HCl pH 7 . 5 , 0 . 5 M NaCl , 1 mM EDTA and 1 mM DTT . Csx1 was concentrated using a centrifugal concentrator , aliquoted and frozen at −80°C . The pET21c-mazEF-Fxa construct was a kind gift from Professor Masayori Inouye ( Robert Wood Johnson Medical School , New Jersey ) . pET21c-mazEF-Fxa was transformed into BL21 ( DE3 ) E . coli cells and expressed as previously published ( Park et al . , 2012 ) , except cells were grown at 25°C for 4 hr after induction with 0 . 5 mM IPTG at OD600 of ~0 . 6 . Cells were harvested by centrifugation at 4000 x g at 4°C for 15 min , suspended in buffer A ( 10 mM Tris-HCl , pH 7 . 5 and 150 mM NaCl ) and lysed by sonicating six times 2 min on ice with 2 min rest intervals . MazEF was purified with a 5 ml HisTrapFF column , washing with 20 CV buffer A and 20 CV 2% buffer B ( 10 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 1 M imidazole ) prior to elution with a linear gradient of buffer B across 15 CV . Size exclusion chromatography ( Sepharose200 26/600; GE Healthcare ) was used to further purify MazEF , eluting protein with buffer A . MazEF was concentrated using a centrifugal concentrator , aliquoted and frozen at −80°C . RNA oligonucleotides were ordered from IDT . Before assays , oligonucleotides were 5′-32P-radiolabelled and purified by denaturing ( 7M urea ) polyacrylamide ( 20% ) gel electrophoresis with 1 x Tris-borate-EDTA ( TBE ) buffer , followed by band excision , gel extraction , ethanol precipitation , as described previously ( Rollie et al . , 2015 ) . The RNA transcript substrate was generated using as template a HindIII linearised pUC19 vector containing a cloned DNA sequence ( between BamHI-HindIII ) harbouring a target sequence ( in bold ) recognised by Csm crRNA A26 and a T7 phage polymerase promoter . Transcripts were obtained using the MEGAscript kit ( Ambion by Life Technologies ) according to the manufacturer instructions . 32P-α-ATP was added to the cold NTPs mix for further visualisation . Transcripts were finally purified using a G50 column ( illustra Dye terminator removal kit-GE Healthcare ) before use . For kinetic analysis , cleaved and uncleaved species were quantified using the Bio-Formats plugin ( Linkert et al . , 2010 ) of ImageJ ( Schneider et al . , 2012 ) as distributed in the Fiji package ( Schindelin et al . , 2012 ) and fitted to a single exponential curve using Kaleidagraph ( Synergy Software ) , as described previously ( Sternberg et al . , 2012 ) . For the transcript cleavage , a single experiment was quantified , but replicates were qualitatively similar . For the target A26 cleavage , the cleavage rate determined for site B1 could be a slight underestimate , as cleavage of B1 products at site B2 would decrease the B1 signal . This appears to be a minor effect however ( see discussion ) . The sequence of the DNA fragment used for the transcript generation , with the A26 target sequence in italics , is shown below: 5’GGATCCTAATACGACTCACTATAGGATCAGATCATATCAGCTACATCGACAGGGTATTATTTGTTTGTTTCTTCTAAACTATAAGCTAGTTCTGGAGAGCATTAGCATGTAGAGGGTACAGTTTGGGTATTGCCGTTCTGGTCCTTATACGAAATGGAGATCGATTCTCGAGAGGGTCGTTGTTAAGAACGACGTTGTTAGAAGTTGGGTATGGTGGAGATGGAAGCTT The assay of target RNA cleavage by Csm was carried out as described previously ( Zhang et al . , 2016 ) unless specified . Briefly , the reaction was carried out under single turn-over conditions with 5 µM of CSM and 25 nM labelled RNA target in a buffer containing 20 mM MES pH 6 . 0 , 100 mM NaCl , 0 . 1 mg/ml BSA , 2 mM MgCl2 and 0 . 5 or 1 mM ATP as mentioned in the text . The reaction was incubated at 70°C for the specified times . Following the reaction , products were phenol-chloroform extracted and run in 50% formamide on a denaturing gel ( 20% acrylamide , 7M Urea , 1X TBE ) to observe backbone cleavage . Alternatively , the same reaction products were chilled in a stop solution containing 2 mM EDTA and 10 µM of DNA trap before adding Ficoll and running on a native gel ( 12% acrylamide containing 50 mM of NaCl and 1X TBE ) . Following electrophoresis , gels were imaged by phosphorimaging as described previously ( Zhang et al . , 2016 ) . For the experiments comparing unmodified and phosphorothioate-containing target RNA , the target RNA cleavage reactions were performed as described above in presence of α-32P-ATP to visualise the formation of cyclic oligoadenylate . Generation of cyclic oligoadenylate ( cOA ) activator was generated by incubating 120 µg S . solfataricus Csm complex with 0 . 5 mM ATP , 1 mM MgCl2 and 100 nM target RNA in Csx1 buffer for 2 hr at 70°C . Reaction product was isolated by phenol-chloroform extraction followed by chloroform extraction and frozen at −20°C . For the experiments where the cOA production was observed over time , 2 nM of 32P-α-ATP was added ( in addition to 1 or 0 . 5 mM of cold ATP ) in the reaction to allow visualisation of the generated product on denaturing gel electrophoresis and phosphorimaging . In the case of the spiking experiment , 32P-α-ATP was added to a final concentration of 2 nM in the reaction ( at 0 , 20 or 40 min ) . The quantification of free 32P-α-ATP versus 32P-α-ATP incorporated in cOA allowed determination of a total ATP ratio used in cyclase reaction that is lower than 10% ( 50 µM ) for the longest time points . MazF was used to generate linear oligoadenylate . Active MazF was initially obtained by incubating 1 mg of MazEF with 0 . 1 units Factor X ( Sigma-Aldrich ) activated in FXa buffer containing 10 mM Tris-HCl pH 8 . 0 and 1 mM DTT for 3 hr at 37°C . MazEF digestion with bovine trypsin ( Promega ) ( 1600:1 ratio ) was found to yield a similar level of active MazF when incubated in the same buffer for 15 min . Therefore , MazF was predominantly generated by trypsin digestion and used immediately . Linear oligoadenylate was generated by incubating MazF with 30 µM A3 , A4 , A5 or A6 RNA ( Table 1 ) in FXa buffer for 2 . 5 hr at 37°C . Products were phenol-chloroform extracted , and then chloroform extracted and frozen at −20°C . Denaturing polyacrylamide gel electrophoresis ( 20% acrylamide , 7M urea and 1x TBE ) followed by phosphorimaging of MazF-cleaved 32P end-labelled A3 , A4 , A5 and A6 RNA was used to assess successful generation of intended linear OA . Csx1 and Csx1 H345N mutant were diluted in buffer containing 20 mM Tris-HCl pH 8 . 0 , 0 . 5 M NaCl , 1 mM DTT and 1 mM EDTA . 250 nM of Csx1 dimer was incubated with 0 . 01% ( v/v ) cOA activator or 3 µM linear oligoadenylate ( A3 , A4 , A5 or A6 ) and 50 nM 32P-5’-labelled substrate RNA in Csx1 buffer containing 20 mM MES pH 5 . 5 , 100 mM K-glutamate and 1 mM DTT . Two control reactions , both containing only 50 nM RNA and buffer were incubated at 4°C or 50°C , respectively . All other reactions were incubated at 50°C and quenched at 5 , 15 or 30 min by the addition of a reaction volume equivalent of 100% formamide prior to freezing at −20°C . RNA cleavage by Csx1 was visualised by phosphorimaging following denaturing gel electrophoresis . Csx1 is stable and active at 70°C , but the choice of 50°C reduced background substrate RNA cleavage . Liquid chromatography-high resolution mass spectrometry ( LC-HRMS ) analysis was performed on a Thermo Scientific Velos Pro instrument equipped with HESI source and Dionex UltiMate 3000 chromatography system . Compounds were separated on a Kinetex 2 . 6 µm EVO C18 column ( 2 . 1 × 100 mm , Phenomenex ) using a linear gradient of 2–15% acetonitrile against 20 mM ammonium bicarbonate pH 8 ( gradient start delay 5 min , gradient length 28 min ) at a flow rate of 350 µl min−1 and column temperature of 40°C . Data were acquired on the FT mass analyzer in negative ion mode with scan range m/z 150–1500 . Source voltage was set to 3 . 5 kV , capillary temperature was 350°C , and source heater temperature was 250°C .	The gene editing tool often known simply as CRISPR has become well known in recent years . Its potential applications are wide ranging , including uses in research , healthcare and agriculture . Yet , the CRISPR system originated in microbes where it helps to protect them from viral infections . Viruses infect by inserting their own genes into a host cell , and – almost like a pair of scissors – the CRISPR system can cut up the virus’s DNA to stop infections . CRISPR experts know the popular form of CRISPR as type II , but there are others . Type III CRISPR is less useful as a genetic tool but does also protect microbes from viruses . In addition to targeting DNA , type III CRISPR targets the related RNA molecules from viruses . When it encounters RNA from a virus , the type III CRISPR produces a small molecule called cyclic oligoadenylate ( or cOA for short ) . The cOA molecule activates enzymes known as non-specific ribonucleases , which can destroy all the RNA in the cell . This defence is a less subtle than that provided type II CRISPR and can also damage the cell by destroying other RNA molecules that the microbes use to survive . As such , proper regulation is essential to prevent the type III system from unnecessarily killing the infected cell . Rouillon et al . studied the control of the type III CRISPR system from the heat-loving microbe Sulfolobus solfataricus , which is found in volcanic springs . This species has been a model for studies of the CRISPR system for many years , in part because its proteins are very stable which makes them easier to work with in the laboratory . The results show that the type III CRISPR makes cOA by combining four molecules of adenosine triphosphate ( ATP ) into a ring . CRISPR responds immediately to viral RNA in the cell . It also detaches from the RNA as soon as it starts to be destroyed . Rapid activation and silencing of the production cOA ensures that the CRISPR system is tightly controlled . These findings reveal that cOA production is tightly linked to the abundance of viral RNA , ensuring a proportional and timely response to infection . Using cOA amplifies the cell's response because it allows a single RNA molecule to activate a larger change . Type III CRISPR systems are widespread in nature , and a better understanding of them could improve the yield of products , like yoghurt , that depend on healthy bacteria; currently viruses cause a lot of economic damage in this industry . Further research in this area could also lead to new antibiotics that over-activate type III CRISPR to destroy bacterial cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2018	Control of cyclic oligoadenylate synthesis in a type III CRISPR system
Platelets are anucleate cells in blood whose principal function is to stop bleeding by forming aggregates for hemostatic reactions . In addition to their participation in physiological hemostasis , platelet aggregates are also involved in pathological thrombosis and play an important role in inflammation , atherosclerosis , and cancer metastasis . The aggregation of platelets is elicited by various agonists , but these platelet aggregates have long been considered indistinguishable and impossible to classify . Here we present an intelligent method for classifying them by agonist type . It is based on a convolutional neural network trained by high-throughput imaging flow cytometry of blood cells to identify and differentiate subtle yet appreciable morphological features of platelet aggregates activated by different types of agonists . The method is a powerful tool for studying the underlying mechanism of platelet aggregation and is expected to open a window on an entirely new class of clinical diagnostics , pharmacometrics , and therapeutics . Platelets are non-nucleated cells in blood whose principal function is to stop bleeding by forming aggregates for hemostatic reactions ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005 ) . In addition to their participation in physiological hemostasis ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005 ) , platelet aggregates are also involved in pathological thrombosis ( Davì and Patrono , 2007; Ruggeri , 2002 ) . Moreover , it is known that a range of diseases or medical conditions , such as inflammation , atherosclerosis , and cancer metastasis , are closely associated with platelet aggregation ( Lievens and von Hundelshausen , 2011; Engelmann and Massberg , 2013; Franco et al . , 2015; Gay and Felding-Habermann , 2011 ) . Also , in patients with artificial lungs due to severe respiratory failure such as the coronavirus disease 2019 ( COVID-19 ) caused by severe acute respiratory syndrome coronavirus 2 ( SARS-CoV-2 ) ( Ramanathan et al . , 2020; Ronco et al . , 2020 ) , the long-term foreign body contact of blood with the artificial devices in the extracorporeal circulation often leads to platelet aggregation and thrombus formation followed by serious complications ( e . g . , myocardial infarction , cerebral infarction ) ( Brodie et al . , 2019; Brodie and Bacchetta , 2011; Oliver , 2009 ) . Here , the aggregation of platelets is elicited by a variety of agonists , which bind to and activate specific receptors expressed on the platelet . This leads to platelet activation and structural and functional changes of glycoprotein IIb/IIIa expressed on the platelet surface . The activated form of the glycoprotein can bind with fibrinogen , enabling platelets to interact with each other and form aggregates ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005; Moser et al . , 2008 ) . Despite the existence of diverse agonist types , platelet aggregates look morphologically similar and have long been thought indistinguishable since the discovery of platelet aggregates in the 19th century ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005 ) . This is because morphological characteristics of platelet aggregates on a large statistical scale have been overlooked as microscopy ( a high-content , but low-throughput tool ) has been the only method to examine them ( Finsterbusch et al . , 2018; Nitta et al . , 2018 ) . In this Short Report , we present an intelligent method for classifying platelet aggregates by agonist type . This is enabled by performing high-throughput imaging flow cytometry of numerous blood cells , training a convolutional neural network ( CNN ) with the image data , and using the CNN to identify and differentiate subtle yet appreciable morphological features of platelet aggregates activated by different types of agonists . Our finding that platelet aggregates can be classified by agonist type through their morphology is unprecedented as it has never been reported previously . The information about the driving factors behind the formation of platelet aggregates is expected to lead to a better understanding of the underlying mechanism of platelet aggregation and open a window on an entirely new class of clinical diagnostics , pharmacometrics , and therapeutics . Our procedure for developing an intelligent platelet aggregate classifier ( iPAC ) is schematically shown in Figure 1A . First , a blood sample obtained from a healthy person was separated into several different portions , into which different types of agonists were added to activate platelets while no agonist was added to the last portion for negative control ( Figure 1—figure supplement 1; Materials and methods ) . Here , adenosine diphosphate ( ADP ) , collagen , thrombin receptor activator peptide-6 ( TRAP-6 ) , and U46619 were used since they are commonly used in platelet aggregation tests ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005 ) . Initially , the concentrations of the agonists were carefully chosen ( 20 µM for ADP , 10 µg/mL for collagen , 13 µM for TRAP-6 , 14 µM for U46619 ) to minimize variations in aggregate size between the different blood sample portions . These images were acquired through six experimental trials ( Figure 1—figure supplement 2 ) to mitigate potential bias in the dataset that may have come from experimental variations ( e . g . , signal-to-noise ratio , fluctuations in optical alignment , hydrodynamic cell focusing conditions , sample preparation ) . Then , four different concentrations of each agonist ( 2 , 5 , 10 , 20 µM for ADP , 1 , 5 , 10 , 20 µg/mL for collagen , 1 , 5 , 13 , 20 µM for TRAP-6 , 2 . 8 , 5 . 6 , 14 , 28 µM for U46619 ) were used for platelet activation to examine the potential influence of agonist concentrations on the ability to differentiate platelet aggregates by agonist type , where the concentrations were chosen by referring to the concentrations of agonists used in light transmission aggregometry and in vitro flow-cytometric platelet aggregation tests ( Koltai et al . , 2017; Granja et al . , 2015 ) . The platelet aggregates were enriched by density-gradient centrifugation to remove erythrocytes from the blood sample portions . To prevent the platelet aggregates from dissolving during imaging flow cytometry , 2% paraformaldehyde was added to the blood sample portions to fix them . In addition to this sample preparation procedure , we tested other procedures such as pipetting , vortexing , fixation , and non-fixation and identified the current procedure to be advantageous over the others in preserving the morphology of platelet aggregates ( Figure 1—figure supplement 3; Materials and methods ) . Second , an optofluidic time-stretch microscope ( Goda et al . , 2009; Jiang et al . , 2017; Lei et al . , 2018; Lau et al . , 2016 ) was employed for high-throughput , blur-free , bright-field image acquisition of events ( e . g . , single platelets , platelet-platelet aggregates , platelet-leukocyte aggregates , single leukocytes , cell debris , remaining erythrocytes ) in each sample portion ( Figure 1—figure supplements 4 and 5; Materials and methods ) . Here , fluorescence image acquisition is not needed because fluorescence images of platelet aggregates would look very similar to their bright-field images ( except for the colors ) . Third , the acquired images of the events were used to train two CNN models that classified the platelets based on their morphological features by agonist type ( Figure 1B ) . Specifically , we first trained a CNN model with images of platelet aggregates activated by certain concentrations of agonists ( 12 , 000 images per agonist type ) in order to examine their morphological changes while minimizing a potential influence of concentration-dependent factors on the morphology of the platelet aggregates . Then , we trained the other CNN model with a dataset in which the images of platelet aggregates activated by different concentrations of the agonists were equally mixed ( 12 , 000 images in total per agonist type ) in order to show that different concentrations of the agonists do not perturb the CNN model’s ability to classify platelet aggregates . We employed the CNN ( Krizhevsky et al . , 2012 ) with an encoder-decoder architecture to disregard insignificant features such as background noise and keep important features in the bottleneck layer and trained it with the data of a single blood donor to ensure that only the morphological features driven by the agonists contributed to the development of the iPAC ( Figure 1C; Materials and methods ) . In comparison , we measured the platelet samples that were prepared under the same procedure using a conventional flow cytometer ( Cytomics FC500 , Beckman Coulter ) which is based on fluorescence measurements for cell classification . As shown in Figure 2 , the flow cytometer was not capable of differentiating them as indicated by their significant overlap ( Figure 2—source data 1; Materials and methods ) . The iPAC is manifested as a confusion matrix with each row representing the examples in a predicted class and each column representing the examples in an actual or true class . As shown in Figure 3A , most of the images were classified into the correct groups in the diagonal line of the confusion matrix . Large separations between the different platelet sample portions in Figure 3B that visualizes the bottleneck layer in the CNN indicate the first CNN model’s ability to discriminate various types of agonist-activated platelet aggregates and negative control ( Figure 3—source data 1 ) . The negative control shows the highest classification accuracy , indicating that large morphological changes were made to the activated platelets . The U46619-treated blood sample portion shows the second highest classification accuracy of all the blood sample portions , indicating that the morphological changes caused by the agonist are very different from those caused by the other agonists . Many platelet-leukocyte aggregates were identified in the U46619-treated sample portion , but few in the other blood sample portions ( Figure 2 ) . This may be because U46619 acted as a thromboxane A2 ( TXA2 ) receptor agonist , which activated TXA2 receptors that are abundantly expressed on platelets , vascular smooth muscle cells , and injured vascular endothelial cells . The activation of TXA2 receptors may affect the morphology of U46619-induced platelet aggregates by promoting the expression of adhesion molecules that favors the adhesion and infiltration of leukocytes ( Michelson , 2012; George , 2000 ) . The low classification accuracy values of the ADP- , collagen- , and TRAP-6-treated blood sample portions are presumably due to the fact that these agonists partially share similar mechanisms in forming platelets aggregates ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005; Li et al . , 2000 ) . For example , since platelets also release ADP themselves during activation ( Michelson , 2012; George , 2000; Michelson , 2003; Harrison , 2005 ) , platelet aggregates produced by other agonists may also share similar morphological features as ADP-activated platelet aggregates . In addition , TRAP-6 activates thrombin receptors while thrombin generation may be amplified by other agonists during platelet activation ( Mann , 2011 ) , which indicates that the low prediction values of TRAP-6 can be attributed to the participation of thrombin in platelet aggregation induced by all types of agonists . Furthermore , it is common that platelets are simultaneously activated by multiple agonists whose effects on platelet aggregation are coupled whereas they are also influenced by other factors such as locally produced inhibitors , vascular endothelial cells , blood flow , and coagulation proteins during activation ( Cattaneo and Lecchi , 2007; Michelson , 2012 ) , thereby leading to the low classification accuracy values of certain agonists , which can be overcome by including the influences into the classification model to cover a wide spectrum of aggregation factors . To demonstrate the reproducibility of the iPAC , we tested it with an independent dataset ( a total of 25 , 000 images of all event types ) , which was performed under the same conditions as shown in Figure 1A . The contribution values over all the agonists are in good agreement with the values in the diagonal elements of the confusion matrix ( Figure 3C ) , which validates the reliability of the iPAC . The iPAC’s ability to classify platelet aggregates by agonist type in a concentration-independent manner is indicated by the confusion matrix shown in Figure 3D with an average diagonal element value of 77% . The results also reveal the existence of the unique morphological features related to each agonist type , which is promising for potential application to diagnosis of thrombotic disorders by tracing back to the leading factors of platelet aggregation . In addition , from a viewpoint of potential clinical applications , while the conventional assays can only evaluate platelet aggregability qualitatively , the iPAC can quantify it with the resolving power to identify the contribution of each agonist type to it . However , it can be recognized from the image library ( Figure 1B ) that U46619-activated platelet aggregates have relatively larger size than those in the other sample portions , which may be captured as a type of morphological features by the CNN , leading to the high classification accuracy of the U46619-activated samples . To demonstrate the diagnostic utility of the iPAC , we applied it to blood samples of four healthy human subjects to predict the contribution of each agonist type to platelet aggregates ( if any ) in the samples ( Figure 4 ) . The blood samples were prepared by following the same procedure as shown in Figure 1A except for the step of adding agonists ( with 2000 images of events in each blood sample ) . The experiment was repeated under the same conditions three times . Over 85% of the total population of platelets in all the samples were identified as single platelets , which indicates the ability of the iPAC to differentiate single platelets and platelet aggregates . Furthermore , the agonist types of the platelet aggregates in each subject’s platelet classification results are consistent between the repeated experiments , indicating that the variations between the subjects resulted from platelet heterogeneity , not test variations . The iPAC’s diagnostic ability to obtain this type of information is an effective tool for studying and elucidating the mechanism of platelet aggregation and holds promise for clinical diagnostics , pharmacometrics , and therapeutics , although the iPAC needs more training with a wide spectrum of diseases and medical conditions for the purpose . For example , the iPAC may provide an important clue to the choice of drugs ( e . g . , aspirin or thienopyridines ) for antiplatelet therapy ( Mauri et al . , 2014; Roe et al . , 2012 ) , the gold standard of the treatment and prevention of atherothrombosis ( e . g . , myocardial infarction , cerebral infarction ) , in that aspirin inhibits the formation of TXA2 whose stable analogue is U46619 while thienopyridines exert an antiplatelet effect by blocking the ADP receptor P2Y12 . Furthermore , the iPAC may be able to identify TRAP-6-activated platelet aggregates in the bloodstream of patients with deep vein thrombosis ( since TRAP-6 interacts with the receptor of thrombin ) and suggest that they come from the venous side . The information about the driving factors behind the formation of platelet aggregates is expected to lead to a better understanding of the underlying mechanisms of platelet aggregation and , thereby , open a window on an entirely new class of clinical diagnostics and therapeutics . For example , antiplatelet therapy is the gold standard of the treatment and prevention of atherothrombosis ( e . g . , myocardial infarction , cerebral infarction ) for which aspirin and thienopyridines ( e . g . , prasugrel and clopidogrel ) are primarily used as antiplatelet drugs worldwide ( Mauri et al . , 2014; Roe et al . , 2012 ) . Aspirin inhibits the formation of TXA2 whose stable analogue is U46619 , whereas thienopyridines exert an antiplatelet effect by blocking the ADP receptor P2Y12 . Accordingly , the ability to identify the type of platelet aggregates in the blood stream may provide an important clue to the choice of a drug for antiplatelet therapy . Furthermore , deep vein thrombosis ( DVT ) is a blood clot that normally occurs in a deep vein where coagulation activation plays an important role . Since TRAP-6 interacts with the receptor of thrombin ( i . e . , the product of the coagulation cascade ) , the ability to identify TRAP-6-activated platelet aggregates in the blood stream may suggest that aggregates come from the venous side . Therefore , the iPAC may pave the way for introducing a novel laboratory testing technique for the management of pathological thrombosis such as atherothrombosis and DVT although further basic and clinical studies are needed . The relation between platelet activation signaling pathways and the formation of platelet aggregates has been extensively studied ( Li et al . , 2010; Michelson , 2012; Brass et al . , 2013 ) . It is known that agonists activate platelets in a selective manner via specific receptors , which is followed by a variety of downstream signaling events ( Li et al . , 2010 ) . For example , collagen interacts with the immune-like receptor glycoprotein VI , which signals through an immunoreceptor tyrosine-based activation motif and activates the tyrosine phosphorylation pathway ( Michelson , 2012; Li et al . , 2010 ) In contrast , soluble agonists such as TRAP-6 , U46619 , and ADP interact with G protein-coupled receptors ( Michelson , 2012; Brass , 2003 ) . Furthermore , each soluble agonist couples with a specific type of G protein , which leads to different aggregation mechanisms ( Rivera et al . , 2009 ) and thus suggests different underlying mechanisms for expressing different morphological features on platelet aggregates . It is challenging , but is expected to be intriguing to study and elucidate the mechanisms for a further understanding of the biology of platelets . The detailed procedure of the sample preparation is shown in Figure 1—figure supplement 1 , where platelets and platelet aggregates were enriched from whole blood by the density-gradient centrifugation to maximize the detection efficiency ( Beakke , 1951 ) . Specifically , blood samples were obtained from a healthy person with 3 . 2% citric acid as the anticoagulant ( Figure 1—figure supplement 1A ) . Although it has a depressed concentration of ionized calcium , 3 . 2% citrate blood is desirable for clinical coagulation tests ( Adcock et al . , 1997; Cazenave et al . , 2004 ) . The other common anticoagulants , such as heparin and ethylenediaminetetraacetic acid ( EDTA ) , are not suitable for this study because they influence the coagulation functions of platelets ( Ludlam , 1981 ) . Platelets were immunofluorescently labeled by adding 20 µL PE anti-human CD61 ( BioLegend , 336405 ) to the blood samples to ensure that platelets would be detected in all images ( Figure 1—figure supplement 1B ) . For each agonist type , 500 µL blood was incubated with 50 µL agonist solution , which contained 20 µM ADP ( BioMed , AP-200–422 ) , 10 µg/mL Collagen ( BioMed , AG005K-CS ) , 13 µM TRAP-6 ( H2936 . 0005 , BACHEM ) , or 14 µM U46619 ( Cayman Chemical , 16450 ) , for 10 min ( Figure 1—figure supplement 1C ) . The labeled , activated platelets were then diluted using 5 mL saline ( Figure 1—figure supplement 1D ) . Next , the platelets were isolated by using Lymphoprep ( STEMCELLS , ST07851 ) , a density-gradient medium , using the protocol provided by the vendor . Specifically , the diluted blood was added on top of the Lymphoprep and then centrifuged at 800 g for 20 min ( Figure 1—figure supplement 1E ) . After the centrifugation , 1 mL of the sample was taken from the mononuclear layer , to which 1 mL of 2% paraformaldehyde ( Wako , 163–20145 ) was added for fixation ( Lanier and Warner , 1981; Figure 1—figure supplement 1F , G ) . The operation of the fixation was performed at 4°C for 30 min while other operations were performed at 25°C room temperature . As shown in Figure 1—figure supplement 2 , we first compared several procedures of preparing blood samples , but most of the procedures either left a large amount of non-target blood cells in the sample , thus decreasing the iPAC’s detection efficiency , or dismantled the agonist-activated platelet aggregates . The current procedure is advantageous over the procedures in preserving the morphology of platelet aggregates while eliminating non-target blood cells . This study was approved by the Institutional Ethics Committee in the School of Medicine at the University of Tokyo [no . 11049- ( 6 ) ] . Written informed consents were obtained from the blood donors . The microfluidic chip was fabricated using standard photolithographic methods ( Whitesides et al . , 2001 ) . A designed pattern of the microfluidic channel was drawn using AutoCAD ( Autodesk ) and printed on a film mask ( UnnoGiken ) . Negative photoresist ( KMPR 1035 , MicroChem ) was spin-coated on a silicon wafer and heated at 100°C for 10 min . Then , the silicon wafer , covered with the film mask , was exposed to ultraviolet ( UV ) light followed by hard baking at 100°C for 5 min and developed using SU-8 developer ( MicroChem ) . After washing with isopropyl alcohol and water , the silicon wafer was heated at 150°C for 15 min . The negative photoresist mold on the silicon wafer was fixed in a Petri dish and then filled with polydimethylsiloxane ( PDMS , Dow Corning ) in which PDMS base and curing reagent were mixed at a ratio of 10:1 ( Figure 1—figure supplement 4A ) . PDMS was heated at 80°C for 15 min , and then a small piece of coverslip was placed on PDMS right above the observation area of the microfluidic channel . This step improved the mechanical strength of PDMS so that the channel ( Figure 1—figure supplement 4B ) was able to resist the pressure inside the channel without deformation . After another heating for more than 1 hr , the PDMS layer was cut into a small piece so that it could fit in the size of a glass slide ( Figure 1—figure supplement 4C ) . The inlets and outlet were punched by a 25G needle ( Figure 1—figure supplement 4D ) . To form permanent bonding between the PDMS channel and the glass slide , both the PDMS device and the glass slide were treated with a plasma cleaner ( Harrick Plasma ) ( Figure 1—figure supplement 4E ) . The dimensions of the microchannel in the observation area are about 80 μm in width and 40 μm in height ( Figure 1—figure supplement 4F ) . The optofluidic time-stretch microscope ( Lei et al . , 2016 ) is schematically shown in Figure 1—figure supplement 5 . A Ti:Sapphire mode-locked femtosecond pulse laser with a center wavelength , bandwidth , and pulse repetition rate of 780 nm , 40 nm , and 75 MHz , respectively , was used as an optical source . Each laser pulse was first stretched temporally by a single-mode dispersive fiber with a group-velocity dispersion of −240 ps/nm ( Nufern 630-HP ) and then dispersed spatially by the first diffraction grating with a groove density of 1200 lines/mm . The stretched laser pulse was focused by the first objective lens ( Olympus , 40× , NA 0 . 6 ) onto a flowing cell in the microfluidic channel . The pulse that contained the spatial profile of the cell on its spectrum was collected by the second objective lens and spatially recombined by the second diffraction grating , followed by photodetection with a high-speed photodetector ( New Focus 1580-B ) with a detection bandwidth of 12 GHz . To ensure imaging of platelet-related events ( i . e . , single platelets , platelet-platelet aggregates , platelet-leukocyte aggregates ) , fluorescence detection was used in conjunction with the optofluidic time-stretch microscope . A 488 nm continuous-wave laser was used to detect CD61 fluorescence signals with a photomultiplier tube ( Hamamatsu H10723-01MOD ) . Only the image signals associated with CD61 fluorescence signals were collected . The image-encoded pulse and fluorescence signal were digitized using a high-speed oscilloscope ( Tektronix DPO 71604B ) with a detection bandwidth of 16 GHz and a sampling rate of 50 GS/s . Pulses were repeated by the mode-locked pulse laser at 75 MHz so that image-encoded pulses detected by the photodetector were digitally stacked to form 2D images using MATLAB R2018b ( MathWorks ) . The pulse intensity profile ( usually Gaussian-shaped with ripples ) was normalized to obtain a flat background . Also , all the images were cropped into 160 × 160 pixels by the same cropping algorithm , by which the cell-contained part was completely included in each image for further analysis . We analyzed agonist-activated platelets with a conventional flow cytometer ( Cytomics FC500 , Beckman Coulter ) that can count and analyze large cell populations via scattering and fluorescence measurements with high throughput . Blood samples were processed using the same procedure as for optofluidic time-stretch microscopy , but labeled with anti-CD61-APC and anti-CD45-FITC antibodies ( Beckman Coulter ) for detecting white blood cells and platelets , respectively . To only detect single platelets and platelets aggregates , gating of cellular size and granularity was applied to the light scatter plots . As shown in Figure 2 , the C1 areas , which correspond to CD61-APC positive and CD45-FITC negative , show events associated with single platelets and platelet-platelet aggregates . The C2 areas , which correspond to CD61-APC/CD45-FITC double positive , show events associated with platelet-leukocyte aggregates . The C3 areas ( CD61-APC/CD45-FITC double negative ) and C4 areas ( CD61-APC negative and CD45-FITC positive ) correspond to events which did not contain any platelets . The details of the CNN with the encoder-decoder architecture are as follows . The encoder was used to extract morphological features of platelet aggregates , while the decoder was used to recover the platelet aggregate images from the morphological features . This two-stage structure forced the encoder to extract features from the cells instead of the background or noise , which helped enhance the reliability and accuracy of classification . The images were normalized to 0-mean and divided into training , validation , and test sets at a ratio of 3:1:1 . The CNN classifier was trained on the training set . The validation loss was calculated with the validation dataset at each epoch to monitor the learning process . The learning rate was reduced when the validation loss stopped descending for more than 3 epochs until it reached 1 × 10−8 . The training was ceased when there was no more decrease in the validation loss for more than 6 epochs . After the training ended , the test set was processed to calculate the final classification accuracy for each agonist type . The CNN classifier was implemented on Keras ( Chollet , 2015 ) with the Tensorflow ( Abadi et al . , 2016 ) backbone . The training of the CNN classifier was optimized by Adam with an initial learning rate of 0 . 001 .	Platelets are small cells in the blood that primarily help stop bleeding after an injury by sticking together with other blood cells to form a clot that seals the broken blood vessel . Blood clots , however , can sometimes cause harm . For example , if a clot blocks the blood flow to the heart or the brain , it can result in a heart attack or stroke , respectively . Blood clots have also been linked to harmful inflammation and the spread of cancer , and there are now preliminary reports of remarkably high rates of clotting in COVID-19 patients in intensive care units . A variety of chemicals can cause platelets to stick together . It has long been assumed that it would be impossible to tell apart the clots formed by different chemicals ( which are also known as agonists ) . This is largely because these aggregates all look very similar under a microscope , making it incredibly time consuming for someone to look at enough microscopy images to reliably identify the subtle differences between them . However , finding a way to distinguish the different types of platelet aggregates could lead to better ways to diagnose or treat blood vessel-clogging diseases . To make this possible , Zhou , Yasumoto et al . have developed a method called the “intelligent platelet aggregate classifier” or iPAC for short . First , numerous clot-causing chemicals were added to separate samples of platelets taken from healthy human blood . The method then involved using high-throughput techniques to take thousands of images of these samples . Then , a sophisticated computer algorithm called a deep learning model analyzed the resulting image dataset and “learned” to distinguish the chemical causes of the platelet aggregates based on subtle differences in their shapes . Finally , Zhou , Yasumoto et al . verified iPAC method’s accuracy using a new set of human platelet samples . The iPAC method may help scientists studying the steps that lead to clot formation . It may also help clinicians distinguish which clot-causing chemical led to a patient’s heart attack or stroke . This could help them choose whether aspirin or another anti-platelet drug would be the best treatment . But first more studies are needed to confirm whether this method is a useful tool for drug selection or diagnosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology" ]	2020	Intelligent classification of platelet aggregates by agonist type
Environmental stress during early development can impact adult phenotypes via programmed changes in gene expression . C . elegans larvae respond to environmental stress by entering the stress-resistant dauer diapause pathway and resume development once conditions improve ( postdauers ) . Here we show that the osm-9 TRPV channel gene is a target of developmental programming and is down-regulated specifically in the ADL chemosensory neurons of postdauer adults , resulting in a corresponding altered olfactory behavior that is mediated by ADL in an OSM-9-dependent manner . We identify a cis-acting motif bound by the DAF-3 SMAD and ZFP-1 ( AF10 ) proteins that is necessary for the differential regulation of osm-9 , and demonstrate that both chromatin remodeling and endo-siRNA pathways are major contributors to the transcriptional silencing of the osm-9 locus . This work describes an elegant mechanism by which developmental experience influences adult phenotypes by establishing and maintaining transcriptional changes via RNAi and chromatin remodeling pathways . Increasing evidence suggests that exposure to stressful environmental conditions during a critical period in the early development of an organism can result in altered gene expression patterns that are maintained into adulthood . In mammals , this 'cellular memory' of early life stress has been shown to correlate with behavioral phenotypes that persist into adulthood ( Weaver et al . , 2004b; McGowan et al . , 2009 ) . For example , transcription levels of the glucocorticoid receptor ( GR ) gene in the mouse hippocampus are programmed by the maternal care behaviors of the mother towards her pups during the first postnatal week ( Fish et al . , 2004; Weaver et al . , 2004a; 2007; Szyf , 2007; 2008 ) . High maternal care behaviors ( such as frequent licking and grooming ) lead to high GR expression and low stress responses in the pups as adults , whereas low maternal care results in increased DNA methylation , transcriptional silencing of the GR gene , and high stress responses . Intriguingly , this mechanism of gene regulation appears to be conserved in humans , as childhood adversity correlates with increased transcriptional silencing of the GR gene and mental disorders in adults ( McGowan et al . , 2008; 2009; Labonté et al . , 2012; Labonte et al . , 2012 ) . However , the molecular mechanisms establishing and maintaining the tissue-specific changes in gene expression in neurons due to environmental inputs during early development remain to be elucidated . We have shown that environmental stress early in development can also modulate gene expression and behavior in Caenorhabditis elegans , thereby establishing nematodes as a model system for investigating the mechanisms underlying developmental programming of gene expression ( Hall et al . , 2010; 2013 ) . The C . elegans life cycle is regulated by environmental cues during a critical period in the first larval stage . If conditions are favorable ( low temperature , high food availability , low pheromone concentrations ) , worms continue to develop through four larval stages to become reproductive adults ( control adults , CON ) . However , if conditions are unfavorable , worms enter the alternative stress-induced dauer diapause stage ( Golden and Riddle , 1982; 1984 ) . The dauer formation ( daf ) decision is regulated by the differential expression of conserved insulin and TGF-β pathways in sensory neurons in response to environmental conditions ( reviewed in Fielenbach and Antebi , 2008 ) . Once conditions improve , dauer animals re-enter the reproductive cycle and continue development to become reproductive adults ( postdauer adults , PD ) . We have shown previously that CON and PD adults retain a cellular memory of their developmental history through changes in gene expression , genome-wide chromatin state , and life history traits ( Hall et al . , 2010 ) . In addition , these altered adult phenotypes are in part dependent on endogenous small interfering RNA pathways ( endo-siRNAs ) ( Hall et al . , 2013 ) . In recent years , our knowledge of C . elegans endo-siRNA pathways and their gene targets has increased substantially . Endo-siRNAs are ~20 to 30 nt in length , antisense to coding transcripts , and are categorized into two major groups based on their biogenesis pathways ( Gent et al . , 2010; Vasale et al . , 2010 ) . Primary endo-siRNAs are 26 nt long , have a characteristic 5’ monophosphate , and are derived from Dicer processing of double-stranded RNA template ( 26G-siRNAs ) ( Vasale et al . , 2010 ) . In somatic tissue , biogenesis of ERGO-1 class 26G-siRNAs is dependent upon the enhanced RNAi ( ERI ) complex , which consists of the core proteins RNA-dependent RNA polymerase ( RdRP ) RRF-3 , Dicer-related helicase DRH-3 , and Tudor domain-containing protein ERI-5 that associate with Dicer DCR-1 ( Duchaine et al . , 2006; Pavelec et al . , 2009; Gent et al . , 2010; Thivierge et al . , 2012 ) . Through an unknown mechanism , 26G-siRNAs stimulate the production of secondary endo-siRNAs , which are 22 nt in length , have a 5’ triphosphate , and are produced through reverse transcription of mRNA templates by RdRPs ( 22G-siRNAs ) ( Smardon et al . , 2000; Simmer et al . , 2002; Maine et al . , 2005; Vought et al . , 2005; Aoki et al . , 2007; Pak et al . , 2007; She et al . , 2009; Vasale et al . , 2010; Pak et al . , 2012 ) . Downstream effector functions of 22G-siRNAs , such as mRNA degradation or chromatin remodeling , are mediated by their associated worm-specific Argonaute protein ( WAGO ) in the cytoplasm or nucleus ( Yigit et al . , 2006 ) . Additionally , Mutator proteins have been shown to be essential for the biogenesis and siRNA amplification of AGO ERGO-1 class 26G- and WAGO class 22G-siRNAs ( Zhang et al . , 2011; Phillips et al . , 2012; 2014 ) . Although components of the Mutator complex are expressed throughout the worm , their functions and cellular localization are specific to the germline or soma ( Phillips et al . , 2012; 2014 ) . Despite our understanding of endo-siRNA biogenesis , how siRNAs target and regulate endogenous genes , particularly in response to environmental cues , is poorly understood . In this study , we show that expression of the OSM-9 TRPV channel is differentially regulated based on the developmental trajectory of C . elegans and describe the molecular mechanisms that regulate expression of this gene as a function of the animal’s developmental history . The osm-9 gene is down-regulated specifically in the ADL chemosensory neurons of PD adults , resulting in a corresponding altered olfactory behavior that is mediated by ADL in an OSM-9-dependent manner . We identify a cis-acting motif bound by the DAF-3 SMAD and ZFP-1 ( AF10 ) proteins that is necessary for the down-regulation of osm-9 in PD adults and demonstrate that both chromatin remodeling and endo-siRNA pathways are major contributors to osm-9 regulation . Our results suggest a mechanism of osm-9 regulation whereby the dauer developmental decision , triggered by environmental stress , results in transcriptional silencing of the osm-9 locus mediated by the TGF-β , chromatin remodeling , and endogenous RNAi pathways . This work describes an elegant mechanism of how tissue-specific changes in gene expression triggered by transient environmental cues can be maintained via developmental programming . Gene expression analyses identified the osm-9 TRPV channel gene as a candidate gene whose expression is regulated as a consequence of passage through the dauer stage ( Hall et al . , 2010 ) . osm-9 is expressed in a subset of head and tail neurons and is required for chemosensory , osmosensory , and mechanosensory behaviors in adult animals ( Colbert et al . , 1997; de Bono et al . , 2002; Jansen et al . , 2002; Tobin et al . , 2002; White et al . , 2007; O'Halloran et al . , 2009; Sassa et al . , 2013; Wang et al . , 2015a ) . To characterize the differential gene expression of osm-9 between CON and PD adult animals in further detail , we examined the expression pattern of an integrated gfp reporter gene driven by 375 bp of osm-9 upstream regulatory sequences and 18 bp of the first exon ( osm-9p::gfp ) ( Figures 1A , 2A ) . We examined gfp expression in CON and PD adults that spent 24 hr in dauer induced by crowding ( see Materials and Methods ) ( Hall et al . , 2010 ) . We observed GFP in the majority of AWA and ADL sensory neurons in CON animals ( Figures 1A , B; Figure 1—figure supplement 1A ) . However , while expression in AWA neurons was unaffected , GFP expression was significantly down-regulated in ADL neurons of PD adults ( Figures 1A , B; Figure 1—figure supplement 1A ) . These results suggest that osm-9 is regulated at the transcriptional level as a consequence of passage through the dauer stage . 10 . 7554/eLife . 11642 . 003Figure 1 . The osm-9 TRPV channel gene is regulated by developmental history . ( A ) An osm-9p::gfp transgene is expressed in ADL and AWA neurons in CON adults , but is down-regulated in ADL neurons in PD adults . ( B ) osm-9p::gfp expression in ADL neurons in CON and PD adults that entered dauer due to crowding ( egg plates ) , exposure to crude pheromone with low population density , or starvation . N ≥ 2 trials; n ≥ 40 animals ( Figure 1—source data 1 ) . p<0 . 005 , p<1 x 10–9 , Student’s t-test . GFP in AWA is unaffected ( Figure 1—figure supplement 1 ) . ( C ) Quantification of osm-9 mRNA in ADL neurons of CON and PD using smFISH . The graphs represent individual mean and maximum fluorescence measurements for N = 3 biologically independent trials; n ≥ 87 neurons ( Figure 1—source data 2 ) . Medians are indicated . p<1 x 10–9 , p = 0 . 006 p = 0 . 056 , Student’s t-test . See Figure 1—figure supplement 2 and Figure 1—figure supplement 3 . ( D ) Ascr#3 avoidance behavior of CON and PD adults normalized to M13 buffer . N ≥ 3 trials; n ≥ 60 animals ( Figure 1—source data 3 ) . Data for ADL-expressed osm-9 rescue strains are averages of two extrachromosomal lines . indicates CON significantly different from PD; & CON or PD significantly different from wild-type; $ CON or PD significantly different from osm-9 ( ky10 ) ; One-way ANOVA with LSD posthoc correction , p<0 . 05 . All error bars represent S . E . M . OSM-9 mediated behaviors modulated by ASH and AWA neurons are unaffected ( Figure 1—figure supplement 1 and Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00310 . 7554/eLife . 11642 . 004Figure 1—source data 1 . Spreadsheet of percentages of animals expressing osm-9p::gfp in ADL neurons of wild-type strains that experienced overcrowding , high pheromone , or starvation conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00410 . 7554/eLife . 11642 . 005Figure 1—source data 2 . Spreadsheet containing mean and maximum florescence intensities of osm-9 smFISH probes in wild-type ADL neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00510 . 7554/eLife . 11642 . 006Figure 1—source data 3 . Spreadsheet containing ascr#3 avoidance values for CON and PD in wild-type , osm-9 ( ky10 ) , and ADL-specific , osm-9 rescue strains . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00610 . 7554/eLife . 11642 . 007Figure 1—figure supplement 1 . OSM-9 expression in AWA was unaffected by developmental programming . ( A ) osm-9p::gfp expression in AWA neurons of CON and PD adults . N > 3 trials; n > 60 animals . ( Figure 1—figure supplement 1—source data 1 ) . ( B ) Chemotaxis index of CON and PD adults in wild-type and osm-9 ( ky10 ) strains in response to diacetyl . Behavioral assays were performed as described previously ( Ward , 1973 ) . Chemotaxis Index was calculated as [ ( # animals at diacetyl ) – ( # animals at control ) ] / total # animals . N = 3 trials; n > 300 animals . Error bars represent S . E . M . ( Figure 1—figure supplement 1—source data 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00710 . 7554/eLife . 11642 . 008Figure 1—figure supplement 1—source data 1 . Spreadsheet containing percentage of animals expressing osm-9p::gfp in AWA neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00810 . 7554/eLife . 11642 . 009Figure 1—figure supplement 1—source data 2 . Spreadsheet containing chemotaxis indices in response to diacetyl for wild-type and osm-9 ( ky10 ) strains . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 00910 . 7554/eLife . 11642 . 010Figure 1—figure supplement 2 . osm-9 mRNA levels are down-regulated in PD ADL neurons . ( A ) Representative images analyzed for smFISH experiment . ( B ) Graphs represent the average mean and maximum florescence levels of CON and PD ADL neurons with osm-9 mRNA smFISH probes in arbitrary brightness units ( A . B . U . ) . ( C ) Graphs representing the average mean and maximum fluorescence levels of worm backgrounds . n ≥ 87 neurons over 3 biologically independent trials . Error bars are standard deviation ( Figure 1—figure supplement 2—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01010 . 7554/eLife . 11642 . 011Figure 1—figure supplement 2—source data 1 . Spreadsheet containing mean and maximum florescence intensities for osm-9 smFISH probes in ADL neurons and worm backgrounds . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01110 . 7554/eLife . 11642 . 012Figure 1—figure supplement 3 . gfp mRNA levels are unaltered between CON and PD ADL neurons . ( A ) Quantification of gfp mRNA in ADL neurons of CON and PD using smFISH . ADL neurons were identified by expression of sre-1p::gfp . The graphs represent individual mean and maximum fluorescence measurements for n ≥ 10 neurons over 3 biologically independent trials . * p = 0 . 042 , Student’s t-test . ( B ) Graphs represent the average mean and maximum florescence levels of CON and PD ADL neurons with gfp mRNA smFISH probes in arbitrary brightness units ( A . B . U . ) . ( C ) Graphs representing the average mean and maximum fluorescence levels of worm backgrounds . Error bars are standard deviation ( Figure 1—figure supplement 3—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01210 . 7554/eLife . 11642 . 013Figure 1—figure supplement 3—source data 1 . Spreadsheet containing normalized and raw mean and maximum florescence intensities for gfp smFISH probes in ADL neurons and worm backgrounds . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01310 . 7554/eLife . 11642 . 014Figure 2 . The PD motif is necessary for the differential regulation of osm-9 . ( A ) Schematic of osm-9 promoter driving gfp expression . Dotted lines indicate deleted sequences . Purple , blue , red , and yellow boxes represent 18 bp osm-9 coding sequence , the conserved motif , DAF-3 binding site , and potential DAF-12 binding site , respectively . ( B ) The PD motif DNA sequences of wild-type and mutated versions . Bold 'G' indicates where wild-type sequence was mutated to guanine . ( C ) GFP expression in CON and PD ADL neurons of strains carrying deletion and mutated versions of osm-9p::gfp . Data for each deletion and mutation represents the average of two extrachromosomal lines . * indicates significantly different from wild-type CON or PD; One-way ANOVA with LSD posthoc correction , p<0 . 05 . All error bars represent S . E . M ( Figure 2—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01410 . 7554/eLife . 11642 . 015Figure 2—source data 1 . Spreadsheet of percentages of animals expressing gfp in ADL neurons in strains carrying deletion and mutation versions of osm-9p::gfp . The conserved sequence formula for the PD motif is also included . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 015 To confirm that the differential expression of osm-9p::gfp was reflective of the endogenous osm-9 gene , we quantified osm-9 mRNA molecules in CON and PD ADL neurons using single molecule florescent in situ hybridization ( smFISH ) ( Raj et al . , 2008; Ji and van Oudenaarden , 2012 ) . We used a transgenic strain expressing GFP in ADL neurons under a promoter whose expression is unaffected upon dauer passage ( sre-1p::gfp ) to identify the correct neuron and to quantify the gfp mRNA molecules present as a control ( Figure 1—figure supplement 3 ) . For both the osm-9 and gfp smFISH probes , we observed that the fluorescent foci within ADL exhibited unequal sizes and intensities; thus , we were not confident that we could resolve the number of individual mRNA molecules simply by counting the number of smFISH loci . Instead , we quantified the fluorescence intensities of ADL neurons as a measure of the osm-9 and gfp mRNA molecules present . We found that ADL neurons in CON adults exhibited significantly greater mean and maximum fluorescence intensities for the osm-9 mRNA probes compared to PD adults ( Figure 1C and Figure 1—figure supplement 2 ) . In contrast , fluorescence intensities for the gfp mRNA probes exhibited overall similar mean and maximum fluorescence levels for PD compared to CON adults ( Figure 1—figure supplement 3 ) . We conclude that the differential regulation of the osm-9p::gfp transgene is reflective of the regulation of the endogenous osm-9 gene . Next , we asked whether passage through the dauer stage was necessary for the observed changes in gene expression , or if exposure to high concentrations of pheromone without dauer entry was sufficient as has been reported for the regulation of chemoreceptor genes in a subset of neurons in C . elegans ( Peckol et al . , 2001; Nolan et al . , 2002 ) . To address this question , we allowed wild-type embryos to hatch on plates containing crude pheromone as previously described ( Neal et al . , 2013 ) . Animals that bypassed dauer entry and animals that entered the dauer stage were recovered separately and examined for osm-9p::gfp expression in adults . We found that animals that bypassed dauer entry when exposed to high concentrations of dauer pheromone continued to express GFP in ADL neurons , similar to CON animals in our egg plate preparation ( Figure 1B ) . However , animals that passed through the dauer stage exhibited the expected decrease in osm-9p::gfp expression in ADL ( Figure 1B ) . Similarly , animals that were transiently starved in early development continued to express GFP in ADL if they bypassed the dauer stage , but exhibited decreased GFP expression following passage through dauer ( Figure 1B ) . These results indicate that the down-regulation of osm-9p::gfp in adults is dependent on passage through the dauer stage regardless of the environmental trigger inducing dauer entry . To examine the phenotypic consequences of the differential regulation of osm-9 , we tested whether an ADL-mediated and OSM-9 dependent behavior is affected upon passage through the dauer stage . Wild-type hermaphrodites avoid high concentrations of the dauer pheromone component ascr#3; this avoidance behavior requires the OSM-9 and OCR-2 TRPV channels in the ADL neurons ( Jang et al . , 2012 ) . We hypothesized that if the endogenous osm-9 gene were down-regulated in ADL neurons of PD animals , these animals would fail to avoid ascr#3 similar to osm-9 mutants . To test this hypothesis , we examined the ability of wild-type and osm-9 mutant CON and PD adults to avoid 100 nM ascr#3 using drop-test assays ( Hilliard et al . , 2002; Jang et al . , 2012 ) . We observed that wild-type PD adults exhibited a significantly reduced avoidance response to ascr#3 compared to CON adults ( Figure 1D ) . However , these animals retained the ability to avoid glycerol , an OSM-9-dependent behavior mediated by the ASH neurons , and were attracted to the odorant diacetyl , an OSM-9-dependent behavior mediated by the AWA neurons ( Figure 1—figure supplement 1B; Figure 3—figure supplement 1C ) ( Sengupta et al . , 1996; Colbert et al . , 1997 ) . These observations are consistent with the observed down-regulation of osm-9 expression in the ADL neurons . To verify that the decreased ascr#3 avoidance phenotype in wild-type PD adults was due to decreased expression of osm-9 in ADL neurons , we restored osm-9 expression specifically in ADL neurons of osm-9 mutants by driving the cDNA under sre-1 upstream regulatory sequences . Since sre-1 expression is not detectably altered upon passage through the dauer stage ( Figure 1—figure supplement 2A and Figure 1—figure supplement 3 ) , we predicted that both CON and PD adults expressing sre-1p::osm-9 cDNA::gfp transgene would continue to avoid ascr#3 . However , while the transgene rescued the osm-9 mutant phenotype in CON animals indicating that the fusion protein is functional , avoidance behavior was again decreased in PD animals ( Figure 1D ) . We also attempted to rescue the decrease in ascr#3 avoidance in wild-type PD adults by expressing the same sre-1p::osm-9 cDNA::gfp transgene in ADL neurons . Again , we observed ascr#3 avoidance behavior similar to wild-type animals without the transgene ( Figure 1D ) . Although our results thus far have indicated that osm-9 is regulated at the transcriptional level in response to developmental history , these results suggest that silencing mechanisms might also target the osm-9 coding sequences or mRNA , or alternatively , additional genes contributing to the ascr#3 behavior . Since transcriptional down-regulation contributes to the decreased osm-9p::gfp expression in ADL in PD animals , we analyzed osm-9 upstream regulatory sequences present in the osm-9p::gfp transgene to identify regulatory motifs contributing to the altered expression . Using bioinformatics-based analyses of osm-9 upstream regulatory sequences , we identified a 29 bp motif ( Figure 2A ) that is present in the upstream regulatory sequences of 977 ( 4 . 9% ) genes in the genome . This conserved sequence was slightly enriched ( 121 genes , 5 . 6%; Z-test , p=0 . 056 ) in the regulatory regions of genes that were differentially expressed between CON and PD adults ( Hall et al . , 2010 ) . We also identified an additional motif , a predicted DAF-3 SMAD binding site , located 22 bp upstream of the conserved sequence ( Figure 2A ) ( Thatcher et al . , 1999 ) . We will henceforth refer to this combined regulatory sequence as the PD motif . To test the functional relevance of this motif , we generated a series of osm-9p::gfp transcriptional fusion constructs lacking portions of the osm-9 regulatory sequences and examined their expression in ADL neurons of CON and PD adults ( Figure 2A , C ) . Deletion of the PD motif in the context of the 375 bp osm-9 regulatory sequences ( Deletion 1 ) resulted in continued expression of gfp in ADL in PD adults without affecting expression in CON animals , indicating that this cis-regulatory motif is required for the down-regulation of osm-9 in PD ADL neurons ( Figure 2A , C ) . Deletion of sequences that did not include the PD motif ( Deletion 2 ) resulted in gfp expression changes in PD ADL neurons similar to that observed with the full length osm-9p::gfp transgene ( Figure 2A , C ) . Deletion 3 , which contained the 18 bp of osm-9 coding sequence in addition to the 375 bp of upstream regulatory sequence , also exhibited down-regulation of gfp expression in ADL neurons of PD adults ( Figure 2A , C ) . These results indicate that the cis-acting PD motif is necessary for the transcriptional regulation of osm-9 due to developmental programming . To further characterize the PD motif sequence necessary for differential regulation of osm-9 , we sequentially replaced a subset of bases within the PD motif with guanines and examined gfp expression in ADL neurons of CON and PD adults ( Figure 2B , C ) . Mutated constructs 3 and 5 exhibited significantly increased GFP in ADL neurons of PD adults compared to wild-type ( Figure 2B , C ) , indicating that these sequences were required for complete down-regulation of osm-9 in PD adults . However , we found that mutated constructs 1 , 2 , 4 , 6 , and 7 resulted in a significant decrease in osm-9p::gfp expression in both CON and PD ADL neurons ( Figure 2B , C ) , indicating that these sequences were required for the positive regulation of osm-9 expression . Together these results suggest a model that , in the absence of the PD motif , the default state of osm-9 expression in ADL neurons is transcriptional activation , and that the PD motif contains sequences necessary for the negative regulation of osm-9 in response to developmental history . To further illuminate the mechanisms regulating osm-9 and ascr#3 avoidance behavior , we sought to identify factors that interact with the upstream regulatory sequences of osm-9 . In addition to the potential DAF-3 SMAD binding site , further scrutiny of the osm-9 cis-regulatory sequences yielded potential binding sites for the DAF-12 nuclear hormone receptor ( NHR ) , which functions downstream of TGF-β and insulin signaling pathways in the regulation of dauer formation ( Figure 2A ) . To investigate a potential role of the TGF-β and insulin/IGF dauer formation pathways in the regulation of osm-9 , we examined osm-9p::gfp expression in CON and PD adults of strains carrying mutations in genes implicated in dauer formation . First , we found that mutations in daf-3 SMAD and daf-5 SNO/SKI significantly increased the gfp expression observed in PD ADL neurons compared to wild-type animals ( Figure 3A ) . DAF-3 binds to DAF-5 ( da Graca et al . , 2004 ) , and may act together as negative regulators at the osm-9 locus . In contrast , the daf-7 TGF-β mutant strain exhibits a partial but significant decrease in osm-9p::gfp expression in ADL neurons of both CON and PD animals compared to wild-type , but expression in PD animals is again down-regulated as compared to CON animals ( Figure 3A ) . This result suggests that dauer formation , and not reduced TGF-β signaling alone , is required for the down-regulation of osm-9 in PD ADL neurons . In addition , since DAF-7 TGF-β signaling antagonizes DAF-3/DAF-5 function , this result is also consistent with increased DAF-3/DAF-5 activity in a daf-7 mutant promoting down-regulation of osm-9 expression ( Vowels and Thomas , 1992; Thomas et al . , 1993; Patterson et al . , 1997 ) . Together these results suggest that DAF-3 and DAF-5 play a role in the developmental programming of osm-9 in PD ADL neurons . 10 . 7554/eLife . 11642 . 016Figure 3 . The TGF-β pathway regulates osm-9 expression in response to developmental history . ( A ) Percent ADL neurons expressing osm-9p::gfp in CON and PD adults in wild-type and strains carrying mutations in TGF-β and insulin signaling pathways . GFP expression for Mutation 2 in daf-3 , daf-5 , and daf-3; daf-5 mutants is found in Figure 3—figure supplement 1 . N ≥ 3 trials; n ≥ 60 animals ( Figure 3—source data 1 ) . * indicates CON or PD significantly different from wild-type; One-way ANOVA with LSD posthoc correction , p<0 . 05 . n . d . , not determined . ( B ) Log2 normalized enrichment of DAF-3 SMAD binding to the osm-9 DAF-3 binding site ( DBS ) in C . elegans developmental stages . daf-8 and daf-14 are positive and negative controls , respectively . Bar graph represents IP-qPCR data , normalized to DAF-3 binding to actin act-2 promoter ( Park et al . , 2010 ) . N ≥ 2 biologically independent trials ( Figure 3—source data 2 ) . * indicates enriched over background ( daf-3 CON ) or between CON and PD as indicated , p<0 . 05 , Student’s t-test . ( C ) Log2 PD/CON ratio of ascr#3 avoidance behavior in wild-type and strains carrying mutations in TGF-β pathway genes . Data for controls is found in Figure 3—figure supplement 2 . * indicates significantly different from wild-type; One-way ANOVA with LSD posthoc correction , p<0 . 05 ( Figure 3—source data 3 ) . All error bars represent S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01610 . 7554/eLife . 11642 . 017Figure 3—source data 1 . Spreadsheet containing percentages of animals expressing osm-9p::gfp in ADL neurons in strains carrying mutations in TGF-β and insulin signaling genes . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01710 . 7554/eLife . 11642 . 018Figure 3—source data 2 . Spreadsheet containing DAF-3 enrichment values in wild-type CON , PD , larval L3 , and dauer animals . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01810 . 7554/eLife . 11642 . 019Figure 3—source data 3 . Spreadsheet containing PD/CON ascr#3 avoidance ratios for strains carrying mutations in TGF-β and insulin signaling genes . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 01910 . 7554/eLife . 11642 . 020Figure 3—figure supplement 1 . DAF-3 and DAF-5 may also indirectly regulate osm-9 transcription . Our osm-9p::gfp expression and ascr#3 avoidance behavior results both suggested that DAF-3 SMAD and DAF-5 SNO/SKI are acting as a negative regulators of osm-9 in PD ADL neurons ( see Figure 3 ) . However , mutation of part of the potential DAF-3 binding site ( Mutation 2 ) resulted in a decrease in osm-9 expression in CON adults . We observed significantly increased GFP expression in ADL neurons of CON adults carrying the Mutation 2 construct in the daf-3 ( mgDf90 ) , daf-5 ( e1386 ) , and daf-3 ( mgDf90 ) ; daf-5 ( e1386 ) mutant strains compared to wild-type . We were unable to obtain PD adults in these mutant strains to examine GFP levels . DAF-3 has been shown to negatively regulate the genes daf-7 , daf-8 , and myosin gene myo-2 by binding to their upstream regulatory sequences ( Thatcher et al . , 1999; Park et al . , 2010 ) . Thus , DAF-3 and DAF-5 may be inhibiting activity of an unknown protein that promotes osm-9 expression when they cannot bind to the PD motif . n ≥ 20 animals over at least 3 biologically independent trials . osm-9p::gfp data in wild-type , daf-3 ( mgDf90 ) , and daf-5 ( e1386 ) and Mutation 2 reproduced from Figures 2C and 3A for reference . Data for strains carrying Mutation 2 are averages of values for two independent extrachromosomal lines . * Significantly different from wild-type , & significantly different from Mutation 2 , One-way ANOVA , LSD posthoc correction , p<0 . 05 . Error bars represent S . E . M . ( Figure 3—figure supplement 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02010 . 7554/eLife . 11642 . 021Figure 3—figure supplement 1—source data 1 . Spreadsheet containing the percentage of animals expressing Mut2 version of osm-9p::gfp in daf-3 ( mgDf90 ) and daf-5 ( e1386 ) strains . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02110 . 7554/eLife . 11642 . 022Figure 3—figure supplement 2 . Strains tested for ascr#3 avoidance exhibited expected behaviors for positive and negative controls . ( A ) Fraction of CON adults from strains in Figures 3C , 4B , and 5B that exhibit ascr#3 avoidance behavior . ( B ) Fraction of CON and PD adults from strains in Figures 1D , 3C , 4B , and 5B that exhibit M13 avoidance behavior ( negative control ) . ( C ) Fraction of CON and PD adults from strains in Figures 1D , 3C , 4B , and 5B that exhibit 1 M glycerol avoidance behavior ( positive control ) . Data for neuron-specific rescue strains is average values for two independent extrachromosomal lines . Error bars represent S . E . M . ( Figure 3—figure supplement 2—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02210 . 7554/eLife . 11642 . 023Figure 3—figure supplement 2—source data 1 . Spreadsheet containing avoidance responses CON and PD adults to M13 and 1 M glycerol . Normalized avoidance responses of CON adults to ascr#3 are also included . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 023 We next asked whether DAF-3 directly interacts with osm-9 cis-regulatory sequences . The potential DAF-3 binding site in the PD motif ( GTCTA; overlapping with Mutations 2 and 3 , Figure 2B ) differs from the previously identified binding site by a single base ( GTCTG ) ( Thatcher et al . , 1999 ) . To test whether DAF-3 does indeed bind to the osm-9 PD motif , we performed protein immunoprecipitation of DAF-3 from whole animals using a commercially available antibody ( Novus Biologicals , Littleton , CO ) , followed by quantitative PCR ( IP-qPCR ) in CON and PD adult , dauer , and larval L2/L3 animals . Our results showed that DAF-3 was enriched at the potential DAF-3 binding site ( DBS ) in the osm-9 promoter in PD adults compared to other tested stages ( Figure 3B ) . We also examined DAF-3 enrichment at the daf-8 and daf-14 promoters as positive and negative controls , respectively , since their regulation by DAF-3 has been characterized previously ( Park et al . , 2010 ) . Although these results do not address DAF-3 enrichment at the PD motif specifically in ADL neurons , nor the precise binding site of DAF-3 ( see Figure 3—figure supplement 1 ) , they are consistent with our model that DAF-3 negatively regulates osm-9 in PD ADL neurons through a direct interaction at the PD motif . We next characterized the role of the TGF-β pathway in modulating ascr#3 avoidance behavior by testing the ascr#3 avoidance responses of CON and PD animals in strains carrying mutations in daf-3 SMAD , daf-7 TGF-β , and daf-5 SNO/SKI genes . Based on their role in the regulation of osm-9 transcription ( Figure 3A ) , we expected to observe an elimination of the difference between the PD and CON ascr#3 avoidance behavior in daf-3 and daf-5 mutant strains and a reduced effect in daf-7 . Indeed , we observed that mutations in daf-3 SMAD and daf-5 SNO/SKI genes eliminated the difference between CON and PD avoidance response to ascr#3 , consistent with DAF-3 and DAF-5 acting as negative regulators of ascr#3 avoidance in PD animals ( Figure 3C ) . In addition , a mutation in the daf-7 TGF-β gene significantly disrupted the PD/CON ascr#3 avoidance levels compared to wild-type , through a decrease in CON adult ascr#3 avoidance ( Figure 3C and Figure 3—figure supplement 2A ) . While most components of the TGF-β pathway exhibit broad expression patterns , DAF-7 TGF-β is expressed in ASI , ADE , and OLQ neurons , suggesting that a molecular signal from one of these sensory neurons may regulate osm-9 expression in ADL ( Ren et al . , 1996; Schackwitz et al . , 1996; Meisel et al . , 2014 ) . Since TGF-β and insulin signaling pathways act in parallel to regulate dauer formation , we examined whether members of the insulin signaling pathway also contribute to the developmental programming of osm-9 . We examined osm-9p::gfp expression in animals carrying mutations in the daf-2 insulin/IGF receptor homolog , age-1 PI3K homolog , pdk-1 3-phosphoinositide-dependent kinase 1 ortholog , and daf-16 forkhead FOXO transcription factor ( Friedman and Johnson , 1988; Vowels and Thomas , 1992; Kenyon et al . , 1993; Gottlieb and Ruvkun , 1994; Morris et al . , 1996; Paradis et al . , 1999 ) . We observed that mutations in daf-2 and age-1 significantly reduced GFP levels in ADL neurons of CON adults compared to wild-type , suggesting that insulin signaling is a positive regulator of osm-9 expression ( Figure 3A ) . However , the pdk-1 mutant strain and two different mutations in daf-16 FOXO , which functions downstream in insulin signaling , exhibited wild-type levels of GFP in ADL neurons ( Figure 3A ) . Furthermore , we tested osm-9p::gfp expression in a daf-12 NHR mutant strain , which has a potential binding site ( GGTGTGAC ) located 362 bp upstream of the osm-9 translational start site ( Figure 2A ) ( Ao and Gaudet , 2004 ) . Interestingly , the daf-12 mutant also exhibited significantly decreased GFP in CON adults compared to wild-type ( Figure 3A ) . We were unable to obtain postdauers from the pdk-1 , daf-16 ( m27 ) , and daf-12 strains to examine PD osm-9p::gfp expression . TGF-β and insulin signaling pathways converge onto DAF-12 , which integrates environmental signals and regulates gene expression programs by binding to different ligands ( Mahanti et al . , 2014; Dansey et al . , 2015; Wang et al . , 2015b ) . Our results suggest that insulin signaling , as well as DAF-12 NHR , does not play a role in the developmental programming of osm-9 , but instead contributes to the positive regulation of constitutive osm-9 expression . Together , these results suggest a model where TGF-β and insulin signaling , as well as DAF-12 NHR , promote constitutive osm-9 expression in ADL neurons during favorable growth conditions , resulting in ascr#3 avoidance in adult animals . In unfavorable conditions , reduced TGF-β signaling in conjunction with dauer formation results in the programmed down-regulation of osm-9 in PD ADL neurons through the action of DAF-3 and DAF-5 , resulting in the failure of adult animals to avoid ascr#3 ( Figure 3C ) . Although our data thus far strongly indicate that the developmental programming of osm-9 is established by upstream regulatory sequences , the inability to rescue the ascr#3 avoidance behavioral defect in PD adults of wild-type or osm-9 mutants upon expression of osm-9 coding sequences suggests that the coding sequence may also be a target of developmental programming ( Figure 1D ) . In plants , fungi , and animals , short interfering RNAs ( siRNAs ) have been shown to act in trans to silence homologous sequences either by inhibiting transcription ( transcriptional gene silencing , TGS ) or through the destruction of mRNA ( post-transcriptional gene silencing , PTGS ) ( reviewed in Castel and Martienssen , 2013 ) . To examine whether endogenous RNAi pathways play a role in the developmental programming of osm-9 , we examined osm-9p::gfp expression in a subset of strains carrying mutations in genes with known functions in RNAi pathways ( Figure 4A and Figure 4—figure supplement 1 ) . These strains included animals mutant for components of the Mutator focus , a protein complex that localizes adjacent to P-granules and is thought to contribute to siRNA amplification in the germline ( Zhang et al . , 2011; Phillips et al . , 2012 ) . However , whether Mutator proteins form a complex in somatic tissue and their functions in neurons remain unclear . We found that mutations in five of the seven proteins known to associate with the Mutator focus ( mut-2/rde-3 , mut-14 , mut-15 , mut-16 , and rrf-1 RdRP ) resulted in a significant increase in GFP expression in PD ADL neurons compared to CON ( Figure 4A ) . In previous work , we identified low abundance siRNAs mapping to the osm-9 locus with a bias for exon sequences ( Hall et al . , 2013 ) , indicative of siRNA biogenesis through the action of RdRPs such as RRF-1 ( Figure 4—figure supplement 2 ) ( Pak et al . , 2007; Sijen et al . , 2007 ) . To confirm that endogenous RNAi pathways are contributing to the down-regulation of osm-9 in PD ADL neurons , we again performed smFISH to measure osm-9 mRNA abundance in ADL neurons of CON and PD adults carrying a mutation in the mut-16 gene . For all three trials performed , we observed a significant increase in the mean and maximum fluorescence intensities of PD ADL neurons in mut-16 ( mg461 ) strain compared to wild-type ( Figures 1C and 4B and Figure 4—figure supplement 3 ) . When the trials were combined , the mut-16 ( mg461 ) strain exhibited only a 7% change in mean fluorescence intensity of PD/CON ADL neurons compared to a 42% difference in wild-type ( Figure 4B ) , indicating that functional MUT-16 is required for the full down-regulation of osm-9 mRNA levels in PD ADL neurons . As a control , we also measured gfp mRNA abundance in the mut-16 strain and observed no significant differences in mean and maximum intensity levels between CON and PD ADL neurons ( Figure 4—figure supplement 4 ) . Together , these results indicate a role for the Mutator proteins in the PD control of osm-9 expression in ADL neurons . 10 . 7554/eLife . 11642 . 024Figure 4 . Endogenous RNAi pathways regulate the developmental programming of osm-9 expression and ascr#3 avoidance behavior . ( A ) Percentage of ADL neurons expressing osm-9p::gfp in wild-type and strains carrying mutations in endo-RNAi genes . Data for additional mutant strains is found in Figure 4—figure supplement 1 . N ≥ 2 trials; n ≥ 40 animals ( Figure 4—source data 1 ) . * indicates mutant CON or PD significantly different from wild-type; One-way ANOVA with LSD posthoc correction , p<0 . 05 . ( B ) Quantification of osm-9 mRNA in ADL neurons of mut-16 ( mg461 ) CON and PD using smFISH . The graphs represent individual mean fluorescence measurements for N = 3 biologically independent trials; n ≥ 129 neurons ( Figure 4—source data 2 ) . Maximum florescent measurements are found in Figure 4—figure supplement 3 . Control gfp mRNA measurements are found in Figure 4—figure supplement 4 . Medians are indicated . Comparison of means with S . E . M . of all trials for CON and PD in wild-type and mut-16 is shown . p<0 . 05 , *p< 10–14 , Student’s t-test . ( C ) Log2 PD/CON ratio of ascr#3 avoidance behavior in wild-type and strains carrying mutations in endo-RNAi genes . Data for ADL and ASI-specific expression of mut-16 rescue are averages of two extrachromosomal lines . N ≥ 3 trials; n ≥ 60 animals ( Figure 4—source data 3 ) . indicates PD/CON ratio significantly different from wild-type; & indicates PD/CON ratio significantly different from mut-16 ( mg461 ) ; One-way ANOVA , LSD posthoc correction , p<0 . 05 . All error bars represent S . E . M . Data for controls is found in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02410 . 7554/eLife . 11642 . 025Figure 4—source data 1 . Spreadsheet containing percentages of animals expressing osm-9p::gfp in ADL neurons in strains carrying mutations in genes with RNAi functions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02510 . 7554/eLife . 11642 . 026Figure 4—source data 2 . Spreadsheet containing raw and normalized mean florescence intensities of osm-9 smFISH probes in ADL neurons in mut-16 ( mg461 ) strain . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02610 . 7554/eLife . 11642 . 027Figure 4—source data 3 . Spreadsheet containing PD/CON ascr#3 avoidance ratios for strains carrying mutations in genes with RNAi functions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02710 . 7554/eLife . 11642 . 028Figure 4—figure supplement 1 . The ERGO-1/NRDE-3 endo-siRNA pathway contributes to the developmental programming of osm-9 gene expression . Percentage of CON and PD adults that exhibited GFP in ADL neurons in additional mutant strains with functions in endo-RNAi and chromatin remodeling is shown . * indicates CON or PD significantly different compared to wildtype; p<0 . 05; One-way ANOVA with LSD posthoc correction . N ≥ 2 trials; n ≥ 40 animals . Error bars represent S . E . M . ( Figure 4—figure supplement 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02810 . 7554/eLife . 11642 . 029Figure 4—figure supplement 1—source data 1 . Spreadsheet containing percentage of animals expressing osm-9p::gfp in ADL neurons of additional strains carrying mutations in genes with RNAi and chromatin remodeling functions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 02910 . 7554/eLife . 11642 . 030Figure 4—figure supplement 2 . osm-9 siRNAs map predominantly to exons . ( A ) Schematic of osm-9 gene and regulatory regions . Exons are indicated by gray boxes . A tRNA ( B0212 . t1 ) and a non-coding RNA ( B0212 . 7 ) are located in intron 12 of the osm-9 locus ( shown in purple ) . SiRNAs were sequenced from CON , PD , L3 , and dauer whole animals in previous work ( Hall et al . , 2013 ) ; those sequences mapping to the osm-9 locus are indicated ( Figure 4—figure supplement 2—source data 1 ) . The black arrow indicates the siRNA mapping to the PD motif . ( B ) Characterization of siRNA length and location mapping to the osm-9 locus for each developmental stage . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03010 . 7554/eLife . 11642 . 031Figure 4—figure supplement 2—source data 1 . Spreadsheet containing siRNA sequences mapping to the osm-9 locus for CON , PD , larval L3 , and dauer small RNA libraries . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03110 . 7554/eLife . 11642 . 032Figure 4—figure supplement 3 . osm-9 mRNA levels quantified in mut-16 ( mg461 ) ADL neurons using smFISH . ( A ) Normalized maximum florescence levels in mut-16 ( mg461 ) CON and PD ADL neurons . Medians are indicated . * indicates p<0 . 05 , ** p<10–4 , n . s . not significant , Student’s t-test . ( B ) Graphs represent the average mean and maximum florescence levels of CON and PD ADL mut-16 ( mg461 ) neurons with osm-9 mRNA smFISH probes in arbitrary brightness units ( A . B . U . ) . ( C ) Graphs represent the average mean and maximum florescence background levels of CON and PD ADL mut-16 ( mg461 ) neurons . N = 3 biologically independent trials; n ≥ 129 neurons . Error bars are standard deviation ( Figure 4—figure supplement 3—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03210 . 7554/eLife . 11642 . 033Figure 4—figure supplement 3—source data 1 . Spreadsheet containing mean and maximum florescence intensities for osm-9 smFISH probes in ADL neurons and worm backgrounds in mut-16 ( mg461 ) strain . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03310 . 7554/eLife . 11642 . 034Figure 4—figure supplement 4 . gfp mRNA levels are unaltered between CON and PD ADL neurons in mut-16 ( mg461 ) strain . ( A ) Quantification of gfp mRNA in ADL neurons of mut-16 ( mg461 ) CON and PD using smFISH . ADL neurons were identified by expression of sre-1p::gfp . The graphs represent individual mean and maximum fluorescence measurements for n = 10 neurons for 3 biologically independent trials . None of the comparisons between CON and PD within a trial are significantly different . ( B ) Graphs represent the average mean and maximum florescence levels of CON and PD ADL neurons in mut-16 ( mg461 ) with gfp mRNA smFISH probes in arbitrary brightness units ( A . B . U . ) . ( C ) Graphs representing the average mean and maximum fluorescence levels of worm backgrounds . Error bars are standard deviation ( Figure 4—figure supplement 4—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03410 . 7554/eLife . 11642 . 035Figure 4—figure supplement 4—source data 1 . Spreadsheet containing mean and maximum florescence intensities for gfp smFISH probes in ADL neurons and worm backgrounds in mut-16 ( mg461 ) strain . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 035 To further investigate the role of Mutator proteins in the regulation of osm-9 expression in PD animals , we tested whether the examined mutants affect the ascr#3 avoidance behavior in CON and PD animals . Indeed , we found that mutations in genes associated with the Mutator complex , mut-15 and mut-16 , eliminated the difference in ascr#3 avoidance between CON and PD adults ( Figure 4C ) . The behavioral responses of mut-2 , mut-14 and rrf-1 mutant strains also showed a trend towards disruption in the PD/CON ratio of ascr#3 avoidance ( Figure 4C ) . These results indicate that a subset of the Mutator proteins contributes to the plasticity of ascr#3 avoidance behavior . Since dsRNAs are actively transported between cells and tissue types in C . elegans ( Feinberg and Hunter , 2003; Jose et al . , 2009; Shih and Hunter , 2011; McEwan et al . , 2012; Devanapally et al . , 2015 ) , we asked whether the Mutators act cell-autonomously to regulate ascr#3 avoidance . To address this question , we expressed the mut-16 cDNA under an ADL ( srh-220p ) or ASI ( gpa-4p ) specific promoter in the mut-16 ( mg461 ) strain and examined their ascr#3 avoidance phenotypes ( Figure 4C ) . We chose to examine MUT-16 function in ASI neurons as well , since daf-7 and other TFG-β pathway genes are expressed in ASI ( Ren et al . , 1996 ) . We observed that cell-specific rescue of MUT-16 in ADL partially rescued the ascr#3 avoidance in CON and PD animals ( Figure 4C ) . In contrast , MUT-16 expression in ASI failed to rescue the avoidance behavior ( Figure 4C ) . This result suggests that MUT-16 functions in ADL neurons to mediate ascr#3 avoidance behavior due to developmental history . Both the ERGO-1 and NRDE-3 endo-siRNA pathways require components of the Mutator focus and the ERI complex for the biogenesis of their associated siRNAs ( Duchaine et al . , 2006; Pavelec et al . , 2009; Gent et al . , 2010; Zhang et al . , 2011; Phillips et al . , 2012; Thivierge et al . , 2012 ) . ERGO-1 AGO associates with 26G-siRNAs to target mRNA transcripts in the cytoplasm and acts upstream of the 22G-siRNA-mediated NRDE nuclear RNAi silencing complex , which has been shown to promote heterochromatin formation at targeted gene loci in somatic tissue ( Guang et al . , 2008; Gent et al . , 2010; Guang et al . , 2010; Vasale et al . , 2010; Burkhart et al . , 2011a; Burkhart et al . , 2011b ) . Interestingly , we also observed a significant increase in PD GFP expression in ADL neurons in ergo-1 AGO and nuclear RNAi silencing complex members , nrde-3 AGO , nrde-4 , and nrde-1 mutants ( Figure 4A ) . Moreover , the ERI complex also plays a role in the down-regulation of osm-9 in PD ADL neurons ( Figure 4—figure supplement 1 ) . These results indicate that the Mutator proteins , along with the endo-siRNA ERGO-1/NRDE pathway are playing a role in the developmental programming of osm-9 . Consistent with a role for the ERGO-1 and NRDE pathways in the regulation of osm-9 expression in PD adults , we found that mutations in ergo-1 and nrde-3 also resulted in the elimination of ascr#3 avoidance differences between CON and PD adults ( Figure 4C ) . Although we have not shown that osm-9 is a direct target of endo-siRNA pathways , these results indicate that the Mutator proteins and ERGO-1/NRDE nuclear silencing pathways play a critical role in the developmental programming of osm-9 and mediating the ascr#3 avoidance behavior due to developmental history . Thus far , our results have suggested a model where osm-9 is regulated via transcriptional gene silencing , which would promote alteration of the chromatin state at the osm-9 locus . To ask whether functional chromatin remodeling pathways are necessary for the differential expression of osm-9 we examined osm-9p::gfp expression in CON and PD animals of strains carrying mutations in genes involved in chromatin remodeling ( Figure 5A and Figure 4—figure supplement 1 ) . Although both SET-2 histone H3K4 methyltransferase and HDA-2 histone deacetylase act broadly to regulate histone modifications in the genome , their functions are antagonistic and correlate with gene activation or silencing , respectively ( Simonet et al . , 2007; Hao et al . , 2011 ) . We found that mutations in set-2 and hda-2 resulted in a significant increase in GFP expression in PD ADL neurons compared to wild-type ( Figure 5A ) . Thus , our results suggest the possibility that SET-2 and HDA-2 are targeting additional gene loci that play a role in the down-regulation of osm-9 in PD ADL neurons . We also examined the role of ZFP-1 in the regulation of osm-9 in PD adults , since it has been shown to bind chromatin and is predicted to play a role in the regulation of actively expressed genes through interactions with both chromatin remodeling and RNAi pathways ( Grishok et al . , 2008; Avgousti et al . , 2013 ) . The known function of ZFP-1 is to negatively regulate RNA polymerase II elongation at loci of highly expressed genes ( Cecere et al . , 2013 ) . Interestingly , we observed that a mutation in the zfp-1 zinc finger protein gene eliminated the developmental programming of osm-9p::gfp expression in ADL neurons by significantly increasing GFP in PD adults ( Figure 5A ) . Thus , ZFP-1 is an interesting candidate to play a role in the connection of the chromatin remodeling and the ERGO-1/NRDE-3 pathways in the developmental programming of osm-9 . 10 . 7554/eLife . 11642 . 036Figure 5 . Chromatin remodeling pathways are required for developmental programming of osm-9 and ascr#3 avoidance behavior . ( A ) Percentage of ADL neurons expressing osm-9p::gfp in wild-type and strains carrying mutations in genes with known chromatin remodeling functions . Data for zfp-1 ( ok554 ) represents average GFP expression of an osm-9p::gfp extrachromosomal array in two independent lines . N ≥ 3 trials; n ≥ 60 animals ( Figure 5—source data 1 ) . * indicates CON or PD significantly different from wild-type; One-way ANOVA with LSD posthoc correction , p<0 . 05 . ( B ) Log2 PD/CON ratio of ascr#3 avoidance behavior in wild-type and chromatin remodeling mutant strains . Data for ascr#3 avoidance behavior of ADL-specific zfp-1 rescue strains is an average of two extrachromosomal lines . N ≥ 3 trials; n ≥ 60 animals ( Figure 5—source data 2 ) . * indicates PD/CON ratio significantly different from wild-type; Student’s t-test , p<0 . 05 . & indicates PD/CON ratio significantly different from zfp-1 ( ok554 ) ; Student’s t-test , p<0 . 01 . Data for controls is found in Figure 3—figure supplement 2 . ( C ) Log2 enrichment of ZFP-1 at the osm-9 DAF-3 binding site ( DBS ) and gene body in ADL neurons of CON and PD adults . Promoter regions for egl-30 and act-3 are positive controls ( Cecere et al . , 2013 ) . Data was normalized to ZFP-1 binding to gst-4 ( negative control ) . N ≥ 2 biologically independent trials ( Figure 5—source data 3 ) . * indicates enrichment significantly greater than background or between PD and CON , Student’s t-test , p<0 . 05 . ( D ) Log2 normalized enrichment of DAF-3 SMAD at the osm-9 DAF-3 binding site ( DBS ) in mut-16 ( mg461 ) and zfp-1 ( ok554 ) strains . daf-8 and daf-14 are positive and negative controls , respectively . Bar graph represents IP-qPCR data , normalized to DAF-3 binding to actin act-2 promoter ( Park et al . , 2010 ) , and adjusted to show enrichment of DAF-3 above background levels in daf-3 ( mgDf90 ) strain . N = 3 biologically independent trials ( Figure 5—source data 4 ) . * indicates enrichment significantly greater than background or between PD and CON , Student’s t-test , p<0 . 05 . All error bars represent S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03610 . 7554/eLife . 11642 . 037Figure 5—source data 1 . Spreadsheet containing percentages of animals expressing osm-9p::gfp in ADL neurons in strains carrying mutations in genes with chromatin remodeling functions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03710 . 7554/eLife . 11642 . 038Figure 5—source data 2 . Spreadsheet containing PD/CON ascr#3 avoidance ratios for strains carrying mutations in genes with chromatin remodeling functions . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03810 . 7554/eLife . 11642 . 039Figure 5—source data 3 . Spreadsheet containing ZFP-1 enrichment values in ADL neurons of wild-type CON and PD adults . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 03910 . 7554/eLife . 11642 . 040Figure 5—source data 4 . DAF-3 immunoprecipitation data for the osm-9 , daf-8 , and daf-14 loci in mut-16 ( mg461 ) and zfp-1 ( ok554 ) strains . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 040 Next , we tested the CON and PD adult ascr#3 avoidance response of strains carrying mutations in the genes hda-2 , set-2 , and zfp-1 . We also examined the ascr#3 avoidance response of Heterochromatin protein 1 homolog mutant , hpl-2 , since HPL-2 has been shown to play a role in the NRDE-mediated silencing of the odr-1 gene during olfactory adaptation ( Juang et al . , 2013 ) . We observed that hpl-2 , set-2 , and zfp-1 mutant strains exhibited an elimination or reversal of the wild-type PD/CON ascr#3 avoidance responses , and hda-2 trended towards a reduction in the PD/CON ascr#3 avoidance behavior ( Figure 5B ) . Together , these results suggest that functional chromatin remodeling pathways are required to maintain the down-regulation of osm-9 in PD adults and mediate the ascr#3 behavior due to developmental programming . Since chromatin remodeling proteins act broadly to regulate genes , we asked whether osm-9 is a direct target of chromatin change . ZFP-1 is known to bind chromatin; thus , we tested whether osm-9 is a direct target of ZFP-1 . We created strains expressing the zfp-1 gene tagged with gfp specifically in ADL neurons in a zfp-1 ( ok554 ) mutant . We found that expression of ZFP-1 in ADL partially rescued ascr#3 avoidance behavioral phenotype compared to the zfp-1 mutant , indicating that the ZFP-1 transgene is functional ( Figure 5B ) . Next , we performed chromatin immunoprecipitation of ZFP-1 from ADL neurons of CON and PD adults using α-GFP antibodies , followed by qPCR to test for enrichment of ZFP-1 at candidate gene loci . We found that ZFP-1 was significantly enriched at the osm-9 DBS in PD adults , and enriched at osm-9 coding sequences in both CON and PD ADL neurons ( Figure 5C ) . We also examined ZFP-1 enrichment at the act-3 and egl-30 loci as positive and negative controls , respectively . These loci have been shown to be ZFP-1 targets using whole worm samples ( Cecere et al . , 2013 ) ; however , ZFP-1 enrichment of these loci in ADL neurons was unknown . These results are consistent with ZFP-1 playing a role in the transcriptional gene silencing of osm-9 in ADL neurons of PD animals . Together , our results indicate a role for DAF-3 SMAD , ZFP-1 , and endogenous RNAi pathways in the down-regulation of osm-9 in ADL neurons due to developmental history ( Figure 6 ) . Mutations in individual genes with TGF-β , chromatin remodeling , and RNAi functions resulted in an increase in osm-9p::gfp expression in PD ADL neurons when compared to wild-type , suggesting that these pathways may be coordinating to silence osm-9 . Since dauer formation is required for osm-9 down-regulation , we hypothesized that DAF-3 SMAD binds to the osm-9 promoter during dauer and recruits ZFP-1 and RNAi machinery to the osm-9 locus to maintain the silencing to adulthood . To test our hypothesis , we again immunoprecipitated DAF-3 in mut-16 ( mg461 ) and zfp-1 ( ok554 ) strains and performed qPCR to examine for DAF-3 enrichment at the osm-9 DBS . If DAF-3 binds to osm-9 first and recruits ZFP-1 and RNAi machinery , we would expect to observe DAF-3 enrichment at the osm-9 PD motif in mutant PD adults . However , our results showed that DAF-3 enrichment was significantly decreased in mut-16 and zfp-1 backgrounds at the osm-9 DBS compared to wild-type ( Figure 5D ) , while remained enriched at the daf-8 promoter ( positive control ) . These results indicate that DAF-3 enrichment at the osm-9 DBS in PD animals is dependent upon functional ZFP-1 and MUT-16 . Although we cannot yet address the interactions of these pathways specifically in ADL neurons , our results are consistent with a model that DAF-3 SMAD , ZFP-1 , and endogenous RNAi pathways are acting cooperatively to down-regulate osm-9 as a result of developmental history . 10 . 7554/eLife . 11642 . 041Figure 6 . DAF-3 , ZFP-1 , and RNAi machinery act cooperatively to down-regulate osm-9 in postdauer ADL neurons . Model for developmental programming of osm-9 gene expression as a result of developmental history . See Discussion for details . DOI: http://dx . doi . org/10 . 7554/eLife . 11642 . 041 In our previous work , we identified 2127 genes that exhibited altered expression levels as a consequence of passage through the dauer stage ( Hall et al . , 2010 ) . One question that arose from this observation was how the dauer developmental signal results in establishment of tissue-specific programmed changes in gene expression . DAF-3 SMAD and DAF-5 SNO/SKI are expressed throughout the worm and are required for dauer formation in unfavorable conditions ( Patterson et al . , 1997; da Graca et al . , 2004 ) . Here , we argue that DAF-3 and DAF-5 not only promote dauer formation , but also establish altered expression levels of genes due to developmental programming . First , we showed that the differential expression of osm-9 is dependent upon functional TGF-β signaling during dauer formation , and not just upon passage through the dauer stage ( Figure 3A ) . In addition , we have shown that DAF-3 binds to the PD motif , and is required for the down-regulation of osm-9 in PD ADL neurons ( Figures 3A , B ) . Direct regulation by DAF-3 is an attractive model for how developmental history can result in established gene expression changes for target genes such as osm-9 . Moreover , competitive binding to the PD motif between DAF-3 and other cell or tissue specific transcription factors can make the developmental programming specific to a cell type ( such as ADL ) without affecting expression in other cell types ( such as AWA ) . A similar model of regulation by DAF-3 has been postulated for the myo-2 gene in the pharynx ( Thatcher et al . , 1999 ) . TGF-β signaling has also been shown to regulate neuronal receptor genes based on environmental cues via DAF-3-dependent and independent mechanisms in ASI , AWC , and interneurons ( Nolan et al . , 2002; Lesch and Bargmann , 2010; McGehee et al . , 2015 ) . Since 977 genes in the genome contain the PD motif and the DAF-3 binding site is enriched in chemoreceptor genes ( McCarroll et al . , 2005 ) , we predict that DAF-3 might play a larger role in broadly establishing gene expression changes in response to environmental cues . We are currently testing our model by identifying additional proteins that bind to the PD motif in the osm-9 gene and by examining expression patterns of other genes containing the PD motif . Our results indicate that endogenous RNAi pathways regulate neuronal gene expression in response to developmental history . Temporal regulation of neuronal gene expression has been shown previously in AWC neurons , whereby DAF-3 establishes expression levels of chemoreceptor genes during early larval stages that are maintained in adults by the homeobox transcription factor ortholog , HMBX-1 ( Lesch and Bargmann , 2010 ) . We propose a similar model where DAF-3/DAF-5 contributes to the down-regulation of osm-9 in animals that enter the dauer stage , which is dependent upon functional RNAi pathways ( Figure 5D ) . At this time , we are unable to distinguish whether the requirement of MUT-16 for DAF-3 enrichment at the PD motif is indirect or direct . Our results are consistent with MUT-16 contributing to the positive regulation of daf-3 expression , silencing of osm-9 expression , or both ( Figures 4 and 5D ) . We detected lowly abundant siRNAs homologous to osm-9 in whole animals ( Hall et al . , 2013 ) ; however , the daf-3 locus is targeted by significant numbers of siRNAs that require MUT-16 and CSR-1 AGO for biogenesis ( Claycomb et al . , 2009; Zhang et al . , 2011 ) , which is predicted to positively regulate gene expression by promoting a euchromatic chromatin state at target loci ( Youngman and Claycomb , 2014 ) . Furthermore , we find that components of the Mutator focus play a significant role in the differential regulation of osm-9 in ADL neurons and modulation of ascr#3 avoidance behavior ( Figure 4 ) . To date , the localization and function of Mutator proteins in somatic tissue has been unclear , and some evidence suggests that Mutator component MUT-14 only functions within the germline ( Zhang et al . , 2011; Phillips et al . , 2014 ) . However , MUT-2 , MUT-7 , and the NRDE silencing complex have been shown to play a role in olfactory adaptation mediated by AWC neurons ( Juang et al . , 2013 ) , and our results indicate that the Mutator proteins , MUT-2 , MUT-14 , MUT-15 , and MUT-16 , and the associated RdRP RRF-1 , function to regulate osm-9 expression in ADL ( Figure 4A ) . In addition , we show that MUT-15 , MUT-16 , and RRF-1 contribute to modulation of ascr#3 avoidance behavior , and that MUT-16 is required in ADL neurons for the plasticity of ascr#3 avoidance due to developmental history ( Figure 4B ) . Moreover , we provide evidence that the ERGO-1/NRDE nuclear silencing pathways contribute to the maintained silencing of osm-9 in postdauer animals ( Figure 4A and Figure 4—figure supplement 1 ) . Since the NRDE complex is associated with transcriptional gene silencing , our results suggest that osm-9 regulation in PD ADL neurons by NRDE is more likely to be a direct interaction . We hypothesize that Mutator-amplified siRNAs direct the targeting of the NRDE nuclear silencing complex to the osm-9 locus , which results in transcriptional gene silencing of osm-9 in PD ADL neurons . Furthermore , we speculate that siRNAs generated to target the endogenous osm-9 locus could also direct targeting of the NRDE complex to the ADL-specific osm-9 rescue transgene , resulting in the transcriptional silencing of the osm-9 coding sequence in the absence of the upstream regulatory sequences ( Figure 1D ) . Although we have not shown directly that NRDE-3 targets osm-9 , regulation by the NRDE complex is an attractive hypothesis , since we have evidence of both RNAi and transcriptional silencing mechanisms playing a role in regulating osm-9 ( Figures 1A , B , 2 , and 4A ) . Our results showing that the ERGO-1/NRDE pathway contributes to the down-regulation of osm-9 in PD adults raises new questions of how RNAi pathways regulate endogenous genes due to developmental history . NRDE-3 must be bound by a siRNA to enter the nucleus and target nascent RNAs ( Guang et al . , 2008; Guang et al . , 2010; Burkhart et al . , 2011a; Burkhart et al . , 2011b ) , which suggests a mechanism for selective siRNA amplification in animals with different developmental histories . This prediction is consistent with our previous findings that PD and CON animals have significantly different small RNA profiles ( Hall et al . , 2013 ) . Further investigation of ERGO/NRDE pathways will be necessary to understand their interactions and roles in the maintenance of gene silencing due to developmental programming . Our results have led us to question why osm-9 is down-regulated in PD animals . Dauer pheromone has been shown to have both primer and releaser effects on C . elegans larval development and adult behaviors , respectively ( reviewed in Ludewig and Schroeder , 2013 ) . We speculate that it is advantageous to down-regulate osm-9 expression in animals that have entered the dauer stage . One possibility is that decreased osm-9 expression during dauer results in a failure to respond to ascr#3 , allowing worms to exit the dauer stage in areas of local high pheromone concentration . C . elegans animals that enter the dauer stage due to high population density and do not disperse would have their dauer exit suppressed by the local high concentrations of pheromone ( Golden and Riddle , 1982; 1984 ) . Although ADL has not been implicated in dauer recovery to date , down-regulation of osm-9 could allow dauer animals to become more sensitive to changes in their environment and allow faster dauer recovery when conditions improve ( Gruner et al . , 2014 ) . Second , failure to avoid high concentrations of ascr#3 in postdauer hermaphrodites may serve as a mechanism for increased outcrossing in stressful environments . C . elegans males recover from long periods in the dauer stage at higher rates than hermaphrodites , increasing the opportunities for outcrossing in a postdauer population ( Morran et al . , 2009 ) . We postulate that down-regulation of osm-9 in postdauer hermaphrodites suppresses their dispersal to avoid high concentrations of pheromone , further increasing the opportunities for outcrossing . Thus , we speculate the differential regulation of osm-9 via developmental programming mechanisms potentially serves to suppress both the primer and releaser effects of pheromone in order to promote survival and reproduction in stressful environmental conditions . All C . elegans strains were generated and maintained by using standard methods ( Brenner , 1974; Stiernagle , 2006 ) . Worm strains used in this study are described in Supplementary file 1 . All strains were grown at 20°C or 15oC ( for temperature sensitive strains ) on NGM plates seeded with E . coli OP50 . Developmentally synchronized control and postdauer adults were staged using egg plates as previously described ( Hall et al . , 2010; Ow and Hall , 2015 ) . Controls were obtained by bleaching gravid adults to NGM plates without egg white . To induce worms by high crude pheromone concentrations , dauer formation assays were performed as previously described ( Neal et al . , 2013 ) . Strains with dauer deficient phenotypes were exposed to a starvation protocol and grown at 25°C to produce dauers . To induce dauer formation by starvation , well-fed worms were plated onto seeded 100mm NGM plates , and were monitored daily for the depletion of food and the appearance of dauers . Dauer larvae were obtained by treating starved plates with 1% SDS for 24 hr . osm-9p::gfp and unc-122p::dsRed were cloned into plasmids and microinjected as an extrachromosomal array at a concentration of 30 ng/μL into N2 and integrated into the genome using UV irradiation . The integrated osm-9p::gfp array ( pdrIs1 ) was genetically crossed into mutant backgrounds for imaging . osm-9p::gfp was directly microinjected as an extrachromosomal array into zfp-1 ( ok554 ) strain since the integrated transgene and gene locus were closely linked . Animals were visualized on a Leica CTR550 microscope with a Hamamatsu C10600 camera . Adult animals were dye-filled with 2 mg/mL DiD for two hours prior to imaging in order to identify ADL neurons . A minimum of twenty animals was examined for CON and PD conditions for each strain over at least three trials on separate days . All animals were imaged as young adults , 24 hr after L4 . Primer sequences for osm-9p::gfp cloning are located in Supplementary file 2 . Mutagenesis of the osm-9p::gfp transgene was performed using the QuickChange XL kit ( Agilent ) following the specifications of the manufacturer . Deletions of the osm-9 promoter were made by cloning individual promoter fragments into a plasmid containing gfp . All deletion and mutagenesis transgenes were microinjected as extrachromosomal arrays with co-injection marker unc-122p::dsRed into N2 at a concentration of 30 ng/μL and imaged as described above . Primer sequences are located in Supplementary file 2 . smFISH experiments were performed essentially as described ( Ji and van Oudenaarden , 2012 ) . Briefly , CON and PD adults of the SH239 otIs24 [sre-1p::gfp] strain and SH265 mut-16 ( mg461 ) I; otIs24 were fixed in 10% formaldehyde for 45 min followed by incubation in 70% ethanol for at least 48 hrs at 4°C . Fixed worms were hybridized overnight at 30°C with a set of 32 Stellaris gfp smFISH probes ( labeled with Quasar 670 flourophores , Biosearch Technologies ) ( a kind gift from Oliver Hobert ) and a set of 48 Stellaris osm-9 smFISH probes ( labeled with Quasar 570 flourophores , Biosearch Technologies ) . osm-9 probe sequences are available upon request . Worms were mounted on slides using VectaShield H-1000 ( Vector Laboratories ) . Images for osm-9 smFISH experiements were acquired on a Leica DM5500 B microscope coupled with a Leica CTR5500 electronic box mounted with a Hamamatsu Digital Camera C10600 ORCA R2 . The intensity of osm-9 smFISH signals in the ADL neuron , as demarcated by otIs24 , from maximally projected images of z-stacks ( 0 . 2 μM stacks ) was quantified using the Leica LAS AF 3 . 1 . 0 software . Images of the gfp smFISH signals were taken on a Zeiss Axio Observer Z1 motorized microscope with Hamamatsu Orca Flash 4 . 0 LT camera followed by quantification on Image J . Exposure times were consistent within a trial , and images were not processed before analysis . Fluorescence quantification was performed blindly by two people . Normalization for individual measurements was performed by dividing each measurement by the median of the control samples for each trial . Since the signal to noise ratio was low ( mean background > ( mean – 2StDev ) ) , we did not subtract background values from fluorescence measurements . The drop test acute avoidance response assay was performed using 100 nM ascr#3 , M13 buffer ( negative control ) , and 1M glycerol ( positive control ) as previously described ( Hilliard et al . , 2002; Jang et al . , 2012 ) . Animals were tested off food . For each strain , a minimum of three trials were performed on separate days , with a minimum of 30 animals assayed for each trial . The ascr#3 avoidance behavior was normalized by subtracting the proportion of animals responding to M13 buffer from the proportion of animals responding to ascr#3 . For Figure 1D , the normalized avoidance response for each trial was averaged across trials for the graphs . For Figures 3C , 4C , and 5B , graphs represent averages of PD/CON ratios calculated for each trial . M13 and 1M glycerol avoidance data for each strain is found in Figure 3—figure supplement 2 . Overlap extension PCR ( Higuchi et al . , 1988 ) with Phusion DNA polymerase ( NEB ) was used to construct sre-1p::mut-16 cDNA::gfp , gpa-4p::mut-16 cDNA::gfp , sre-1p::zfp-1 cDNA::gfp , gpa-4p::zfp-1 cDNA::gfp , and sre-1p::osm-9 cDNA::gfp transgenes . To make sre-1p::mut-16 cDNA::gfp , 3 kb of the sre-1 promoter , the mut-16 cDNA ( using a cDNA library prepared from total RNA of a mixed N2 population as the template ) , and the gfp gene ( using Fire vector pPD95 . 75 as the template ) were amplified separately . A final fused PCR was amplified using primers containing XmaI and SbfI sites , digested with XmaI and SbfI and cloned into pUC19 . gpa-4p::mut-16 cDNA::gfp was constructed with the same method , except with 3 kb of the gpa-4 promoter . ADL and ASI-specific zfp-1 transgenes were cloned into pCR TOPO XL ( Life Technologies ) following the instructions of the manufacturer . The fragment consisting of sre-1p::osm-9 cDNA::gfp was microinjected as a linear PCR at 30 ng/μl into N2 and osm-9 ( ky10 ) . Plasmids containing the zfp-1 and mut-16 neuron-specific rescues were microinjected ( Mello et al . , 1991 ) into N2 , mut-16 ( mg461 ) , or zfp-1 ( ok554 ) animals at 50 ng/μl along with an unc-122p::dsRed co-injection marker ( 30 ng/μl ) or genetically crossed into the desired mutant background . Primer sequences are located in Supplementary file 2 . A packed pellet of worms ( ~100 μL ) consisting of staged larval L2 , dauer , CON , or PD animals were subjected to chromatin immunoprecipitation as described using 2% paraformaldehyde as the crosslinking reagent ( Hall et al . , 2010; Hall et al . , 2013 ) . Preparation of the lysate was done using a Sonic Dismembrator Model 100 ( Fisher Scientific ) and immunoprecipitation was performed using 5 μl of α-DAF-3 antibody ( Novus Biologicals NB100-1924 ) . DAF-3 IP was performed for three biologically independent samples for each condition . To determine if DAF-3 was bound to the osm-9 PD motif , quantitative PCR was performed with 2–3 μl of the DAF-3 IP DNA using iTaq Universal SYBR Green Supermix ( BioRad ) following the recommendations of the manufacturer . Ct values were normalized using the act-2 upstream regulatory sequences as previously described ( Park et al . , 2010 ) . Primer sequences are located in Supplementary file 2 . A packed pellet of worms ( ~500 μL ) consisting of control and postdauer young adult populations of SH198 zfp-1 ( ok554 ) III; pdrEx22 [sre-1p::zfp-1 cDNA::gfp; unc-122p::dsRed] , SH261 mut-16 ( mg461 ) I; pdrEx22 , and SH262 daf-3 ( mgDf90 ) X; pdrEx22 were subjected to chromatin immunoprecipitation as described previously using 2% paraformaldehyde as the crosslinking agent ( Hall et al . , 2010; Hall et al . , 2013 ) . Lysates were prepared using a Sonic Dismembrator Model 100 ( Fisher Scientific ) . ZFP-1 immunoprecipitations were performed using 25 μl of GFP-nAb Magnetic beads ( Allele Biotech ) following the recommendations of the manufacturer . To determine ZFP-1 enrichment at gene loci , quantitative PCR was performed using 5 μL of the precipitated DNA using iTaq Universal SYBR Green Supermix ( BioRad ) . Ct values for ZFP-1 enrichment were normalized to the gst-4 gene , which is not a known target of ZFP-1 . Promoter sequences for egl-30 and act-3 were positive controls for ZFP-1 enrichment since they have been identified as ZFP-1 targets ( Cecere et al . , 2013 ) . Primer sequences are located in Supplementary file 2 .	Increasing evidence suggests that experiencing stressful environments early on in life can have profound effects on the health and behavior of adults . For example , stressful conditions in the womb have been linked to adult depression and metabolic disorders . These effects are thought to be the result of changes in the way that genes in specific tissues are regulated in the individuals that have experienced the stress . However , it is not clear how a particular stress can cause long-term changes in gene activity in specific tissues . A microscopic worm called Caenorhabditis elegans is often used as a simple animal model to study how animals develop and behave . Previous studies have shown that adult worms that experienced stress early in life show differences in behavior and gene activity compared to genetically identical worms that did not experience the stress . Here , Sims , Ow et al . asked what signals are required for these changes to happen . The experiments show that a gene called osm-9 – which plays a role in the nervous system – is less active in sensory nerve cells in worms that experienced stress early on in life . This loss of activity resulted in the worms being unable to respond to a particular odor . Two proteins called DAF-3 and ZFP-1 are able to bind to a section of DNA in the osm-9 gene to decrease its activity in response to stress . These proteins are similar to human proteins that are important for development and are associated with some types of leukemia . Further experiments show that small molecules of ribonucleic acid in the “RNA interference” pathway also help to decrease the activity of osm-9 after stress . Together , Sims , Ow et al . ’s findings suggest that environmental conditions in early life regulate the osm-9 gene through the coordinated effort of DAF-3 , ZFP-1 and the RNA interference pathway . The next steps are to investigate how these molecules are able to target osm-9 and to identify other proteins that regulate gene activity in response to stress in early life .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "neuroscience" ]	2016	Developmental programming modulates olfactory behavior in C. elegans via endogenous RNAi pathways
Neural progenitors undergo temporal patterning to generate diverse neurons in a chronological order . This process is well-studied in the developing Drosophila brain and conserved in mammals . During larval stages , intermediate neural progenitors ( INPs ) serially express Dichaete ( D ) , grainyhead ( Grh ) and eyeless ( Ey/Pax6 ) , but how the transitions are regulated is not precisely understood . Here , we developed a method to isolate transcriptomes of INPs in their distinct temporal states to identify a complete set of temporal patterning factors . Our analysis identifies odd-paired ( opa ) , as a key regulator of temporal patterning . Temporal patterning is initiated when the SWI/SNF complex component Osa induces D and its repressor Opa at the same time but with distinct kinetics . Then , high Opa levels repress D to allow Grh transcription and progress to the next temporal state . We propose that Osa and its target genes opa and D form an incoherent feedforward loop ( FFL ) and a new mechanism allowing the successive expression of temporal identities . During brain development , neural stem cells ( NSCs ) generate large numbers of highly diverse neuronal and glial cells in chronological order ( Cepko et al . , 1996; Gao et al . , 2014; Greig et al . , 2013; Holguera and Desplan , 2018 ) . Through a phenomenon known as temporal patterning , NSCs acquire properties that change the fate of their progeny over time ( Kohwi et al . , 2013; Mattar et al . , 2015; Okamoto et al . , 2016 ) . Importantly , temporal patterning of NSCs is an evolutionary conserved process and has been observed across species ranging from insects to mammals ( Alsiö et al . , 2013; Livesey and Cepko , 2001; Toma et al . , 2014 ) . During mammalian brain development , neural progenitors in the central nervous system ( CNS ) undergo temporal patterning by relying on both extrinsic as well as progenitor-intrinsic cues . Wnt7 , for example , is an extracellular ligand required for the switch from early to late neurogenesis in cortical progenitors ( Wang et al . , 2016 ) , Ikaros ( the ortholog of the Drosophila Hunchback ) , in contrast , is an intrinsic factor specifying early-born neuronal fates ( Mattar et al . , 2015 ) . Like Ikaros , intrinsic temporal identity factors in vertebrates are often homologous to factors described in Drosophila ( Naka et al . , 2008; Ren et al . , 2017; Syed et al . , 2017 ) . How these factors are involved in neuronal fate specification and how they are regulated remain unknown . Drosophila has been crucial to understanding stem cell biological mechanisms and in particular distinct temporal patterning processes ( Homem and Knoblich , 2012 ) . During embryonic neurogenesis , Drosophila NSCs , called Neuroblasts ( NBs ) , undergo temporal patterning through a cascade of transcription factors ( Isshiki et al . , 2001 ) . During larval neurogenesis , NB temporal patterning relies on opposing gradients of two RNA-binding proteins ( Liu et al . , 2015; Syed et al . , 2017 ) . Temporal patterning is also seen in intermediate neural progenitors ( INPs ) , the transit-amplifying progeny of a discrete subset of larval NBs called type II NBs ( Bayraktar and Doe , 2013 ) . Once they arise from an asymmetric division of a type II NB , newborn INPs undergo several maturation steps before they resume proliferation: they first turn on earmuff ( erm ) , and Asense ( ase ) , and finally Deadpan ( Dpn ) expression to become mature INPs ( mINP ) ( Bello et al . , 2008; Boone and Doe , 2008; Bowman et al . , 2008; Janssens et al . , 2014; Walsh and Doe , 2017 ) . Then mINPs divide 3–6 times asymmetrically to generate ganglion mother cells ( GMCs ) , which in turn divide to generate a pair of neurons or glia . Analogous to embryonic NBs ( Isshiki et al . , 2001 ) , recent reports suggest that a transcription factor cascade regulates temporal patterning of INPs ( Bayraktar and Doe , 2013 ) . Indeed , the sequential expression of Dichaete ( D ) , Grainyhead ( Grh ) and Eyeless ( Ey ) is required to generate different neurons: D+ INPs produce Brain-specific homeobox ( Bsh ) + neurons , while Ey+ INPs produce Toy+ neurons ( Bayraktar and Doe , 2013 ) . The three temporal identity factors are regulated through various regulatory interactions ( Bayraktar and Doe , 2013; Doe , 2017 ) : D is necessary , but not sufficient , for activating Grh . Grh instead is required for repression of D and activation of Ey ( Bayraktar and Doe , 2013 ) . Therefore , INP temporal patterning is thought to be regulated by a ‘feedforward activation and feedback repression’ mechanism ( Figure 1A ) . Intriguingly however , INP temporal patterning also critically requires the SWI/SNF chromatin remodeling complex subunit Osa ( Eroglu et al . , 2014 ) . Although Osa is not considered a specific temporal identity factor , it is required to initiate temporal patterning by activating the initial factor D . While the Osa target gene hamlet is required for the Grh-to-Ey transition ( Eroglu et al . , 2014 ) , regulation of the first transition is less well understood . This result suggests that in addition to feedforward activation and feedback repression , temporal switch genes are required to ensure correct INP temporal patterning . Nevertheless , D and ham double knock down ( k . d . ) phenotypes do not recapitulate the complete loss of temporal patterning initiation observed in Osa-depleted type II NB lineages , suggesting the contribution of additional unidentified factors . Here , we describe a FACS-based method to isolate INPs from three different temporal identities . By comparing the transcriptomic profiles of each set of INPs , we identify odd-paired ( opa ) , a transcription factor whose expression is enabled by direct binding of Osa to its TSS , as a regulator of temporal patterning and repressor of D . Though Osa enables both D and Opa expression , Opa’s slower activation kinetics allow D to function in a short time window before being repressed by Opa . This mode of action resembles an incoherent feedforward-loop ( FFL ) motif , where an upstream gene directly activates the target gene , meanwhile indirectly repressing it by activating its repressor ( Alon , 2007; Mangan and Alon , 2003 ) . Thus , we uncover a novel mechanism controlling temporal patterning during neurogenesis . To obtain a comprehensive list of temporally regulated genes in INPs , we used FACS to purify INPs at each of their three temporal states: D+ , Grh+ and Ey+ ( Figure 1B ) . For this , we generated fly lines expressing tdTomato under an INP specific promoter ( erm-Gal4 >CD8::tdTomato ) and expressing GFP-fusions of one of the temporal identity factors ( D-GFP , Grh-GFP and Ey-GFP , Figure 1—figure supplement 1A ) . Although D-GFP flies were generated with CRISPR/Cas9 method to knock-in GFP into the endogenous locus , Grh-GFP and Ey-GFP flies were generated as BAC clones insertions ( Spokony and White , 2012 ) . To test if extra copies from BAC clones cause overexpression effects , numbers of each temporal state were quantified in control versus GFP-tagged brains ( Figure 1—figure supplement 1A ) . After dissection and dissociation of third instar larval brains , GFP-positive INP populations ( D-GFP+ , Grh-GFP+ and Ey-GFP+ ) were identified ( Figure 1B and Figure 1—figure supplement 1B ) as the largest cells with highest GFP and tdTomato expression ( Figure 1—figure supplement 1B ) . Using immunofluorescence ( IF ) , these cells were verified to be mature INPs ( Figure 1—figure supplement 1C-D ) . All sorted cells within the INP populations expressed Dpn , indicating a 100% mature INP identity , while unsorted cells showed a mixture of Dpn+ and Dpn- cells ( Figure 1—figure supplement 1C-D ) . We validated the temporal identity of the progenitors by performing IF for their respective temporal identity markers ( Figure 1C–F and Figure 1—figure supplement 2 ) . Importantly , each GFP+ sorted INP population was 100% positive for its respective temporal marker ( Figure 1F ) . In contrast , the unsorted cells consisted of mixed cell populations containing various temporal identities ( Figure 1—figure supplement 2B ) . Lastly , we tested for the presence of sorted cells expressing markers of two temporal identities , which reflects transition states of INP temporal patterning as occurs in vivo . Analyzing Grh IF on D-GFP+ and Ey-GFP+ sorted cells , and Ey IF on Grh-GFP+ sorted cells revealed that sorted populations contained only 4–6% of such double-positive cells ( Figure 1C–F , and Figure 1—figure supplement 2A-C ) , suggesting we can isolate almost pure populations of different temporal states . Collectively , we established the genetic tools and methodology to precisely sort INPs into separate populations according to their three distinct temporal states . Since our stringent FACS sorting conditions led to low RNA yields , we generated cDNA libraries using DigiTag ( Landskron et al . , 2018; Wissel et al . , 2018 ) . With this RNA sequencing strategy , we found 458 genes expressed differently between D+ and Grh+ INPs , and 466 genes between Grh+ and Ey+ INPs ( FDR 0 . 05 , log2foldchange > 1 , and Rpm ( reads per million mapped reads ) >10 in one of three samples/D+ , Grh+ or Ey+ INPs ) . Hierarchical clustering identified genes specifically expressed in certain temporal states , and therefore potentially involved in temporal patterning ( Figure 1G ) . First , we confirmed the quality of our dataset by examining the transcriptional changes of temporal identity genes with quantitative PCR ( qPCR ) ( Figure 1—figure supplement 1E ) . As expected , each temporal state had high expression levels of their own temporal identity genes . Second , we confirmed the expression of known temporal identity genes ( Figure 1—figure supplement 1F ) . FACS-purified Grh+ INPs expressed high levels of Ey mRNA . However , immunofluorescent analysis showed that Grh+ INPs expressed only low levels of Ey protein , suggesting that post-transcriptional modifications regulate the Grh-to-Ey transition ( Figure 1C–F and Figure 1—figure supplement 1F ) . Third , we performed GO-term analysis on the identified gene clusters . Genes upregulated in D+ INPs showed enrichment for mitochondrial translation , cellular nitrogen compound metabolic process and gene expression ( Figure 1—figure supplement 2D ) . Genes upregulated in Grh +INPs were enriched for protein binding and system development ( Figure 1—figure supplement 2E ) . Finally , genes upregulated in Ey +INPs were enriched for neurogenesis and sequence-specific DNA binding ( Figure 1—figure supplement 2F ) . Interestingly , we observed that the glial identity-promoting factor glial cell missing ( gcm ) and cell cycle inhibitor dacapo ( dap ) were upregulated in Ey+ INPs ( Figure 1G—figure supplement 1F ) . These observations support previous findings indicating that INPs begin producing glia cells instead of neurons during their later cell divisions , and that Ey is required for cell cycle exit ( Baumgardt et al . , 2014; Bayraktar and Doe , 2013; Ren et al . , 2018; Viktorin et al . , 2013 ) . To identify genes that regulate transitions of temporal patterning , we focused on genes with a dynamic expression pattern between INP populations . To this end , we focused on genes with a log2foldchange > 1 in either the D-to-Grh or Grh-to-Ey transition . From this list , we excluded genes with a log2foldchange < 0 . 5 in the remaining transition . We applied a cut-off of Rpm ( reads per million mapped reads ) >50 in one of the three temporal identity states due to the fact that all the other temporal identity factors , along with osa and ham , had high expression levels . With these criteria , we identified 71 genes ( Supplementary file 1 and Supplementary file 2 ) , 49 of which displayed an expression pattern of high in D + INPs , low in Grh +INPs , and finally higher in Ey +INPs . Among these genes , odd-paired ( opa ) was ranked as the 5th hit that is most downregulated in Grh+ INPs ( Figure 1G–I , Supplementary file 1 ) . Since Osa binds to the TSS of opa in order to prime its expression ( Eroglu et al . , 2014 ) , we investigated in detail the potential role of Opa in regulating INP temporal patterning . Opa is a transcription factor containing five zinc finger domains and is essential for para-segmental subdivision of Drosophila embryos ( Benedyk et al . , 1994; Mizugishi et al . , 2001 ) . During development , Opa ensures the timely activation of the transcription factors engrailed and wingless ( Benedyk et al . , 1994 ) . To test if opa regulates INP temporal patterning , we depleted opa using RNAi expressed specifically in INPs with ermGal4 . Opa knockdown slightly increased the total number of INPs ( Dpn+ cells ) , but drastically increased the number of D+ INPs while decreasing the number of both Grh+ and Ey+ INPs ( Figure 2A–D ) . We confirmed this result by performing mosaic analysis with a repressible cell marker ( MARCM ) to create mosaic opa ( -/- ) mutant or control opa ( +/+ ) GFP+ cell clones ( Lee and Luo , 1999 ) . Control clones were indistinguishable from WT , whereas opa mutant clones contained predominantly D+ INPs , at the expense of the other two temporal states ( Figure 2E–F ) . The RNAi and mosaic mutant analysis both indicate that loss of Opa causes a shift in INP temporal state identity such that the early generated D+ INPs are increased while the later generated Grh+ and Ey+ INPs are decreased . These results suggest that opa is regulating the D-to-Grh transition by either repressing D or activating Grh . Since it has been previously shown that Grh is not sufficient for D repression ( Bayraktar and Doe , 2013 ) , we tested whether the main role of opa is to repress D . For this , we depleted opa in DM1 lineages , which undergo temporal patterning by expressing only D and then Ey ( Figure 2—figure supplement 1 ) . Opa knock-down in DM1 lineages caused a significant increase in the number of D+ INPs at the expense of Ey+ INPs , suggesting that opa is required for D repression ( Figure 2—figure supplement 1 ) . Finally , we tested if opa regulates processes upstream of temporal patterning during the stages of initial INP maturation with a type II-specific driver line . When expressing opa RNAi specifically in type II NBs , we observed no effect on INP maturation ( Figure 2—figure supplement 2A ) as observed with sequential activation of Ase and Dpn , but immunofluorescent analysis of INPs for D , Grh and Ey expression showed the same phenotype as INPs depleted for opa ( Figure 2—figure supplement 2B-D ) . Collectively , these data suggest that opa inhibits D expression . Furthermore , similar to hamlet , Opa appears to act as a temporal identity switch gene , controlling the transition from a D+ to a Grh+ state . To test if opa knock-down impairs INP asymmetric cell division leading to the disruption in temporal patterning , we analyzed the expression of Mira , a known scaffolding protein that localizes asymmetrically during cell division , and aPKC , which localizes to apical cortex ( Figure 2—figure supplement 2E ) . Opa-depleted INPs can asymmetrically segregate Mira and aPKC , which suggests that asymmetric division is normal . Thus , opa is indeed a temporal switch factor required for the D-to-Grh state . INP temporal patterning results in the production of different neuronal subtypes at distinct periods of neurogenesis . For instance , ‘young’ , D+ INPs produce Brain-specific homeobox ( Bsh ) + neurons and ‘old’ , Ey+ INPs produce Toy+ neurons ( Bayraktar and Doe , 2013 ) . Since the progression of INP temporal identity is disrupted in opa-depleted INPs , we tested whether this disrupted identity affects the production of different types of neurons . INP-driven opa RNAi displayed a significant increase in Bsh+ neurons , at the expense of Toy+ neurons ( Figure 3A–C ) . In addition , opa-depleted MARCM clones also contained increased numbers of Bsh+ neurons compared to wild-type counterparts ( Figure 3D ) . This result confirms that shifting the INP identity toward a D+ identity leads to a concomitant increase in the Bsh+ neurons produced by D+ INPs . Thus , altering the temporal identity progression of neural progenitors can alter the proportions of neuronal subtypes in the brain . We next investigated whether altering the proportions of neuronal subtypes leads to a defect on brain morphology and function . The adult central complex ( CCX ) brain region relies on type II NB neurogenesis ( Bayraktar et al . , 2010; Izergina et al . , 2009 ) . Opa-depletion in INPs caused major alterations in the gross morphology of the adult CCX . The fan-shaped body ( FB ) was enlarged , the noduli ( NO ) and ellipsoid body ( EB ) only partially formed , and the protocerebral bridge ( PB ) appeared fragmented ( Figure 3E ) . Since the CCX is required for adult motor functions ( Callaerts et al . , 2001; Young and Armstrong , 2010 ) , we tested whether altered CCX morphology affected motor behavior . Compared to control flies , INP-driven opa RNAi caused impaired negative geotaxis performance ( Figure 3F ) . Thus , opa is a temporal switch gene required for neuronal subtype specification , which is required for the correct assembly and function of the adult central complex . Thus , the temporal identity specification of neural progenitors is crucial for proper neural cell complexity , and brain function . If opa is required for the D-to-grh transition , what is the molecular mechanism of this transitional regulation ? To answer this question , we first confirmed that opa is indeed a target of Osa in type II NB lineages by analyzing opa protein expression within the NB lineage , and whether this expression is regulated by Osa . We generated healthy , homozygous , endogenously C-terminally tagged opa::V5 knock-in flies ( Figure 4—figure supplement 1A ) . Through immunofluorescent analysis of V5 tag expression , we observed that Opa is expressed throughout the type II lineage in INPs ( marked with Dpn and Ase ) and , GMCs ( Pros+ cells ) and neurons , but not in NBs ( Dpn+ ) or immature INPs ( Dpn-/Ase- or Dpn-/Ase+ cells ) ( Figure 4—figure supplement 1B-D ) . Opa is also expressed in the DM1 lineage , even though DM1 lineages display a temporal patterning lacking Grh expression ( Figure 1—figure supplement 1E ) . To check the specificity of the opa-V5 line , we depleted opa specifically in type II lineages using RNAi . As expected , opa-V5 expression decreased with opa-RNAi ( Figure 4—figure supplement 1E-F ) . The proper expression of opa is dependent on Osa , since Osa-knockdown in type II NBs resulted in a loss of Opa ( Figure 4—figure supplement 2A and B ) . Since both D and opa are direct Osa targets , we next compared the expression pattern of D and opa ( Figure 4A ) . Without exception , D+/opa- INPs appeared before D+/opa+ cells in the lineage ( Figure 4A ) . However , in later temporal states , all Grh+ and Ey+ INPs expressed opa ( Figure 4B , and Figure 4—figure supplement 3A ) . Our transcriptome data suggest that opa expression fluctuates throughout the three different INP populations . To confirm this hypothesis , we calculated the intensity of the opa-V5 signal among these three populations ( Figure 4C–D , and Figure 4—figure supplement 3B ) . Indeed , we found that D+ INPs express the highest opa protein levels ( Figure 4C ) , while Grh+ INPs express the lowest ( Figure 4D and Figure 4—figure supplement 3B ) . Since D expression precedes opa expression , it is possible that D activates opa . However , upon type II NB specific D knockdown , opa localization was unchanged ( Figure 4E ) . Interestingly , D knockdown alone also did not prevent later temporal stages , Grh and Ey , to appear ( Bayraktar and Doe , 2013 ) , suggesting that other factor ( s ) are required to maintain temporal identities in INPs . Since Osa-depleted type II NB lineages fail to initiate temporal patterning ( Eroglu et al . , 2014 ) , we hypothesized that one of these unidentified factors could be a target of Osa that remains expressed in D-depleted INPs , such as opa . To test this hypothesis , we examined the epistatic genetic interactions between D and Opa . Double knock down of D and opa by type II NB-specific RNAi produced type II lineages containing fewer Dpn+/Ase+ INPs compared to controls ( Figure 4F–G ) . This result suggests that even though D and opa are Osa targets , two of them alone cannot fully account for Osa tumor suppressor role ( Figure 4F–G ) . Importantly , all known temporal identity markers on the remaining cells were absent , suggesting a complete loss of temporal identity in these INPs ( Figure 4F–G ) . However , since these cells also lost their INP identity due to lack of Dpn and Ase , they exhibit a different phenotype than Osa knockdown . Therefore , our data suggest that opa is required for the repression of D , the activation of Grh , and thus the progression of temporal identities in INPs . If Opa suppresses D , one puzzling aspect of our data is the presence of double-positive D+/opa+ INPs ( Figure 4A ) . To better understand this paradox , we overexpressed opa in type II NBs during a period before D is normally expressed . Overexpression of opa resulted in shorter lineages ( Figure 5—figure supplement 1A-B ) , decreased total INP numbers ( Figure 5—figure supplement 1A ) , and a loss of type II NBs ( marked by Dpn or Mira ) ( Figure 5—figure supplement 1A-B ) . Co-expressing the apoptosis inhibitor p35 did not prevent NB loss or shortened lineages , suggesting that opa overexpression does not induce cell death , but causes premature differentiation instead ( Figure 5—figure supplement 1C ) . NBs and INPs overexpressing opa successfully segregated Mira and aPKC , excluding that asymmetric cell division was altered ( Figure 5—figure supplement 1D-E , and Figure 2—figure supplement 2E ) . Overexpressing opa in type II NB lineages caused complete loss of D+ INPs , but the few remaining INPs could still activate Grh and Ey ( Figure 5A–C ) , which is similar to D knockdown phenotype ( Bayraktar and Doe , 2013 ) . To exclude that these could result from altered NB patterning , we next overexpressed opa in an INP-specific manner during a stage where D is normally expressed . Opa overexpression caused a decrease in D+ INPs ( Figure 5D–F ) , and a concomitant increase in both Grh+ and Ey+ INP populations ( Figure 5D–F ) . This result further indicates that Opa represses the early D+ temporal identity , but also activates later Grh+ temporal identity . We also overexpressed opa in DM1 lineages in an INP-specific manner , which resulted in a decrease in D+ INP numbers and an increase in Ey+ INPs ( Figure 5—figure supplement 2A and C ) . However , ectopic Grh expression was undetectable ( Figure 5—figure supplement 2B ) , suggesting opa mis-expression does not cause ectopic Grh expression . Collectively , these results show that opa-mediated repression of D depends on Opa expression levels . Having established an interaction between opa and D , we next wondered if opa and ham , two temporal switch genes , can recapitulate the Osa loss-of-function phenotype , a more upstream regulator of lineage progression in type II NBs . Osa knock-down causes INPs to revert back to the NB-state due to a failure to initiate temporal patterning , while single depletion of opa or ham leads to either an increase in D+ or Grh+ cells , respectively ( Figure 2; Eroglu et al . , 2014 ) . Co-expressing opa RNAi with ham shmiR in an INP-specific manner caused supernumerary Dpn+ , Ase+ INPs ( Figure 6—figure supplement 1A ) . In addition , the number of D+/Dpn+ and Grh+/Dpn+ INPs were also increased , which is in contrast to single depletion of opa or ham ( Figure 6A–B , Figure 2; Eroglu et al . , 2014 ) . Thus , opa and ham loss-of-function phenotypes are additive . Importantly , despite inducing over-proliferation of mature INPs ( Ase+/Dpn+ ) , depleting both opa and ham in type II NBs could not recapitulate the Osa loss-of-function phenotype because imINPs could mature and express Ase , and therefore did not revert into ectopic NBs ( Figure 6—figure supplement 1B ) . This suggests that Osa regulates temporal patterning in two levels: initiation by D activation , and progression by opa and ham . Temporal patterning is a phenomenon where NSCs alter the fate of their progeny chronologically . Understanding how temporal patterning is regulated is crucial to understanding how the cellular complexity of the brain develops . Here , we present a novel , FACS-based approach that enabled us to isolate distinct temporal states of neural progenitors with very high purity from Drosophila larvae . This allowed us to study the transitions between different temporal identity states . We identified odd-paired ( opa ) , a transcription factor that is required for INP temporal patterning . By studying the role of this factor in temporal patterning , we propose a novel model for the regulation of temporal patterning in Drosophila neural stem cells . We establish two different roles of the SWI/SNF complex subunit , Osa , in regulating INP temporal patterning . Initially , Osa initiates temporal patterning by activating the transcription factor D . Subsequently , Osa regulates the progression of temporal patterning by activating opa and ham , which in turn downregulate D and Grh , respectively ( Figure 6C ) . The concerted , but complementary action of opa and ham ensures temporal identity progression by promoting the transition between temporal stages . For instance , opa regulates the transition from D to Grh , while ham regulates the transition from Grh to Ey . We propose that opa achieves this by repressing D and activating grh , as indicated by the lack of temporal patterning in D and opa-depleted INPs ( Figure 4C–D , Figure 6C ) . Loss of opa or ham causes INPs to lose their temporal identity and overproliferate . Moreover , we propose that D and opa activate Grh expression against the presence of ham , which represses Grh expression . As D and opa levels decrease as INPs age and become Grh positive , ham is capable of repressing Grh later on in temporal patterning ( Figure 6C ) . This explains how opa and ham act only during specific stages even though they are expressed throughout the entire lineage . An open question pertains to the fact that the double knock-down of opa and ham , as well as that of D and opa , failed to recapitulate the Osa phenotype . Even though opa and ham RNAi caused massive overproliferation in type II lineages , we could not detect any Dpn+ Ase- ectopic NB-like cells ( as occurs in Osa mutant clones , Eroglu et al . , 2014 ) . We propose that this is caused by D expression which is still induced even upon opa/ham double knockdown , but not upon Osa knock-down where D expression fails to be initiated . Thus , the initiation of the first temporal identity state may block the reversion of INPs to a NB-state . In the future , it will be important to understand the exact mechanisms of how opa regulates temporal patterning . We further demonstrate that Osa initiates D expression earlier than opa expression . Osa is a subunit of SWI/SNF chromatin remodeling complex , and it guides the complex to specific loci throughout the genome , such as the TSS of both D and opa . The differences in timing of D and opa expression may be explained by separate factors involved in their activation . Previous work suggests that the transcription factor earmuff may activate ( Janssens et al . , 2014; Janssens et al . , 2017 ) . However , it remains unknown which factor activates opa expression . One possibility is that the cell cycle activates opa , since its expression begins in mINPs , a dividing cell unlike imINPs , which are in cell cycle arrest . We propose that balanced expression levels of D and opa regulates the timing of transitions between temporal identity states . Indeed , Osa initiates D and opa , the repressor of D , at slightly different times , which could allow a time window for D to be expressed , perform its function , then become repressed again by opa . Deregulating this pattern , for example by overexpressing opa in the earliest INP stage , results in a false start of temporal patterning and premature differentiation . This elegant set of genetic interactions resembles that of an incoherent feedforward loop ( FFL ) ( Kim et al . , 2008; Mangan and Alon , 2003 ) . In such a network , pathways have opposing roles . For instance , Osa promotes both the expression and repression of D . Similar examples can be observed in other organisms , such as in the galactose network of E . coli , where the transcriptional activator CRP activates galS and galE , while galS also represses galE ( Shen-Orr et al . , 2002 ) . In Drosophila SOP determination , miR-7 , together with Atonal also forms an incoherent FFL ( Li et al . , 2009 ) . Furthermore , mammals apply a similar mechanism in the c-Myc/E2F1 regulatory system ( O'Donnell et al . , 2005 ) . The vertebrate homologues of opa consist of the Zinc-finger protein of the cerebellum ( ZIC ) family , which are suggested to regulate the transcriptional activity of target genes , and to have a role in CNS development ( Elms et al . , 2004; Elms et al . , 2003; Gaston-Massuet et al . , 2005; Inoue et al . , 2004; Inoue et al . , 2007 ) . In mice , during embryonic cortical development , ZIC family proteins regulate the proliferation of meningeal cells , which are required for normal cortical development ( Inoue et al . , 2008 ) . In addition , another member of the ZIC family , Zic1 , is a Brn2 target , which itself controls the transition from early-to-mid neurogenesis in the mouse cortex ( Urban et al . , 2015 ) . Along with these lines , it has been shown that ZIC family is important in brain development in zebrafish ( Maurus and Harris , 2009; Sanek and Grinblat , 2008 ) . Furthermore , the role of ZIC has been implicated in variety of brain malformations and/or diseases ( Aruga et al . , 2010; Blank et al . , 2011; Hatayama et al . , 2011 ) . These data provide mere glimpses into the roles of ZIC family proteins in neuronal fate decisions in mammals , and our study offers an important entry point to start understanding these remarkable proteins . Our findings provide a novel regulatory network model controlling temporal patterning , which may occur in all metazoans , including humans . In contrast to existing cascade models , we instead show that temporal patterning is a highly coordinated ensemble that allows regulation on additional levels than was previously appreciated to ensure a perfectly balanced generation of different neuron/glial cell types . Together , our results demonstrate that Drosophila is a powerful system to dissect the genetic mechanisms underlying the temporal patterning of neural stem cells and how the disruption of such mechanisms impacts brain development and behavior . The following Drosophila stocks were used: UAS-opaRNAi ( VDRC , TID: 101531 ) , UAS-mcherryshmiR ( BL35785 ) , UAS-DRNAi ( VDRC , TID: 49549 , 107194 ) , UAS-osaRNAi ( VDRC , TID: 7810 ) , UAS-hamshmiR ( BL32470 ) , UAS-osashmiR ( Eroglu et al . , 2014 ) , UAS-p35 , UAS-opa ( Lee et al . , 2007 ) , PBac{grh-GFP . FPTB}VK00033 ( BL42272 ) , PBac{EyGFP . FPTB}VK00033 ( BL42271 ) ( Spokony and White , 2012 ) , D::GFP ( generated in this study ) , opa::V5 ( generated in this study ) . GAL4 driver lines used: UAS-cd8::tdTomato; ermGal4 , UAS-cd8::GFP; ermGal4 ( Pfeiffer et al . , 2008; Weng et al . , 2010 ) , UAS-dcr2; worGal4 , aseGal80; UAS-cd8::GFP ( Neumüller et al . , 2011 ) , UAS-dcr2; UAS-cd8::GFP; VT17-Gal4 ( VDRC , TID: 212057 , discarded ) . Mutant fly strains used for clonal analysis were FRT82B , opa7 ( Lee et al . , 2007 ) . Clones were generated by Flippase ( FLP ) /FLP recombination target ( FRT ) -mediated mitotic recombination , using the elavGal4 ( C155 ) ( Lee and Luo , 1999 ) . Larvae were heat shocked for 90 min at 37°C and dissected as third-instar wandering larvae ( 120 hr ) . RNAi crosses were set up and reared at 29°C , and five days later , third-instar wandering larvae were dissected . w118 was used as control for comparison with RNAi lines , whereas UAS-mcherryshmiR was used as control for comparison with shmiR lines , and experiments involving UAS-transgenes . For both genes , the guides were cloned as overlapping oligos into linearized pU6-BbsI-chiRNA ( Addgene 45946 , Gratz et al . , 2013 ) and injected at 100 ng/μl into actCas9 flies ( BL 54590 , Port et al . , 2014 ) . Donors ( either oligos or plasmid ) were co-injected at 250 ng/μl . For opa , donors were Ultramer Oligos from IDT with around 60nt homology arms on either side . For D , homology arms were 800 bp and 900 bp long . Donor plasmid contained GFP , V5 , 3xFlag , and dsRed . They were screened for dsRed eyes and then , the selection cassette was removed with hsCre ( BL 851 ) . opa gRNA GATGCATCCCGGCGCAGCGA opa donor GAACCCGCTGAACCATTTCGGACACCATCACCACCACCACCACCTGATGCATCCCGGCGCgGCaACcGCGTATggtaagcctatacctaaccctcttcttggTCTAGAtagcacgTGAGAGTGGGAGAACTGGTGGCCCGAGGAGGCGCCACCGCCGGCCGCCCAACCGA D gRNA GTGCTCTATTAGAGTGGAGT Negative geotaxis assay was used as described before ( Ali et al . , 2011 ) , where the percentage of flies passing the 8 . 5 cm mark in 10 s was assessed . For each genotype and gender , 10 two-day old adult flies in 10 biological replicates were measured and for each replicate , 10 measurements were performed with 1 min rest period in between . Larval or adult brains were dissected in 1X PBS , and then fixed for 20 min at room temperature ( RT ) in 5% paraformaldehyde in PBS and washed once with 0 . 1% TritonX in PBS ( PBST ) . The brains were incubated for 1 hr at RT with blocking solution ( 5% normal goat serum or 1% BSA in PBST ) . Blocking was followed by overnight incubation at 4°C with primary antibodies in blocking solution . Then , the brains were washed three times with PBST , and incubated for 1 hr at RT with secondary antibodies ( 1:500 , goat Alexa Fluor , Invitrogen ) in blocking solution . After secondary antibody , brains were washed three times with PBST , and mounted in Vectashield Antifade Mounting Medium ( Vector Labs ) . Antibodies used in this study were: guinea pig anti-Deadpan ( 1:1000 , Eroglu et al . , 2014 ) , rat anti-Asense ( 1:500 , Eroglu et al . , 2014 ) , guinea pig anti-Miranda ( 1:500 , Eroglu et al . , 2014 ) , rat anti-Grh ( 1:1 , 000; Baumgardt et al . , 2009 ) ; rabbit anti-D ( 1:1 , 000; gift from Steve Russell ) ; mouse anti-Ey ( 1:10; DSHB ) ; guinea pig anti-Toy ( gift from Uwe Walldorf ) , guinea pig anti-Bsh ( gift from Makoto Sato ) , mouse anti-Bruchpilot nc82 ( 1:10 , DSHB ) , mouse anti-V5 ( 1:500 , Sigma Aldrich , V8012 ) , mouse antiV5 IgG2a ( Thermo Fisher Scientific , R960-25 , used in Figure 4—figure supplement 1D ) , rabbit anti-V5 ( Abcam , ab9116 , used in Figure 4—figure supplement 3A ) , mouse anti-Pros ( 1:100 , Developmental Studies Hybridoma Bank ) , mouse anti-pH3 ( Ser10 ) ( 1:500 , Cell Signaling Technologies , 9701S ) , rabbit anti-aPKC ( 1:500 , Santa Cruz Biotechnology , sc-216 ) . Throughout the paper , for every quantification , dorsomedial 2 and 3 type II NB lineages ( DM2 and 3 ) were considered , if not stated otherwise . FACS-sorted cells from ~300 larval brains ( UAS-cd8::tdTomato , ermGal4 ) or their unsorted control matches were plated on cover glass ( Labtek II Chambered Coverglass , 8-well , 155409 , Thermo Fisher Scientific ) into Schneider’s medium ( Homem et al . , 2013 ) . The dishes were placed onto ice and cells were incubated for 1 hr to settle down . Cells were then fixed with 5% PFA in PBS at RT and washed three times with 0 . 1% PBST . After washes , cells were incubated for 1 hr at RT with blocking solution ( 5% normal goat serum in 0 . 1% PBST ) . The cells were then incubated overnight at 4°C with primary antibodies in blocking solution , which was followed by three washes with 0 . 1% PBST , and secondary antibody ( 1:500 , goat Alexa Fluor , Invitrogen ) incubation for 1 hr at RT . Cells were again washed three times with 0 . 1% PBST , and then mounted in in Vectashield Antifade Mounting Medium with Dapi ( Vector Labs ) . Confocal images were acquired with Zeiss LSM 780 confocal microscopes . Embryos were collected and dechorionated , then boiled in 2x Laemmli buffer and loaded on 4–12% gradient Bis-Tris gels ( NuPAGE , Invitrogen ) . After SDS-PAGE according to Invitrogen’s protocol , proteins were transferred to a Nitrocellulose membrane ( 0 . 22 µm , Odyssey LI-COR ) for 2 hr at 100V , blocked with 5% milk powder in blocking solution ( PBS with 0 . 2% Tween ) for 1 hr , overnight incubation with primary antibody in blocking solution at 4°C , 3x washed with washing solution ( PBS with 0 . 1% Tween ) and followed by 1 hr incubation with secondary antibody ( 1:15000 , goat IRDye , LI-COR ) in blocking solution . After three washes with washing solution , the membranes were air-dried , and fluorescent signal were detected with Odyssey CLx imaging system ( Odyssey CLx LI-COR ) . Antibodies used were: mouse anti-V5 ( 1:1000 , Sigma Aldrich , V8012 ) , anti-alpha tubulin ( 1:10000 , Sigma Aldrich , T6199 ) . For intensity measurements of opa-V5 signal , cells expressing Dpn and temporal identity markers ( D , Grh or Ey ) were circled with selection tools . Raw integrity density ( sum of gray values of all selected pixels ) was measured using FIJI . In each image , five temporal identity positive INP and five temporal identity negative INP were measured for raw integrity density along with three background circles with no opa-V5 signal , ( eg . D+ vs D- INPs ) . Then , corrected total cell fluorescent ( CTCF ) were calculated with ‘Integrated density – ( Area of selected cells X Mean fluorescence of background readings ) ’ ( McCloy et al . , 2014 ) . Then , the mean of temporal identity positive versus negative cells were calculated and the values were normalized to means of background for each brain . Statistical analyses were performed with GraphPad Prism 7 . Unpaired two-tailed Student’s t-test was used to assess statistical significance between two genotypes . Experiments were not randomized , and investigator was not blinded . Sample sizes for experiments were estimated on previous experience with similar setup which showed significance , thus , no statistical method was used to determine sample size . Cell dissociation and FACS were performed as previously described with minor changes ( Berger et al . , 2012; Harzer et al . , 2013 ) . UAS-cd8::tdTomato; ermGal4 driver line was used to induce expression of membrane bound tdTomato in INPs . In addition to the driver lines , temporal identity factors were tagged with GFP . Flies expressing both fluorophores were dissected at L3 stage , and then dissociated into single cell suspension . Decreasing levels of tdTomato were observed in differentiated cells due to lack of driver line expression . Thus , biggest cells with highest tdTomato expression and highest GFP expression were sorted . For RNA isolation , cells were sorted directly in TRIzol LS ( 10296010 , Invitrogen ) , while for cell staining , they were sorted on coated glass-bottomed dishes and stained as previously described ( Berger et al . , 2012 ) . RNA was isolated using TRIzol LS reagent ( 10296010 , Invitrogen ) from FACS sorted cells . Then RNA samples were used as template for first-strand cDNA synthesis with random hexamer primers ( SuperScriptIII , Invitrogen ) . qPCR was done using Bio-Rad IQ SYBR Greeen Supermix on a Bio-Rad CFX96 cycler . Expression of each gene was normalized to Act5c , and relative levels were calculated using the 2-ΔΔCT method ( Livak and Schmittgen , 2001 ) . Primer used were: act5c AGTGGTGGAAGTTTGGAGTG , GATAATGATGATGGTGTGCAGG D ATGGGTCAACAGAAGTTGGGAG , GTATGGCGGTAGTTGATGGAATG grh TCCCCTGCTTATGCTATGACCT , TACGGCTAGAGTTCGTGCAGA ey TCGTCCGCTAACACCATGA , TGCTCAAATCGCCAGTCTGT ham ATAGATCCTTTGGCCAGCAGAC , AGTACTCCTCCCTTTCGGCAAT opa CTGAACCATTTCGGACACCATC , CCAGTTCTCCCACTCTCAATAC For each experiment 6000–7000 FACS-sorted D+ , Grh+ or Ey+ INPs were isolated by TRIzol purification . Three replicates from each temporal state were analyzed . RNA samples were reverse transcribe into first-strand cDNA using SuperScriptIII Reverse Transcriptase ( Invitrogen ) with oligo- ( dT ) 2- primers . Then the second-strand cDNA were generated . It was followed by library preparation with Nextera DNA Library Preparation Kit ( Illumina ) as previously described ( Landskron et al . , 2018; Wissel et al . , 2018 ) . Libraries were purified with Agencourt AMPure XP beads . Purified libraries were then subjected to 50 base pair Illumina single-end sequencing on a Hiseq2000 platform . Genes are filtered by the indicated log2fc and an adjusted P value < 0 . 05 in at least one pairwise comparison . In addition , a minimal expression of 10 RPM in at least one condition was required . The tree cut into four clusters ( different cluster numbers were tested; Kolde and Package , 2015 , 202AD ) . GO analysis was performed with FlyMine ( Lyne et al . , 2007 ) , Holm-Bonferroni correction with max p-value 0 . 05 was used . Biological process and molecular function were the ontologies . The Gene Expression Omnibus accession number for the RNA-sequencing data reported in this paper is GSE127516 . Gene Ontology ( GO ) enrichment analysis were performed on www . flymine . org/with Holm-Bonferroni correction with max p-value 0 . 05 . Biological process and molecular function were the ontologies .	The brain consists of billions of neurons that come in a range of shapes and sizes , with different types of neurons specialized to perform different tasks . Despite their diversity , all of these neurons originate from a single population known as neural stem cells . As the brain develops , each neural stem cell divides to produce two daughter cells: one remains a stem cell , which can then divide again , and the other becomes a neuron . A longstanding question in developmental biology is how a limited pool of neural stem cells can generate so many different types of neurons . The answer seems to lie in a process known as temporal identity , whereby neural stem cells of different ages give rise to different types of neurons . This requires neural stem cells to keep track of their own age , but it is still unclear how they can do so . Abdusselamoglu et al . have now uncovered part of the underlying mechanism behind temporal identity by studying fruit flies , an insect in which the early stages of brain development are similar to the ones in mammals . A method was developed to sort fly neural stem cells into groups based on their age . Comparing these groups revealed that a protein called Opa make neural stem cells switch from being 'young' to being 'middle-aged' . Another protein , Osa activates Opa , which in turn represses a protein called Dichaete . As Dichaete is mainly active in young neural stem cells , the actions of Osa and Opa push neural stem cells into middle age . Fruit flies are therefore a valuable system with which to study the mechanisms that regulate neural stem cell aging . Revealing how the brain generates different types of neurons could help us study the way these cells organize themselves into complex circuits . This knowledge could then be harnessed to understand how these processes go wrong and disrupt development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2019	The transcription factor odd-paired regulates temporal identity in transit-amplifying neural progenitors via an incoherent feed-forward loop
The microtubule cytoskeleton is critical for muscle cell differentiation and undergoes reorganisation into an array of paraxial microtubules , which serves as template for contractile sarcomere formation . In this study , we identify a previously uncharacterised isoform of microtubule-associated protein MAP4 , oMAP4 , as a microtubule organising factor that is crucial for myogenesis . We show that oMAP4 is expressed upon muscle cell differentiation and is the only MAP4 isoform essential for normal progression of the myogenic differentiation programme . Depletion of oMAP4 impairs cell elongation and cell–cell fusion . Most notably , oMAP4 is required for paraxial microtubule organisation in muscle cells and prevents dynein- and kinesin-driven microtubule–microtubule sliding . Purified oMAP4 aligns dynamic microtubules into antiparallel bundles that withstand motor forces in vitro . We propose a model in which the cooperation of dynein-mediated microtubule transport and oMAP4-mediated zippering of microtubules drives formation of a paraxial microtubule array that provides critical support for the polarisation and elongation of myotubes . Skeletal muscle fibre formation requires a coordinated programme of morphological and biochemical changes in the differentiating cells . Upon differentiation , mono-nucleated myoblasts withdraw from the cell cycle and fuse to form syncytial myotubes ( Wakelam , 1985 ) . The microtubule cytoskeleton is required for these processes ( Bischoff and Holtzer , 1968; Holtzer et al . , 1975; Toyama et al . , 1982 ) and undergoes reorganisation into an array of paraxial microtubules ( Warren , 1974 ) , which serves as template for contractile sarcomere formation ( Antin et al . , 1981; Pizon et al . , 2005 ) . During myogenesis , microtubules are rearranged from a dynamic radial array to a parallel array of stable posttranslationally modified microtubules within the elongating cell ( Warren , 1974; Tassin et al . , 1985; Saitoh et al . , 1988; Gundersen et al . , 1989 ) . Preventing detyrosination of tubulin or interfering with microtubule stabilisation by depletion of EB1 or MURF impairs expression of myogenic markers ( Spencer et al . , 2000; Chang et al . , 2002; Zhang et al . , 2009 ) , thus suggesting that signalling through modified microtubules might control myogenic differentiation . On the other hand , changes in the regulation of microtubule dynamics at the cell cortex that neither affect the content of tubulin modifications nor the expression of differentiation markers can result in cell polarisation and fusion defects as caused by the depletion of EB3 ( Straube and Merdes , 2007 ) . This suggests a dual function of microtubules during the early stages of muscle cell differentiation to control ( 1 ) morphological changes and ( 2 ) biochemical composition , before later serving as structural templates for myofibrillogenesis . A number of microtubule-associated proteins ( MAPs ) have been implicated in the organisation of microtubules into bundles . MAP2 and tau determine the spacing between microtubules in dendrites and axons , but do not control the directionality of microtubules within those bundles ( Chen et al . , 1992 ) . Proteins of the PRC1/MAP65/Ase1 family preferentially bundle microtubules in antiparallel orientation and are responsible for the stabilisation of the antiparallel microtubule overlaps in the spindle midzone during mitosis ( Loiodice et al . , 2005; Gaillard et al . , 2008; Subramanian et al . , 2010 ) . Microtubule–microtubule sliding by Eg5 , kinesin-1 , kinesin-14 , and dynein has been implicated in microtubule organisation and force generation ( Kapitein et al . , 2005; Fink and Steinberg , 2006; Straube et al . , 2006; Braun et al . , 2009; Fink et al . , 2009; Lu et al . , 2013; Tanenbaum et al . , 2013 ) , and we begin to understand how the interplay of motors and MAPs organises particular microtubule arrangements ( Janson et al . , 2007; Bieling et al . , 2010; Braun et al . , 2011 ) . Here , we show that the microtubule array in myoblasts is highly motile and becomes increasingly parallelised and immobilised as cells progress through the differentiation programme . We identify a previously uncharacterised differentially regulated isoform of microtubule-associated protein MAP4 , called oMAP4 , as a key organiser of microtubules in differentiating cells . Depletion of oMAP4 results in microtubule misalignment and increased microtubule motility in differentiating muscle cells . This results in defects in myogenic progression , cell polarisation , and cell–cell fusion . We further show that oMAP4 zippers preferentially antiparallel microtubules in vitro and restricts motor-driven microtubule sliding in differentiating muscle cells . Based on our own data , we propose a model whereby the cooperation of motor-driven microtubule–microtubule sliding and oMAP4-mediated zippering organises the microtubule cytoskeleton in differentiating muscle cells to support and govern cell polarisation and differentiation . To characterise microtubule organisation in differentiating muscle cells , we determined filament orientation , motility , and growth characteristics of microtubules at different stages during differentiation of C2C12 cells . As reported previously , we found that microtubule organisation changes from a radial , centrosome-dominated array in undifferentiated cells , to a paraxial array in myotubes ( Figure 1A ) ( Warren , 1974; Tassin et al . , 1985 ) . Microtubule growth directionality relative to the longitudinal axis of the cell was determined by tracking EB3-tdTomato-labelled microtubule ends . The asymmetry in the distribution of microtubule growth angles increases significantly during the first two days of differentiation ( Figure 1B–D , Figure 1—figure supplement 1 ) . This suggests that guided growth of microtubules ( probably along existing microtubules ) contributes to the progressively more ordered parallel microtubule array in differentiating cells . Furthermore , microtubules within the array are highly motile in undifferentiated myoblasts as visualised by the photoactivation of paGFP-Tubulin or conversion of mEos2-Tubulin ( Figure 1E , F , Videos 1 , 2 ) . In differentiating muscle cells , microtubules become very stable and static as seen by the low frequency and speed of microtubule-sliding movements and by diminished loss of microtubules from the photoactivated region due to depolymerisation ( Figure 1E–H , Figure 1—figure supplement 2 , Video 3 ) . The reduction in microtubule movements could be due to the differential regulation of motors that drive microtubule sliding . Conventional kinesin as well as dynein has been implicated in microtubule–microtubule sliding and microtubule movement along the cell cortex in other cell systems ( Rusan et al . , 2002; Fink and Steinberg , 2006; Straube et al . , 2006; Bicek et al . , 2009; Jolly et al . , 2010; Samora et al . , 2011; Lu et al . , 2013 ) . Both motors contribute to microtubule movements in myoblasts as the frequency of microtubule movements was reduced more than threefold upon depletion of either dynein heavy chain or Kif5b ( Figure 1E–H , Figure 1—figure supplement 3 , Videos 4 , 5 ) . However , Kif5b and dynein were continuously expressed during muscle differentiation ( Figure 1I ) , suggesting that other factors prevent motor-dependent microtubule sliding and promote parallel microtubule array formation during myogenesis . 10 . 7554/eLife . 05697 . 003Figure 1 . Microtubules are arranged in stable paraxial arrays during muscle cell differentiation . ( A ) Structured illumination microscopy of anti-tubulin-stained C2C12 cells pre/post induction of muscle differentiation as indicated . Microtubule filaments have been manually traced to highlight arrangement . Scale bar 10 μm . ( B ) Tracks of EB3-GFP in C2C12 cells at different stages of differentiation . Directionality is colour-coded ( green and red: ±45° to longitudinal cell axis , blue and yellow: perpendicular to cell axis ±45° ) . Cell outlines are indicated with dashed white line . Scale bars 20 μm . ( C ) Cumulative distribution of MT growth angles for example cells shown in ( B ) . Kuiper statistics ( K–S ) is calculated as a measure for microtubule alignment as the sum of the maximum deviations d1 and d2 from a random distribution . See Figure 1—figure supplement 1 for angular histograms . ( D ) Average MT asymmetry of differentiating myoblasts . Data show mean ± SEM for 4–9 cells with >5000 microtubule tracks per condition . ( E and F ) Motility of paGFP-Tubulin-labelled microtubule segments in myoblasts pre/post induction of differentiation and after depletion of dynein ( shDHC ) and kinesin-1 ( shKHC ) . Bar-shaped patterns were activated perpendicular to the microtubule orientation using mCherry-Tubulin as marker ( F ) . An individual bar-shaped activation pattern is shown in ( E ) for each condition . Arrows highlight microtubule-sliding events . Scale bars are 5 μm . See supplementary Videos 1–5 . ( G and H ) Frequency and velocity of microtubule sliding events observed following photoactivation of tubulin segments . Data show mean ± SEM , n = 17–51 activated patterns . Asterisks indicate significant difference from undifferentiated control cells ( p < 0 . 05 , p < 0 . 005 , *p < 0 . 0005 ) . ( I ) Immunoblotting of C2C12 cell extracts pre/post induction of differentiation for DHC and KHC . Tubulin serves as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00310 . 7554/eLife . 05697 . 004Figure 1—figure supplement 1 . Microtubule growth orientation . Angular distribution of microtubule growth data obtained from EB3-GFP tracks for the cells shown in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00410 . 7554/eLife . 05697 . 005Figure 1—figure supplement 2 . Dissipation of photoconverted microtubule labelling . Dissipation of photoconverted regions of mEos2-Tubulin in C2C12 cells before ( undiff ) and 48 hr post induction of differentiation . Cells were treated with Taxol to stop dissipation by depolymerisation , or azide to stop dissipation by motor-driven movement . Data have been corrected for bleaching by substraction of data from cells treated with both Taxol and azide . Data show mean ± SEM , n = 10–20 cells from two experiments . Half-life was determined from exponential decay curve fitted to the data . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00510 . 7554/eLife . 05697 . 006Figure 1—figure supplement 3 . Verification of dynein and kinesin depletion . Immunoblotting of C2C12 cell extracts from GFP-positive FACS-sorted cells treated with individual short hairpin RNAs ( shRNAs ) for 72 hr probed with antibodies as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00610 . 7554/eLife . 05697 . 007Video 1 . Photoconversion of mEos2-Tubulin in an undifferentiated C2C12 myoblast showing converted ( left panel and magenta in right panel ) and non-converted channels ( middle panel and green in right panel ) . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00710 . 7554/eLife . 05697 . 008Video 2 . Photoactivation of bar-shaped patterns of paGFP-Tubulin in an undifferentiated C2C12 myoblast . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00810 . 7554/eLife . 05697 . 009Video 3 . Photoactivation of bar-shaped patterns of paGFP-Tubulin in a 94-hr differentiated C2C12 myoblast . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 00910 . 7554/eLife . 05697 . 010Video 4 . Photoactivation of bar-shaped patterns of paGFP-Tubulin in an undifferentiated C2C12 myoblast treated with shRNA against dynein heavy chain . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01010 . 7554/eLife . 05697 . 011Video 5 . Photoactivation of bar-shaped patterns of paGFP-Tubulin in an undifferentiated C2C12 myoblast treated with shRNA against kinesin heavy chain ( Kif5b ) . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 011 Structural MAPs are candidates for a role in organising parallel microtubule arrays and regulating motor activity . MAP4 , the only non-neuronal member of the MAP2/Tau family , has been reported to stabilise and bundle microtubules ( Aizawa et al . , 1991; West et al . , 1991; Ookata et al . , 1995; Nguyen et al . , 1997 , 1998; Hasan et al . , 2006 ) . Our previous finding that MAP4 can limit force generation by dynein motors ( Samora et al . , 2011 ) suggested that MAP4 might prevent dynein-driven microtubule motility in muscle cells . Mouse skeletal muscle cells express tissue-specific isoforms of MAP4 ( Mangan and Olmsted , 1996 ) . To investigate the differential regulation of experimentally confirmed and predicted MAP4 transcripts ( Figure 2A ) , we performed RT-PCR analysis of total RNA isolated from differentiating C2C12 cells . We confirmed the continuous expression of the ubiquitous isoform uMAP4 and upregulation of muscle-specific mMAP4 24 hr post induction of differentiation ( Figure 2B ) . Interestingly , we found that a previously uncharacterised isoform , with a unique 48 kD projection domain , was also upregulated after 24 hr of differentiation . In addition to muscle , this isoform is also highly expressed in brain tissue ( Figure 2—figure supplement 1 ) . We refer to this isoform as oMAP4 . Further analysis revealed that uMAP4 , mMAP4 , and oMAP4 transcripts were each expressed as variants with three , four , or five tau-like microtubule binding repeats due to alternative splicing of exons 14 and 15 ( Figure 2—figure supplement 2 ) . To confirm that these transcripts encode proteins , we generated specific antibodies for mMAP4 and oMAP4 and probed whole cell lysates of C2C12 cells on Western blots . Both mMAP4 and oMAP4 protein levels increase between 24 and 48 hr after differentiation ( Figure 2C , D ) . As expected , GFP-fusion proteins of uMAP4 , mMAP4 , and oMAP4 decorated microtubules along their length in C2C12 cells ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 05697 . 012Figure 2 . oMAP4 is required for myoblast elongation and fusion . ( A ) Domain organization of MAP4 isoforms . Green and red boxes represent isoform-specific regions in the projection domains of mMAP4 and oMAP4 , respectively . Note that all three isoforms are expressed with 3 , 4 , or 5 tau-like MT binding repeats . ( B ) RT-PCR-based MAP4 expression analysis of RNA samples isolated from 0–60 hr differentiated C2C12 cells . Glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) was used as loading control . ( C and D ) Immunoblotting of extracts from 0 to 96 hr differentiated C2C12 cells using antibodies against the N-terminus of oMAP4 ( C ) or the muscle-specific insertion of mMAP4 ( D ) . Tubulin was used as a loading control . ( E ) Examples of shRNA-treated cells co-expressing GFP-Tubulin 48 hr after induction of differentiation . Scale bar 50 μm . ( F , H ) Distribution of cell lengths after 48 hr differentiation following treatment with indicated shRNAs . 1000–1900 cells were measured from three independent experiments for each condition . Small triangles represent mean values . ( G , I ) Myoblast fusion analyses of C2C12 cells treated with indicated shRNAs after 56 hr differentiation . 1000–1200 cells were analysed for each condition and only cells with three or more nuclei were scored as fused . Data are collected from three experiments for each condition and are presented as mean ± S . E . M . Asterisks indicate significant difference from shControl treatment or between pairs of samples as indicated ( p < 0 . 05 , p < 0 . 005 , *p < 0 . 0005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01210 . 7554/eLife . 05697 . 013Figure 2—figure supplement 1 . Expression of major MAP4 isoforms . Relative transcript abundances of MAP4 isoforms during myoblast differentiation as determined by analysis of RNA sequencing data from undifferentiated and 60 hr differentiated C2C12 cells ( Trapnell et al . , 2010 ) . Comparative expression levels of MAP4 isoforms in several mouse tissues obtained by analysis of Affymetrix exon array data ( Pohl et al . , 2009 ) . Data are shown as mean ± SD from three experiments for each tissue . See previously generated data sets 1–2 in supplement . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01310 . 7554/eLife . 05697 . 014Figure 2—figure supplement 2 . C2C12 cells express three major MAP4 isoforms with variable numbers of MT binding repeats . RT-PCR from total RNA isolated from C2C12 cells before ( 0 hr ) and 48 hr post induction of differentiation with upstream primers specific for projection domains of uMAP4 , mMAP4 , and oMAP4 and downstream primer at 3′ end of the MAP4 coding sequence to determine number of microtubule binding repeats . Digestion of PCR from 48 hr differentiated cells with exon-specific restriction nucleases to determine spliced exons that give rise to variants with different numbers of microtubule binding repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01410 . 7554/eLife . 05697 . 015Figure 2—figure supplement 3 . Microtubular localisation of eGFP-tagged MAP4 isoforms in C2C12 cells . Tubulin is used as a marker for microtubules . Scale bars are 20 μm , insets 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01510 . 7554/eLife . 05697 . 016Figure 2—figure supplement 4 . Verification of mMAP4 depletion by immunofluorescence . Protein depletion was quantified by counting cells that expressed mMAP4 in GFP-Tubulin and shRNA co-expressing cells . Bars are 20 μm . 40–50 cells each were analysed for shControl and sh-mMAP4 transfections . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01610 . 7554/eLife . 05697 . 017Figure 2—figure supplement 5 . Verification of MAP4 depletion by immunoblotting . Immunoblotting of C2C12 cell extracts treated with individual shRNAs as indicated for 70–74 hr . Tubulin serves as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01710 . 7554/eLife . 05697 . 018Figure 2—figure supplement 6 . Verification of oMAP4 depletion and RNAi-protected rescue construct by immunoblotting . Immunoblot of whole cell extracts of HeLa cells co-transfected with shRNAs and control or rescue plasmids . Tubulin serves as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 018 To investigate their involvement in microtubule organisation in muscle cells , we depleted each MAP4 isoform using a vector-based RNA interference approach . GFP-Tubulin was co-expressed with the short hairpin RNAs ( shRNAs ) to serve as a marker to detect successful transfection . Efficient depletion was confirmed by Western blotting of FACS-sorted GFP-positive cells and by immunofluorescence ( Figure 2—figure supplements 4 , 5 ) . Muscle differentiation phenotypes were assessed as ( 1 ) the ability of single nucleated cells to elongate and ( 2 ) the efficiency of cells to fuse and form syncytia containing three or more nuclei ( Straube and Merdes , 2007 ) . 48 hr after differentiation , the average length of myoblasts treated with control shRNA was 107 ± 40 μm . Depletion of uMAP4 did not significantly affect this ( 105 ± 36 μm , p = 0 . 4; Figure 2E , F ) . Depletion of mMAP4 allowed myoblasts to elongate to an average length of 143 ± 58 μm , which is significantly longer than control cells . Myoblast fusion efficiency was not significantly affected following the depletion of uMAP4 or mMAP4 ( Figure 2G ) . In contrast , the depletion of oMAP4 caused severe defects in myoblast elongation and fusion . oMAP4-depleted cells had an average length of 82 ± 39 μm , which is significantly shorter than control myoblasts ( p << 0 . 001 , Figure 2E , F ) . Furthermore , cell–cell fusion efficiency was reduced more than fourfold following depletion of oMAP4 ( Figure 2G ) . Importantly , co-transfection of an RNAi-resistant version of oMAP4 ( FLAG-oMAP4RIP ) rescued the myoblast elongation and fusion defects caused by oMAP4 shRNA ( Figure 2H , I , Figure 2—figure supplement 6 ) . The overexpression of uMAP4 was not able to rescue oMAP4-depletion phenotypes ( Figure 2I ) . Our data collectively demonstrate that oMAP4 is specifically required for morphogenesis and cell–cell fusion during muscle cell differentiation . To understand which processes in the myogenic programme depend on oMAP4 , we analysed the timing of myogenic events . We find a severe delay and reduction in the expression of embryonic myosin , a structural protein required for sarcomere formation , in oMAP4-depleted cells ( Figure 3A–C ) . Likewise , the relocation of centrosomal proteins to the nuclear surface of myoblasts , a characteristic event during myogenesis ( Tassin et al . , 1985; Srsen et al . , 2009 ) , is severely delayed in oMAP4-depleted cells ( Figure 3D , E ) . These results confirm that progression through the myogenic differentiation programme depends on oMAP4 . Previous work has shown that interference with microtubule stability and the acquisition of posttranslational modifications of tubulin result in similar myogenesis defects ( Spencer et al . , 2000; Chang et al . , 2002; Zhang et al . , 2009 ) . Depletion of oMAP4 did not result in a reduction of tubulin acetylation ( Figure 3F ) , perhaps because other microtubule stabilisers were still present . Nevertheless , the microtubule cytoskeleton appeared disorganised in oMAP4-depleted cells ( Figure 3A , B ) . To quantitatively assess microtubule network organisation and distinguish this from effects due to different morphology and differentiation status , we selected mono-nucleated elongated myoblasts treated with control and oMAP4 shRNA and manually traced GFP-Tubulin-labelled microtubules in equivalent sections of these cells ( Figure 3G , Videos 6 , 7 ) . We found that the microtubule network was largely parallel in control cells with ∼80% of microtubules oriented ±15° of the longitudinal cell axis . This was reduced to ∼50% in oMAP4-depleted cells ( Figure 3H ) . Microtubule orientation was fully rescued by expression of an RNAi-resistant oMAP4 construct ( Figure 3—figure supplement 1 ) . Depletion of mMAP4 increased paraxial microtubule alignment and uMAP4 depletion slightly reduced it ( Figure 3—figure supplement 1 ) . We confirmed the microtubule alignment defect caused by oMAP4 depletion results by tracking growing microtubule ends labelled with EB3-tdTomato ( Figure 3I , Videos 8 , 9 ) . While microtubule growth speed and duration were only slightly affected , the orientation of microtubule growth deviated more from the longitudinal cell axis when oMAP4 was depleted compared to control cells ( Figure 3J–L , Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 05697 . 019Figure 3 . oMAP4 is required for the parallel arrangement of microtubules in differentiating muscle cells . ( A and B ) C2C12 myoblasts 48 hr after induction of differentiation treated with shRNA as indicated and stained for Myogenin ( a marker for differentiating myoblasts , yellow ) , PCM-1 ( red ) and DAPI ( blue ) . GFP-Tubulin ( green ) indicates successful transfection with shRNA . Insets show higher magnification of microtubule arrangement . Scale bars 25 μm , 5 μm in insets . ( C ) Timecourse of expression of embryonic myosin , a marker of myogenic differentiation . ( D ) Timecourse of relocalisation of PCM-1 from a focus around the centrosome to the surface of the nucleus . Cells with a complete nuclear ring were scored as positive . Data in C , D show mean ± SD , n = 30–50 cells from 2 experiments . ( E ) RNAi rescue experiment of delayed PCM-1 relocalisation phenotype . Data show mean ± SEM of 3 experiments with 50–60 cells each . ( F ) Timecourse of accumulation of cells with acetylated tubulin during muscle cell differentiation . Data in show mean ± SD , n = 30–50 cells from 2 experiments . ( G ) Manual tracing of microtubule filaments ( yellow ) relative to the longitudinal cell axis ( red ) in elongated mono-nucleated cells selected after 48 hr differentiation and shRNA treatment as indicated . Scale bars 10 μm . See supplementary Videos 6 , 7 . ( H ) Microtubule directionally from data as in g shown as cummulative frequency distribution . Data show mean ± SD , n = 3 experiments , 1522–1958 MTs . ( I ) Automatic tracking of EB3-tdTomato comets in elongated mono-nucleated cells selected after 48 hr differentiation and shRNA treatment as indicated . Direct lines from start to end of track are shown and colour-coded for direction relative to longitudinal axis of cell as indicated in legend . Scale bars 10 μm . See supplementary Videos 8 , 9 . ( J ) Distribution of microtubule growth angles obtained from data as in H . Data show mean ± SD , n = 3 experiments , 6251–6382 tracks . ( K and L ) Microtubule growth speed and duration was determined from EB tracks as in I . Pooled data shown as statistical box plots with percentiles as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 01910 . 7554/eLife . 05697 . 020Figure 3—figure supplement 1 . Microtubule orientation in depleted cells . Manual segmentation of microtubules ( white/black ) , their angular distribution relative to main cell axis ( red ) , and cumulative frequency of microtubule orientation following depletion of individual MAP4 isoforms and in oMAP4 RNAi rescue experiment . n ≥ 10 cells . Kuiper statistic relative to random distribution is calculated as in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02010 . 7554/eLife . 05697 . 021Figure 3—figure supplement 2 . Microtubule growth orientation in depleted cells . Angular distribution of microtubule growth data from Figure 3J . To aid visualisation of data tails , bin size increases progressively between 0 and ±60° and is then constantly 30° . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02110 . 7554/eLife . 05697 . 022Video 6 . Microtubule orientation in a 48 hr differentiated C2C12 myoblast treated with shControl co-expressing GFP-Tubulin . Manual tracing of microtubule cytoskeleton ( yellow lines ) and main cell axis ( red line ) shown as used for analysis . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02210 . 7554/eLife . 05697 . 023Video 7 . Microtubule orientation in a 48-hr differentiated C2C12 myoblast treated with sh-oMAP4 co-expressing GFP-Tubulin . Manual tracing of microtubule cytoskeleton ( yellow lines ) and main cell axis ( red line ) shown as used for analysis . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02310 . 7554/eLife . 05697 . 024Video 8 . Growing microtubules in a 48-hr differentiated C2C12 myoblast treated with sh-Control co-expressing EB3-tdTomato . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02410 . 7554/eLife . 05697 . 025Video 9 . Growing microtubules in a 48-hr differentiated C2C12 myoblast treated with sh-oMAP4 co-expressing EB3-tdTomato . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 025 We hypothesised that a failure in the guidance of microtubule assembly along existing filaments could be the underlying cause of the disorganised microtubule network . Alternatively , microtubule sliding and looping as observed in undifferentiated cells could disorganise the microtubule network in oMAP4-depleted cells . To distinguish between these possibilities , we analysed microtubule motility in differentiated , shRNA-treated cells , again selecting elongated , mono-nucleated cells to exclude effects due to morphology alone . As described above , microtubule motility was strongly suppressed in differentiating muscle cells ( Figures 1E–H , 4a ) . Depletion of oMAP4 resulted in a more than fourfold increase in the frequency of microtubule sliding events , and these were significantly faster than those in control cells ( Figure 4A–D ) . In addition , the rare sliding events in differentiated control cells were usually restricted to movements parallel to the long axis of the cell , while microtubules were observed to loop in oMAP4-depleted cells , similarly to undifferentiated cells ( Figure 4E , Videos 1–3 , 10 , 11 ) . These observations suggest that oMAP4 acts to maintain the parallel microtubule network by restricting microtubule movement , especially those that are off-axis . 10 . 7554/eLife . 05697 . 026Figure 4 . oMAP4 prevents microtubule sliding in cells . ( A ) Microtubule motility ( arrows ) observed after photoconversion of mEOS2-Tubulin in 48 hr differentiated cells treated with shRNA as indicated . Scale bars 5 μm . See supplementary Videos 10 , 11 . ( B ) Frequency of microtubule motility events relative to average sliding velocity in 48-hr differentiated cells treated with shRNA as indicated . Data pooled from three experiments , n = 15–23 cells . ( C and D ) Frequency of microtubule motility events: all events are shown in C , while only fast sliding events are shown in D . The latter were defined as movement faster than 700 nm/s . Data show mean ± SD , n = 3 experiments , 15–23 cells . ( E ) Directionality of microtubule motility events . Paraxial movement is defined as occurring parallel to the long cell axis with a deviation less than 45° . Looping microtubules changed direction by more than 90° during the movement event . Data show mean ± SD , n = 3 experiments , 15–23 cells . ( F and G ) Expression of GFP-oMAP4 ( upper row ) or GFP-uMAP4 ( lower row ) in undifferentiated C2C12 cells stained with tubulin antibodies . The number of strong microtubule bundles ( arrows ) of at least 2-μm length and threefold intensity of an individual microtubule was scored . Data in g show mean ± SEM , n = 13–23 cells . oMAP4 causes statistically significant change ( p = 0 . 02 ) . Scale bar 20 μm , 2 μm in insets . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02610 . 7554/eLife . 05697 . 027Figure 4—figure supplement 1 . Microtubule orientation in depleted cells . Examples of microtubule traces ( yellow ) in dynein depleted and dynein + oMAP4-depleted 48 hr differentiated elongated myoblasts , see Figure 3G for examples for control and oMAP4 depletion . Angular distribution and cumulative frequency of microtubule orientation relative to the main cell axis ( red ) is shown for data pooled from three experiments . Kuiper statistic relative to random distribution is calculated as in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02710 . 7554/eLife . 05697 . 028Video 10 . Photoconversion of bar-shaped patterns of mEOS2-Tubulin in a 48-hr differentiated C2C12 myoblast treated with shControl . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 02810 . 7554/eLife . 05697 . 029Video 11 . Photoconversion of bar-shaped patterns of mEOS2-Tubulin in a 48-hr differentiated C2C12 myoblast treated with sh-oMAP4 . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 029 As we demonstrated in a previous study that uMAP4 limits force generation by dynein to prevent excessive movement of astral microtubules along the cell cortex ( Samora et al . , 2011 ) , we wondered whether oMAP4 prevents microtubule sliding by inhibiting dynein function directly or whether oMAP4 crosslinks microtubules and acts as a brake to all motor-driven sliding . In order to distinguish between these possibilities , we co-depleted dynein from oMAP4-depleted cells . If oMAP4 would act exclusively through dynein , we posited this should rescue the depletion phenotypes . Depletion of dynein heavy chain alone did not significantly affect the already low-microtubule sliding frequencies in differentiated cells , although the velocity of microtubule sliding was reduced ( Figure 4B–D ) . Cells co-depleted for dynein and oMAP4 still showed a high frequency of microtubule-sliding events ( Figure 4B , C ) , although fast sliding events and looping were reduced ( Figure 4D , E ) . Consequently , dynein co-depletion slightly alleviated the microtubule disorganisation caused by oMAP4 depletion , but did not fully rescue it ( Figure 4—figure supplement 1 ) . These results suggest that oMAP4 prevents microtubule sliding by crosslinking microtubules rather than by inhibiting dynein force generation per se . Indeed , expression of GFP-oMAP4 in undifferentiated myoblasts resulted in a significant increase in strong microtubule bundles , while expression of GFP-uMAP4 had no effect ( Figure 4F , G ) . To directly test the hypothesis that oMAP4 is a crosslinker , we recombinantly expressed and purified oMAP4 and GFP-oMAP4 from Escherichia coli ( Figure 5A ) . Using in vitro microtubule co-sedimentation assays , we confirmed microtubule-binding activity of the purified proteins ( Figure 5B ) . When Taxol- or GMP-CPP-stabilised microtubules were incubated with 60-nM oMAP4 , we frequently observed microtubule bundles and structures with crossovers ( Figure 5C–F ) . This confirmed that oMAP4 has microtubule cross-linking activity . We next asked whether oMAP4 has the ability to organise dynamic microtubules into antiparallel bundles in vitro . To do this , we used total internal reflection ( TIRF ) microscopy to visualise microtubules assembled from biotinylated microtubule seeds immobilised on streptavidin-coated coverslips . In control chambers , microtubules continued growing without changing direction when they encountered other microtubules and microtubules only overlapped when they happened to grow in the same direction ( Figure 6A , B , Video 12 ) . The addition of GFP-oMAP4 promoted zippering of those growing microtubules that encountered each other at shallow angles ( Figure 6A–C; Video 13 ) . To assess whether oMAP4 was specific for the orientation of the microtubules , we determined microtubule polarity based on the growth characteristics of the microtubule ends observed in the video ( Figure 6C ) and determined the rate of microtubule zippering relative to the incident angle of the two microtubules . No microtubule-zippering events were observed at angles between 25° and 150° , suggesting that oMAP4 can only generate forces to bend microtubules by up to 30° . Furthermore , oMAP4 showed a strong preference for zippering antiparallel-oriented microtubules ( Figure 6B , C ) . 10 . 7554/eLife . 05697 . 030Figure 5 . oMAP4 bundles microtubules in vitro . ( A ) SDS-PAGE analysis of oMAP4 protein purification . N-terminally 6xHis tagged full-length oMAP4 was purified by Ni2+-NTA affinity chromatography followed by ion exchange chromatography . ( B ) Microtubule co-sedimentation analysis of oMAP4 protein . Purified protein from ( A ) was incubated with Taxol-stabilised microtubules prior to centrifugation . Separate centrifugations of microtubules alone and oMAP4 protein alone were used as controls . Pellet and supernatant fractions were analysed by SDS-PAGE and Coomassie staining . ( C and D ) Analysis of microtubule bundling by oMAP4 in vitro . Taxol- or GMP-CPP-stabilised microtubules were incubated with buffer or 60 nM purified oMAP4 protein before spreading on glass for imaging . Insets show various microtubule crosslinking and bundling events . Scale bars 20 μm , 5 μm in insets . ( E and F ) Comparison of bundling efficiency of oMAP4 with negative control ( EB1 ) and positive control ( PRC1 ) . Representative image and data quantified from >300 microtubules are shown . Scale bar 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 03010 . 7554/eLife . 05697 . 031Figure 6 . oMAP4 zippers dynamic microtubules with a bias for antiparallel arrangements . ( A ) Dynamic Rhodamine-labelled microtubules ( greyscale ) assembled from immobilised Hilyte640-labelled seeds ( red ) in vitro are zippered in the presence of oMAP4 . Arrows highlight bundled microtubules . Scale bar 10 μm . ( B ) Histogram showing proportion of microtubule encounters leading to MT zippering relative to microtubule encounter angles . Antiparallel encounters are observed at angles above 90° . Note no zippering occurs between 25 and 150° . Data for GFP-oMAP4 , EB1-GFP and PRC1 , each at 80 nM concentration are shown for comparison . n ≥ 400 microtubule encounters for control and oMAP4 , 136 encounters for PRC1 and 184 encounters for EB1 . ( C ) Example of zippering events ( arrows ) in the presence of 80 nM oMAP4-GFP . Microtubule polarity is indicated with ( + ) at the dynamic plus end . See supplementary Videos 12 , 13 . Scale bar 10 μm . ( D and E ) Immunoblot of oMAP4 in fractions from glycerol gradients to determine its sedimentation coefficient with volume from top of gradient indicated ( D ) . Calibration curve using standard proteins with known sedimentation coefficients and linear regression curve ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 03110 . 7554/eLife . 05697 . 032Figure 6—figure supplement 1 . Microtubule length distribution in zippering experiments . For microtubule encounters between 10 and 30° incident angle in ( near ) parallel or ( near ) antiparallel orientation , the length of each microtubule was measured to the attached seed or the closest crossover point . None of the distributions are significantly different . n = 22–54 encounters . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 03210 . 7554/eLife . 05697 . 033Video 12 . TIRF-based assay showing dynamic Rhodamine-labelled microtubules assembled from immobilised seeds . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 03310 . 7554/eLife . 05697 . 034Video 13 . TIRF-based assay showing dynamic microtubules in the presence of 80 nM GFP-oMAP4 . Note that antiparallel microtubule encounters result in zippering into an antiparallel bundle in most cases . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 034 Another known antiparallel microtubule-bundling protein , PRC1 is a dimer that accumulates specifically in antiparallel-microtubule overlaps in the spindle midzone ( Subramanian et al . , 2010 ) . We observed that PRC1 was more potent to bundle-stabilised microtubules free in solution than oMAP4 ( Figure 5E , F ) . However , PRC1 was not able to zipper microtubules in our assays using dynamic microtubules growing from immobilised seeds ( Figure 6B ) . This is in agreement with the literature that described PRC1 to specifically bind to antiparallel-microtubule overlaps that form when microtubules ‘occasionally encountered each other in a plus end-to-plus end configuration’ ( Bieling et al . , 2010 ) rather than PRC1 itself causing the formation of the overlaps . We do not observe a substantial enrichment of GFP-oMAP4 in antiparallel or parallel overlaps ( Figure 6C ) . To determine whether the bias in zippering towards antiparallel-oriented microtubules could be due to differences in the length of microtubules that encountered each other , we measured the length dependence of MAP4-mediated zippering . We found similar distributions of microtubule lengths for encounters that resulted in zippering and those encounters that occurred at similarly shallow-incipient angles but not led to zippering ( p = 0 . 22 ) . Likewise , there was no difference between parallel and antiparallel encounters that were zippered ( p = 0 . 59 ) and those that did not result in zippering ( p = 0 . 88 ) ( Figure 6—figure supplement 1 ) . Therefore , our results demonstrate that oMAP4 is a microtubule-organising factor , which can arrange microtubules into antiparallel and with lesser efficiency parallel bundles . This function is consistent with the depletion phenotypes observed in differentiating muscle cells , where oMAP4 helps to arrange paraxial microtubules and thereby supports cell differentiation . PRC-1 is dimer that forms highly ordered crosslinks that do not substantially limit microtubule–microtubule sliding at moderate concentrations ( Subramanian et al . , 2010 ) . As we proposed that oMAP4 contributes to microtubule alignment by preventing microtubule sliding in differentiating muscle cells , we next asked how oMAP4-crosslinks microtubules and whether these can withstand motor forces . We considered whether the unique projection domain in oMAP4 confers microtubule cross-linking activity by dimerisation . Proteins of the MAP2/tau family are highly elongated , structurally disordered monomers ( Hernandez et al . , 1986; Devred et al . , 2004 ) . Using the GOR secondary structure prediction method , we find that oMAP4 , uMAP4 , and tau share a similar structure of over 60% random coil , about 20% helical , and 13% extended strand ( Garnier et al . , 1996 ) . Furthermore , no significant coiled-coil domain was predicted . To confirm this , we performed sedimentation analysis of bacterially expressed oMAP4 and GFP-oMAP4 in comparison with a number of standard proteins ( Figure 6D , E ) . We obtained sedimentation constants of 3 . 6 ± 0 . 25 S and 4 . 3 ± 0 . 21 S , respectively . Given the molecular weight of the monomers being 99 kD and 131 kD , respectively , we can calculate the frictional ratio ƒ/ƒmin = 2 . 1 , suggesting that oMAP4 is a highly elongated monomer with a shape comparable to tau ( ƒ/ƒmin = 1 . 8 , [Devred et al . , 2004] ) . We , therefore , predict that oMAP4 most likely bundles microtubules using a second microtubule-binding region in its projection domain . Single kinesin-1 and dynein molecules can generate forces of up to 7 pN ( Nishiyama et al . , 2002; Toba et al . , 2006 ) . To determine whether oMAP4 bundling can indeed withstand the forces exerted by several motors tugging the microtubules apart , we performed gliding assays with kinesin-1 . The presence of 80 nM oMAP4 did only slightly affect the velocity of kinesin-mediated movements of single microtubules ( Figure 7A , B ) . However , antiparallel-microtubule bundles formed in solution and landing on the kinesin surface were often static or their movement was slow and non-persistent ( Figure 7C–E , Video 14 ) . Given their predominantly antiparallel arrangement , forces on the microtubules within a bundle would cancel each other out as long as the linkage between the microtubules is maintained . Occasionally , bundles were driven apart . This usually occurred in bundles where extensive lateral forces were generated on one microtubule that was significantly longer than other microtubules in the bundle ( Figure 7C ) . The antiparallel nature of the oMAP4-generated bundles was confirmed in these cases from the opposing direction of their movement after separation ( Figure 7C , magenta arrows ) . Most importantly , we observed that more than 75% of microtubule bundles withstand motor forces for the entire duration of our 7 . 5 min videos ( Figure 7C , E , F , yellow arrows ) , suggesting that oMAP4-mediated cross-linking is indeed able to prevent motor-driven microtubule sliding as we hypothesised . 10 . 7554/eLife . 05697 . 035Figure 7 . oMAP4 bundles can withstand motor forces . ( A ) Microtubule gliding assay on a Drosophila kinesin-1-coated surface . Time colour-coded projections over 30 s are shown for buffer control and 80 nM oMAP4 . ( B ) Instantaneous speeds of microtubule motility were determined from tracks of 50 microtubules and shown as histograms . The presence of 80 nM oMAP4 reduces velocity of single microtubules by about 5% . ( C ) Microtubule bundles formed in solution by oMAP4 and landing on the kinesin-coated surface tend not to move ( yellow filled arrows ) , while single microtubules do ( green arrows ) . In about 25% of cases , bundles are driven apart by motor forces ( magenta filled arrows ) with microtubules gliding apart at normal speed once separated ( magenta arrows ) . See supplementary Video 14 . ( D ) Time colour-coded projection of gliding assay showing a microtubule bundle ( arrow , appears white due to averaging of all time points ) and individual microtubules over 447 s in the presence of 80 nM oMAP4 . ( E ) Instantaneous speeds of microtubule motility were determined from tracks of 20 single and bundled microtubules and shown as histograms . ( F ) Time between start of observation and either end of observation or time when bundle was separated by motor forces . Bundles that moved out of the field or dissociated as intact bundle were not scored . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 03510 . 7554/eLife . 05697 . 036Video 14 . Microtubule gliding assay on a kinesin-1-coated surface in the presence of 80 nM GFP-oMAP4 . Note that single microtubules move persistently , while bundles don't until they are driven apart . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 036 Why is a highly organised microtubule cytoskeleton required to undergo muscle differentiation ? Microtubules are the stiffest of the cytoskeletal polymers that can bear high-compressive loads , especially when reinforced laterally ( Brangwynne et al . , 2006 ) . Our data support a model in which microtubules fulfil a structural role during the elongation of muscle cells . If so , one would expect that the degree of alignment of microtubules will correlate with a cell's ability to elongate . We tested this by plotting the Kuiper statistic as a measure of microtubule orderliness against the mean cell length of 48 hr differentiated cells for different RNAi treatments used in this study and found indeed a linear correlation ( Figure 8A ) . Likewise , the depletion of mMAP4 , which increases the length of differentiating myoblasts also increases the paraxial microtubule alignment beyond that of control cells ( Figure 3—figure supplement 1 ) . These data support the idea that MT orientation is strongly linked to the morphological changes required for muscle differentiation . As observed for oMAP4 , our previously published data on EB3 depletion ( Straube and Merdes , 2007 ) and a number of unpublished observations , myoblasts that fail to elongate are also impaired in cell–cell fusion . While we don't yet understand the relationship between cell elongation and fusion , a mechanism to prevent the fusion of myoblasts that did not complete the previous step of differentiation , makes intuitively sense . oMAP4-depleted cells show in addition to impaired morphological changes , also a delay in the expression of myogenic markers , such as myogenin and embryonic myosin , further suggesting that a signalling step in the differentiation programme has not been completed . Microtubules have been implicated in a signalling role during myogenesis based on the observation that altering the level of posttranslational tubulin modifications either through chemical inhibition or depletion of microtubule-stabilising MAPs leads to similar delays in the expression of differentiation markers and impaired formation of myotubes ( Spencer et al . , 2000; Chang et al . , 2002; Zhang et al . , 2009 ) . The depletion of oMAP4 does not negatively affect tubulin acetylation ( Figure 3F ) but might impact other aspects of this putative microtubule-dependent signalling event . It will be a future challenge to elucidate the pathway that couples microtubule organisation and chemical modification to the timing of the myogenic protein expression programme . 10 . 7554/eLife . 05697 . 037Figure 8 . oMAP4 and dynein co-operate in the organisation of the paraxial microtubule network in differentiating muscle cells . ( A ) Kuiper statistics of traced microtubule filaments as in Figures 3G , 4 , Figure 4—figure supplement 1 as a measure for microtubule orderliness is plotted against the mean cell length of 48 hr differentiated C2C12 myoblasts treated with shRNAs as indicated . A linear fit to the data is shown . ( B ) We propose that the correlation between the precision of paraxial alignment and cell elongation suggests that ordered microtubule arrays confer higher mechanical stability to counteract contractile forces . ( C ) Model of cooperation of oMAP4-mediated zippering with motor-driven microtubule sliding/transport in the formation of a highly ordered paraxial microtubule network . Microtubules are indicated in green , oMAP4 in magenta , dynein in blue , and kinesin in purple . DOI: http://dx . doi . org/10 . 7554/eLife . 05697 . 037 We show that oMAP4 and PRC1 are both microtubule-bundling proteins , but with very different properties . While PRC1 is an antiparallel dimer that forms ordered crosslinks of a defined distance and specifically enriches in antiparallel-microtubule overlaps ( Bieling et al . , 2010; Subramanian et al . , 2010 ) , oMAP4 has little preference for binding to bundles , but instead is able to zipper microtubules . Zippering of microtubules growing from surface-attached seeds requires bending of microtubules . We consider that the lateral forces required to bring two microtubules close together against the rigidity of the microtubule have to be borne by a single crosslinking molecule . Thus , our data suggest that single molecules of oMAP4 can resist higher loads than PRC1 crosslinks . When we apply longitudinal forces as experienced by pre-formed bundles in kinesin-gliding assays , the shear forces are probably shared between several crosslinking molecules , thus enabling oMAP4 to withstand counteracting motor forces on the bundled microtubules . Bundles that contain one very long microtubule that is driven laterally on the kinesin surface , eventually splay apart as lateral forces cannot be shared between the crosslinkers . In contrast , PRC1 cannot resist motor forces and at moderate levels only slightly slows motor-driven sliding of microtubules ( Bieling et al . , 2010; Subramanian et al . , 2010 ) . Thus , oMAP4 has the required properties to fulfil the role of a microtubule organiser , justifying its name as organising MAP4 , although the structural basis for oMAP4's antiparallel zippering remains to be elucidated . The motors dynein and kinesin move microtubules and cause apparent disorder that is limited by oMAP4 crosslinking . However , both motors have also been reported to bundle and organise microtubules . Indeed , dynein depletion alone results in reduced microtubule alignment ( Figure 4—figure supplement 1 ) . This is consistent with the finding that dynein controls muscle length in Drosophila ( Folker et al . , 2012 ) and earlier reports of dynein involvement in the self-organisation of microtubule networks and its ability to crosslink and slide antiparallel microtubules as well as transporting microtubules along the cell cortex ( Heald et al . , 1996; Adames and Cooper , 2000; Merdes et al . , 2000; Fink and Steinberg , 2006; Samora et al . , 2011; Tanenbaum et al . , 2013 ) . As oMAP4 is only able to efficiently zipper microtubules at incident angles of less than 30° if antiparallel and less than 10° if parallel , we propose that dynein-mediated looping and buckling of microtubules ( Figure 4E; Fink and Steinberg , 2006; Tanenbaum et al . , 2013 ) brings microtubules into a favourable position for oMAP4-mediated zippering . As oMAP4-mediated bundling resists motor-driven sliding , dynein can only move those microtubules that are not yet aligned to the paraxial network . Thus , dynein and oMAP4 are likely to cooperate in the formation of the highly ordered microtubule arrangement in differentiating muscle cells ( Figure 8C ) . In the absence of oMAP4 , excessive motor-driven microtubule motility disorganises microtubules . In the absence of dynein , oMAP4 might stabilise high-angle microtubule crossovers , but will not be able to align them into the network . If the activity of both oMAP4 and dynein is reduced , oMAP4 zippering is missing and kinesin-mediated sliding and bundling ( Straube et al . , 2006; Jolly et al . , 2010 ) prevails ( Figure 8B ) . In agreement with this model , some disorganisation of microtubules has been observed in kinesin-1-depleted myotubes ( Wilson and Holzbaur , 2012 ) . Microtubule–microtubule sliding has recently been implicated in driving neurite outgrowth ( Lu et al . , 2013 ) , and paraxial microtubule arrangements have been shown to drive dorsal closure during embryonic development ( Jankovics and Brunner , 2006 ) . Thus , the mechanisms we reveal here for motor and MAP cooperation in the formation of paraxial microtubule networks are likely to be of importance beyond muscle cells . Indeed , oMAP4 is highly expressed in brain ( Figure 2—figure supplement 1 ) , suggesting that it might be required to support the paraxial microtubule arrays in dendrites and the axon . While the zippering model described above explains how a paraxial array is maintained in the presence of microtubule turnover , it does not explain how the symmetry of the radial microtubule cytoskeleton in the myoblast is broken in the first place . We think that a bipolar elongating myoblast can be compared to a migrating cell with two fronts . Symmetry breaking in cell migration occurs through protrusion mediated by the actin cytoskeleton . Microtubules support and stabilise cell protrusions , and microtubule plus ends are selectively stabilised at the leading edge ( Waterman-Storer et al . , 1999; Kaverina and Straube , 2011 ) . This establishes and reinforces the new polarity axis . Similarly , in differentiating myoblasts dynamic capture of microtubule plus ends occurs at the cell tips ( Straube and Merdes , 2007 ) . This leads to the selective stabilisation of paraxial microtubules to break the symmetry in the microtubule network . The zippering mechanism described here would promote shorter microtubules to align to longer and already bundled microtubules and thus amplify the directional bias and increase stability and orderliness in the paraxial array . It is likely that the guidance of microtubule growth through the cooperation of plus end tracking proteins and motors also contributes to maintenance of the ordered array ( Mattie et al . , 2010 ) . It will be interesting to dissect the contributions of these different pathways to microtubule network reorganisation , not only at the onset of myogenesis , but also during later stages when the parallel network is replaced by the grid-like lattice found in adult muscle ( Oddoux et al . , 2013 ) . RNA was extracted from 0 to 60 hr differentiated C2C12 cells using Trizol reagent ( Invitrogen , Life Technologies , UK ) , and random-primed cDNA was synthesized using RevertAid H Minus M-MuLV Reverse Transcriptase ( Fermentas , Fisher Scientific , UK ) according to manufacturer's protocol . PCR reactions were carried out using cDNA timecourse samples as templates , isoform-specific upstream primers GCCAGCCTTCTGAGCCTTG ( for uMAP4 ) , GAGATCCAAGATGTTCAAGTC ( for mMAP4 ) , and CTGTTGGAAGAGACCCCAC ( for oMAP4 ) and downstream primers CAGCTGGCACTGAGCCTG ( to determine relative expression levels ) and GAAGGGCCTCACTGCCAC ( to determine number of microtubule binding repeats ) . As control , the coding sequence of glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) was amplified using the primers CCCACTTGAAGGGTGGAG and CAGGCGGCACGTCAGATC . For differential expression analyses of MAP4 isoforms , the mouse genome assembly ( release date July 2007 ( NCBI37/mm9 ) ) was examined using the UCSC Genome Browser at http://genome . ucsc . edu/ . For RNA sequencing analyses , raw data from previously described C2C12 data sets ( Trapnell et al . , 2010 , see Supplementary file 1 ) of the Mouse ENCODE project were exported . Reads per million ( RPM ) values were then summed and divided by the length of sequenced regions to obtain RPM/Kb values . Expression levels of MAP4 isoforms in various tissues were compared by exporting and analysing published Affymetrix mouse exon array data sets ( Pohl et al . , 2009 , see Supplementary file 2 ) from the Genome Browser . For cloning uMAP4 ( accession number M72414 ) and oMAP4 ( accession number BC042645 ) , cDNA was amplified using primers CAGGTCGACAGAATGGCCGACCTCAG and GACCGCGGACGAGACCAGAATGTCATC for uMAP4 , and GAGGTCGACATGGACTCCCGGAAAGAAATC and CCCGCGGTCTCAATTTGTCTCCTGG for oMAP4 . PCR products were cloned into the MscI site of pCAPs ( Roche , UK ) and confirmed by sequencing , digested with SalI and SacII , and transferred into pEGFP-C1 ( Clontech , France ) . To clone mMAP4 ( assembled from sequences under accession numbers M72414 and U08819 ) , two overlapping fragments encoding its N-terminal and C-terminal domains were generated by PCR using primers CAGGTCGACAGAATGGCCGACCTCAG and CTGCAATCAGCAAGCCCAC , and CCTCTGGGAGATCACCATC and CCCGCGGTCTCAATTTGTCTCCTGG , respectively , and each cloned into pCAPs and sequenced . Plasmids carrying N-terminal and C-terminal coding sequences were then digested with SalI + SpeI and SpeI + SacII , respectively , and cloned into the SalI + SacII sites of pEGFP-C1 . To clone FLAG-oMAP4 , a 5X-FLAG tag ( DYKDADLDKDDDDK ) was amplified by PCR and digested with NgoMIV and SalI . It was then inserted into pEGFP-oMAP4 that was opened with AgeI and SalI , thereby replacing the eGFP ORF . FLAG-oMAP4RIP was generated using FLAG-oMAP4 as a template and primers GAGGTCGACATGGACTCCCGGAAAGAAATC , CTCCTCGAGTTAAGCAGTTGGTACCTGAG , and CCCAGCTGTGAACTTGGTTGATAAGTACCCCTG in 3-step mutagenesis PCR . Bacterial expression vector for N-terminally 6xHis-tagged oMAP4 was generated by SalI + MfeI digestion of pEGFP-oMAP4 ( four MTB repeats ) and insertion of the coding sequence into the XhoI + EcoRI sites of pRSET-A . N-terminally 6x-His-eGFP-tagged oMAP4 expression construct was cloned by transferring EGFP-oMAP4 with NcoI-MfeI from pEGFP-oMAP4 ( five MTB repeats ) into the NcoI-EcoRI sites of pRSET-B vector ( Invitrogen ) . To clone GST fusions of MAP4 fragments for polyclonal antibody generation , regions specific to mMAP4 ( amino acids 631–1060 ) and oMAP4 ( amino acids 22–350 ) were obtained by PCR using pEGFP-mMAP4 ( primers GCCAGCCTTCTGAGCCTTG and CTCCTCGAGTTAAGTAATGGCCCCTGGTTG ) and pEGFP-oMAP4 ( primers GAGGGTCAATTGAATGAAATCGGGCTGAATG and CTCCTCGAGTTAAGCAGTTGGTACCTGAG ) . mMAP4 and oMAP4 PCR products were then digested with EcoRI + XhoI and MfeI + XhoI , respectively , and inserted into the EcoRI + XhoI sites pGEX-6P-1 vector ( GE Healthcare , UK ) . Depletion constructs were based on pSUPERneo . gfp ( Oligoengine , Seattle , WA ) . shRNA-target sequences were chosen using Oligo Retriever ( http://cancan . cshl . edu/RNAi_central/RNAi . cgi ? type=shRNA ) and were as follows: shControl ( targeting Luciferase ) : CGTACGCGGAATACTTCGA; sh-uMAP4: GCCTTGCTCAGGAGTATCC; sh-mMAP4: CAGAGAGTTTGGATAAGAA; sh-oMAP4: GCTGTGAATCTTGTCGATAAG; shDHC: CAATTACAGTCTGGAGTTA; shKHC ( targeting Kif5b ) : CAATTGGAGTTATAGGAAA . The neo . gfp cassette was replaced by GFP-Tubulin , mCherry-Tubulin , or mEos2-Tubulin . EB3-tdTomato , paGFP-Tubulin , and mEos2-Tubulin were described previously ( Rusan and Wadsworth , 2005; McKinney et al . , 2009; Samora et al . , 2011 ) . Rabbit polyclonal antibodies against specific regions in oMAP4 and mMAP4 were raised against affinity purified GST-oMAP422–350 and GST-mMAP4631–1060 , respectively by Absea Biotechnology Ltd . ( China ) . Polyclonal antibodies were purified from sera using the same GST fusion proteins . To do this , purified protein samples were run on SDS-PAGE , transferred to nitrocellulose membranes , protein bands were stained with Ponceau S and excised using a sterile scalpel . Membrane pieces were blocked with 5% wt/vol milk in TBST , cut into small pieces and incubated over night at 4°C in 1 ml TBST and 1 ml antiserum . Membrane pieces were washed thrice with TBST and antibodies were eluted by adding 200 μl 100 mM Glycine-HCl ( pH 2 . 5 ) and vortexing for 1 min . Purified antibodies were then transferred to fresh tubes and neutralised immediately by adding 10 μl 1 M Tris-Base . Antibodies were stored at −20°C after adding Glycerol to a final concentration of 50% vol/vol . Mouse C2C12 myoblasts were cultured on rat tail collagen ( Sigma , UK ) in DMEM-GlutaMAX ( Invitrogen ) supplemented with 10% fetal bovine serum ( FBS ) , 2 mM L-Glutamine , 100 U/ml penicillin , and 100 μg/ml streptomycin with 5% CO2 in a humidified incubator . For localisation analyses , myoblasts were transfected with 1 μg plasmid DNA using Fugene 6 reagent ( Roche ) and analysed 48 hr post-transfection . For myoblast elongation and fusion analysis , 10 , 000 or 20 , 000 C2C12 cells , respectively , were seeded onto collagen-coated coverslips . Cells were transfected 24 hr later with a total amount of 1 . 5 μg shRNA constructs or shRNA and rescue constructs using Lipofectamine Plus reagent ( Invitrogen ) in OptiMEM ( Invitrogen ) , which was replaced with growth medium 4–6 hr after transfection . Muscle cell differentiation was induced by replacing growth medium with differentiation medium ( DMEM , 0 . 1% FBS , 2 mM L-glutamine , 5 μg/ml insulin , 5 μg/ml , transferrin , 100 U/ml penicillin and 100 μg/ml streptomycin ) 20 hr after shRNA transfection . DMEM was changed daily and analysis of shRNA transfected cells was consistently performed 72–78 hr after transfection . Live cells were imaged at 37°C and 5% CO2 in a stage top incubator ( Tokai Hit , Fujinomiya , Japan ) using a 100× or a 60× oil NA 1 . 4 objective on a Deltavision system ( Applied Precision , LLC , Issaquah , WA ) using Chroma filter sets and a Coolsnap HQ camera controlled by SoftWorx ( Applied Precision , LLC ) . For all analysis of microtubule phenotypes , mono-nucleated , elongated cells were selected . For analyses of microtubule dynamics during myoblast differentiation , fluorescence images of EB3-tdTomato were acquired with 500 ms exposure at a temporal resolution of 3 s for 120 s . EB comets were tracked using plusTipTracker and further analysed for microtubule directionality using custom MATLAB code ( see supplementary file MTdirectionality . m ) . For microtubule motility experiments , cells were transiently transfected with pSuper-mEOS2-Tubulin plasmids or co-transfected with paGFP-Tubulin and pSuper-mCherry-Tubulin plasmids . The microtubule cytoskeleton was focussed and bar-shaped regions of interest perpendicular to the microtubule network were selected using the non-converted mEOS2-Tubulin or mCherry-Tubulin signal . After photoactivation or photoconversion of these regions of interest using 10% 406 nm with the Deltavision photokinetics module laser power , images were acquired with 500 ms exposure at a temporal resolution of 1 . 6 s per frame for 60 s or lower temporal resolution for up to 6 min . Images were deconvolved with low noise filtering method for 10 iterations using SoftWorx ( Applied Precision , LLC ) . Microtubule motility events were scored when an activated fragment moved for more than 0 . 5 μm from its original location . The velocity of microtubule movements was determined as the average velocity during each motility event , which lasted on average 15–20 s . Directionality of microtubule motility was scored as paraxial if within 45° of the long axis of the cell . Looping microtubules were defined as those that underwent a directional change of more than 90° during the movement event . For analysis of the contribution of microtubule dynamics and motor-driven microtubule movement to microtubule loss from photoactivated regions , undifferentiated or 48 hr differentiated C2C12 cells expressing mEOS2-Tubulin were treated with either 10 μM Taxol ( to suppress MT depolymerisation ) or 5 mM azide in 1 mM 2-deoxyglucose ( to suppress motor activity ) or a mixture of Taxol and azide ( as a control for photobleaching ) for 30 min before imaging . Signal intensity of photoconverted regions at each time point recorded using ImageJ . After background subtraction , photobleaching correction and normalisation of fluorescence intensity , the half-life of signal dissipation was determined by fitting a single exponential curve to the fluorescence intensity data over time . For immunofluorescence experiments , cells were fixed in −20°C cold methanol for 24–48 hr . Fixed cells were rehydrated by washing with PBS for 5 min and blocked with 0 . 5% BSA wt/vol in PBST for 5 min . Coverslips were then incubated with primary antibodies overnight at 4°C or for 4 hr at RT and secondary antibodies for 1 hr in 0 . 5% BSA wt/vol in PBST . To stain DNA , coverslips were incubated with 4′ , 6-Diamidino-2-phenylindole dihydrochloride ( #D9542; DAPI; Sigma ) for 2 min . Coverslips were then mounted using Vectashield mounting medium ( #H-1000; Vector Labs , UK ) . Primary antibody dilutions were as follows: rabbit anti-mMAP4 ( 1:1000 ) , rabbit anti-PCM-1 ( 1:500 , Dammermann and Merdes , 2002 ) , mouse anti-embryonic myosin ( 1:50; F1 . 652 , DSHB , University of Iowa ) , mouse anti-α-tubulin ( 1:1000; DM1A , Sigma ) , mouse anti-acetylated tubulin ( 1:1000; 6-11B-1 , Sigma ) , mouse anti-myogenin ( 1:50; F5D , Santa Cruz ( Dallas , TX ) ) . Alexa488- , Alexa594- or Alexa647-conjugated anti-mouse or anti-rabbit secondary antibodies ( Molecular Probes , Life Technologies , UK ) were used at 1:500 dilution . For myoblast elongation analysis , cells were imaged using a 20× objective on the Deltavision system as above . Cell lengths of GFP-Tubulin-positive mono-nucleated cells were measured as straight-line distance from tip-to-tip using Image-Pro Analyzer 7 . 0 ( Media Cybernetics , UK ) . For cell fusion analysis , cells with two or less nuclei ( visualised as holes in GFP-Tubulin signal ) and those with 3 or more nuclei were counted on the entire coverslip directly on the Deltavision microscope using a 20× objective and GFP filter sets . To analyse microtubule filament arrangement , GFP-Tubulin was imaged on a Perkin–Elmer UltraView spinning disk confocal system using a 488 nm laser and an Orca-R2 camera ( Hamamatsu , Japan ) under the control of Volocity software ( Perkin–Elmer , Waltham , MA ) . Filaments were manually traced using ImageJ and deviation from the longitudinal cell axis calculated in MATLAB . To determine the number of bundles per cell , undifferentiated C2C12 cells were transiently transfected with either GFP-oMAP4 or GFP-uMAP4 , fixed after 24 hr and stained with tubulin antibodies . Bundles were scored when elongated regions of more than 2 . 3-μm length had an intensity of more than threefold of a single microtubule . Superresolution images of microtubules labelled with anti-tubulin and Alexa488-conjugated anti-mouse antibodies and embedded in ProLong Gold reagent ( Molecular Probes ) were obtained on a N-SIM system ( Nikon , Japan ) at The Babraham Institute using 3D SIM mode with 15 individual images collected for reconstruction . For validation of depletion , shRNA-transfected and 52–54 hr differentiated cells were detached from culture dishes with Trypsin- EDTA . Trypsinised cells were then transferred to PBS +2% FBS and GFP expressing cells were sorted on FACSDiva or Influx instruments ( BD Biosciences , UK ) . Collected cells were gently pelleted at 300×g for 2 min , resuspended to 10 , 000 cells/μl in 1× sample buffer and incubated at 95°C for 5 min . Immunoblotting was performed as described previously ( Samora et al . , 2011 ) . Primary antibody dilutions were as follows: mouse anti-α-tubulin ( 1:10 , 000; DM1A , Sigma ) , mouse anti-FLAG ( 1:5000; FLAG-M2 , Sigma ) , mouse anti-DHC ( 1:500; R-325 , Santa Cruz ) , mouse anti-uKHC ( 1:750; H-50 , Santa Cruz ) , rabbit anti-uMAP4 ( 1:1 , 000 , H-300 , Santa Cruz ) , rabbit anti-mMAP4 ( 1:5000 ) , rabbit anti-oMAP4 ( 1:5000 ) , mouse anti-pentahis ( 1:3000 , Qiagen ( UK ) ) . Horseradish peroxidase conjugated anti-mouse or anti-rabbit secondary antibodies ( Promega , UK ) were used at 1:4000 dilution . 6xHis-oMAP4 and 6xHis-GFP-oMAP4 were expressed in E . coli strain BL21-CodonPlus- ( DE3 ) and expression was induced with 0 . 5 mM isopropyl-β-D-thiogalactoside at 37°C . Bacteria were lysed in binding buffer ( 50 mM NaPO4 buffer , pH 8 . 0; 300 mM NaCl; 2 mM β-mercaptoethanol; 15% glycerol ) by sonication . oMAP4 was bound to Ni-NTA resin ( Qiagen ) and eluted with 250 mM imidazole in binding buffer . After twofold dilution with low-salt buffer ( 20 mM MES , pH 6 . 8; 1 mM EGTA; 0 . 5 mM MgCl2 ) the MAP4-containing fractions were loaded on SP Fast Flow Sepharose ( GE Healthcare ) , washed with low-salt buffer and eluted with a step gradient of high-salt buffer ( 20 mM MES , pH 6 . 8; 1 mM EGTA; 0 . 5 mM MgCl2; 1 M NaCl ) . All purification was carried out at room temperature to minimise protein aggregation . Proteins were analysed by SDS–polyacrylamide gel electrophoresis and SimplyBlue staining ( Invitrogen ) . Buffer exchange to BRB80 ( 80 mM PIPES , pH 6 . 8; 1 mM MgCl2; 1 mM EGTA ) was carried out using Vivaspin spin columns ( Sartorius , Germany ) according to the manufacturer's protocol . GST fusion proteins for antibody generation and purification were bound to Glutathione-Agarose ( Sigma ) , washed with PBS and eluted with GST elution buffer ( 50 mM Tris-HCl , pH-8 . 0; 150 mM NaCl; 2 mM 2-Mercaptoethanol; 10 mM Glutathione ) . Full-length Drosphila kinesin-1 was purified previously ( Braun et al . , 2009 ) . EB1-GFP-6xHis was purified as described previously ( Grimaldi et al . , 2014 ) . 6xHis-PRC1 was purified using Ni-NTA as described for 6xHis-oMAP4 above . To determine sedimentation properties , cleared extracts of BL21 cells expressing 6xHis-oMAP4 and 6xHis-GFP-oMAP4 in BRB80 plus complete protease inhibotors ( Roche ) were loaded onto 5-ml 10–40% vol/vol glycerol gradients prepared with a Gradient Master ( Biocomp , Canada ) and spun at 45 , 000×g in a SW55Ti rotor for 14 hr . Standard proteins were loaded at 5 mg/ml individually on separate gradients . Gradients were fractionated by pipetting from the top in 250 μl aliquots and analysed by measuring the absorbance at 280 nm or analysing band intensity on coomassie-stained polyacrylamide gels or immunoblots using pentahis antibodies ( Qiagen ) . The frictional ratio was determined ƒ/ƒmin = Smax/S with Smax = 0 . 00361·M2/3 in Svedbergs for a protein of mass M in Daltons ( Erickson , 2009 ) . Tubulin was prepared from pig brains according to published protocols ( Gell et al . , 2011 ) . Labelled tubulin was from Cytoskleleton Inc . ( Denver , CO ) , nucleotides were from Jena Biosciences ( Germany ) and all other chemicals were from Sigma unless indicated . Microtubule seeds were assembled from tubulin , biotin-tubulin , and Hilyte647-tubulin at a molar ratio of 25:1:2 in the presence of 1 mM GMP-CPP in MRB80 ( 80 mM PIPES , pH 6 . 8 with KOH , 1 mM EGTA , 4 mM MgCl2 ) for 1 hr at 37°C , diluted 20-fold with MRB80 + 2 μM Paclitaxel and stored at RT . A 100-μm deep flow chamber was made from a slide and a hydrochloric acid-treated coverslip using double-sided tape ( Scotch 3M , UK ) . For MT dynamics assays , the flow chamber was passivated with PLL-PEG-50% biotin ( Susos AG , Zurich , Switzerland ) . Seeds were attached to this surface using streptavidin and blocked with 1 mg/ml κ-casein . A reaction mix containing 15 μM tubulin , 1 μM X-Rhodamine tubulin , 50 mM KCl , 1 mM GTP , 0 . 6 mg/ml κ-casein , 0 . 2% methyl cellulose , 4 mM DTT , 0 . 2 mg/ml catalase , 0 . 4 mg/ml glucose oxidase , 50 mM glucose in MRB80 , supplemented with 80 nM GFP-oMAP4 or buffer was clarified for 8 min at 190 , 000×g in an airfuge ( Beckman Coulter , UK ) , the supernatant added to the flow chamber and sealed with candle wax . For gliding assays , 3 nM Drosophila kinesin in MRB80 , supplemented with 10 mM β-Mercaptoethanol and 0 . 1 mM MgATP were flown into the glass chamber , before blocking with 1 mg/ml κ-casein . The gliding mix containing X-rhodamine- and Hilyte647-labelled microtubules , 1 mM ATP , 4 mM MgCl2 , oxygen scavenger system ( 4 mM DTT , 0 . 2 mg/ml catalase , 0 . 4 mg/ml glucose oxidase , 50 mM glucose ) , ATP regeneration system ( 5 mM phosphocreatine , 7 U/ml creatine phosphokinase ) , 80 nM GFP-oMAP4 in MRB80 was added to the flow chamber and sealed with candle wax . Microtubule assembly and gliding assays were observed on an Olympus TIRF system using a 100× NA 1 . 49 objective , 1 . 6× additional magnification , 488 nm , 561 nm , and 640 nm laser lines , a Hamamatsu ImageEM-1k back-illuminated EM-CCD camera under the control of xcellence software ( Olympus , Germany ) . Microtubule gliding speeds were determined from tracks obtained using ImageJ plugin MTrackJ . Microtubule encounters were classified as zippering when they resulted in bundling for a minimum length of 2 μm away from the initial point of contact . For parallel bundling , zippering would need to occur towards the minus end of both microtubules . For antiparallel zippering , both aligned microtubule ends would need to continue growth for at least 2 μm along the lattice of the other microtubule . Microtubule lengths were measured to address whether the bias in microtubule zippering by oMAP4 arises from differences in lengths between parallel and antiparallel microtubules . For microtubule encounters in either orientation that resulted in successful zippering , distance from the point of encounter to the microtubule seed or to the next microtubule crossover point was measured . Microtubule lengths were measured similarly for unsuccessful encounters but these were restricted to shallow angle encounters at which oMAP4-mediated zippering typically occurs , that is , 10°–30° . Microtubules were polymerised in BRB80 buffer from 40 to 50 μM pig-brain tubulin in the presence of 1 mM GTP by incubation at 37°C for 30 min and stabilised by addition of 20 μM Paclitaxel ( Sigma ) . To remove non-polymerised tubulin , microtubules were pelleted by centrifugation at 45 , 000 RPM for 25 min at 27°C , washed with and resuspended in BRB80 and 20 μM Paclitaxel . To test binding of MAP4 proteins to microtubules , 1 . 5 μM of Taxol-stabilised microtubules was mixed with purified 60 nM oMAP4 in the presence of 20 μM Paclitaxel in BRB80 in a total volume of 50 μl . Reactions were then incubated for 20 min at 37°C and centrifuged at 35°C , 50 , 000 RPM for 15 min . After recovery of the supernatant , the pellet was washed with and then resuspended in 50 μl BRB80 and 20 μM Paclitaxel . After addition of SDS-PAGE sample buffer , pellet and supernatant fractions were analysed by SDS-PAGE . For microtubule bundling experiments , 40 μl drops of 0 . 1 mg/ml Poly-L-Lysine ( Sigma ) were placed onto 70% ethanol washed and air-dried 22 × 22 mm glass coverslips ( Menzel-Gläser , Germany ) . After removal of the Poly-L-Lysine solution using a bench top coverslip centrifuge ( Technical Video Ltd . , Port Townsend , WA ) , coverslips were rinsed three times with 150 μl ddH2O using the same procedure and air dried . To test if oMAP4 can bundle microtubules , Rhodamine-labelled microtubule seeds stabilised with GMP-CPP or Taxol-stabilised microtubules were mixed in BRB80 in a total volume of 10 μl and incubated with buffer or with 60 nM oMAP4 at 22°C for 10–15 min . 1 μl samples of reactions were then placed onto slides , mounted with Poly-L-Lysine coated coverslips , and imaged using the Deltavision widefield system as above . For comparative analysis of EB1- , PRC1- , and oMAP4-mediated bundling , freshly polymerised GMPCPP-stabilised Hilyte647-labelled microtubules were incubated with 80 nM protein in MRB80 for 10 min before loading into a 10 μm flow chamber made from a slide and a poly-lysine-coated coverslip . At least 30 random fields were imaged using TIRF microscopy at critical angle . Bundles were counted if at least three times as bright as single microtubules . Statistical data analyses and graphing were performed using Origin Pro 8 . 5 ( OriginLab , Northampton , MA ) or MATLAB ( Mathworks , Natick , MA ) . Image preparation for publication was done using deconvolution in Softworx ( Applied Precision ) , ImageJ and Adobe Illustrator . All statistical significance analyses were carried out using two-tailed two-sample t-tests assuming equal variance . Kuiper statistics was calculated from cumulative distributions of data compared to a random distribution using custom MATLAB code ( see Source code 1 MTdirectionality . m ) .	Skeletal muscles—which enable animals to move—are made up of large elongated muscle cells that span the entire length of the muscle . These cells contain stacks of structures called sarcomeres that enable the cells to contract and generate the force required for movement . Cells called myoblasts elongate and fuse together at their tips to make the muscle cells . Within the myoblasts , long filaments called microtubules are arranged in an overlapping linear pattern . The filaments act as a template that helps the sarcomeres to align as the muscle cells form . A family of microtubule-associated proteins ( or ‘MAPs’ for short ) bind to microtubules and assist in organising the filaments , but it is not clear how they work . Mogessie et al . used microscopy to observe the formation of the microtubule filaments in living myoblasts . The experiments show that the filaments progressively become more ordered as the myoblasts develop into muscle cells . Mogessie et al . identified a new member of the MAP family that is produced in myoblasts as soon as they start to form muscle fibres , and named it oMAP4 . The microtubules in cells that make smaller amounts of this protein were more disorganised , and these cells were unable to fuse with each other to form muscle cells . The experiments also found that oMAP4 can create links between different microtubules and act as a brake to prevent the filaments being moved excessively by motor proteins . Therefore , Mogessie et al . suggest that oMAP4 contributes to the formation of a strong and stable arrangement of filaments . This , in turn , allows the muscle cells to become very long . Making more oMAP4 alone is not sufficient to form the elongated muscle cells . Therefore , the next challenge is to understand how other processes—such as the selective stabilisation of some microtubules and the movement of cell materials along the microtubules—cooperate to control muscle fibre formation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	A novel isoform of MAP4 organises the paraxial microtubule array required for muscle cell differentiation
Eukaryotic cells rapidly reduce protein synthesis in response to various stress conditions . This can be achieved by the phosphorylation-mediated inactivation of a key translation initiation factor , eukaryotic initiation factor 2 ( eIF2 ) . However , the persistent translation of certain mRNAs is required for deployment of an adequate stress response . We carried out ribosome profiling of cultured human cells under conditions of severe stress induced with sodium arsenite . Although this led to a 5 . 4-fold general translational repression , the protein coding open reading frames ( ORFs ) of certain individual mRNAs exhibited resistance to the inhibition . Nearly all resistant transcripts possess at least one efficiently translated upstream open reading frame ( uORF ) that represses translation of the main coding ORF under normal conditions . Site-specific mutagenesis of two identified stress resistant mRNAs ( PPP1R15B and IFRD1 ) demonstrated that a single uORF is sufficient for eIF2-mediated translation control in both cases . Phylogenetic analysis suggests that at least two regulatory uORFs ( namely , in SLC35A4 and MIEF1 ) encode functional protein products . Protein synthesis , as one of the most energy consuming processes in the cell , is under stringent regulation . In eukaryotes , the activity of many components of the translational machinery is modulated by various post-translational modifications in order to adjust either global or mRNA-specific translation . One of the better studied cases of translational control is the phosphorylation of eukaryotic initiation factor 2 ( eIF2 ) ( Sonenberg and Hinnebusch , 2009 ) . eIF2 forms the ternary complex ( TC ) with GTP and Met-tRNAi and is loaded onto the 40S ribosome to enable it to recognize a start codon , after which eIF2*GDP is released . GDP is then recycled to GTP by guanine exchange factor ( GEF ) , eIF2B , to enable another round of initiation . During various stress conditions the cell triggers the integrated stress response ( ISR ) by activating any of four kinases , EIF2AK1 ( also known as [a . k . a . ] HRI ) , EIF2AK2 ( a . k . a . PKR ) , EIF2AK3 ( a . k . a . PERK ) , or EIF2AK4 ( a . k . a . GCN2 ) , that phosphorylate the alpha subunit of eIF2 at Ser51 ( Baird and Wek , 2012 ) . Instead of a rapid recycling , eIF2B forms a stable complex with phosphorylated eIF2 . The concentration of eIF2 is higher than that of eIF2B , therefore even phosphorylation of a modest number of eIF2 molecules rapidly reduces the pool of active eIF2B resulting in the general inhibition of total protein synthesis ( Hinnebusch , 2014 ) . While the general suppression of translation conserves cellular resources , the active synthesis of certain factors is required to respond to the consequences of stress . Mammalian genes whose expression is known to evade translational arrest triggered by eIF2 phosphorylation include ATF4 ( Lu et al . , 2004; Vattem and Wek , 2004 ) , PPP1R15A ( a . k . a . GADD34 ) ( Lee et al . , 2009 ) , ATF5 ( Watatani et al . , 2008; Zhou et al . , 2008; Hatano et al . , 2013 ) , and DDIT3 ( a . k . a . CHOP ) ( Jousse et al . , 2001; Chen et al . , 2010 ) . ATF4 and ATF5 are believed to be regulated through the mechanism known as delayed reinitiation , initially characterized for the yeast GCN4 ( a functional analogue of ATF4 ) ( Hinnebusch , 1997 ) . This requires at least two upstream open reading frames ( uORFs ) . In ATF4 mRNA , after translation termination at the first uORF , the 40S resumes scanning albeit without the TC . The distance scanned by this ribosome subunit before it reacquires the TC depends on TC availability . Under normal conditions most of the 40S is quickly reloaded with TC and therefore can reinitiate at the second uORF . Under stress conditions ( i . e . , low eIF2 availability ) , a larger fraction of 40S subunits scan past the second uORF initiation codon before binding of the TC , thereby enabling reinitiation at the next ORF . A different mechanism of translational resistance , relying on the translation of a single uORF in the 5′ leader , has been proposed for DDIT3 ( Palam et al . , 2011 ) . A fraction of scanning ribosomes recognize and initiate at the uORF initiation codon in a weak Kozak context . Under normal conditions with a high initiation rate , the translation of this uORF inhibits leaky scanning by the obstruction of scanning ribosomes . Under stress conditions , the reduced ribosomal loading results in an alleviation of this obstruction . For the examples mentioned above , translational control is based on the reduced availability of TC . In specific cases initiation can occur without eIF2 . Some viral mRNAs harbour internal ribosome entry sites ( IRES ) that allow translation initiation to take place by recruiting alternative factors , that is eIF5B ( Pestova et al . , 2008; Terenin et al . , 2008 ) , eIF2D , and a complex of MCTS1 ( a . k . a . MCT-1 ) , and DENR ( Dmitriev et al . , 2010; Skabkin et al . , 2010 ) , or even initiate without Met-tRNAi and any initiation factors ( Wilson et al . , 2000 ) . However , the existence of ‘viral-like’ IRESs in mammalian mRNAs remains controversial ( Shatsky et al . , 2010; Jackson , 2013 ) . The present work uses ribosome profiling ( Ingolia et al . , 2009 ) to explore the immediate effect of sodium arsenite treatment ( NaAsO2 ) , which results in a rapid phosphorylation of eIF2 , on protein synthesis . This technique provides a snapshot of translating ribosomes over the entire transcriptome with subcodon resolution ( see reviews by Michel and Baranov , 2013; Ingolia , 2014 ) . In order to generate the most informative conditions for characterizing eIF2-dependent mechanisms of translation regulation , it was important to minimize the transcriptional response and induce significant but not complete inhibition of translation . For this purpose , we chose to treat cells with sodium arsenite for a short time period and to monitor the immediate translational response . Sodium arsenite is a well-known potent inducer of eIF2 phosphorylation that activates EIF2AK1 ( McEwen et al . , 2005 ) . We examined changes in the phosphorylation status of eIF2 and EIF4EBP1 during arsenite treatment to identify suitable conditions for ribosome profiling . The phosphorylated form of eIF2 ( p-eIF2 ) progressively accumulates during the first 2 hr of stress ( Figure 1A ) . After 0 . 5 hr of arsenite treatment , p-eIF2 reaches 30–40% of maximal levels of phosphorylation ( Figure 1A ) . The dephosphorylation of EIF4EBP1 becomes evident 1 hr after treatment and does not revert ( Figure 1A ) . We did not detect changes in phosphorylation of p70 S6 kinase and its substrate RPS6 during arsenite treatment ( Figure 1A ) . A robust accumulation of ATF4 is evident 30 min after treatment . These results suggest that 0 . 5 hr post-treatment is likely to be suitable for examining eIF2 inhibition whilst minimizing possible arsenite-induced side effects . Furthermore , the number of ribosomes in polysome fractions was reduced by ∼4 . 5-fold under these conditions ( Figure 1B ) . 10 . 7554/eLife . 03971 . 003Figure 1 . Analysis of differential gene expression under conditions of oxidative stress induced with sodium arsenite treatment . ( A ) Western blotting time series analysis of several protein components of HEK293T lysates after treatment with 40 µM sodium arsenite . ( B ) Sucrose density gradient profiles of HEK293T cells untreated and treated for 30 min with sodium arsenite at 40 µM . ( C ) Distribution of raw read counts over mRNA functional regions . ( D ) Metagene analysis: short reads from all mRNAs are aligned around 5′ and 3′ ends of CDS , transcript read density ( RNA ) is shown using curves , ribosome density ( number of footprint reads ) is shown using columns corresponding to the alignment locations of the read 5′ ends . ( E ) Differential gene expression analysis . Scatter plots compare ribosome occupancy ( top ) , transcript levels ( middle ) , and translation efficiency ( TE ) ( bottom ) between treated and untreated conditions . To avoid error due to uORF translation the number of ribo-seq reads aligning to the CDS only was used to determine the ribosome occupancy and TE . The x axis represents the normalized number of reads corresponding to the experiment/condition of minimal expression ( see ‘Materials and methods’ ) . The threshold used to denote differentially expressed genes ( Z-score of 4 ) is indicated in orange . Certain genes of interest are indicated with numbers , followed by their gene symbols . ( F ) Two heat maps displaying the fold change and Z-score for the top 22 most stress resistant and bottom 9 most stress sensitive genes , as estimated based on statistical significance of the change of their ribosome occupancy ( ribo-seq Z-score ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 00310 . 7554/eLife . 03971 . 004Figure 1—source data 1 . Read counts and statistics of gene expression response for RNA-seq and ribo-seq experiments for control and stress conditions for each individual transcript . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 00410 . 7554/eLife . 03971 . 005Figure 1—figure supplement 1 . Additional characteristics of ribosome profiling data . ( A ) Reproducibility between biological replicas , from top to bottom: ribosome profiling under control conditions; ribosome profiling under arsenite induced stress; mRNA-seq fragments , non-treated; mRNA-seq fragments under stress . ( B ) Distribution of ribosome protected fragments and fragmented RNA relative to the annotated CDS depending on their length ( x axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 00510 . 7554/eLife . 03971 . 006Figure 1—figure supplement 2 . Analysis of differential gene expression . ( A ) The Z-score based normalization approach for the identification of differentially expressed genes . Z-score is used to mitigate expression variance for the genes expressed at different levels . ( B ) and ( C ) Left panels: correlation of Z-scores between replicas calculated for the changes in RNA levels ( B ) and translation efficiencies ( C ) . Right panels: the most significantly regulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 006 Therefore , HEK293T cells were treated with 40 µM sodium arsenite for 30 min before harvesting for ribosomal profiling which was carried out according to the Ingolia et al . protocol ( Ingolia et al . , 2012 ) with some modifications ( see ‘Materials and methods’ ) . Figure 1C–E shows the general characteristics of the ribo-seq and mRNA-seq datasets . As expected , in both conditions most ribo-seq reads were mapped to coding regions . The distribution of ribo-seq 5′-end reads , but not of mRNA-seq 5′-end reads ( same size randomly fragmented ‘naked’ mRNA isolated from cytoplasmic lysate ) , exhibits the characteristic triplet periodicity ( Figure 1—figure supplement 1 ) . Over 7000 mRNA sequences were uniquely mapped with at least 100 ribo-seq reads . Owing to the stochastic nature of massively parallel sequencing , the accuracy of an estimate of the level of expression of a gene is dependent on its sequencing depth . Therefore the estimated expression levels of weakly expressed genes have greater variability than highly expressed genes . To mitigate this effect we used a Z-score transformation ( see review by Quackenbush , 2002 ) . Genes were first ordered based on their lowest read depth ( minimum expression ) . The parameters of the distribution of expression changes for the genes with similar expression levels were used to calculate Z-scores of differential expression for individual genes ( see ‘Materials and methods’ and Figure 1—figure supplement 2 ) . We used a Z-score of 4 as an arbitrary threshold of statistical significance for differentially regulated genes to minimize the false discovery rate . The estimated time required for mRNA maturation ( Femino et al . , 1998; Audibert et al . , 2002 ) is comparable to the duration of the arsenite treatment , therefore we did not expect significant changes in mRNA levels due to a transcriptional response . However , it is conceivable that the arsenite treatment could affect the stability of specific mRNAs . Indeed , the treatment was found to significantly alter the transcript levels of 24 genes ( Figure 1E ) . The most pronounced effect observed is the accumulation of JUN mRNA , a transcript that is short-lived under normal conditions ( Elkon et al . , 2010; Rabani et al . , 2011 ) . JUN encodes a subunit of the AP-1 transcription factor implicated in response to a myriad of physiological and pathological stimuli ( reviewed in Hess et al . , 2004 ) . AP-1 transcription factors consist of homo- or heterodimers of different subunits and its composition is crucial for its specificity . As displayed in Figure 1F , despite a significant decrease in its translational efficiency , the overall expression of JUN is almost unaffected upon arsenite treatment because of the increase in its mRNA levels ( Figure 1E , F ) . According to previous observations , AP-1 may induce apoptosis upon treatment with arsenite ( Huang et al . , 1999; Namgung and Xia , 2000 ) . A median 5 . 4-fold reduction of ribosomal occupancy ( translational efficiency , TE ) was observed with the profiling data , a value that is consistent with the reduced number of ribosomes in the polysome fraction ( Figure 1B ) . A relatively small subset of mRNAs displayed exceptional sensitivity to translational inhibition ( see Figure 1—source data 1; the most prominent ones are displayed in Figure 1F ) . Among these is ODC1 which codes for ornithine decarboxylase , the rate-limiting enzyme of the polyamine biosynthesis pathway . An interesting case of potential downregulation is in EIF2AK2 ( a . k . a . PKR ) , which encodes one of the four eIF2 kinases ( we refer to it as potential because its TE and ribo-seq Z-scores did not pass the threshold of statistical significance , but are close to it ) . Among other extremely sensitive genes are several that encode RNA-binding proteins ( PABPC1 , PCBP2 , RPL12 , and CSDE1 ) and cyclins G1 ( CCNG1 ) and I ( CCNI ) . To explore the possible activation of the mTOR signalling axis after the course of 0 . 5 hr arsenite treatment , we analysed the translation of mRNAs that were reported to be strongly downregulated upon pharmacological inhibition of mTOR ( Hsieh et al . , 2012 ) . Almost all of them have negative Z-scores with their TE decrease upon arsenite treatment ∼25% greater than the average ( Figure 1—figure supplement 2C ) . Thus , while arsenite treatment may affect the mTOR pathway , its impact on translation control is not substantial in comparison with eIF2 inhibition . Several genes that were previously reported to resist the translation inhibition caused by eIF2 phosphorylation were also found to be resistant in our study ( Figure 1F ) . This includes the well-studied ATF4 , ATF5 , and PPP1R15A . We did not observe translational resistance for either SRC ( Figure 1—source data 1 ) , which was reported to be translated in an eIF2-independent mode ( Allam and Ali , 2010 ) , or PRNP , which was also reported to escape eIF2 associated repression ( Moreno et al . , 2012 ) . The general repression of eIF2 may be expected to promote translation of cellular IRESs if they enabled an eIF2-independent mode of translation as reported for XIAP mRNA ( Thakor and Holcik , 2012 ) . However , we found no evidence of resistance by any genes with putative IRES elements according to the IRESite database ( Mokrejs et al . , 2010 ) , see Table 1 including XIAP mRNA . It is important to note that many of the genes from the IRESite are not expressed at levels sufficient for detecting resistance . 10 . 7554/eLife . 03971 . 007Table 1 . Translation response of mRNAs with reported IRES from IRESiteDOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 007Gene_nameIRES nameORF#Minimal expressionTE fold changeTE Z-scoreAGTR1AT1R_var1140 . 461 . 12AGTR1AT1R_var2140 . 461 . 12AGTR1AT1R_var3140 . 461 . 12AGTR1AT1R_var4140 . 461 . 12APAF1Apaf-11102 . 90 . 291 . 8AQP4AQP411 . 50 . 982 . 47ATAD5ELG111083 . 40 . 220 . 75BAG1BAG1_p36delta236 nt BAG1_p3641376 . 50 . 1−1 . 31BCL2BCL21100 . 18−0 . 22BIRC2c-IAP1_285-1399 c-IAP1_1313-146211470 . 13−0 . 94CCND1CCND112130 . 17−0 . 15CDK11APITSLRE_p5810NANACDKN1Bp27kip1112520 . 17−0 . 3CSDE1UNR116 , 657 . 50 . 05−5 . 12DCLRE1AhSNM111065 . 20 . 17−0 . 24EIF4G1eIF4G112 , 9370 . 210 . 65EIF4G1eIF4GI-ext112 , 9370 . 210 . 65EIF4G2DAP5121 , 727 . 60 . 261 . 56EIF4G3eiF4GII11305 . 60 . 13−1 . 46EIF4G3eIF4GII-long11305 . 60 . 13−1 . 46FGF1FGF1A10NANAFMR1FMR11137 . 80 . 13−1 . 35HSPA1Ahsp7010NANAHSPA5BiP_-222_-311819 . 30 . 231IGF2IGF2_leader210NANALAMB1LamB1_-335_-1111410 . 13−1 . 52LEF1LEF112480 . 15−0 . 8MNTMNT_75-267 MNT_36-16011440 . 281 . 32MYBMYB1154 . 20 . 09−1 . 47MYCc-myc219460 . 14−1 . 2MYCL1L-myc10 . 5NANAMYCNn-MYC10NANANKRFNRF_-653_-17110640 . 18−0 . 01PDGFBPDGF2/c-sis10NANAPIM1Pim-111130 . 12−1 . 1RUNX1AML1/RUNX11100 . 16−0 . 38RUNX1T1MTG8a11630 . 13−1 . 08XIAPxIAP_5-464 XIAP_305-46612169 . 60 . 12−1 . 67 The mRNAs encoding ATF4 , PPP1R15A , SLC35A4 , C19orf48 , ATF5 , and HOXB2 were found to be ‘preferentially translated’ ( defined as having a TE >4 and a fold change >1 ) , while mRNAs encoding IFRD1 , PTP4A1 , PCNXL4 , and UCP2 were found to be ‘resistant’ ( TE >4 and a fold change <1 ) . Due to the small number of preferentially translated and resistant mRNAs , we analysed their properties together and , for simplicity , we refer to them as resistant for the remainder of this text . Examination of individual mRNA profiles frequently revealed the presence of extensively translated uORFs in resistant mRNAs ( Figure 2 ) . Indeed , with the exception of a single weakly expressed gene HOXB2 , all mRNAs found to be resistant ( TE Z-score >4 ) contained uORF ( s ) that are translated under normal conditions . However , ribosome profiling data suggest that 8% of other expressed mRNAs also have translated uORFs ( Figure 3A ) . Hence , the mere presence of translated uORFs is a poor predictor of resistance . We therefore investigated further the features of uORFs that are associated with stress resistance . As can be seen in Figure 3B , uORFs in the resistant mRNAs are usually efficiently translated under normal conditions , though yet again there is a large absolute number of non-resistant mRNAs that contain similarly efficiently translated uORFs . Figure 3C shows the relationship between the TE of the CDS and the resistance: the CDS of the most resistant mRNAs is weakly translated under normal conditions . The ratio of the ribosome densities in uORFs and in CDS provides a much better criterion for discriminating between resistant and non-resistant mRNAs ( Figure 3D ) . The length of all the translated uORFs ( in the resistant mRNAs ) was found to exceed 20 codons , although this is of limited predictive value as there are many long uORFs in non-resistant mRNAs ( Figure 3E ) . Based on these findings we expected that , upon arsenite treatment , the ribosome density for resistant mRNAs would shift from the 5′ leaders to CDS . Such a shift is indeed observable ( Figure 3F ) . 10 . 7554/eLife . 03971 . 008Figure 2 . Upstream open reading frame ( uORF ) conservation and ribosome density for the eight top most stress resistant mRNAs in terms of their translation efficiency and also for mRNAs of UCP2 , PPP1R15B , AZIN1 , and MIEF1 . Bottom plots for each mRNA show counts of mRNA-seq reads ( grey ) and ribosome reads ( blue and red ) as columns ( control: positive values; arsenite treatment: negative values ) . The annotated CDS region is highlighted in yellow . Translated conserved ORFs in the 5′ leaders are highlighted in violet . Read counts above the cut-off are shown with numbers above corresponding columns . Top plots represent conservation of uORF features within the leaders of the orthologous mRNAs ( upstream of annotated CDS ) obtained from the analysis of genomic alignments of the 46 vertebrates using the human sequence as a reference . Each box corresponds to one of the three reading frames where AUG codons are shown as pink dots and stop codons as navy dots in each of the genomic sequences used in the alignments . Regions of multiple sequence alignment corresponding to translated conserved uORFs are highlighted in violet . Introns and gaps were removed from the alignments . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 00810 . 7554/eLife . 03971 . 009Figure 2—figure supplement 1 . Multiple alignments of codon sequences from 100 vertebrate genomes aligned to the region of the conserved SLC35A4 upstream open reading frame ( uORF ) in three different frames . Codons are represented as coloured bricks according to the following scheme: pink ( ATG ) , navy ( stop codons: TAA , TAG or TGA ) . The rest are coloured depending on the nature of substitution relative to the human sequence as follows: white: no substitution; light green: synonymous; green: positive ( BLOSUM62 > 0 ) ; red: negative ( BLOSUM62 ≤ 0 ) ; deletions: grey . Locations of ORFs ( from ATG to a stop ) present in humans are shown under each alignment as a thicker dark blue bar . Regions of alignment with a high number of light green or green bricks ( synonymous and positive substitutions ) indicate protein coding evolution . The codon substitution analysis was carried out using CodAlignView ( Jungreis I , Lin M , Kellis M . CodAlignView: a tool for visualizing protein-coding constraint ) and processed with a python script to remove sequences of codons . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 00910 . 7554/eLife . 03971 . 010Figure 2—figure supplement 2 . Multiple alignments of codon sequences from 100 vertebrate genomes aligned to the region of conserved MIEF1 upstream open reading frame ( uORF ) in three different frames . See Figure 2—figure supplement 1 for the explanation of the colour scheme used for the alignment visualization . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01010 . 7554/eLife . 03971 . 011Figure 2—figure supplement 3 . Publicly available ribosome profiling data in GWIPS-viz for SLC35A4 and MIEF1 . Ribosome profiling data aligned to the SLC35A4 ( A ) and MIEF1 ( B ) loci of the human genome from nine studies available in the GWIPS-viz Browser . The positions of the conserved upstream open reading frames ( uORFs ) are shown with a red bar below the blue bars representing corresponding RefSeq transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01110 . 7554/eLife . 03971 . 012Figure 3 . Relationship between mRNA stress resistance and upstream open reading frames ( uORFs ) . ( A ) Frequency of AUG initiating uORF occurrence and their translation in stress resistant and other mRNAs . Relationship between stress resistance ( y axis ) and translation efficiency of uORFs ( B ) , CDS ( C ) , and uORF/CDS ratio ( D ) . Translationally resistant genes ( shaded in blue ) have a high uORF translational efficiency ( TE ) and a low CDS TE . ( E ) Relationship between uORF length ( x axis ) , their TE ( y axis ) , and the level of stress resistance ( differential colouring ) . ( F ) Relationship between stress resistance ( y axis ) and shift of ribosome density in the 3′ direction . These plots indicate that , under normal conditions , resistant mRNAs tend to display a high uORF TE and a low CDS TE while , under stress conditions , resistant mRNAs are associated with a shift of ribosome density in the 3′ direction owing to a reduced ratio of ribosome density between uORFs and CDS . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01210 . 7554/eLife . 03971 . 013Figure 3—figure supplement 1 . Analysis of 5’ leader and upstream open reading frame ( uORF ) features in the resistant mRNAs . ( A ) WebLogo representation of information content within translation initiation sequences ( from position −4 to position +3 ) for uORF starts in the resistant mRNAs . ( B ) Comparison of frequencies of various translation initiation sequences ( −4 to +3 ) for annotated ORFs ( x axis ) and AUG present in 5' leaders ( y axis ) . Translation initiation sequences of uORFs in the resistant mRNAs are shown in blue . ( C ) Scatter plot representing relationship between translation response ( y axis ) and the length of 5′ leaders ( x axis ) . ( D ) Relationship between translational response ( y axis ) and free energy of potential RNA secondary structures within the first 240 nt of 5′ leaders ( x axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 013 We also compared various sequence features of 5′ leaders and uORFs between the resistant mRNAs and the remaining expressed mRNAs . We explored the nucleotide ( nt ) context surrounding uORF start codons ( mostly AUG but also CUG ) in resistant mRNAs but found no evidence for selection for a particular context ( Figure 3—figure supplement 1A ) ; the frequency of individual heptameric initiation sequences ( −3 to +4 ) is equally variable across uORFs of resistant and non-resistant mRNAs as well as at annotated starts of CDS ( Figure 3—figure supplement 1B ) . We found that the average 5′ leader length of resistant mRNAs is longer ( 378 . 5 nt ) than that of other mRNAs ( 169 . 0 nt ) , but there is significant variation within both distributions ( Figure 3—figure supplement 1C ) . We also found that 5′ leaders of resistant mRNAs have lower potential for RNA secondary structure formation within the first 240 nt based on free energy estimates of potential structures predicted with RNAfold ( Lorenz et al . , 2011 ) . Yet , the difference is small and RNA secondary structure potential does not correlate well with resistance ( Figure 3—figure supplement 1D ) . Based on this analysis we concluded that uORFs are ubiquitous in all highly resistant mRNAs expressed in HEK293T cells and the efficient ( and perhaps inhibitory ) translation of these uORFs most likely plays a crucial role in the mechanism of resistance . The mere presence of an uORF and its translation is insufficient to provide the resistance to eIF2 inactivation . We focused our attention on newly identified uORF-bearing mRNAs whose translation was refractory to eIF2 inactivation . For some of these the regulatory function of uORFs has been described previously , but its implication in eIF2-dependent translational control was not shown . This was true for IFRD1 ( a . k . a . PC4 and TIS7 ) , an interferon-related developmental regulator that was reported to be a modifier gene for cystic fibrosis ( Gu et al . , 2009 ) . It has been reported previously that one of the two IFRD1 transcript variants possesses a 51 codon uORF which triggers mRNA decay upon termination under normal conditions but not under conditions of Unfolded Protein Response mediated by tunicamycin ( Zhao et al . , 2010 ) . We did not observe a significant change in the transcript level upon arsenite treatment . A second case is UCP2 , which codes for a mitochondrial anion carrier protein that increases the proton conductance of the mitochondrial membrane in response to reactive oxygen species production ( Baird and Wek , 2012 ) . It was shown that expression of UCP2 is upregulated at the translational level upon oxidative stress ( Adam et al . , 2006 ) . The UCP2 5′ leader contains a 36 codon uORF that inhibits translation of UCP2 mRNA under normal conditions ( Hurtaud et al . , 2006 ) . To our knowledge the other newly identified mRNAs have not been shown to be regulated at the translational level before . One of the most unusual examples is the mRNA encoding the probable UDP-sugar transporter protein SLC35A4 ( Figure 2 ) . Its 5′ leader contains 11 AUG codons , most of which are not conserved; however , one AUG that initiates a 102 codon uORF is highly conserved across vertebrates . This uORF encodes a peptide sequence containing PFAM domain DUF4535 , ID PF15054; moreover , the pattern of its conservation is consistent with protein coding evolution ( Figure 2—figure supplement 1 ) , suggesting that this uORF likely encodes a functional protein . This alternative protein ( EMBL accession HF548106 ) was recently detected by mass spectrometry analysis of cultured cells and human tissues ( Vanderperre et al . , 2013; Kim et al . , 2014 ) . We examined translation of this mRNA in other publicly available ribosome profiling datasets using GWIPS-viz ( Michel et al . , 2014 ) and found that this uORF is translated in all datasets ( Figure 2—figure supplement 3A ) . How ribosomes reach the 12th AUG codon upon arsenite treatment is unclear and merits further investigation . Notably , one of the resistant mRNAs found in our study is the PPP1R15B gene that encodes a phosphatase that dephosphorylates eIF2 , PPP1R15B ( a . k . a . CReP ) ( Novoa et al . , 2001 ) . Sustained translation of PPP1R15B mRNA under conditions of eIF2 inactivation represents a feedback loop for reactivation of eIF2 during recovery from stress . Although they did not pass our stringent criteria for a resistant gene , other candidates which we identified ( based on the gene function and their profiles ) are AZIN1 ( TE Z-score 2 . 76 ) and MIEF1 ( TE Z-score 2 . 88 ) . AZIN1 encodes an inhibitor of ODC1 ( ornithine decarboxylase ) antizymes . Antizymes are proteins that target ODC1 for degradation , and AZIN1 is highly similar to ODC1 but lacks ornithine decarboxylation enzymatic activity . This makes it a competitive inhibitor of antizymes ( Murakami et al . , 1988 ) . It has been shown that an uORF initiated with a non-cognate AUU codon mediates sensitivity of AZIN1 mRNA translation to polyamine levels ( Ivanov et al . , 2008 ) . MIEF1 , mitochondrial elongation factor 1 , is another candidate bicistronic mRNA that we have identified ( the other is SLC35A4 ) . Similar to SLC35A4 , we observed evidence of protein coding evolution within its uORF ( Figure 2—figure supplement 2 ) . Its translation is also supported by multiple ribo-seq datasets available in GWIPS-viz ( Michel et al . , 2014 ) , see Figure 2—figure supplement 3B . Examination of the sequence encoded by its uORF revealed that it contains a conserved domain that belongs to a PFAM family Complex 1 protein ( LYR family ) , ID PF05347 . To examine the role of the 5′ leaders in modulating the resistance to eIF2 inhibition of resistant mRNAs revealed in this study ( IFRD1 , PPP1R15B , UCP2 , PTP4A1 , and SLC35A4 ) , we designed reporter constructs and prepared capped and polyadenylated mRNAs with sequences of 5′ leaders upstream of a Firefly luciferase ( Fluc ) coding region . We used the 5′ leader of ATF4 mRNA and the HCV IRES as positive controls and , as a negative control , we used a non-specific 63 nt leader from the vector pGL3 . mRNAs along with a control mRNA encoding Renilla luciferase ( Rluc ) were transfected into HEK293T cells and simultaneously treated with 40 μM arsenite or vehicle ( Andreev et al . , 2009 ) . Under normal conditions the translation of mRNAs bearing the 5′ leaders of IFRD1 , PPP1R15B , UCP2 , and PTP4A1 was about sevenfold lower than that of the control mRNA with the simple non-specific leader ( pGL3 ) , whereas SLC35A4 was even lower ( Figure 4A and Figure 4—figure supplement 1A ) . Arsenite treatment resulted in significant inhibition of pGL3 and control Rluc translation , while translation of other mRNAs did not change considerably and even slightly increased for SLC35A4 and HCV IRES . Similar results were observed in the Huh7 hepatocarcinoma cell line ( Figure 4—figure supplement 1C ) . To address the effect of arsenite treatment on ongoing translation , which may be more relevant than conditions applied for ribosome profiling , reporter mRNAs were transfected and 1 hr later cells were treated with either non-specific translational inhibitor cycloheximide or arsenite . As expected , both inhibitors efficiently blocked translation of control Rluc mRNA but only cycloheximide was able to arrest translation driven by leaders of resistant mRNAs . Surprisingly , during arsenite treatment the reporter mRNA with the SLC35A4 5′ leader was able to produce 15 times more luciferase than after treatment with cycloheximide ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 03971 . 014Figure 4 . Upstream open reading frame ( uORF ) involvement in modulation of IFRD1 and PPP1R15B mRNAs stress resistance . ( A ) Firefly luciferase ( Fluc ) activity produced by expression of mRNA containing different 5′ leaders 2 hr after arsenite treatment ( red bars ) and in untreated cells ( blue bars ) . Relative units correspond to Fluc activity normalized to the median Renilla luciferase ( Rluc ) activity derived from a co-transfected Rluc mRNA . The green text represents fold change calculated from the same data . The ORF organization of examined mRNAs is outlined on the left . Bars represent standard deviations . ( B ) Time series analysis of Fluc expression in cells treated at the time of transfection with sodium arsenite to a concentration of 40 µM ( dotted lines ) or vehicle ( solid lines ) . Fluc activity of vehicle at 1 hr was taken as 100% for each mRNA and experimental condition . ( C ) Effect of start codon identity in IFRD1 and PPP1R15B 5′ leaders on Fluc activity . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01410 . 7554/eLife . 03971 . 015Figure 4—figure supplement 1 . ( A ) Resistance of additional reporters with 5′ leaders of ATF5 , UCP2 , PTP4A1 , and SLC35A4 to arsenite treatment . ( B and C ) Resistance of different 5′ leaders to arsenite treatment in HEK293T cells ( B ) and to arsenite or dithiothreitol ( DTT ) treatment in Huh7 cells ( C ) . Western blot on panel B demonstrates ubiquitous phosphorylation of eukaryotic initiation factor 2 ( eIF2 ) upon DTT treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01510 . 7554/eLife . 03971 . 016Figure 4—figure supplement 2 . Effect of arsenite treatment on ongoing reporter translation . Reporter firefly luciferase ( Fluc ) mRNAs along with control Renilla luciferase ( Rluc ) mRNA were co-transfected into HEK293T and 1 hr later cells were treated either with vehicle or with 100 µg/ml cycloheximide or 40 µM arsenite . Two hours after treatment the cells were harvested and luciferase activities were measured . Normalized luciferase values for each sample treated with cycloheximide were set as 100% . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01610 . 7554/eLife . 03971 . 017Figure 4—figure supplement 3 . ( A ) Effect mutations that improve initiation Kozak context for uAUG in the IFRD1 leader during arsenite treatment . ( B ) Effect of Torin-1 treatment on translation of different reporter mRNAs in HEK293T . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 01710 . 7554/eLife . 03971 . 018Figure 4—figure supplement 4 . Top: Effect of GADD34-Flag ( PPP1R15A ) overexpression on activity of firefly luciferase under control of the IFRD1 mRNA leader ( Fluc , green bars ) and on mRNA encoding Renilla luciferase ( Rluc , light pink bars ) . Bottom: Western blots showing the presence of the GADD34-Flag protein product and the phosphorylation level of eukaryotic initiation factor 2 ( eIF2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 018 For several other reporter mRNAs with different 5′ leaders which possess low TEs under normal conditions , translation was significantly downregulated upon arsenite treatment ( data not shown ) . This rules out the possibility that low TE of reporter mRNA results in resistance to stress . We also measured the kinetics of protein synthesis to rule out the possibility that our observations can be explained by mRNA silencing or destabilization ( Andreev et al . , 2013 ) . For this purpose cells were treated with arsenite ( or a vehicle ) immediately after transfection and luciferase activity was measured over time . Both IFRD1 and PPP1R15B reporters showed that luciferase activity increases over time with no indication of a plateau that would be expected upon mRNA destabilization ( Figure 4B ) . Treatment with a potent inhibitor of mTOR kinase , torin-1 ( Thoreen et al . , 2009 ) , led to inhibited translation of these reporters to the same degree as the control pGL3 mRNA ( Figure 4—figure supplement 3 ) . Thus , IFRD1 or PPP1R15B leaders do not provide translational resistance to the stress response that involves sequestration of the cap-binding protein eIF4E . It was conceivable that translational resistance was due to side effects of arsenite treatment rather than its direct effect of eIF2 inactivation . To directly address the impact of eIF2 phosphorylation on reporter mRNA translation during arsenite stress , we carried out an experiment where cells were pre-transfected with a plasmid encoding the full length human PPP1R15A ( a . k . a . GADD34 ) phosphatase subunit which is able to reverse eIF2 phosphorylation ( Brush et al . , 2003 ) . Arsenite-induced eIF2 phosphorylation , as expected , was almost completely alleviated in the presence of GADD34 ( residual phosphorylation probably reflects less than 100% efficient plasmid transfection ) , see Figure 4—figure supplement 4 . As a result , the downregulation of control Rluc mRNA was only twofold in comparison with a more than sixfold reduction in cells not transfected with the GADD34 plasmid . Translation of the IFRD1 reporter was not affected under either condition . We therefore concluded that translational inhibition caused by arsenite treatment is predominantly due to the phosphorylation of eIF2 . Also , we treated cells with 2 . 5 mM dithiothreitol ( DTT ) , triggers the Unfolded Protein Response and results in eIF2 phosphorylation ( Prostko et al . , 1993 ) . We found that the leaders of IFRD1 and PPP1R15B provide translational resistance under these conditions as well ( Figure 4—figure supplement 1B ) . While both IFRD1 and PPP1R15B mRNAs possess uORFs with high TE ( Figure 2 ) , the architecture of their uORFs is markedly different ( schematic organization of the 5′ leaders is depicted in Figure 4 ) . IFRD1 mRNA contains a single highly conserved 53 codon uORF that starts 19 nt from the 5′ end and ends 43 nt upstream of the main ORF . PPP1R15B mRNA contains two in-frame uORFs separated by 21 nt . The first uORF is eight codons long and is 127 nt from the 5′ end . The second 52 codon long uORF is 75 nt upstream of the CDS . Substitution of the IFRD1 uORF AUG with the AUA codon increased the reporter expression eightfold but made the translation susceptible to eIF2 inhibition . For PPP1R15B , substitution of the first uORF AUG with AUA slightly reduced the reporter activity under normal conditions but , surprisingly , further increased the reporter resistance to the stress . A similar substitution of the second uORF start codon significantly reduced the resistance to stress , as in the case with IFRD1 ( Figure 4C ) . We conclude that neither of these genes is regulated by delayed reinitiation ( as in ATF4 and GCN4 ) , since in both cases a single uORF is sufficient for eIF2-mediated translational control . A single uAUG in IFRD1 mRNA is in a suboptimal initiation context ( A in −3 position but U in +4 ) . To explore how the context may affect the resistance we introduced +4U/G mutation that improves the context . We found a slight inhibition of the IFDR1 main ORF translation under normal conditions ( presumably due to an increased inhibitory effect of the uORF translation ) . However , this mutation did not alter the sensitivity of the main ORF translation to arsenite stress ( Figure 4—figure supplement 3A ) . We observed earlier that most stress resistant mRNAs possess efficiently translated uORFs . We hypothesized that some features of the 5′ leaders upstream of uORFs may be important for resistance . To address this issue we created two additional reporters based on control pGL3 and IFRD1 , where we added a 5′ terminal stem loop of intermediate stability ( Figure 5 ) . As expected , the addition of this stem loop resulted in a threefold to fourfold decrease in the activity of both reporters under normal conditions . Interestingly , when arsenite stress was induced , the SL-IFRD1 construct did not exhibit resistance while translation of SL-PGL3 was downregulated as much as the pGL3 construct . Therefore we propose that efficient loading of pre-initiator complexes to the uORF is necessary for stress resistance in IFRD1 . Next we addressed the question of whether the specific sequence upstream of the IFRD1 uORF is required for regulation . We substituted it with an artificial single stranded ( CAA ) 6 sequence of the same length . This mutation did not alter stress resistance . Thus , we hypothesize that , for resistant translation , the uORF has to be preceded with a leader allowing a high initiation rate at the uORF . 10 . 7554/eLife . 03971 . 019Figure 5 . Features of the IFRD1 5′ leader required for resistance . ( A ) Firefly luciferase ( Fluc ) activity produced by expression of mRNA containing pGL3 and IRFD1 leaders ( outlined on the left ) with and without an additional stem loop at the 5′ end under different conditions . Blue ( normal conditions ) and red ( stress conditions ) bars correspond to leaders lacking the stem loop while light blue ( normal ) and yellow ( stress ) correspond to leaders with the stem loop . The fold change of Fluc activity in response to stress is indicated above by green arrows . ( B ) Effect of ( CAA ) 6 addition to the 5′ leader of IFRD1 on Fluc activity in response to stress . DOI: http://dx . doi . org/10 . 7554/eLife . 03971 . 019 eIF2 phosphorylation is a key event in the response of cells to various stresses and is also involved in the cell cycle . The global downregulation of protein synthesis triggered by eIF2 inactivation has two main purposes . The first is to conserve cellular resources , and the second is to provide a delay to evaluate the severity of the damage and , depending on its level , reprogram gene expression either towards apoptosis or to a pro-survival repair response . This necessarily requires activation of genes involved in the ISR . mRNAs of genes involved in the ISR ought to be translated under conditions of eIF2 inactivation ( see reviews by Luo et al . , 1995; Baird and Wek , 2012 ) . In order to identify such mRNAs we utilized ribosome profiling , a technique that generates a snapshot of ribosome locations on the entire set of mRNAs ( Ingolia et al . , 2009 ) . We applied this technique to HEK293T cells 30 min after treatment with sodium arsenite , a well-known inducer of eIF2 phosphorylation . This enabled us to study the early stress response at the level of translation . Under our strict criteria of statistical significance ( Z-score TE >4 ) , translation of 10 mRNAs was found to be resistant . Translation of six mRNAs ( encoding ATF4 , PPP1R15A , SLC35A4 , C19orf48 , ATF5 , HOXB2 ) was increased and translation of four ( encoding IFRD1 , PTP4A1 , PCNXL4 , UCP2 ) was reduced only slightly in comparison with a global reduction in translation . Seven mRNAs ( encoding CCNG1 , CCNI , CSDE1 , ODC1 , PABPC1 , PCBP2 , RPL12 ) were found to be particularly sensitive . We confirmed translational resistance of some genes previously reported and identified novel stress resistant genes . Features found to be common amongst the resistant mRNAs are their low levels of CDS translation ( Figure 3C ) and low levels of transcription , except ATF4 ( Figure 1E ) . It is unclear , however , whether these features are common due to the resistance-providing mechanism or due to their function . mRNAs resistant to eIF2 are expected to encode components of cell signalling ( e . g . transcription factors , kinases , phosphatases ) and therefore are not required in large quantities . We also found that , with one exception , all mRNAs significantly resistant to the stress conditions possess translated uORFs in their 5′ leaders . Their number , mutual organization , and depth of phylogenetic conservation vary . IFRD1 has only a single uAUG in its 5′ leader while SLC35A4 has 11 ( Figure 2 ) . For these mRNAs at least one of the uORFs is efficiently translated under normal conditions and is longer than 20 codons . We confirmed that 5′ leaders of some of these mRNAs confer translation resistance to a reporter in synthetic RNA constructs . Owing to the diversity of the uORF organisation in these 5′ leaders , the mechanism of uORF-mediated resistance may vary and should be studied individually . We chose to explore how the sequence properties of IFRD1 and PPP1R15B 5′ leaders affect the resistance of downstream ORFs . For this purpose we carried out site-directed mutagenesis of 5′ leader sequences in the constructs containing a luciferase reporter . We found that translation of only one uORF is sufficient to provide resistance to eIF2 inhibition for IFRD1 and PPP1R15B . Therefore it is likely that the resistance for these transcripts is provided by the mechanism that resembles alleviation of scanning ribosomes obstruction rather than delayed reinitiation ( ATF4-like cases ) . The substitution of IFRD1 leader upstream of an uORF with an artificial single stranded sequence does not affect stress resistance , ruling out the requirement for a specific nucleotide sequence . However it is likely to enable rapid ribosome loading at the uORF since the addition of a 5′ terminal stem loop of intermediate stability abolished stress resistance . This observation may explain why translation of many mammalian mRNAs possessing uORFs is not resistant to eIF2 phosphorylation . Translation of these uORFs might be inhibited to such an extent that they would be unable to provide any resistance . While it is possible that uORFs provide mRNAs with stress resistance in the same manner in all resistant mRNAs detected here , we think it is more likely that uORFs are an essential component of the diverse mechanisms . We mentioned earlier two such mechanisms , delayed initiation and alleviation of scanning ribosomes obstruction . However , even an IRES-mediated resistance would likely require translation of an uORF as it would prevent IRES structure melting by the scanning ribosomes . The genes that we found to be resistant to eIF2 inhibition may participate in the ISR . This is the case with PPP1R15B . Similar to PPP1R15A , it encodes a subunit of the phosphatase that dephosphorylates eIF2 , thus providing feedback preventing complete translation suppression and also enabling recovery from the stress-induced translational arrest ( He et al . , 1997; Jousse et al . , 2003 ) . Expression of PPP1R15A is tightly regulated , its basal level of expression is almost undetectable ( Novoa et al . , 2001 ) , whereas PPP1R15B is constitutively present in cells ( Jousse et al . , 2003 ) . PPP1R15B is a short-lived protein whose half-life is approximately 45 min ( Jousse et al . , 2003 ) . Therefore it needs to be continuously synthesized in order to dephosphorylate eIF2 . It has previously been found to remain present in the cell upon arsenite and tunicamycin treatments ( Jousse et al . , 2003 ) . It is also possible that some of the translationally resistant genes are not directly implicated in the ISR . Their resistance could be related to other eIF2-mediated regulatory mechanisms , for example during cell-cycle progression or development ( Datta et al . , 1999; Harding et al . , 2009 ) . At least one of the newly identified stress resistant mRNAs encoding oncogenic phosphatase PTP4A1 ( a . k . a . PRL-1 ) may be directly implicated in malignant transformation since it downregulates expression of p53 tumour suppressor ( Min et al . , 2009 ) . Translational resistance of some genes to eIF2 phosphorylation may also be a consequence of the ORF organisation of their mRNAs which serves a different purpose . This could be the case for the candidate bicistronic mRNAs identified in this study ( MIEF1and SLC35A4 ) . The ratio between uORF and main ORF translation changes in both upon eIF2 inactivation . Coding for two functionally related proteins in the same mRNA may be advantageous for coordination of their expression . To summarize , our work expands the list of mRNAs which are known to be persistently translated under conditions of eIF2 phosphorylation , although it suggests that the number of such mRNAs is very low . The analysis of ribosome densities on mRNAs resistant to eIF2 phosphorylation accentuates the vital role of uORF translation in providing the resistance . The ribosomal profiling technique was carried out according to Ingolia et al . ( 2012 ) but with important modifications described below . HEK293T cells were grown in DMEM supplemented with alanyl-glutamine and 10% FBS and replated to 150 mm dishes ( two dishes per sample ) . After the cells reached 70–80% confluency , sodium arsenite ( or vehicle ) was added at 40 µM and 30 min later the cells were harvested: dishes were immediately chilled on ice and washed with PBS + cycloheximide ( 100 µg/ml ) . Importantly , cells were not pre-treated with cycloheximide to avoid artificial accumulation of initiation complexes at translation initiation starts ( Gerashchenko and Gladyshev , 2015 ) . Cells were then lysed with buffer containing 20 mM Tris–HCl ( pH 7 . 5 ) , 250 mM NaCl , 1 . 5 mM MgCl2 , 1 mM DTT , 0 . 5% Triton X-100 , 100 µg/ml cycloheximide ( Sigma-Aldrich ) , 20 U/ml TURBO DNAse ( Ambion , Waltham , MA ) . Note that the buffer contains low magnesium ( 1 . 5 mM ) , since high magnesium concentrations stabilize secondary structures in mRNAs which may hamper digestion by RNAse I ( RNAse I itself does not require divalent cations for its activity ) and lowering magnesium concentration in 5 mM to 15 mM range has been shown to improve footprint resolution ( Ingolia et al . , 2012 ) . Cell lysates were incubated on ice for 10 min , centrifuged at 16 , 000×g at +4°C for 10 min , and the supernatant was divided into two parts for ribo-seq and ‘naked’ mRNA-seq library preparation . One part of the lysate , usually 10–20 A260 , was treated with RNAse I ( Ambion ) with 100 U per 3 . 1 A260 of lysate at 23°C for 50 min . The digestion was then stopped with the appropriate amount of SUPERASE inhibitor ( Ambion ) . The treated lysate was loaded on 10–60% ( m/v ) sucrose density gradient containing 20 mM Tris–HCl ( pH 7 . 5 ) , 250 mM NaCl , 15 mM MgCl2 , 1 mM DTT , 100 μg/ml cycloheximide and centrifuged in SW-41 rotor at 35 , 000 rpm for 3 hr . Sucrose density gradients were prepared as described previously ( Stone , 1974 ) . Briefly , 5 . 5 ml of 10% sucrose was slowly layered onto the same volume of 60% sucrose , gradient tubes were then sealed with parafilm , slowly placed horizontally for 4 hr to allow spontaneous gradient formation and then slowly returned to a vertical position . This method of gradient formation is highly reliable and reproducible and does not require any special equipment . Total RNA from the fractions corresponding to 80S peak was then extracted with phenol/chloroform followed by ethanol precipitation . For the mRNA-seq control , the second lysate aliquot was processed with Trizol-LS ( Life Technologies , Waltham , MA ) according to the manufacturer's protocol . mRNA from total RNA was isolated using the Oligotex mRNA kit ( Qiagen , Netherlands ) . Two rounds of polyA ( + ) -mRNA selection instead of one were applied to decrease rRNA contamination to approximately 3% . Purified mRNA was then subjected to alkaline hydrolysis as described by Ingolia et al . ( 2009 ) . Both ribo-seq and mRNA-seq samples were loaded onto a 15% denaturing urea PAGE ( containing 1× TBE , 7 M urea , and AA:bis-AA in the ratio 20:1 ) . Bands corresponding to nucleic acid fragments of 28–34 nt were excised for both ribo-seq and mRNA-seq samples . RNA was extracted by overnight incubation in a shaker using buffer containing 0 . 3 M NaOAc ( pH 5 . 1 ) , 1 mM EDTA , and 0 . 1% SDS followed by precipitation with one volume of isopropanol and 2 µl of GlycoBlue ( Life Technologies ) . The same quantity of both ribo-seq and mRNA-seq fragments ( usually 100 ng ) were mixed with 1:10 , 000 of unspecific RNA oligonucleotide 5′AUGUACACGGAGUCGACCCGCAACGCGA 3′ which serves as a ‘spike-in’ control . The library preparation was carried out as previously described ( Ingolia et al . , 2012 ) with the following modifications . First , the circularization reaction was performed for 2 hr . Second , during PCR library amplification , the temperature ramping speed was set as 2 . 2°C/s to reduce bias associated with GC content ( Aird et al . , 2011 ) . Two independent biological replicates were carried out . Libraries were sequenced either on an Illumina MiSeq genome analyser at the TrinSeq genomic facility ( Dublin ) or on an Illumina HiSeq 2000 system at the Beijing Genomics Institute ( BGI ) . Reporter DNA constructs were prepared on the basis of the pGL3R vector ( Stoneley et al . , 1998 ) . Plasmids containing the 5′ leaders of test mRNAs were cloned between SpeI and NcoI . pGL3R-HCV and pRluc plasmids are described in Andreev et al . ( 2009 ) . The 5′ leader of IFDR1 mRNA ( NM_001550 . 3 ) was shortened by 38 nt at the 5′ end to correspond to the location of the likely predominant transcription start based on the analysis of available ESTs ( Expression Sequence Tags ) for this region . The 5′ leader of PPP1R15B ( NM_032833 . 3 ) was extended at the 5′ end by 1 nt , which is present in the majority of available EST sequences . For the same reasons , the 5′ leader of UCP2 mRNA ( NM_003355 . 2 ) was shortened by 16 nt and the 5′ leader of PTP4A1 mRNA ( NM_003463 . 4 ) by 547 nt; both 5′ leaders were cloned into pGL3R between Spe1 and Nco1 . The 5′ leader of SLC35A4 mRNA ( NM_080670 . 2 ) was extended by 11 nt . To prepare pGL3-ATF4 , the human ATF4 5′ leader was obtained by RT-PCR with primers GGGTAATACGACTCACTATAGGGTTTCTACTTTGCCCGCCCACAG and GGCGCCATGGTTGCGGTGCTTTGCTGGAATCG . The resulting product was digested with NcoI ( underlined ) and inserted into pGL3 vector at SmaI-NcoI sites . pGL3-ATF5 was prepared with the leader of ATF5 mRNA ( NM_001193646 . 1 ) shortened by 62 nt . The HCV-Fluc plasmid contained the T7 promoter , the entire viral 5′ leader , and the first 33 codons of viral ORF fused to Fluc and without its initiator codon and entire viral 3′ UTR . Full length human ppp1R15A ( GADD34 ) sequence fused with N-terminal FLAG tag was cloned in pcDNA 3 . 1 construct between HindIII and XbaI to prepare pcDNA GADD34 construct . mRNA preparation was carried out as described by Dmitriev et al . ( 2007 ) . Briefly , PCR products were obtained with a forward primer containing the T7 promoter ( either the universal primer which anneals to the vector sequence immediately upstream of insertions , CGCCGTAATACGACTCACTATAGGGAGCTTATCGATACCGTCG or the T7 promoter-containing gene specific primer ) and reverse primer containing an oligo ( dT ) stretch of 50 nt T50AACTTGTTTATTGCAGCTTATAATGG . To introduce stem loop structure , PCR products were obtained with forward primer containing the T7 promoter: CGCCGTAATACGACTCACTATAGGGAGTGGACTTCGGTCCACTCCCAGCTTATCGATACCGTCG . To introduce the CAA6 sequence upstream of the IFRD1 uORF , the following primer was used: CGCCGtaatacgactcactataGGGCAACAACAACAACAACAACAACATGTATCGTTTTCGATCACAGCTC . The PCR products were then purified and used as templates for T7 RNA polymerase using in vitro RNA transcription by T7 RiboMAX Large Scale RNA Production kit ( Promega , Fitchburg Center , WI ) . For preparation of m7G-capped transcripts the 3′-O-Me-m7GpppG ( ARCA cap analogue , New England Biolabs , Ispwich , MA ) was added to the transcription mix without GTP for 5 min to prime transcripts with cap followed by the addition of GTP ( at a ratio of ARCA:GTP 10:1 ) . The resulting RNAs were purified by LiCl precipitation and examined for integrity by PAGE . Experiments with mRNA transfection were performed as described in Andreev et al . ( 2012 ) . Briefly , the mixture of 0 . 2 µg m7G-capped Fluc mRNA and 0 . 01 µg m7G-capped Rluc mRNAs per 1 well of 24-well plate was transfected to the cells at 70–80% confluency either with Lipofectamin 2000 ( Invitrogen , Waltham , MA ) or Unifectin 56 ( Rusbiolink , Russian Federation ) . Simultaneously with transfection , cells were treated with either 40 µM sodium arsenite , 2 . 5 mM DTT or 250 nM Torin-1 ( Torics Biosciences , Minneapolis , MN ) . Two hours later ( or at the specified time interval ) , cells were harvested and luciferase activities were analysed with the Dual Luciferase Assay kit ( Promega ) . For experiments with GADD34 overexpression , cells were transfected with either pcDNA-GADD34 or control pcDNA3 . 1 one day prior to mRNA transfection . Plasmids were transfected with Fugene 6 ( Promega ) according to the manufacturer's instructions . For western blotting , cells were rapidly lysed with buffer containing 1% SDS and 20 mM Tris–HCl pH 6 . 8 followed by brief sonication of the lysates . This was done to prevent post-translational modifications of proteins of interest during the lysis . Antibodies used in the study were: rabbit anti-EIF4EBP1 ( AB3251; Chemicon International , Germany ) , rabbit anti-GAPDH ( PTG10494-1-AP; Proteintech , Chicago , IL ) , rabbit anti-ATF4 ( 10835-1-AP; Proteintech ) , rabbit anti-phospho-p70 S6 kinase ( Thr389 ) ( 9205S; Cell Signalling , Danvers , MA ) , rabbit anti-phospho-S6 ribosomal protein ( Ser235/236 ) ( 2211S; Cell Signalling ) , rabbit anti-S6 ribosomal protein ( 2217S , Cell Signalling ) , rabbit anti-phospho eIF2 ( S51 ) ( SA-405; Enzo , New York , NY ) , and anti-FLAG-M2 ( SigmaAldrich , St . Louis , MO ) . To remove non-specific binding , phospho-eIF2 antibodies ( 1:2500 ) were incubated along with 10% fetal bovine serum in TBS-T . Cutadapt ( Martin , 2011 ) was used to remove the 3′ adapter of the reads ( TGGAATTCTCGGGTGCCAAGG for the first replicate and CTGTAGGCACCATCAATAGATCGGAAGAGCACACGTCTGAACTCCAGTCAC for the second ) . Reads that did not map to either the ‘spike-in’ ( ATGTACACGGAGTCGACCCGCAACGCGA ) or rRNA sequences were aligned to the RefSeq catalogue ( Pruitt et al . , 2014 ) downloaded from the NCBI website on 15 August 2013 . The alignment was carried out with Bowtie version 1 . 0 . 0 ( Langmead et al . , 2009 ) , with parameters -a -m 100 –norc ( all read mappings to the positive strand were taken with exception of those with more than 100 mappings ) . Reads that mapped to transcripts of more than one gene or multiple times to a transcript were discarded . In order to maximize the genuine ribosome footprints aligning to the transcriptome , ribo-seq reads with a length typical for monosomes ( 29–35 inclusive ) were used for further analysis . In the case of multiple transcript variants , among the transcripts annotated as protein coding , the one with the highest ribo-seq read density in control conditions was brought forward for differential expression analysis . The raw read count data were rescaled to normalize for the differences in the total number of reads mapped with a rescale factor F . For the first replicate the rescaled factor F for each sample was calculated as the difference by which the total number of mapped reads exceeds the lowest total number of mapped reads out of two conditions , that is:Fn=∑ixniminixni , where xni is the number of mapped reads from sample n in the condition i . This normalization was carried out independently for ribo-seq and mRNA-seq . The second replicate was sequenced on Illumina Miseq and Hiseq 2000 instruments and obtained sequence reads were aggregated . The number of ‘spike-in’ reads was used to rescale the read counts with a similar approach as for replicate 1 . The raw read count of each sample was divided by the rescaling factor F calculated as above with the only difference that xni represents the number of ‘spike-in’ reads from sample n in the condition i . This rescaling was also implemented for the ribosomal profiles of individual transcripts shown in Figure 2 . The normalized read counts of ribo-seq reads aligning to the coding regions ( as determined by inferred locations of the A-site codons ) and of mRNA-seq reads aligning to the entire transcript were used for the differential expression analysis . For ribo-seq reads the A-site codon of the elongating ribosome was inferred to be the 17th or 18th nt of the read from its 5′ end depending on the read length . Transcripts were binned based on the number of mapped reads ( expression/coverage level ) in one of the conditions where this value is the minimal . For the analysis of differential translation efficiency the minimum value ( referred to as the minimum expression level ) was taken from four conditions while , for the analysis of differential RNA level , only RNA-seq reads obtained under control and stress conditions were used . With the minimum expression level threshold of two reads , transcripts were sorted in ascending order and arranged in bins of size 300 . Each bin had transcripts with a similar number of mapped reads and was analysed independently . The mean and standard deviation of change in expression of the transcripts within each bin was used to determine a Z-score for each transcript . For the remaining transcripts of insufficient number to be binned ( <300 ) , the mean and standard deviation was obtained from the previous bin . The Z-score determined for each transcript enabled comparison between bins . The analysis of translation of mRNA leaders was carried out for the transcripts with at least two normalized read counts in each of all four experiments/conditions . An uORF was defined as a sequence of sense codons uninterrupted with a stop codon and beginning with an AUG codon located upstream of the annotated CDS . In the case of uORFs overlapping CDS , the 5′ end of CDS was considered as the end of the uORF in order to avoid ambiguity in assigning ribo-seq reads to one of the two overlapping ORFs . Nested uORFs ( those contained within uORFs in the same frame ) were excluded for the same reason . The TE of an uORF was estimated as the average density of ribo-seq reads in the uORF divided by the average density of the mRNA-seq reads for the corresponding mRNA . An uORF was considered to be translated if more than five ribo-seq reads aligned to it . For transcripts with more than one translated uORF , the properties of the uORF with the highest number of aligned ribo-seq reads were used . For the purpose of the analysis represented in Figure 3C , the centre of ribosome density was defined as the minimal mRNA coordinate for which the number of ribo-seq reads aligning 5′ of the corresponding location is the same or greater than the number of ribo-seq reads aligning 3′ of the corresponding location . This value was determined for genes under arsenite and control conditions . The difference in ribosome footprint density was divided by the CDS length to prevent skewing of results in favour of transcripts with longer coding regions . The list of human cellular IRES was obtained from IRESite ( Mokrejs et al . , 2010 ) on 2 August 2014 . We identified the most probable translation initiation sites by manually examining the ribo-seq profiles of eight translationally resistant genes ( PPP1R15A , IFRD1 , SLC35A4 , C19ORF48 , PTP4A1 , PCNXL4 , UCP2 , PPP1R15B ) . ATF4 and ATF5 were not included as these appeared to be regulated by an alternative method . uORF initiation sites with an AUG or CUG were selected based on their ability to fit with the observed profile upon manual examination . AUG codons were preferred , but the surrounding consensus sequence was not considered . A sequence logo of the initiation sites ( −4 to +3 ) was produced with WebLogo ( Crooks et al . , 2004 ) . For Figure 3—figure supplement 1B , the sequences of all coding transcripts were included to determine the frequency of the initiation sites for both annotated start sites and for AUG sites in the leaders . The relative frequency was obtained by dividing the number of occurrences of a particular sequence by the total number . For example , the sequence GGCCATGG , the most common Kozak sequence in CDS sites occurring 640 times in 35 , 851 transcripts , has a relative frequency of ( 640/35 , 841 ) × 100 = 1 . 78% . The free energy of leaders was estimated with RNAfold ( Lorenz et al . , 2011 ) . The first 240 nt was used ( transcripts with shorter leaders were excluded ) as free energy of RNA is related to its length . We chose 240 nt as this was the length of the shortest leader of resistant mRNAs with a translated uORF . Sequences of ribosome profiling libraries have been deposited into the NCBI Gene Expression Omnibus portal under the accession number GSE55195 .	Proteins carry out essential tasks for living cells and genes contain the instructions to make proteins within their DNA . These instructions are copied to make a molecule of mRNA , and a molecular machine known as a ribosome then reads and translates the mRNA to build the protein . The first step in the translation process is called ‘initiation’ and requires a protein called eIF2 to work together with the ribosome . This step involves identifying an instruction called the start codon that marks the beginning of the mRNA's coding sequence . The section of an mRNA molecule before the start codon is not normally translated by the ribosome and is hence called the 5′ untranslated region . Building proteins requires energy and resources , and so it is carefully regulated . If a cell is stressed , such as by being exposed to harmful chemicals , it makes fewer proteins in order to conserve its resources . This down-regulation of protein production is achieved in part by the cell chemically modifying its eIF2 proteins to make them less able to initiate translation . However , stressed cells still continue to make more of certain proteins that help them to combat stress . The mRNA molecules for some of these proteins contain at least one other start codon in the 5′ untranslated region . The sequence that would be translated from such a start codon is known as an upstream open reading frame ( or uORF for short ) —and this feature is thought to help certain proteins to still be expressed despite low levels of active eIF2 . Andreev , O'Connor et al . have now analysed which mRNAs are translated in human cells that have been treated with a chemical that induces stress and makes the eIF2 protein less able to initiate translation . To do so , a technique called ribosome profiling was used to identify all of the mRNA molecules bound to ribosomes shortly after treatment with this chemical . Overall translation of most mRNAs in stressed cells was reduced to a quarter of the normal level . However , Andreev , O'Connor et al . observed that the translation of a few mRNAs continued almost as normal , or even increased , after the chemical treatment . Notably , most of these mRNAs encoded regulatory proteins , which are not required in large amounts . With one exception , all of these resistant mRNAs contained uORFs . In unstressed cells , these uORFs were efficiently translated , while the same mRNA's coding sequences were translated less efficiently . Andreev , O'Connor et al . suggest that these two features could be used to identify mRNAs that are still translated into working proteins when cells are stressed . Further work is now needed to explore the mechanisms by which translation of these uORFs allows mRNAs to resist the stress .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	Translation of 5′ leaders is pervasive in genes resistant to eIF2 repression
Developed areas are thought to have low species diversity , low animal abundance , few native predators , and thus low resilience and ecological function . Working with citizen scientist volunteers to survey mammals at 1427 sites across two development gradients ( wild-rural-exurban-suburban-urban ) and four plot types ( large forests , small forest fragments , open areas and residential yards ) in the eastern US , we show that developed areas actually had significantly higher or statistically similar mammalian occupancy , relative abundance , richness and diversity compared to wild areas . However , although some animals can thrive in suburbia , conservation of wild areas and preservation of green space within cities are needed to protect sensitive species and to give all species the chance to adapt and persist in the Anthropocene . Global loss of biodiversity leads to disruption of ecosystem services around the world , ultimately threatening human well-being ( Cardinale et al . , 2012 ) . Vertebrate species loss is typically considered to be worst in the most developed landscapes , where urbanization serves as an intense and long-term disturbance that permanently alters habitat and truncates food webs ( Lombardi et al . , 2017; McKinney , 2006 ) . However , for some species , urbanization can offer abundant nutrient-rich food that is less ephemeral compared to wild areas ( Bateman and Fleming , 2012; Wang et al . , 2017 ) . Whether this food is enough to counteract the negative effects of disturbance ( i . e . higher road mortality , fragmentation ) depends on a species’ ability to adapt to the stressors of urban living ( Witte et al . , 1982 ) . Mammal species , especially those with large home ranges , are arguably most at risk from development , leading some to suggest that developed areas have a dearth of predators , and that prey species could benefit by using humans as a shield ( Crooks , 2002; Ordeñana et al . , 2010 ) . Previous studies have shown cities to be depauperate of bird life , supporting the traditional view that development and biodiversity cannot coexist ( Keast , 1995; Strohbach et al . , 2014 ) . However , recent evidence has shown that some mammal species previously thought mal-adapted to urban landscapes ( i . e . mountain lion [Puma concolor] , fisher [Martes pennanti] ) are thriving in them ( Bateman and Fleming , 2012; LaPoint et al . , 2013 ) , suggesting an evolutionary trend that could be important for conservation in the Anthropocene . Existing research on mammal communities across urbanization gradients has focused on single cities , yielding conflicting results , perhaps due to variation in city structure and characteristics ( Lombardi et al . , 2017; Saito and Koike , 2013 ) . Given the rapid expansion of urban areas worldwide , and the recent case studies of urban adaptations by wildlife ( LaPoint et al . , 2013; Riley et al . , 2014; Wang et al . , 2017 ) , more large-scale studies are needed to evaluate the response of wildlife communities to urban development if we are to understand urban ecology , conservation , and evolution in the Anthropocene . Here , we present the results of a large-scale mammal survey of two urban-wild gradients . Our objectives were to determine how diversity , richness , detection rate , and occupancy of the mammal community change as a function of human disturbance . We hypothesized that the availability of supplemental food at higher levels of development would positively affect mammalian populations and outweigh the negative effects of disturbance , except for the most sensitive species . Specifically , we predicted that mammalian relative abundance would increase with developmental level but that species richness and diversity would decrease . Furthermore , we predicted that occupancy of the most sensitive species ( i . e . large and medium carnivores ) would be highest in wild areas both in our study area and around the world . Washington , District of Columbia , USA ( hereafter DC ) is a city of approximately 177 km2 with an estimated human population size of 681 , 000 , thus a density of 3847 people/km2 . Our study spanned a 56 , 023 . 7 km2 area around the city with a mean of 4 . 4 houses/km2 and matrix of agriculture ( ~21 . 3% ) and forest ( ~54 . 1% ) . Raleigh , North Carolina , USA ( hereafter Raleigh ) is approximately 375 km2 with an estimated human population size of 459 , 000 , thus a density of 1278 people/km2 . Our study spanned a 66 , 640 km2 area around the city with a mean of 17 . 7 houses/km2 and matrix of agriculture ( ~24 . 3% ) and forest ( ~52 . 3% ) . From 2012–2016 , 557 trained volunteers deployed 1427 unbaited camera traps across an urban-wild gradient around Raleigh and DC . Each individual camera was considered a 'camera site' and volunteers ran cameras at an average of two sites each . Following Hammer et al . ( 2004 ) , we used the Silvis housing density dataset with 1km grid cells to define five development levels of the gradient for sampling stratification ( excluding open water ) : urban ( >1000 houses/km2 ) , suburban ( 147 . 048–1000 houses/km2 ) , exurban ( 12 . 64–147 . 047 houses/km2 ) , rural ( 0 . 51–12 . 63 houses/km2 ) and wild ( <0 . 5 houses/km2 ) . Within those gradient levels , camera placement was also stratified between residential yards , open areas ( >0 . 001 km2 absent of trees ) , small forest fragments ( ≤1 km2 ) and large forest fragments ( >1 km2 ) Supplementary file 1 . Forest fragment size was verified using the 2006 US National Landcover Dataset ( NLCD ) and Landscape Fragmentation Tool v2 . 0 ( Vogt et al . , 2007 ) in ArcMap ( Version 10 . 1 , ESRI , Redlands , CA , USA ) which defines forest fragments by size . Yards were not available for sampling in the urban or wild levels of the gradient . Urban areas were not sampled in Raleigh and open areas were not sampled in DC . All adjacent cameras were spaced at least 200 m apart . Camera placement was randomized as much as possible using ArcMap ( Version 10 . 1 ) to randomly generate points within polygons while following certain rules . For example , we selected sites within forests that volunteers were permitted to access and were within a reasonable hiking distance ( i . e . < 11 km hike round trip ) with terrain that was not too steep to traverse safely ( i . e . <45 degree slope ) . Within yards , cameras were placed as randomly as possible while avoiding the highest human traffic areas ( i . e . walkways , doors , gates and driveways ) . No explicit power analysis was used to predetermine sample size . Our sample size goal was 20 spatial replicates ( equating to ~420 trap nights ) , which has been found to maximize precision for estimating detection rate ( Kays et al . , 2010; Rowcliffe et al . , 2008 ) . Camera sites are biological replicates , parallel measurements capturing random biological variation . This study did not include technical replicates . Volunteers used Reconyx ( RC55 , PC800 , and PC900 , Reconyx , Inc . Holmen , WI ) and Bushnell ( Trophy Cam HD , Bushnell Outdoor Products , Overland Park , KS ) camera traps attached to trees at 40 cm above the ground . Cameras were deployed for three weeks and then moved to a new location without returning , with sampling taking place continuously throughout the year . Cameras recorded multiple photographs per trigger , at a rate of 1 frame/s , re-triggering immediately if the animal was still in view . We grouped consecutive photos into on sequence if they were <60 s apart , and used these sequences as independent records , counting animals in the sequence , not individual photos ( Parsons et al . , 2016 ) . We then collapsed these independent records into daily detection/non-detection for occupancy modeling . Initial species identifications were made by volunteers using customized software ( available freely from eMammal . org , source code proprietary ) and all were subsequently reviewed for accuracy before being archived at the Smithsonian Digital Repository ( McShea et al . , 2016 ) . We used package iNEXT ( Hsieh et al . , 2016 ) in R ( Version 3 . 1 . 0; R Development Core Team . , 2008 ) via R Studio ( RStudio Team , 2015 ) to calculate Hill numbers ( i . e . the effective number of species , incorporating relative abundance and richness ) of species richness and Shannon diversity ( Chao et al . , 2014 ) between gradient levels ( urban-suburban-exurban-rural-wild ) and plot types ( yard , open , small forest , large forest ) . iNEXT calculates the Shannon diversity as Hill number q = 1 , equal to the exponential of Shannon's entropy index , thus the natural log of those results was used for display purposes . We used detection/non-detection data to compute diversity estimates and the associated 95% confidence intervals via rarefaction , plotting the diversity estimates while accounting for sample size . We fit a curve to diversity estimates between gradient levels using a generalized additive model with a polynomial term . We modeled variation in occupancy ( ψ ) and detection rate using 13 covariates ( Supplementary file 2 ) representing development level , the amount of core forest , small scale forest cover , prey relative abundance and whether hunting was allowed . We added year as a covariate to account for population changes over time . We used the Landscape Fragmentation Tool v2 . 0 ( Vogt et al . , 2007 ) and the NLCD ( 2006 ) land use dataset in ArcMap ( Version 10 . 1 ) to create a landcover layer representing the percent of large core forest ( forest patches larger than 1 km2 ) in a 5 km radius around camera locations which we considered best approximated the home range size of our target species ( Bekoff , 1977; Fritzell and Haroldson , 1982; Lariviere and Pasitschniak-Arts , 1996; Lariviere and Walton , 1997 ) . Forest patches did not necessarily fall entirely within the buffer . We considered road density as an additional covariate at the 5 km scale but initial evaluations showed it to be highly correlated with housing density ( 87 . 1% ) so we chose to eliminate it from the analysis . We used a 100 m radius for small-scale forest cover to best represent small forest patches within suburban neighborhoods ( e . g . small vacant lots with trees , greenways ) . We represented deer and rodent+lagomorph relative abundance using site-specific detection rate ( the number of detections divided by the total number of camera-nights ) . We included an indicator ( 0/1 , no hunting/hunting ) to categorize whether a site allowed hunting or not . We modeled detection probability ( p ) using five covariates ( Supplementary file 2 ) . Because both ambient temperature and undergrowth can affect the camera’s ability to detect an animal , we included daily covariates for temperature and NDVI ( Moderate Resolution Imaging Land Terra Vegetation Indices 1 km monthly , an average value over the month ( s ) the camera ran ) obtained from Env-DATA ( Dodge et al . , 2013 ) . To complement NDVI , we also considered site-specific detection distance , a measure of how far away the camera was able to detect a human , which is influenced by both understory and site topography . We included an indicator ( 0/1 , not yard/yard ) to categorize whether a site was a residential yard or not . In Raleigh , two different camera models were used ( both Reconyx and Bushnell ) so we added a 0/1 ( Bushnell/Reconyx ) covariate to account for potential difference in detection probability between the two brands . We diagnosed univariate correlations between covariates using a Pearson correlation matrix , and used a restrictive prior for beta coefficients where correlation was >0 . 60 ( i . e . logistic ( 0 , 1 ) ; a prior with reduced variance to induce shrinkage , similar to ridge regression; Hooten and Hobbs , 2015 ) . All covariates were mean-centered . We used a Poisson count model ( e . g . Kays et al . , 2017 ) to assess differences in total mammal detection rate ( i . e . the intensity with which a site was used , count/day ) between the five gradient levels ( urban , suburban , exurban , rural , wild ) and four plot types ( large forest , small forest , open , yard ) . We fit a curve to total detection rate estimates between gradient levels using a generalized additive model . No other covariates were used in this model . We then ran separate count models for four predator species ( coyote ( Canis latrans ) , gray fox ( Urocyon cinereoargenteus ) , red fox ( Vulpes vulpes ) and bobcat [Lynx rufus] ) to evaluate covariates of detection rate , running one fully-parameterized model ( Supplementary file 2 ) to evaluate which explained the most variation in detection rate . We assessed model fit with posterior predictive checks ( PPC ) ( Gelman et al . , 2014; Kery and Schaub , 2012 ) by calculating the sum of squared Pearson residuals from observed data ( T ( y ) ) and from data simulated assuming the fully parameterized model was the data-generating model ( T ( ysim ) ) . We calculated a Bayesian p-value as pB = Pr ( T ( ysim ) >T ( y ) ) from posterior simulations and assumed adequate fit if 0 . 1 < pB < 0 . 9 ( Supplementary file 3 ) . We fit the detection rate model in OpenBUGS v3 . 2 . 3 ( Lunn et al . , 2009 ) via R2OpenBUGS v3 . 2 ( Sturtz et al . , 2005 ) in R ( Version 3 . 1 . 0 ) via R Studio . We based inference on posterior samples generated from three Markov chains , using trace plots to determine an adequate burn-in phase . All models achieved adequate convergence ( R^≤1 . 1 ) ( Gelman et al . , 2014 ) by running for 50 , 000 iterations following a burn-in phase of 1000 iterations , thinning every 10 iterations . We based significance on whether parameter 95% credible intervals overlapped zero . We used the multispecies occupancy model of Rota et al . ( 2016 ) to estimate the probability of occupancy of four predator species: bobcat , coyote , red fox and gray fox . Although we are using the term occupancy , because data were collected from camera traps estimates are more analogous to ‘use’ than true occupancy ( Burton et al . , 2015 ) . This model is distinct from the classic multispecies community models ( Dorazio and Royle , 2005;Dorazio et al . , 2006; Gelfand et al . , 2005 ) and is rather a generalization of the single-season occupancy model ( MacKenzie et al . , 2002 ) to accommodate two or more interacting species . It contains single-species ( first order ) occupancy models for each interacting species alone as well as pairwise ( second order ) models for the co-occurrence of each pair of species ( Rota et al . , 2016 ) . For each species and pairwise interaction , the model estimates detection probability ( p ) , defined as the probability of detecting an occurring species at a camera site , and occupancy ( ψ ) , defined as the probability that a given camera site is occupied , for each species . The latent occupancy state of each species at a site is modeled as a multivariate Bernoulli random variable such that ( assuming 2 interacting species ) :Z~MVB ( ψ11 , ψ10 , ψ01 , ψ00 ) Where ψ11is the probability that both species occupy a site , ψ10 is the probability that only species 1 occupies a site , ψ01 is the probability that only species 2 occupies a site and ψ00 is the probability that neither species occupies a site . We assumed all species occurred independently and considered the same set of five covariates for the detection probability models and 13 covariates in the occupancy model of each species ( Supplementary file 2 ) . We considered interactions ( i . e . city*covariate ) between each occupancy covariate and city ( 0/1 , DC/Raleigh ) . We estimated occupancy for each species across levels of the development gradient ( urban , suburban , exurban , rural , wild ) and plot types ( yard , open , small forest , large forest ) within each city separately by including development level and plot type as categorical covariates in our model . We fit models in STAN ( Version 2 . 15 . 1; Stan Development Team , 2015b ) via the RSTAN ( Version 2 . 15 . 1; Stan Development Team , 2015a ) interface in R ( Version 3 . 4 . 0 ) via R Studio ( Version 1 . 0 . 143 ) . We based inference on posterior samples generated from two Markov chains , using trace plots to determine an adequate burn-in phase and subsequently running chains until they reached adequate convergence ( R^>1 . 1 ) ( Gelman et al . , 2014 ) . All models achieved adequate convergence by running for 3000 iterations following a burn-in phase of 1000 iterations . We based predictor significance on whether beta coefficient 95% credible intervals overlapped zero . We assessed model fit with posterior predictive checks ( PPC ) ( Gelman et al . , 2014; Kery and Schaub , 2012 ) by calculating the sum of squared Pearson residuals from observed data ( T ( y ) ) and from data simulated assuming the fully parameterized model was the data-generating model ( T ( ysim ) ) . We calculated a Bayesian p-value as pB = Pr ( T ( ysim ) >T ( y ) ) from posterior simulations and assumed adequate fit if 0 . 1 < pB < 0 . 9 . To our knowledge , the squared Pearson’s residual has not been derived in the context of occupancy models , so we present our derivation of this test statistic in Supplementary file 4 . We added a random effect on detection/non-detection for the coyote portion of the model since initial assessments of fit for this species were inadequate ( i . e . pB >0 . 9 ) . We assessed differences in occupancy between gradient levels for each species using overlapping 95% confidence intervals . We removed omnivores from the dataset of Rich et al . ( 2017 ) to better compare with carnivore occupancy from our own dataset . Where species occupancy was estimated from multiple studies in the Rich et al . dataset , we calculated averages to compare to occupancy estimates from our own study . We summarized occupancy estimates of Rich et al . and our own study within each developmental level using a box and whisker plot and assessed statistically significant differences based on whether or not interquartile ranges overlapped . Raw detections data have been deposited in Data Dryad , doi:10 . 5061/dryad . 11rf64v . The software used for initial species identifications is available via eMammal . org . To download and use the software , users must first create an account on eMammal and become associated with an existing project . This can be done by using the 'Join' button on the project's homepage , or by emailing the contact person , also listed on the project homepage . Usually the user will also have to pass an online or in person training , depending on the project requirements , to be approved to download the software . Working with citizen scientist volunteers , we obtained 53 , 273 detections of 19 mammal species at 1427 sites along an urban-wild gradient in Washington , DC and Raleigh , NC , USA , sampling both private and public lands . In DC , we detected 17 mammal species with mean naïve occupancy of 0 . 19 ( min = 0 , max = 0 . 93 ) and mean detection rate of 0 . 09 detections/day ( min = 0 , max = 1 . 05 ) . In Raleigh , we detected 17 mammal species with mean naïve occupancy of 0 . 14 ( min = 0 , max = 0 . 79 ) and mean detection rate of 0 . 08 detections/day ( min = 0 , max = 0 . 09 ) . We found no significant decline of species diversity or richness from suburban to wild gradient levels ( Figure 2—figure supplement 1 , Figure 1 ) . However , Shannon diversity was significantly lower at the urban level in DC , possibly due to low sampling ( Figure 2 , Supplementary file 1 ) . Diversity in yards was significantly higher or not statistically different from large and small forest fragments in both cities ( Figure 2—figure supplements 2 , 3 ) . Most ( 92 . 3% ) of the 13 mammal species detected >20 times occupied all levels of development below the urban level . Two of the largest predators , coyotes and bobcats , were absent from the highest development level ( urban ) but were detected at all other levels in both cities . Black bears ( Ursus americanus ) , which are actively discouraged from colonizing central North Carolina ( North Carolina Wildlife Resources Commission , 2011 ) , were not detected in Raleigh and were detected in DC at all levels of the gradient except suburban and urban , though were predominantly in the wild level . These results indicate that the extant mammal guild exploits all levels of the urban-wild gradient and that no species are entirely relegated to the wild gradient level . However , some species appear less adapted to habitation in human-dominated areas , spending most of their time at the wild levels of the gradient ( i . e . bobcat , bear; Figure 1 ) . We recognize that the current community represents species that survived the initial arrival of high-density human settlement . In particular , two large predators ( wolves ( Canis lupus ) and cougars [Puma concolor] ) were extirpated from our study area a century ago . However , even cougars and wolves have recently shown surprising adaptability in the face of development at other sites ( Bateman and Fleming , 2012; Wang et al . , 2017 ) suggesting that , given enough time and protection from persecution , many of the most ‘wild’ of species may adapt to human development . Predators are thought to be the most at risk from urbanization ( Crooks , 2002 ) , therefore , we evaluated predictors for occupancy ( MacKenzie et al . , 2002 ) and detection rate ( Kays et al . , 2017 ) for four carnivores: coyote , gray fox , red fox , and bobcat . Both of our models fit well , with Bayesian p-values between 0 . 1 and 0 . 9 ( Supplementary file 3 ) . Suburban and urban occupancy probabilities were not statistically different from wild for any of the species ( Figure 3—figure supplement 1 ) and we noted a decreasing trend in occupancy from urban to wild ( Figure 3 ) . We compared the occupancy estimates from our study to those reported for carnivores in protected areas around the world ( Rich et al . , 2017 ) and found no significant difference ( Figure 3 ) , suggesting that the ecological function of predators in this urban system is not substantially reduced from the current wild state , excepting the historical extirpation of the two largest native predators from the region . Our occupancy and detection rate models yielded similar results ( Supplementary file 5–7 ) demonstrating that green space is important to carnivore species that are less-adapted to human-altered landscapes . These models show a greater association of carnivores with green space when housing density is high ( e . g . coyote and gray fox , Supplementary file 6 , 7 ) , consistent with other studies finding urban green space important in maintaining biodiversity in cities ( Gallo et al . , 2017; Lombardi et al . , 2017; Matthies et al . , 2017 ) . It is likely that shyer species are not avoiding regions of high human density , but require patches of forest to navigate residential areas that are freely used by more commensal species , such as red foxes ( Tigas et al . , 2002 ) , which we frequently detected in yards . Indeed , we found a gradient of responses in carnivore use of human-dominated environments , from red fox which are the most urban adapted ( i . e . negatively associated with local large forest fragments and the only species to have a positive association with yards ) to bobcats which appear to be the most human-averse ( i . e . rarely detected in the suburban level of the gradient ) ( Figure 1; Figure 3—figure supplement 1 ) . Contrary to expectations , we found no evidence for a negative impact of suburban and exurban development on extant native mammal diversity , richness , and occupancy and detection rate of carnivores . In fact , all metrics were significantly greater than , or equal to , wild areas . We suspect that developed areas offer good food resources for wildlife through direct and indirect feeding ( i . e . bird feeders supplementing prey , pets ) , accidental feeding ( i . e . garbage ) , and ornamental plantings ( for herbivores ) , but testing this hypothesis will require additional diet studies in urban landscapes ( Contesse et al . , 2004 ) . Furthermore , the structure of mature suburbia ( i . e . older , established neighborhoods with large trees , wooded riparian areas , small parks ) contributes to a more diverse and varied landscape than wild areas with more homogenous forest cover , which is potentially beneficial for many generalist species . Developed areas where hunting is limited or prohibited also offer a safe haven for game species , presuming they can navigate the road networks ( Collins and Kays , 2011 ) and avoid direct human conflict . Our discovery of a wild suburbia suggests high levels of adaptation by mammals to developed landscapes over the last few decades , including predators and prey . The resilience of these species gives hope for wildlife in the Anthropocene , but the generality of this pattern needs to be tested in other cities to understand how habitat type , development patterns , apex predators , and hunting regulations influence urban mammal communities , as there are examples of far more drastic impacts of urbanization on other taxa and in other places around the globe ( Keast , 1995; McKinney , 2008 ) . Indeed , in Tokyo , Japan , the relative abundance of mammals declined with urbanization ( Saito and Koike , 2013 ) and avian communities in Quebec , Canada and Rennes , France showed a similar decline in richness ( Clergeau et al . , 1998; Saito and Koike , 2013 ) . This suggests that city structure , size and human density may influence mammalian distribution along urban-wild gradients with large , sprawling New World cities showing different patterns than the smaller more concentrated cities of the Old World . Although our study provides a less dire picture of urban ecosystem function than previously thought , we do not suggest abandoning mitigation of urbanization’s negative impacts , or conservation of completely wild areas . Factors such as urban green space , connectivity and availability of completely wild areas give species the time and space to adapt to changing habitats and climates . Further understanding of how urban wildlife navigates human-dominated areas and factors that contribute to the adaptation of species to the Anthropocene will be critical to maintaining diversity in a rapidly urbanizing world .	Humans transform natural ecosystems worldwide into towns and cities , replacing natural habitat with human-built surfaces . This loss of habitat and increase in human activity make suburban areas difficult for some species to survive in , raising concerns that developed areas become ecologically unbalanced as they lose biodiversity . However , the preservation of urban green space and lack of hunting could also open the door for some species to thrive in the midst of large human populations . Indeed , some animals , mammals in particular , have grown more tolerant of humans and appear to have adapted to suburban landscapes around the world . Some species that have been exclusively living in the wilderness , such as a small carnivore called the fisher , are even moving back into cities . Research into how mammals are coping with the urbanization of their habitats has produced conflicting results . Studies that explore a variety of cities and habitats would help to clear up this confusion . Parsons et al . worked with citizen scientist volunteers to survey the mammals at 1 , 427 sites across Washington DC and Raleigh , North Carolina . The volunteers set up motion-triggered cameras in these sites , which covered a full range of urban and wild habitats , including back yards and large nature preserves . The cameras detected similar or higher numbers of mammal species in suburban sites compared to wild areas . Indeed , most species appear to use suburban areas at least as much as wild land . Urban green space is especially important; it is used by less urban-adapted species like coyotes to navigate areas that are densely populated by humans . The results presented by Parsons et al . suggest that many mammals have indeed adapted to the suburban environment over the last few decades , resulting in more balanced urban ecosystems . More testing in other cities will help to determine how general this pattern of adaptation is , and provide us with knowledge that could help us to conserve many different species . However , some species were still most abundant in wild areas , emphasizing the need to also conserve wildlands and to minimize our impact on natural ecosystems .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "and", "discussion" ]	[ "ecology", "short", "report" ]	2018	Mammal communities are larger and more diverse in moderately developed areas
The only true living endothermic vertebrates are birds and mammals , which produce and regulate their internal temperature quite independently from their surroundings . For mammal ancestors , anatomical clues suggest that endothermy originated during the Permian or Triassic . Here we investigate the origin of mammalian thermoregulation by analysing apatite stable oxygen isotope compositions ( δ18Op ) of some of their Permo-Triassic therapsid relatives . Comparing of the δ18Op values of therapsid bone and tooth apatites to those of co-existing non-therapsid tetrapods , demonstrates different body temperatures and thermoregulatory strategies . It is proposed that cynodonts and dicynodonts independently acquired constant elevated thermometabolism , respectively within the Eucynodontia and Lystrosauridae + Kannemeyeriiformes clades . We conclude that mammalian endothermy originated in the Epicynodontia during the middle-late Permian . Major global climatic and environmental fluctuations were the most likely selective pressures on the success of such elevated thermometabolism . One key adaptation enabling tetrapods to cope with fluctuating climatic conditions was the acquisition of endothermy ( Paaijmans et al . , 2013 ) . This character is defined here as the ability to actively produce body heat through metabolic activity ( Cannon and Nedergaard , 2004 ) . Its development and anchoring in populations constitutes a major step in vertebrate evolution because it modified the energy relationships between organisms and their environments . By actively raising and maintaining body temperature within a narrow range that allows optimal physiological and biochemical functioning , endothermic vertebrates are able to colonise environments with extreme thermal conditions , for example freezing at high latitudes and altitudes ( Day et al . , 2015 ) . Endothermy is commonly associated with homeothermy , being the capacity to regulate the body heat through metabolic activity as well . This combination corresponds to one end of a gradient of thermoregulatory strategies observed in living animals . The other end of the spectrum is ectothermy combined with poïkilothermy which animals use as a thermoregulatory strategy to increase their body temperature toward optimal levels by using external heat sources . Their body temperature therefore traces that of their surroundings and is the most commonly occurring energy saving strategy . Amongst modern vertebrates , various thermoregulatory strategies have been adopted between these two end-members , such as regional endothermy ( Bernal et al . , 2001; Katz , 2002 ) or inertial homeothermy ( McNab and Auffenberg , 1976 ) , and only birds and mammals fall within the endothermy end of the spectrum . It has been proposed that bird thermoregulation originated within non-avian dinosaurs ( Seebacher , 2003; Amiot et al . , 2006; Grady et al . , 2014 ) , or even earlier within basal archosauriforms ( Farmer and Carrier , 2000; de Ricqlès et al . , 2003; Seymour et al . , 2004; Summers , 2005; Gower et al . , 2014; Legendre et al . , 2016 ) . Various approaches have been tried by many researchers to assess the origin of mammalian endothermy ( McNab , 1978; Bennett and Ruben , 1986; Farmer , 2000; Hillenius , 1992 , 1994; Kemp , 2006a; Khaliq et al . , 2014; Owerkowicz et al . , 2015; Benoit et al . , 2016b; Crompton et al . , 2017 ) . Some consider the appearance of endothermy to have either occurred during the transition from basal synapsid ‘pelycosaurs’ to therapsids , and to be either due to a shift in foraging ecology ( Hopson , 2012 ) or due to a response to the availability of a seasonally arid , savanna-like biome by the end of the Early Permian ( Kemp , 2006b ) . How and why endothermy evolved in mammals remains a contentious issue , mostly because of the very low fossilization potential of anatomical and behavioural features associated with thermoregulation . Amongst the latter features , the presence of hair as an insulating integument is unequivocally associated with endothermy in all extant mammals . The oldest synapsid fossils preserved with fur imprints are Castorocauda ( Ji et al . , 2006 ) and Megaconus ( Zhou et al . , 2013 ) . These early relatives of mammals were recovered from the Middle-Late Jurassic of China , implying that hair and fur appeared before ~165 Ma . The occurrence of retracted , fully ossified and non-ramified infraorbital canals ( a structure associated with the presence of maxillary vibrissae ) within non-mammaliaform Prozostrodontia , implies an older age of approximately 240 to 246 Ma for the occurrence of fur and hair ( Benoit et al . , 2016b ) . Another anatomical character interpreted as associated with endothermy is the bony secondary palate . This is a feature associated with efficient respiratory capabilities considered to be linked to the high energy required for elevated metabolic rates . In some Triassic non-mammaliamorph therapsids , bauriid therocephalians and cynodonts , a bony secondary palate is fully developed ( Abdala et al . , 2014 ) . It is noteworthy that a complete bony secondary palate is also present in dicynodonts , however it is primarily formed by the premaxilla ( King , 1988 ) and not the maxilla as documented in therocephalians , cynodonts and extant mammals . Although a secondary osseous palate is ubiquitous in mammals , it also occurs in a few ectotherms ( crocodiles , scincid lizards ) , thus questioning its direct link to endothermy ( Bennett and Ruben , 1986 ) . Almost all extant endotherms possess nasal turbinate bones covered with mucosa that reduce heat loss and moisten air during respiration ( Owerkowicz et al . , 2015 ) . This feature , absent in extant ectotherms ( Witmer , 1995 ) , may have been present in therocephalian , cynodont and dicynodont therapsids , as postulated from bony ridges in the nasal cavities interpreted as supports for the turbinate complex ( Hillenius , 1992 , 1994; Crompton et al . , 2017 ) . A peculiar histological structure of fast-growing bone associated with highly vascularised woven-fibred matrix and primary osteons known as fibrolamellar bone ( FLB ) , is another feature often used as evidence of a high metabolic activity ( Montes et al . , 2010; Legendre et al . , 2016 ) . Accordingly , several bone palaeohistological studies have addressed the quest for the presence of FLB in therapsids ( de Ricqlès , 1972 , 1979; Botha , 2003; Botha and Chinsamy , 2001 , 2004; Ray et al . , 2004; Olivier et al . , 2017 ) . Ray et al . ( 2004 ) and Olivier et al . ( 2017 ) analysed several therapsid groups ( anomodont , gorgonopsian , therocephalian , cynodont ) and found FLB in some genera ( Aelurognathus , Pristerognathus , Tritylodon , Oudenodon , Lystrosaurus , Moghreberia ) , suggesting sustained fast growth , and thus elevated metabolic activity . The presence of FLB has also been demonstrated in some earlier non-therapsid synapsids such as Sphenacodon , Dimetrodon or even Ophiacodon ( Huttenlocker et al . , 2010; Shelton et al . , 2012; Shelton and Sander , 2017 ) . However , FLB also occurs in a few ectotherms such as in some turtles and crocodilians , and is absent in small mammals and passerine birds ( Bouvier , 1977 ) , showing that FLB is mostly correlated with high growth rates , which does not always correlate to high metabolic rates . Therefore , these characters alone cannot be considered as definitive evidence of endothermy , leaving the question of the timing and selection pressure for the origin of mammal endothermy still heavily debated . Because the oxygen isotope fractionation between bone or tooth phosphate and body fluids is temperature dependent , and phosphate has a strong resistance to diagenetic alteration , oxygen isotope compositions of therapsid apatite phosphate ( δ18Op ) has been used in this pilot study to investigate the origin of mammalian endothermy . Indeed the δ18Op value of vertebrate apatite ( bone , tooth ) reflects both the oxygen isotope composition of the animal body water ( δ18Obw ) and its body temperature ( Tb ) . Body water derives mainly from drinking meteoric water or plant water ( D’Angela and Longinelli , 1990; Kohn , 1996a ) , and the δ18O value of this water in turn depends on climatic parameters such as air temperature , hygrometry , and amount of precipitation ( Dansgaard , 1964; von Grafenstein et al . , 1996; Fricke and O'Neil , 1999 ) . Variations in the δ18O values of ectotherm apatite , along with increasing latitude , are expected to reflect decreasing air temperatures as their body temperatures follow those of the environment . In contrast endotherms , which have a constant body temperature , should not be affected by environmental temperatures changes . Moreover , physiological adaptation to specific habitat use ( aquatic , semi-aquatic or terrestrial ) affects the δ18Obw value by controlling the magnitude of body input and output oxygen fluxes , some of them being associated with oxygen isotopic fractionations ( Amiot et al . , 2010 ) . Therefore , co-existing endotherms and ectotherms should have distinct apatite δ18Op values reflecting their body temperature and ecological differences . By comparing apatite δ18Op values of therapsids with those of co-existing ectotherms of known ecologies at various palaeolatitudes , it should be possible to infer therapsid thermophysiology , a methodology that has previously been applied to non-avian dinosaurs ( Fricke and Rogers , 2000; Amiot et al . , 2006 ) . Following the protocol of previous research undertaken to establish the Permo-Triassic climatic conditions that prevailed during which South African tetrapods , including therapsids , radiated ( Rey et al . , 2016 ) , this study aims to investigate thermophysiological strategies developed by various Permo-Triassic therapsid groups using the stable oxygen isotope compositions of their phosphatic remains . Our results add new data to the discussion of the origin of mammalian endothermy and its link to global climatic change . The 13 sampled South African Permian therapsids come from three different assemblage zones ( AZ ) of the Beaufort Group: the lower Tapinocephalus AZ , the Tropidostoma AZ and the lower Daptocephalus AZ ( Viglietti et al . , 2016 ) . Oxygen isotope compositions of three therapsid genera ( Dicynodon , Diictodon and Oudenodon ) from the youngest assemblage zone ( lower Daptocephalus AZ ) and seven therapsid genera ( Aelurosaurus , Diictodon , Ictidosuchoides , Oudenodon , Rhachiocephalus , Tropidostoma and a basal cynodont ) from the Tropidostoma AZ were respectively compared with one co-occuring Rhinesuchus and one co-occuring rhinesuchid . Differences between all the Permian therapsids and ectothermic temnospondyl range from +1 . 1 ± 0 . 6‰ to +8 . 0 ± 0 . 9‰ ( Figure 1A ) , encompassing the expected range for which therapsids are considered ectothermic . 10 . 7554/eLife . 28589 . 003Figure 1 . δ18Op differences between Permian therapsids and other tetrapods . Differences in δ18Op values between therapsids and stereospondyls ( white symbols ) and between therapsids and parareptiles ( black symbols ) from the same localities are plotted against their corresponding palaeolatitude . A theoretical framework based on modern temperature gradient ( 0 . 6 ± 0 . 1°C/°Lat; see Appendix 1 ) and phosphate-water-temperature oxygen isotope fractionation ( Lécuyer et al . , 2013 ) predicts various δ18Op value differences . The lighter orange and red areas correspond to the uncertainty of the temperature gradient . Dap . : Daptocephalus; Tap . : Tapinocephalus . DOI: http://dx . doi . org/10 . 7554/eLife . 28589 . 003 In addition , δ18Op values of the therapsids Dicynodon , Diictodon and Oudenodon are compared to those of the supposedly semi-aquatic parareptile Pareiasaurus ( Ivakhnenko , 2001; Kriloff et al . , 2008 ) ( see Appendix 1 ) , with an observed range of +4 . 3 ± 0 . 4‰ to +6 . 8 ± 0 . 5‰ ( Figure 1A ) which is similar to that measured between therapsids and amphibians . This also supports the ectothermic status of Dicynodon , Diictodon and Oudenodon . Anteosaurus , Criocephalosaurus , Struthiocephalus , Glanosuchus and a titanosuchid from the lower Tapinocephalus AZ have also been compared to two co-occuring basal pareiasaurs which are attributed to either Embrithosaurus , Nochelosaurus or Bradysaurus ( Lee , 1997 ) and are considered to have been terrestrial animals ( Canoville et al . , 2014 ) . From Figure 1B , the δ18Op differences range from −1 . 4 ± 0 . 6‰ to 0 . 7 ± 1 . 0‰ , also supporting the ectothermic status of these therapsids . From the Middle Permian of China , one anteosaurid Sinophoneus yumenensis from the low palaeolatitude locality of Dashankou has a δ18Op value 4 . 4 ± 0 . 3‰ lower than the co-existing bolosaurid parareptile Belebey chengi , which is considered to have been a terrestrial ectotherm ( Berman et al . , 2000; Müller et al . , 2008 ) . This difference between only two values would suggest that Sinophoneus was endothermic , but it is also very close to the expected ranges for ectothermic therapsids ( Figure 1B ) . Considering Sinophoneus as semi-aquatic , as has been suggested for some anteosaurids ( Boonstra , 1955 , Boonstra , 1962 ) , the δ18Op difference would imply an ectothermic thermophysiology for this therapsid . This hypothesis needs to be tested with a larger number of samples , which are not yet available . From the Cynognathus AZ ( subzone B ) of South Africa , differences between the therapsids Kannemeyeria , Cynognathus and Diademodon and the temnospondyl amphibians Xenotosuchus and Microposaurus range from −1 . 5 ± 1 . 1‰ to +0 . 9 ± 1 . 5‰ ( Figure 2A ) , which fit within the range predicting endothermic therapsids . Interestingly , these therapsids have values ranging from 0 . 0 ± 1 . 6‰ to +1 . 8 ± 1 . 6‰ higher than the coexisting terrestrial archosauriform Erythrosuchus ( Botha-Brink and Smith , 2011 ) , a range suggesting that they shared a similar thermophysiology ( Figure 2B ) . Therefore δ18Op values imply that , as in the case of the therapsids , Erythrosuchus was also endothermic which is consistent with the elevated growth rates implied by its palaeohistology ( de Ricqlès et al . , 2008; Botha-Brink and Angielczyk , 2010 ) . 10 . 7554/eLife . 28589 . 004Figure 2 . δ18Op differences between Early to Middle Triassic therapsids and other tetrapods . Differences in δ18Op values between therapsids and stereospondyls ( white symbols ) and between therapsids and archosauriforms ( black symbols ) from the same localities are plotted against their corresponding palaeolatitude . A theoretical framework based on a lower-than-today thermal gradient ( 0 . 4 ± 0 . 1°C/°Lat; see Appendix 1 ) and phosphate-water-temperature oxygen isotope fractionation ( Lécuyer et al . , 2013 ) predicts various δ18Op value differences . The lighter orange and red areas correspond to the uncertainty of the temperature gradient . DOI: http://dx . doi . org/10 . 7554/eLife . 28589 . 004 Also from South Africa , five Lystrosaurus specimens from the lower Lystrosaurus AZ have δ18Op values similar to those of the co-existing semi-aquatic stereospondyl Lydekkerina ( Schoch , 2008; Canoville and Chinsamy , 2015 ) . In addition , an indeterminate lystrosaurid from the Induan Jiucaiyuan Formation of the Xinjiang Province has a δ18Op value similar ( with a difference of −0 . 1 ± 0 . 6‰; Figure 2B ) to that of the proterosuchid ‘Chasmatosaurus’ yuani , a basal archosauriform considered terrestrial and possessing an intermediate thermometabolsim based on a palaeohistological study ( Botha-Brink and Smith , 2011 ) . The combined results from South Africa and China suggest that the analysed lystrosaurids were terrestrial endotherms ( Figure 2; see Appendix 1 ) . From the Ermaying Formation of the Shanxi Province ( China ) , the therapsids Shansiodon wangi and Parakannemeyeria youngi have respectively δ18Op values of 2 . 0 ± 0 . 7‰ and 1 . 7 ± 0 . 7‰ . These are both lower than the sampled erythrosuchid archosauriform Shansisuchus shansisuchus , which fall within two theoretical overlapping ranges ( Figure 2B ) . As for the South African erythrosuchids , if we consider Shansisuchus as a terrestrial endotherm-like animal and the low palaeolatitude of this part of China region , then the two therapsids also fall within the range of endotherms . The late Anisian cynodont Diademodon and the kannemeyeriiform Kannemeyeria , from the Cynognathus AZ ( subzone C ) , have both lower δ18Op values than those of the contemporary semi-aquatic stereospondyls Paracyclotosaurus and Xenotosuchus with differences ranging from −3 . 9 ± 2 . 7‰ to −0 . 5 ± 0 . 6‰ ( Figure 3A ) . This pattern fits within the main range predicting endothermic therapsids . 10 . 7554/eLife . 28589 . 005Figure 3 . δ18Op differences between Middle to latest Triassic therapsids and other tetrapods . Differences in δ18Op values between therapsids and stereospondyls ( white symbols ) and between therapsids and archosauriforms ( black symbols ) from the same localities are plotted against their corresponding palaeolatitude . A theoretical framework based on a lower-than-today thermal gradient ( 0 . 5 ± 0 . 1°C/°Lat; see Appendix 1 ) and phosphate-water-temperature oxygen isotope fractionation ( Lécuyer et al . , 2013 ) predicts various δ18Op value differences . The lighter orange and red areas correspond to the uncertainty of the temperature gradient . DOI: http://dx . doi . org/10 . 7554/eLife . 28589 . 005 The Moroccan kannemeyeriiform Moghreberia nmachouensis from the early middle Carnian of the Argana Basin has a mean δ18Op value 2 . 0 ± 0 . 5‰ higher than the co-existing aquatic stereospondyl Almasaurus habbazi ( Figure 3A ) , thus implying that Moghreberia nmachouensis was also endothermic . An indeterminate cynodont from the Rhaetian Lower Elliot Formation of Lesotho has a δ18Op value 2 . 1 ± 0 . 3‰ , higher than that of an indeterminate basal sauropodomorph . The suspected endothermy and terrestriality of both dinosaurs ( Amiot et al . , 2006; D'Emic , 2015 ) and cynodonts are in agreement with their δ18Op difference that falls within the expected range predicting similar thermophysiology between the two ( Figure 3B ) . In order to investigate the origin of mammal endothermy amongst the Permo-Triassic therapsids , stable oxygen isotope compositions of apatite phosphate and carbonate from therapsids and associated taxa recovered from several palaeolatitudes were analysed . The following results are highlighted: Nineteen new fossil apatite samples were analysed to determine stable oxygen isotope compositions of apatite phosphate and carbonate , along with 89 samples for which oxygen isotope compositions have already been published ( Rey et al . , 2016; Supplementary file 1 ) . This sample total comprises 41 teeth and 65 bones of 90 individual tetrapods ( Therapsida , Archosauriformes , Parareptilia and Stereospondyli ) recovered from Permian and Triassic deposits of South Africa , Lesotho , Morocco and China . All the sample localities are correlated to the marine biostratigraphic stages using the absolute ages accepted by the International Commission on Stratigraphy ( Cohen et al . , 2013 ) ; updated 12/2016 ) , with the Permo-Triassic and Guadalupian-Lopingian boundaries now respectively considered to be at 251 . 90 ± 0 . 02 Ma ( Burgess et al . , 2014 ) and 259 . 1 ± 0 . 5 ( Zhong et al . , 2014 ) Ma . South African samples comprise Permian and Triassic bones and teeth of therapsids , pareiasaurs , archosauriforms and stereospondyls recovered from 10 localities in the Beaufort Group ( Karoo Supergroup ) , and housed in the collections of the Iziko South African Museum , Cape Town ( SAM , Supplementary file 1 ) and at the Evolutionary Studies Institute , University of the Witwatersrand , Johannesburg ( ESI , Supplementary file 1 ) . Permian biozone ages of South African localities were taken from ( Rubidge et al . , 2013; Day et al . , 2015 ) , whereas Triassic age determination has been achieved by biostratigraphic correlation with Laurasian sequences ( Hancox et al . , 1995; Rubidge , 2005; Abdala and Ribeiro , 2010 ) . Lesotho samples comprise a cynodont therapsid and a basal sauropodomorph dinosaur from a Triassic locality near the town of Pokane , and are part of the Paul Ellenberger Collection at the Institut des Sciences de l’Evolution , University Montpellier , France ( ISEM , Supplementary file 1 ) . The locality belongs to the ‘Red Beds inférieurs a or b’ of the lower Elliot Formation which is currently regarded as latest Triassic ( late Rhaetian ) ( Knoll , 2004 ) . Moroccan samples comprise therapsid and stereospondyl bones recovered from the ‘Locality 11’ of the Argana Group ( Jalil , 1999 ) near the village of Alma , and housed at the Museum National d’Histoire Naturelle , Paris , France ( MNHN , Supplementary file 1 ) . The locality is biostratigraphically correlated to the upper Timezgadiouine Formation , considered to be Middle to early Late Carnian ( Jalil , 1999 ) . Chinese samples are from Permian and Triassic localities situated in Gansu , Shanxi and Xinjiang provinces and comprise therapsids found in association with archosauriforms or parareptiles . These remains are curated at the Institute of Vertebrate Paleontology and Paleoanthropology in Beijing , China ( IVPP , Supplementary file 1 ) . The Dashankou locality , from Gansu Province , is biostratigraphically dated as Early Roadian ( Liu et al . , 2009; Liu , 2010 ) . From Shanxi Province , sampled fossils originate from three localities in the Ermaying Formation which is considered to be Anisian ( Liu et al . , 2013 ) . From Xinjiang Province , two localities in the Jiucaiyuan Formation have been sampled and are considered Early Triassic ( Metcalfe et al . , 2009 ) . Calculation of palaeogeographic coordinates of the sampling sites was performed after careful selection of the magnetic poles of West Gondwana ( Muttoni et al . , 2001 ) , North China ( et al . , 1992 ) , South Jungar ( Choulet et al . , 2013 ) and the Alashan terrane ( Meng , 1992; Yuan and Yang , 2015 ) . The Apparent Polar Wander Path ( APWP ) of South Africa ( Torsvik et al . , 2012 ) was used to constrain the palaeolatitudinal position of the South African and Lesotho fossil sites . Palaeolatitudes and associated uncertainties ( A95 ) are shown in Supplementary file 1 . To measure the oxygen isotope composition of the apatite phosphate group , the phosphate ions were isolated using acid dissolution and anion-exchange resin applying a standard protocol ( Lécuyer , 2004 ) . Silver phosphate was quantitatively precipitated in a thermostatic bath set at a temperature of 70°C . After filtration , washing with double deionized water and drying at 50°C , an aliquot of 300 μg of Ag3PO4 was mixed with 300 μg of nickelised carbon in a silver reaction capsule . Silver phosphate was then reduced into CO to measure its 18O/16O ratio ( Lécuyer et al . , 2007; Fourel et al . , 2011 ) . Each sample was heated at 1450°C by pyrolysis using a VarioPYROcube EA system ( Elementar ) interfaced to an IsoPrime isotope ratio mass spectrometer working in continuous flow mode at the UMR CNRS 5276 LGLTPE , University Claude Bernard Lyon 1 . Isotopic compositions are quoted in the standard δ notation relative to V-SMOW . Silver phosphate precipitated from standard NBS120c ( natural Miocene phosphorite from Florida ) was repeatedly analysed ( δ18O = 21 . 71 ± 0 . 20‰; n = 30 ) along with the silver phosphate samples derived from the tetrapod remains . For the oxygen isotope analysis of apatite carbonate , about 10 mg of tooth or bone powder was pre-treated ( Koch et al . , 1997 ) . Powders were washed with a 2% NaOCl solution to remove organic matter , then rinsed five times with double deionized water and air-dried at 40°C for 24 hr . Potential secondary carbonate was removed by adding 0 . 1 M acetic acid and leaving for 24 hr , after which the powder was again rinsed five times with double deionized water and air-dried at 40°C for 24 hr . The powder/solution ratio was kept constant at 0 . 04 g mL−1 for both treatments . Stable isotope ratios were determined by using a Thermo Finnigan Gasbench II at the geochemistry laboratory of the Institute of Geology and Geophysics ( Chinese Academy of Sciences , China ) . For each sample , an aliquot of 2 mg of pre-treated apatite was reacted with 5 drops of supersaturated orthophosphoric acid at 72°C for one hour under a He atmosphere before starting 10 measurement cycles of the isotopic composition of the CO2 produced with a Finnigan MAT 253 continuous flow isotope ratio mass spectrometer . The measured oxygen isotopic compositions were normalized relative to the NBS-19 calcite standard and have a reproducibility index better than ±0 . 2‰ . Isotopic compositions are quoted in the standard δ notation relative to V-SMOW . Analysed materials consist of bone or tooth dentine , which is more porous than enamel with small and less densely inter-grown apatite crystals ( Mills , 1967 ) . Thus , their original stable isotope compositions are more prone to diagenetic alteration that may have taken place through precipitation of secondary minerals within and at the surface of bioapatite crystals , adsorption of ions on the surface of apatite crystals , or dissolution and recrystallization with isotopic exchange . The samples from South Africa have been previously tested for primary preservation through comparison between their δ18Op values , δ18Oc values and carbonate content on the basis of the following considerations: ( 1 ) the carbonate content in apatite of modern vertebrates typically ranges from less than 1% up to 13 . 4% . Thus , samples that have a carbonate content exceeding 13 . 4 wt% likely contain additional inorganic carbonate precipitated from diagenetic fluids , and would result in potentially biased δ18Oc values of apatite carbonate ( Figure 5 ) ; ( 2 ) In modern vertebrates , the oxygen isotope composition of apatite carbonate is higher than that of co-occurring apatite phosphate ( 7–9 ‰ in mammals ) , and up to 14 . 7‰ in sharks ( Vennemann et al . , 2001 ) . Experimental ( et al . , 1967 ) and empirical studies ( Zazzo et al . , 2004b ) have shown that microbially-mediated diagenetic alteration of apatite phosphate results in a greater difference between δ18Oc and δ18Op values . Therefore , fossil samples exhibiting δ18Oc-δ18Op differences larger than 14 . 7‰ are most likely altered and can be disregarded ( Figure 5 ) . Inorganic alteration at low temperature has little effect on the δ18Op values of phosphates , even at geological time scales ( Lecuyer et al . , 1999 ) , so samples affected by inorganic diagenetic alteration of carbonates , ( resulting either in a high overall carbonate content or anomalous δ18Oc-δ18Op differences ) , may still preserve the original oxygen isotope composition of their phosphate ( Figure 5 ) . Using these two assessments , newly measured δ18Op values are considered to have preserved their original isotopic signatures and can be interpreted in terms of ecologies and physiologies . 10 . 7554/eLife . 28589 . 007Figure 5 . Isotopic preservation assessment . δ18Oc-δ18Op differences between teeth and bones plotted against the structural carbonate content ( wt% ) of apatite . Samples that have δ18Oc-δ18Op differences higher than 14 . 7‰ are considered doubtful as regards potential diagenetic alteration ( see text ) . For carbonate contents ( wt% ) higher than 13 . 4% , the δ18Oc values are considered to be inherited from inorganic diagenetic processes . A high difference between δ18Oc and δ18Op is interpreted as the result of a microbially-mediated alteration of the apatite phosphate or too high δ18Oc values resulting from the addition of inorganic carbonate or isotopic exchange with an external source of inorganic carbon . The grey crosses refer to previously published South African bone and tooth samples ( Rey et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28589 . 007 For all localities , average δ18Op values were calculated for each tetrapod species . Differences in δ18Op values between therapsid species and co-occurring non-therapsid tetrapods ( amphibians , parareptiles or archosauriforms ) were calculated and plotted against their corresponding palaeolatitude for three time intervals: the middle to late Permian ( Figure 2 ) , the Early to Middle Triassic ( Figure 3 ) and the Middle Triassic to latest Triassic ( Figure 4 ) . These differences were compared to the following four theoretical areas of values represented as coloured areas in Figures 1–3 . To construct those theoretical areas , both the phosphate-water temperature scale from Lécuyer et al . , 2013 and the differences of stable oxygen compositions between mammals of various ecologies from Cerling et al . ( 2008 ) have been used . ( see Appendix 1 for their construction details ) . Orange and green areas in Figures 1A , 2A and 3A represent expected δ18Op value differences between terrestrial therapsids and semi-aquatic stereospondyls ( white symbols ) or parareptiles ( black symbols ) ; red and blue areas in Figures 1B , 2B and 3B represent expected δ18Op value differences between terrestrial therapsids and terrestrial Permian parareptiles or Triassic archosauriforms ( black symbols ) . Oblique orange and red areas in Figures 1–3 represent expected δ18Op value differences between an endotherm and an ectotherm . Vertical green and blue areas in Figures 1–3 represent expected δ18Op value differences between animals having similar thermophysiology .	School textbooks often refer to “cold-blooded” and “warm-blooded” animals , but these terms are misleading . Rather than being cold , animals like reptiles have body temperatures that are mostly determined by their external environment and can actually achieve high body temperatures , for example , by basking in the sun . By contrast , “warm-blooded” mammals produce their own heat and typically maintain a body temperature that is warmer than their environment . As such , so-called warm-blooded animals are more accurately referred to as “endotherms” and cold-blooded animals as “ectotherms” . Endothermic animals share several characteristics , including insulating layers – like fur or feathers – that keep the body warm , and a secondary palate that separates the mouth and nose for continuous breathing , even while eating . Many of these traits are seen in fossils belonging to a group of animals called the therapsids . Also known as the “mammal-like reptiles” , these animals are descended from ectothermic reptiles but are the ancestors of the endothermic mammals . They dominated the land between 270 and 220 million years ago , during periods of time called the Permian and the Triassic . They also survived two major mass extinction events , including the most devastating mass extinction in all of Earth’s history . However , when the ancestors of mammals became truly endothermic remains an open question . Previous studies that have tried to determine this by focusing on the physical characteristics of therapsids have not yet given a consistent date . Rey et al . took a new approach to answer when endothermy first evolved in the mammal-like reptiles , and instead looked at the chemical makeup of minerals in over 100 fossils . Oxygen can exist in different forms called stable isotopes: oxygen-16 and the rarer and heavier oxygen-18 . The ratio of these two isotopes in a fossil will depend on , among other things , where the animal lived and , importantly , its body temperature . Therefore , Rey et al . compared oxygen-containing minerals in the bones and teeth of therapsids to those of other animals that lived alongside them to look for signatures that indicated differences in body temperature and how it was regulated . It appears that two different branches of the therapsid’s family tree independently became endothermic . One branch includes the mammals and their direct ancestors , while the second is more distantly related to mammals . Both became endothermic towards the end of the Permian Period , between about 259 and 252 million years ago . Based on these findings , Rey et al . suggest that endothermy allowed these animals to better cope with fluctuating climates , which helped them to be among the few species that survived the mass extinction event at the end of the Permian . Going forward , these new findings can help scientists to understand which physical characteristics were necessary for endothermy to first develop and which helped to optimize it afterwards . Furthermore , they also suggest that endothermic animals are more able to survive fluctuations in climate , which could guide efforts to protect modern-day endangered species that are most at risk from the ongoing effects of climate change .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "ecology" ]	2017	Oxygen isotopes suggest elevated thermometabolism within multiple Permo-Triassic therapsid clades
Management of salivary gland hypofunction caused by irradiation ( IR ) therapy for head and neck cancer remains lack of effective treatments . Salivary glands , especially the parotid gland , actively uptake dietary nitrate and secrete it into saliva . Here , we investigated the effect of dietary nitrate on the prevention and treatment of IR-induced parotid gland hypofunction in miniature pigs , and elucidated the underlying mechanism in human parotid gland cells . We found that nitrate administration prevented IR-induced parotid gland damage in a dose-dependent manner , by maintaining the function of irradiated parotid gland tissue . Nitrate could increase sialin expression , a nitrate transporter expressed in the parotid gland , making the nitrate-sialin feedback loop that facilitates nitrate influx into cells for maintaining cell proliferation and inhibiting apoptosis . Furthermore , nitrate enhanced cell proliferation via the epidermal growth factor receptor ( EGFR ) –protein kinase B ( AKT ) –mitogen-activated protein kinase ( MAPK ) signaling pathway in irradiated parotid gland tissue . Collectively , nitrate effectively prevented IR-induced xerostomia via the EGFR–AKT–MAPK signaling pathway . Dietary nitrate supplementation may provide a novel , safe , and effective way to resolve IR-induced xerostomia . Head and neck cancer ( HNC ) is the sixth most common malignancy worldwide ( Lindsey et al . , 2012 ) . Irradiation ( IR ) is an important treatment approach for HNC . Salivary glands , which are often included in the IR field , are highly radiosensitive , and IR often results in salivary gland hypofunction . IR destroys salivary gland acinar cells , the sole site of fluid transport in the gland parenchyma , resulting in xerostomia , which severely impacts quality of life for affected patients ( Bhide et al . , 2009; Davies and Thompson , 2015; Mercadante et al . , 2017 ) . Acinar cells in salivary glands include both serous and mucous cell types . Serous acinar cells are much more radiosensitive than mucous acinar cells . Notably , the parotid gland consists solely of serous acinar cells , which makes the parotid gland more radiosensitive than the submandibular and sublingual gland ( Grundmann et al . , 2009 ) . As such , the parotid gland provides an excellent model organ to study IR-induced salivary gland hypofunction . Many studies have attempted to treat salivary hypofunction using gene transfer , stem cell transplantation , or salivary gland regeneration . These studies have elucidated the mechanisms involved in xerostomia , and allowed for the development of new treatments for xerostomia ( Vissink et al . , 2010; Vissink et al . , 2015 ) . However , most of these treatment modalities are therapeutic and thus beneficial for patients with established xerostomia ( i . e . , when salivary glands are severely damaged ) and have not been adequately resolved IR-induced xerostomia ( Vissink et al . , 2010 ) . In addition , there have been no studies on the prevention of IR-induced xerostomia . New approaches for the prevention of IR-induced xerostomia are urgently required to maintain the tissue architecture of the parotid gland even after IR treatment . Inorganic nitrate is a component of the human diet , and is found in vegetables , fruits , and drinking water . At least 25% of circulating nitrate is actively taken up by salivary glands ( via sialin ) and secreted into saliva , such that the concentration of nitrate in saliva is approximately tenfold greater than that in blood ( Omar et al . , 2016 ) . Nitrate has been shown to offer protection against diseases such as obesity , diabetes mellitus , and heart disease ( Gilchrist et al . , 2014; Khambata et al . , 2017; Kina-Tanada et al . , 2017; Lundberg et al . , 2018; Ma et al . , 2018; McNally et al . , 2016; Omar et al . , 2016 ) . Sialin ( Slc17a5 ) is a transmembrane protein expressed most abundantly in the acinar cells of the parotid gland ( Qin et al . , 2012 ) . Sialin transports nitrate and other essential cellular substances such as glutamate and aspartate , and contributes to the maintenance of acinar cells physiological function ( Qin et al . , 2012 ) . It is unclear why sialin is most strongly expressed in the parotid gland and facilitates nitrate influx into the gland . The interaction of sialin and nitrate is unique to the parotid gland and not found in other organs . It is highly likely that the interaction between nitrate and sialin contributes to the maintenance of parotid gland homeostasis . The parotid glands of mice and rats are not similar to those of humans ( Wang et al . , 1998 ) . We previously demonstrated that the parotid glands of miniature pigs have similar anatomical and physiological characteristics to those of human parotid glands . Then we established miniature pig model of IR-induced parotid gland hypofunction using single-dose or fractionated IR ( Guo et al . , 2014; Li et al . , 2005 ) . In this study , we investigated the preventive and therapeutic effects of dietary nitrate supplementation on IR-induced salivary gland hypofunction using an established miniature pig model , and characterized the mechanism underlying these effects . First , we investigated whether inorganic nitrate could prevent or treat IR-induced xerostomia . Miniature pigs were administered a single dose of 20 Gy ( biological dose equal to 60 Gy ) ( Figure 1a and d and Figure 1—figure supplement 1a , b ) , and the effects of preventive and therapeutic nitrate administration were evaluated ( Figure 1a–f ) . The IR control group exhibited a sharp decrease in mean salivary flow rate ( SFR ) ( Figure 1b and e ) . In the preventive nitrate group , the mean SFR was significantly decreased at 1-month post-IR , but recovered to nearly normal levels at 4 months post-IR . Long-term observation ( 2 years post-IR ) showed that SFR was at 80% of pre-IR levels ( Figure 1b ) . However , the mean SFR in the therapeutic nitrate group did not show significant recovery and remained similar to that in the IR control group ( Figure 1e ) . The mean salivary concentrations of amylase , sodium , and chloride in saliva were significantly higher in the preventive nitrate group than in the IR control group ( Figure 1—figure supplement 1c ) . In contrast , the mean potassium concentration was significantly lower following nitrate administration ( Figure 1—figure supplement 1c ) . The mean change in local blood flow rate in irradiated parotid glands displayed a pattern similar to that of the mean SFR . Local blood flow rate was stable in the preventive nitrate group post-IR , but decreased sharply with time in the therapeutic and IR control groups ( Figure 1c and f ) . Four months post-IR , SFR reduction and parotid gland tissue degeneration had stabilized ( the latter was determined using terminal deoxynucleotidyl transferase dUTP nick-end labeling in situ assay ) ( Figure 1—figure supplement 1d ) . Histological analyses showed that far more acinar cells remained and less fibrosis was present at 4 months post-IR in the preventive nitrate group compared to the therapeutic and control groups ( Figure 1g ) . Next , we studied the dose-dependent effects of nitrate on IR-induced salivary gland hypofunction using a fractionated IR model that mimics clinical conditions ( Figure 2a ) . Exogenously administered nitrate was well absorbed , as evidenced by significantly higher salivary and serum nitrate concentrations in the nitrate groups compared with those in the control groups . This difference was more pronounced at higher nitrate doses ( Figure 2—figure supplement 1a , b ) . The mean SFR gradually decreased with time after unilateral parotid gland fractionated IR ( ‘IR control’ in Figure 2b ) . Four months post-IR , the mean SFR in the IR control group was approximately 20% of that observed pre-IR ( Figure 2b’ ) . Conversely , mean SFRs were protected from IR-induced changes in a dose-dependent manner ( Figure 2b ) . The highest mean SFR was observed in the group that received 2 mmol/kg∙day nitrate , with a mean SFR of nearly 85% of the pre-IR SFR . The mean SFRs were 65% , 50% , and 30% of those observed before IR in response to nitrate doses of 1 , 0 . 5 , and 0 . 25 mmol/kg∙day , respectively ( Figure 2b’ ) . The mean changes in local blood flow rate showed a pattern similar to those of the mean SFRs . The mean local blood flow rate was highest for the group that received the highest nitrate dose , and this effect decreased in a dose-dependent manner ( Figure 2c ) . In addition , nitrate reduced irradiated parotid gland weight in a dose-dependent manner , and there was no significant difference between parotid gland weights in the highest-dose group ( 2 mmol/kg∙day ) and the sham group ( Figure 2d ) . Several salivary constituents ( amylase , calcium , and potassium ) were significantly altered following nitrate administration , particularly at 4 months post-IR ( Figure 2—figure supplement 1c , e ) . At 4 months post-IR , hematoxylin and eosin ( H&E ) staining , and Masson staining , showed that most acinar cells were lost in parotid glands in the IR control group . Moreover , the ductal system was irregularly expanded , bent , or blocked , and exhibited extensive fibrosis , which was typical of IR-induced salivary gland damage ( Figure 2e ) . However , parotid gland morphology was generally maintained in the nitrate groups in a dose-dependent manner . In the highest-dose group ( 2 mmol/kg∙day ) , the morphology was nearly identical to that of the sham group ( Figure 2e ) . These results demonstrated that nitrate contributed to the preservation of function and morphology in irradiated parotid glands in a dose-dependent manner . Nitrate administration significantly increased the proliferation of acinar and ductal cells as determined by Ki67 expression ( Figure 3a ) . Furthermore , microvessel density ( MVD; measured by CD31 expression ) was significantly increased following nitrate administration , whereas the IR control group showed marked MVD loss ( Figure 3b ) . Aquaporin 5 ( AQP5 ) , a water channel essential for saliva secretion , was highly expressed in the acini of the nitrate group ( Figure 3c ) , further suggesting that the secretory function of the remaining parotid gland tissue was preserved in response to nitrate administration . Surprisingly , sialin expression was increased after nitrate administration ( Figure 3d ) , suggesting there is a close interaction between sialin and nitrate . Moreover , the results of nitrate administration to human parotid gland cells ( hPGCs ) in vitro confirmed that nitrate had a dose-dependent positive effect on sialin expression ( Figure 3e and f , Figure 3—source data 1 and Figure 3—source data 2 ) . Overall , these findings suggested that preventive nitrate administration protected the parotid gland against IR damage via nitrate-sialin interaction . Because sialin is known to facilitate the influx of nitrate and other essential cellular substances , we next investigated the roles of sialin in parotid gland cells . First , we observed nitrate administration on cell proliferation and apoptosis to hPGCs in vitro under post-IR ( 5 Gy ) culturing conditions or physiological conditions ( Figure 4a and Figure 4—figure supplement 1a ) . We found that nitrate administration prior to IR ( Figure 4a ) significantly increased cell proliferation ( Figure 4b and c ) , but little effect on cell apoptosis ( Figure 4d ) . Moreover , nitrate administration to hPGCs in vitro resulted in a dose-dependent increase in sialin expression ( Figure 4e and Figure 4—source data 1 ) . Consistently , nitrate administration to hPGCs under physiological conditions also exhibited increased cell proliferation and elevated sialin expression in a dose-dependent manner ( Figure 3f and Figure 4—figure supplement 1a , d ) . We knocked down ( with short interfering RNA; siRNA ) or overexpressed sialin ( Slc17a5 ) in hPGCs in vitro ( Figure 4f , and Figure 4—figure supplement 1e , h , Figure 4—figure supplement 1—source data 1 , Figure 4—figure supplement 1—source data 2 ) to decrease or increase nitrate influx in cells . We found that Slc17a5 knockdown led to significantly reduced cell proliferation , and Slc17a5 overexpression resulted in a small , but significant , increase in proliferation ( Figure 4g and h ) . These findings indicated that sialin-mediated nitrate transportation promoted hPGCs proliferation . Cell cycle analysis indicated that Slc17a5 knockdown prolonged the G1 phase , and shortened the S and G2/M phases ( Figure 4i ) , resulting in G1 to S phase arrest . These results suggested that sialin-mediated nitrate transportation promoted hPGCs proliferation through the regulation of cell cycle transitions . Furthermore , Slc17a5 knockdown resulted in increased apoptosis ( Figure 4j ) . Cells derived from the submandibular glands of Slc17a5 knockdown mice ( sgRNA two-cell embryo ) showed decreased nitrate secretion and exhibited a significantly lower proliferation rate that that of cells derived from wild-type ( WT ) mice ( Figure 4k and l ) ; this further supports the conclusion that sialin-mediated nitrate transportation promoted cell proliferation . We further investigated whether nitrate or sialin played a dominant role in IR protection . SialinH183R mutation has a defect in nitrate transportation; we overexpressed sialinH183R ( Figure 4m ) and found that sialinH183R could not increase cell proliferation similar to WT sialin ( Figure 4n ) . These results demonstrated that nitrate could increase sialin expression in parotid gland cells , creating a nitrate-sialin feedback loop , and that sialin-mediated nitrate transportation is essential to maintain cell survival , promote parotid gland cell proliferation , and inhibit apoptosis . Sialin acts as a transporter in IR protection in which nitrate plays a dominant role . It is the sialin-meditated nitrate transportation that protects salivary glands from IR damage . IR of hPGCs significantly impaired proliferation ( Figure 5a–c ) , and increased apoptosis ( Figure 5d ) . Sialin expression progressively and significantly decreased in hPGCs over time following IR , which showed that sialin is markedly affected by IR ( Figure 5e and Figure 5—source data 1 ) . Slc17a5 knockdown with siRNA and subsequent 5 Gy IR ( Figure 5—figure supplement 1a , b , Figure 5—figure supplement 1—source data 1 ) significantly impaired cell proliferation ( Figure 5—figure supplement 1c , d ) and increased apoptosis ( Figure 5—figure supplement 1e ) . Conversely , Slc17a5 overexpression ( Figure 5e , f , g , Figure 5—source data 2 ) that facilitated nitrate transportation preserved cell proliferation ( Figure 5h and i ) and counteracted IR-induced cell apoptosis ( Figure 5j ) . These results suggested that reduced sialin expression worsened IR damage , and that maintenance of sialin expression for nitrate transportation could promote cell proliferation and inhibit IR-induced apoptosis . After nitrate administration , EGFR was significantly upregulated , which indicated that the EGFR signaling pathway was activated ( Figure 6a and Figure 6—source data 1 ) . Protein kinase B ( AKT ) and mitogen-activated protein kinase ( MAPK ) are downstream effectors of the EGFR pathway . They participate in maintenance of cell survival and promotion of cell proliferation , respectively . Cell proliferation is triggered by the MAPK pathway through activation of extracellular regulated protein kinase ( ERK ) . We found that nitrate administration increased phosphorylation of EGFR , AKT , and ERK ( Figure 6a and Figure 6—source data 1 ) . Blockade of EGFR attenuated nitrate-mediated promotion of cell proliferation ( Figure 6b ) . Moreover , blockade of EGFR inhibited phosphorylation of AKT and ERK; nitrate administration reversed this effect ( Figure 6c and Figure 6—source data 2 ) , which indicated that EGFR was upstream of both the AKT and MAPK signaling pathways . Notably , knockdown of sialin prevented nitrate-induced activation of EGFR suggesting that nitrate-mediated regulation of the EGFR signaling pathway required sialin ( Figure 6a and Figure 6—source data 3 ) . Nitrate could be metabolized into nitrite ( NO2− ) and nitric oxide ( NO ) . PTIO ( 2-Phenyl-4 , 4 , 5 , 5-tetramethylimidazoline-3-oxide-1-oxyl ) is the NO scavenger . Sialin expression was decreased after NO was blocked by PTIO ( Figure 6d and Figure 6—source data 4 ) . We also found that EGFR phosphorylation was decreased after NO scavenging ( Figure 6d and Figure 6—source data 4 ) . We have demonstrated that nitrate increases sialin expression and EGFR phosphorylation via the NO pathway and thus regulates EGFR-AKT-MAPK signaling pathway . Moreover , we have also observed significantly elevated phosphorylation of EGFR , AKT , and ERK in the nitrate group in vivo compared with the IR control and sham groups that were consistent with our in vitro study ( Figure 6e and Figure 6—source data 5 ) . These results suggested that a nitrate-sialin feedback loop could mediate cell proliferation to protect the parotid gland from IR damage via the EGFR–AKT–MAPK signaling pathway ( Figure 6f ) . In this study , we demonstrated that dietary nitrate administration protected the parotid gland against IR damage in a dose-dependent manner , by using an established miniature pig model of IR-induced parotid gland hypofunction . We found that nitrate supplementation enhanced hPGCs cell proliferation via EGFR–AKT–MAPK signaling pathway . Therefore , dietary nitrate administration may be an effective novel approach for the prevention of IR-induced xerostomia . In the present study , nitrate increased serous acinar and ductal cell proliferation rates , and increased MVD . Each of these parameters is crucial for maintenance of salivary gland homeostasis and restoration of function ( Aure et al . , 2015; Ekström et al . , 2017; Weng et al . , 2018 ) . We showed a close interaction between nitrate and sialin in vivo using a miniature pig model and in vitro using hPGCs , demonstrating a nitrate-sialin feedback loop in which the nitrate increased sialin expression and sialin facilitated nitrate influx into cells . It is the sialin-mediated nitrate transportation that plays a critical role in maintaining proliferation and survival of parotid gland epithelial cells; nitrate plays a dominant role in preventing IR damage and sialin acts as a nitrate transporter . Preventive nitrate administration increased sialin expression , promoted acinar and ductal cell proliferation , and reduced apoptosis via the EGFR–AKT–MAPK signaling pathway . However , nitrate administration could not reverse IR-induced damage when sialin expression was sharply reduced . These findings explained why preventive , but not therapeutic , nitrate administration was effective in our in vivo large animal study . Several treatments have shown beneficial effects against IR-induced xerostomia , such as antioxidants ( e . g . , amifostine ) , gene transfer of human aquaporin 1 cDNA ( hAQP1 ) ( Baum et al . , 2017; Baum et al . , 2009; Delporte et al . , 1997; Gao et al . , 2011; Shan et al . , 2005; Vitolo and Baum , 2002 ) , fibroblast growth factor-2 ( Guo et al . , 2014 ) , or sonic hedgehog ( Hu et al . , 2018 ) , intraperitoneal injection of the immune inhibitor rapamycin ( Zhu et al . , 2016 ) , stem cell transplantation , and salivary gland regeneration . Clinical trials showed positive effects of hAQP1 gene transfer therapy ( Alevizos et al . , 2017; Baum et al . , 2012; Baum et al . , 2010 ) , which exhibited beneficial effects in patients with established xerostomia ( i . e . , when the salivary glands already were severely damaged ) , but did not totally resolve IR induced xerostomia . However , cost , uncertain efficacy , potential side effects , tolerance issues , immunogenicity , and/or ethical issues have delayed translation of these therapeutic strategies to the clinic ( Grundmann et al . , 2009; Vissink et al . , 2015 ) . Our study showed that administration of inorganic nitrate significantly blunted IR-induced parotid gland hypofunction when administered prior to IR . In miniature pigs , nitrate administration resulted in preservation of approximately 85% of secretory function and maintenance of nearly normal morphology at the highest dose provided in this study ( 2 mmol/kg∙day ) . In addition , nitrate administration contributed to maintenance of levels of saliva constituents ( e . g . , amylase ) . Furthermore , inorganic nitrate lacks immunogenicity , is not subject to tolerance , is not toxic , and does not induce significant side effects at the doses employed ( Omar et al . , 2012 ) . Although nitrate is a natural substance present in the human diet , it has been regarded as a potential carcinogen for decades , and is generally believed to be harmful . However , recent studies failed to provide evidence for this assumption ( Bryan et al . , 2012; Hezel et al . , 2015; Khambata et al . , 2017; McNally et al . , 2016; Wu et al . , 2013; Xie et al . , 2016 ) . In 2011 , the World Health Organization published a guideline regarding nitrate and nitrite in drinking water which stated that nitrate is not carcinogenic , based on both laboratory animal studies and epidemiological studies ( Organization , 2011 ) . Nitrate occurs in two forms: organic and inorganic . Dietary nitrate is in the inorganic form ( Xia et al . , 2015 ) . Although organic nitrate has been used for the treatment of cardiovascular disease ( Omar et al . , 2016 ) , it has several notable disadvantages that limit its application in the clinic ( Omar et al . , 2012 ) , which include poor absorption , low oral bioavailability due to first-pass metabolism in the liver , better effectiveness via sublingual , transdermal , or rectal administration , short half-life , development of tolerance resulting from long-term administration , and side effects such as headache and postural hypotension . Conversely , inorganic nitrate has a simple ionic structure that is easily absorbed , exhibits high oral bioavailability , has a long half-life , and causes no side effects ( Khambata et al . , 2017; Qin et al . , 2012; Xia et al . , 2015 ) . Therefore , inorganic nitrate appears to comprise a safe and effective treatment for IR-induced salivary gland hypofunction . The best dose of sodium nitrate used in the miniature pig in the in vivo study was 2 mmol/kg·day . The coefficient of drug administration from pigs to humans was 0 . 73; thus , the dose converted to human was 1 . 46 mmol/kg·day . For example , an adult weighing 60 kg might have to take 7 g sodium nitrate per day , which is considered a relatively high dose . However , the parotid gland of miniature pigs received 60 Gy IR doses in our study , and the period of irradiation lasted for only 5 days receiving 7 . 5 Gy each day . When HNC patients receive IR , the IR period lasts for two months and they receive only 2 Gy daily , and the total dose would not be given to the parotid gland as most doses are focused on the tumor . In this case , the parotid gland of patients with HNC could not receive as strong a dose as the experimental animals in this study . Taken together , although the nitrate intake from diet food was 3–7 mg/kg per day for clinical treatment ( an adult weighing 60 kg takes 0 . 18–0 . 42 g sodium nitrate per day ) ( Ashworth and Bescos , 2017 ) , it is considered that the dose for clinical treatment could be well tolerated by humans . Dietary nitrate can be metabolized into nitric oxide ( NO ) via the nitrate–nitrite–NO pathway ( Lundberg et al . , 2008 ) . NO is an important gaseous signaling molecule that participates in many physiological processes , including stimulation of reparative angiogenesis and reduction of oxidative stress ( Sengupta et al . , 2004; Sessa , 2009 ) . Under hypoxic or acidic conditions , NO synthesis primarily occurs via the nitrate–nitrite–NO exogenous pathway . Notably , IR induces a hypoxic and acidic environment within salivary glands . Dietary nitrate supplementation results in NO production and reduces hypoxia by inducing a prolonged increase in blood flow ( Lundberg et al . , 2008 ) , and increased gland microvascularization , which is beneficial for production of saliva by acinar cells . Furthermore , nitrate-mediated NO formation also increases sialin expression and upregulates EGFR-AKT-MAPK signaling pathways , the classical pathways that are responsible for promoting cell proliferation , maintaining cell survival , and preventing cell apoptosis ( Sabbah et al . , 2020; Sun et al . , 2015; Zhang et al . , 2011 ) . However , nitrate is a stable substance that cannot be all metabolized to NO under IR , while sialin expression increases after nitrate administration , which helps sialin transport more nitrate into cells . It is considered that , except for NO , there are other mechanisms involved in preventing IR damage . It has been revealed that nitrate is related to mitochondrial respiration , activation of key metabolic regulatory pathways , and reduction of oxidative stress ( Lundberg et al . , 2011; Lundberg et al . , 2018 ) . Thus , more detailed mechanisms , including nitrate regulation of mitochondrial functions , should be the focus of future studies . In conclusion , nitrate administration preserves miniature pig parotid gland function and protects the parotid gland against IR damage . The underlying mechanism involves the sialin-mediated nitrate transportation that maintains salivary gland homeostasis and prevents IR damage via the EGFR–AKT–MAPK signaling pathways . Dietary nitrate is therefore a potential treatment modality for IR-induced salivary gland hypofunction and it appears worthwhile to test in clinical studies . Healthy miniature pigs ( 8–12 months of age , 40–60 kg body weight ) were purchased from the Institute of Animal Science of Chinese Agricultural University ( Beijing , China ) . All animals were maintained under conventional conditions with free access to water and food . Food stock ( 200–250 g , mixed with water ) was supplied twice per day , at 08:30 and 17:00 . Animal studies were conducted according to the NIH’s Guide for the Care and Use of Laboratory Animals , and approved by the Animal Care and Use Committee of Capital Medical University ( Beijing , China; approval no . AEEI-2015-098 ) . Parotid gland saliva and blood samples were collected at different time points ( Figure 1A and D; Figure 2A ) . Briefly , miniature pigs first received general anesthesia , after which pilocarpine ( 0 . 1 ml/kg body weight ) was injected intramuscularly to stimulate parotid gland saliva secretion . A modified Lashley cup was used to collect saliva for 10 min . Saliva flow rates were expressed as volume ( ml ) per 10 min per gland . Blood samples were simultaneously collected from the precaval vein . The collected saliva and blood samples were analyzed by standard clinical chemistry and—for blood—hematology procedures , as previously described ( Gao et al . , 2011; Zhu et al . , 2016 ) . Blood flow rate of the targeted parotid gland was measured after saliva collection . In the parotid gland area , three , aseptic 2 . 5 mm deep holes were made using an epidural anesthesia needle . Thereafter , local blood flow rate was measured aseptically using a 3-mm laser Doppler blood flow probe ( Moor Instruments Ltd , Axminster , UK ) for 3 min , at each of the three positions ( Guo et al . , 2014; Xu et al . , 2010 ) . Saliva and blood samples were immediately centrifuged at 500×g for 20 min at room temperature . Nitrate concentrations were detected using the Total Nitric Oxide and Nitrate/Nitrite Parameter Assay Kit ( PKGE001; R&D Systems , Minneapolis , MN ) , following standard experimental procedures provided by the manufacturer . Animals were sacrificed at the end points of our experimental designs ( Figure 4a and d; Figure 5a ) . Parotid glands were collected , fixed in 4% paraformaldehyde , dehydrated in gradient ethanol solutions , embedded in paraffin , and sectioned at 4 μm thickness . H&E staining was performed to analyze morphological changes . Immunohistochemical staining was employed to determine the expression of CD31 ( Abcam , Cambridge , UK , RRID:AB_726362 ) , Ki67 ( Abcam , RRID:AB_443209 ) , AQP5 ( Thermo Fisher Scientific , Waltham , MA , RRID:AB_2553573 ) , and sialin ( Thermo Fisher Scientific , RRID:AB_2577049 ) . CD31 and Ki67 were used to analyze the MVD and cell proliferation , respectively . AQP5 is expressed in apical membranes of acinar cells and plays a key role in saliva secretion . Thus , AQP5 expression was assessed to determine the saliva secretory function of the parotid gland on a histological level . To analyze cell apoptosis in parotid gland tissue , a terminal deoxynucleotidyl transferase dUTP nick-end labeling ( TUNEL ) in situ assay was performed , in accordance with the manufacturer’s instructions ( RiboBio , Guangzhou , China ) . In brief , parotid gland sections were incubated with proteinase K ( 20 µg/ml ) at 37°C for 30 min , and subsequently subjected to terminal deoxynucleotidyl transferase ( TdT ) and Alexa Fluor 567-conjugated EdUTP at 37°C for 1 hr . Nuclei were stained with Hoechst 33 , 342 ( RiboBio ) for 30 min . Stained sections were assessed by confocal microscopy ( Olympus , Tokyo , Japan ) . C57 mice were generated in which the gene for sialin ( Slc17a5 ) was knocked out by two-cell embryo microinjection ( sgRNA two-cell embryo ) , as previously described ( Wang et al . , 2017; Wu et al . , 2019 ) . Submandibular glands were dissected from the sgRNA two-cell embryo and WT mice; the gland cells were cultured in vitro , following the same procedure as for hPGCs described above . Animal studies were conducted according to the NIH’s Guide for the Care and Use of Laboratory Animals , and approved by the Animal Care and Use Committee of Capital Medical University ( Beijing , China; approval no . AEEI-2017-009 ) . Proliferation rates of hPGCs and cells derived from mouse submandibular glands were measured with a 5′-ethynyl-2′-deoxyuridine ( EdU ) Staining Kit ( RiboBio ) and with Cell Counting Kit-8 ( CCK-8; Dojindo , Shanghai , China ) respectively , in accordance with the manufacturers’ instructions . For the EdU assay , cells were incubated with EdU solution in a 24-well plate for 2 h , fixed in 4% paraformaldehyde at room temperature for 30 min , rinsed with PBS , and stained with Apollo solution ( RiboBio ) . Nuclei were stained with Hoechst 33 , 342 . Hoechst-stained cells and EdU-positive cells were counted using Image-Pro Plus 6 . 0 software ( Media Cybernetics , Bethesda , MD ) . For the CCK-8 assay , cells were incubated with 10 µl CCK-8 solution in a 96-well plate for 1 hr . Absorbance was measured at 450 nm by using a microplate reader ( BioTek Instruments , Winooski , VT ) ; the absorbance level at 12 hr served as baseline . For gain- and loss-of-function approaches , Slc17a5 was overexpressed or knocked down in hPGCs . In transfection experiments , a plasmid containing sialin full-length cDNA or sialinH183R mutant DNA , which did not transport nitrate ( Obio Technology , Shanghai , China ) was used for overexpression , whereas sialin siRNA ( Invitrogen , Carlsbad , CA ) was used for knockdown experiments . Empty plasmid ( Obio Technology ) and Scrambled Stealth siRNA ( Invitrogen ) served as the respective controls . Sialin siRNA nucleotide sequences were: siRNA-1 , TCCTGGAGGATATGTTGCCAGCAAA , and siRNA-2 , CATCACAAATACATTTGCCACTATT; the siRNA transfection protocol was previously described ( Feng et al . , 2013 ) . Transfected cells were irradiated and supplemented with nitrate . For the plasmid transfection experiments , Lipofectamine 3000 Reagent ( Invitrogen ) was used , in accordance with the manufacturer’s instructions . Briefly , Lipofectamine 3000 was diluted in Opti-MEM I Reduced Serum Medium ( Invitrogen ) in one tube and incubated at room temperature for 5 min . Plasmid was added to Opti-MEM I Medium ( Thermo Fisher Scientific ) in another tube . This diluted plasmid was then added to the diluted Lipofectamine 3000 solution and incubated at room temperature for 20 min . The final transfection complex was dripped gently onto plates with 70% confluent hPGCs , and the medium was changed 24 hr after transfection . Total RNA was extracted by using TRIzol Reagent ( Thermo Fisher Scientific ) , and the residual DNA was removed by RNase-Free DNase ( Promega , Madison , WI ) . Total RNA samples ( 1 μg ) were reverse transcribed to synthesize cDNA by AMV Reverse Transcriptase ( Promega ) , in accordance with the manufacturer’s instructions . RT-PCR was performed by using an ABI Prism 7000 Sequence Detection System ( Applied Biosystems , Thermo Fisher Scientific ) with SYBR Green Reagent ( Roche , Basel , Switzerland ) . The primer sequences used are listed in Table S1 in Supplementary file 1 . Relative gene expression levels were calculated using the comparative cycle threshold method ( 2−ΔΔCT ) and normalized to the level of β-actin . Total protein was extracted by radioimmunoprecipitation assay ( RIPA ) lysis buffer ( Applygen Technologies , Beijing , China ) , and the protein concentration was measured with a Bicinchoninic Acid ( BCA ) Protein Assay Kit ( Thermo Fisher Scientific ) . Total protein samples ( 20 μg ) were separated by 10% sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) , and transferred to polyvinylidene difluoride ( PVDF ) membranes ( Millipore , Beijing , China ) . Membranes were blocked by 5% skim milk and incubated with the following primary antibodies: rabbit anti-sialin ( Thermo Fisher Scientific , RRID:AB_2577049 ) , rabbit anti-phospho-EGF Receptor ( Tyr1068 ) ( Cell Signaling Technology [CST] , Shanghai , China , RRID:AB_2096270 ) , rabbit anti-EGF Receptor ( D38B1 ) ( CST , RRID:AB_2246311 ) , rabbit anti-phospho-AKT ( Ser473 ) ( CST , RRID:AB_2315049 ) , rabbit anti-AKT ( CST , RRID:AB_2246311 ) , rabbit anti-phospho-ERK1/2 ( Thr202/Tyr204 ) ( CST , RRID:AB_2315112 ) , rabbit anti-ERK1/2 ( CST , RRID:AB_390779 ) , and mouse anti-β-actin ( Abcam , RRID:AB_2305186 ) . Thereafter , membranes were incubated with goat anti-rabbit or anti-mouse secondary antibodies ( both Abcam ) and visualized by enhanced chemiluminescence ( BD Biosciences , Franklin Lakes , NJ ) . To analyze cell cycle and apoptosis , flow cytometric analysis was performed . Briefly , for cell cycle phase distribution analysis , cells were harvested , fixed in 70% ethanol at 4°C overnight , then washed with PBS and incubated with RNase A ( Sigma-Aldrich ) at room temperature for 30 min . Cells were then stained with propidium iodide ( Thermo Fisher Scientific ) and analyzed by flow cytometry . Cell apoptosis was detected by using the Annexin V-FITC Apoptosis Detection Kit ( BD Biosciences ) . In brief , cells were trypsinized and suspended with bonding buffer . After addition of Annexin V-FITC solution and incubation for 15 min , 5 μl propidium iodide was added , and analyzed by flow cytometry as above . The number of animals or cell cultures used is indicated for each experiment . All cell cultures were performed in duplicate . Data are presented as the mean ± SEM . A one-way ANOVA was used to analyze study results , assuming equal variances . The levels of statistical significance were as follows: p<0 . 05 , p<0 . 01 , **p<0 . 001; Δ<0 . 05 , ΔΔ<0 . 01 . Statistical analysis was performed using SPSS 18 . 0 software ( SPSS Inc , Chicago , IL ) .	Head and neck cancers are commonly treated using radiotherapy , where a beam of high-energy radiation is targeted at the tumour . This often severely damages the surrounding salivary glands , leading to chronic dry mouth and impairing a patient’s sense of taste , nutrient intake , speech and immune system . Despite this significant impact on quality of life , there is no effective treatment yet for this side effect . In the body , salivary glands are one of the primary users of a compound known as nitrate , which is commonly found in the diet . In the glands , it is ushered into cells thanks to a protein known as sialin . The nutrient supports the activity and maintenance of the glands , before it is released in the saliva . Feng , Wu et al . therefore decided to test whether nitrate could offer protection during neck and head radiotherapy . The experiments used miniature pigs , which have similar salivary glands to humans . The animals that received sodium nitrate before and after exposure to radiation preserved up to 85% of their saliva production . By comparison , without any additional nitrate , saliva production fell to 20% of pre-radiation levels . To understand how this protective effect emerged , Feng , Wu et al . added nitrate to cells from a human salivary gland known as the parotid . This led to the cells producing more sialin , creating a feedback loop which increases the amount of nitrate in the salivary glands . Further examination then showed that the compound promotes growth of cells and reduce their death . These findings therefore suggest that clinical studies may be worthwhile to test if nitrate could be used to prevent dry mouth in head and neck cancer patients who undergo radiotherapy .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2021	Dietary nitrate supplementation prevents radiotherapy-induced xerostomia
Both proteins and RNAs can misfold into non-functional conformations . Protein chaperones promote native folding of nascent polypeptides and refolding of misfolded species , thereby buffering mutations that compromise protein structure and function . Here , we show that RNA chaperones can also act as mutation buffers that enhance organismal fitness . Using competition assays , we demonstrate that overexpression of select RNA chaperones , including three DEAD box RNA helicases ( DBRHs ) ( CsdA , SrmB , RhlB ) and the cold shock protein CspA , improves fitness of two independently evolved Escherichia coli mutator strains that have accumulated deleterious mutations during short- and long-term laboratory evolution . We identify strain-specific mutations that are deleterious and subject to buffering when introduced individually into the ancestral genotype . For DBRHs , we show that buffering requires helicase activity , implicating RNA structural remodelling in the buffering process . Our results suggest that RNA chaperones might play a fundamental role in RNA evolution and evolvability . Protein chaperones can buffer the effects of mutations that affect protein stability and/or folding , as evidenced by the release of cryptic genetic variation upon inhibition of Hsp90 ( Rutherford and Lindquist , 1998; Queitsch et al . , 2002; Rohner et al . , 2013 ) , increased enzyme evolvability in Escherichia coli strains overexpressing GroEL ( Tokuriki and Tawfik , 2009 ) , accelerated rates of evolution in habitual chaperone clients ( Bogumil and Dagan , 2010; Warnecke and Hurst , 2010; Williams and Fares , 2010 ) , lower mutational penetrance in Caenorhabditis elegans larvae with higher Hsp90 titres during embryonic development ( Burga et al . , 2011 ) , and increased fitness of E . coli mutator strains following GroEL overexpression ( Fares et al . , 2002 ) . Whether RNA chaperones play a similarly pervasive role in buffering mutations that affect RNA structure or folding , however , has not been addressed empirically . RNA misfolding is common ( Herschlag , 1995 ) and frequently produces long-lived alternate structures ( Downs and Cech , 1996 ) that require the assistance of RNA-binding proteins for timely resolution . Like their protein-folding counterparts , RNA chaperones can promote orderly structural transitions towards and subsequently stabilize the native fold or—as exemplified by classic work on the Neurospora crassa CYT-19 protein ( Mohr et al . , 2002; Bhaskaran and Russell , 2007 ) —facilitate the refolding of misfolded species ( Russell , 2008 ) . Of particular interest in this regard are DEAD box RNA helicases ( DBRHs ) , which can alleviate folding errors by unwinding short RNA helices , thus enabling renewed exploration of the folding landscape ( Pan and Russell , 2010 ) . By the same token , DBRH activity may also counteract mutations that precipitate a deleterious increase in RNA stability . Many DBRHs , including RhlB ( a component of the E . coli degradosome [Py et al . , 1996] ) and the eukaryotic translation initiation factor eIF4A are involved in the folding and unfolding of diverse RNAs and might therefore act as broad-spectrum mutation buffers . We tested whether DBRHs buffer deleterious mutations in vivo using a fitness rescue paradigm ( Fares et al . , 2002 ) . Briefly , an E . coli strain propagated under conditions of weak selection is expected to accumulate deleterious mutations and experience a concomitant decline in fitness compared with its ancestor . Similarly , strains adapting to a novel environment will accumulate not only beneficial but also deleterious mutations , which may hitchhike along with beneficial alleles and can be exposed as deleterious in the old environment ( Fares et al . , 2002 ) . In both scenarios , buffering can be inferred if ( over ) expression of a candidate chaperone leads to a greater fitness gain in the low-fitness evolved strain compared with its ancestor . We therefore performed pairwise competition experiments ( Lenski , 1991 ) between the E . coli REL606 strain , which is the ancestor of the long-term evolution experiment ( LTEE ) , and two evolved mutS mutator strains that were sampled from a lineage after ∼20 , 000 ( 20k ) and ∼40 , 000 ( 40k ) generations of adaptation to a minimal glucose-limited medium ( Sniegowski et al . , 1997 ) ( Figure 1A ) . Performing 24-hr competitions in the alternative LB medium , we observed reduced fitness relative to the REL606 ancestor for the 40k but not for the 20k genotype ( Figure 1B ) . Reduced fitness in the 40k strain cannot be proximately attributed to the mutS mutation , which underlies the mutator phenotype; this mutation arose ∼3000 generations into the LTEE ( Sniegowski et al . , 1997 ) and is already present in the 20k strain . Thus , the mutation ( s ) responsible for reduced fitness emerged later . 10 . 7554/eLife . 04745 . 003Figure 1 . Relative fitness of Escherichia coli REL606-derived strains . ( A ) Relationships between strains used in different competition assays . Short names of competed strains are given in bold; gen . : generations . ( B ) Relative fitness of the 20k and 40k genotypes , each competed against their REL606 ancestor . ( C ) Relative fitness of ancestral and evolved genotypes overexpressing one of three DEAD box RNA helicases ( DBRHs ) compared with identical strains carrying the empty control plasmid . E166K , E157K , and E158K: competitions in the 40k background where plasmids carried mutated versions of the respective DBRH . In each case , the central glutamic acid residue of the DEAD motif has been recoded to lysine , compromising the helicase activity . Bar heights indicate mean relative fitness across four biological replicates , with each mean derived by averaging over four technical replicates . Error bars represent standard errors of the mean . *p < 0 . 01 , p < 0 . 05 ( one-sample t-test ) . Additional results for competitions terminated in mid-exponential phase ( after 2 hr ) are shown in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 00310 . 7554/eLife . 04745 . 004Figure 1—figure supplement 1 . Relative fitness in competition experiments terminated in mid-exponential phase ( REL606 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 004 Having established the presence , in the 40k strain , of deleterious mutations potentially amenable to buffering , we introduced plasmids carrying a specific E . coli DBRH gene ( either csdA , rhlB , or srmB ) into each of the three genetic backgrounds ( ancestor , 20k , and 40k ) . We then competed each transformed strain against a strain of the same genotype but bearing an empty control plasmid . Whereas overexpression in the ancestral and 20k backgrounds had limited effects on competitive fitness , overexpression of either DBRH enhanced fitness of the mutationally compromised 40k genotype ( Figure 1C ) . For each DBRH tested , fitness gains were abolished when we introduced mutations that rendered the respective helicase domain catalytically inactive ( Figure 1C ) , suggesting that helicase and therefore RNA remodelling activities are essential for buffering . Protein chaperones like GroEL and Hsp90 target misfolded substrates by recognizing exposed hydrophobic patches that are buried in the native state ( Hartl et al . , 2011 ) . This generic mechanism allows buffering to occur across a broad range of substrates , differentiating these chaperones from gene- or pathway-specific suppressors . To determine whether DBRH-mediated buffering encompasses diverse target substrates , we performed the same suite of fitness rescue experiments in a second , independently evolved strain . This MG1655-derived mutH deletion ( ΔmutH ) strain was sampled after a shorter period of laboratory evolution ( ∼500 generations ) but also exhibited reduced fitness compared with its ancestor ( Figure 2A ) and experienced fitness gains upon DBRH overexpression ( Figure 2B ) . We confirmed that fitness effects were not directly related to the mutH deletion by deleting mutH in the ancestral MG1655 background de novo . Fitness of the de novo ΔmutH strain was not reduced compared with the deletion-free ancestor ( Figure 2A ) . 10 . 7554/eLife . 04745 . 005Figure 2 . Relative fitness of Escherichia coli MG1655-derived strains . ( A ) Relative fitness of the evolved and de novo-constructed ΔmutH strains , each competed against their MG1655 ancestor . ( B ) Relative fitness of ancestral , evolved , and de novo ΔmutH genotypes overexpressing one of three DEAD box RNA helicases compared with identical strains carrying the empty control plasmid . E166K , E157K , and E158K , bar heights and error bars are as described in Figure 1 . *p < 0 . 01 , p < 0 . 05 ( one-sample t-test ) . Additional results for competitions terminated in the mid-exponential phase ( after 2 hr ) are shown in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 00510 . 7554/eLife . 04745 . 006Figure 2—figure supplement 1 . Relative fitness in competition experiments terminated in mid-exponential phase ( MG1655 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 006 To rule out buffering of identical mutations across strains , we sequenced the genomes of the evolved ΔmutH strain and its laboratory ancestor . As expected , we found significantly fewer mutations in the evolved ΔmutH compared with the 40k strain ( Table 1 , Supplementary files 1–3 ) . More importantly , there were no identical point mutations or indels in the two evolved strains , implying buffering of independent mutations . This might be indicative of a general rather than gene- or pathway-specific buffering mechanism and is consistent with DBRHs being broad-spectrum catalysts of RNA remodelling that recognize and target misfolded substrates through a non-specific mechanism of action ( Jarmoskaite et al . , 2014 ) . 10 . 7554/eLife . 04745 . 007Table 1 . Number of mutations in evolved mutator strains compared with their respective ancestorsDOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 007StrainSNPs ( CDS/total ) Small indels† ( CDS/total ) Large deletions20k667/75586/129140k1163/1291128/1834Evolved ΔmutH12/120/20‡CDS , coding sequence; SNP , single-nucleotide polymorphism . †≤4 bp . ‡Excluding the mutH deletion itself . To lay the groundwork for unravelling the molecular basis of buffering , we sought to identify mutations that are individually deleterious and whose effects on fitness are ameliorated by chaperone overexpression . In the absence of strong a priori candidates for DBRH-mediated buffering—neither strain harboured mutations in known structural RNAs ( Supplementary files 2 , 3 ) —the evolved ΔmutH strain , carrying only 12 point mutations ( Table 1 ) , affords us the rare opportunity to pinpoint such mutations systematically . Using a recombineering approach ( see ‘Materials and methods’ ) , we individually introduced each mutation located in a gene of known function ( N = 7 , five mutations are located in y-genes ) into the MG1655 genome . We then competed each of these strains , isogenic except for a single mutation , against MG1655 . Only one strain , carrying a mutation in the lamB gene , exhibited both reduced fitness and evidence for buffering ( Figure 3A ) . Although primarily known for its role in maltose uptake , lamB is a more general glycoporin that becomes derepressed under glucose-limiting conditions to maximize sugar uptake ( Death et al . , 1993 ) . LamB deletion mutants are outcompeted by reference strains when grown on glucose ( Death et al . , 1993 ) , suggesting a possible cause of fitness loss in our strain . Growth phase-related fitness patterns were mirrored by the evolved ΔmutH strain ( Figure 2—figure supplement 1 ) consistent with lamB being a dominant driver of fitness loss in this strain . 10 . 7554/eLife . 04745 . 008Figure 3 . Fitness effects and buffering of individual mutations . ( A ) Relative fitness of strains carrying single point mutations introduced into the relevant ancestral background competed against the respective ancestor . The mutations correspond to those listed in Supplementary files 2 , 3 for the respective genes . Initial screening for fitness defects involved two biological replicates ( grey diamonds ) . For the two mutations where the initial screen suggested a measurable fitness deficit , lamB and rplSsyn , all competitions were carried out in quadruplicate . Bar heights and error bars are as described in Figure 1 . *p < 0 . 01 , p < 0 . 05 ( one-sample t-test ) . OE: overexpression; EP: empty plasmid . Additional results for competitions terminated in mid-exponential phase ( after 2 hr ) are shown in Figure 3—figure supplements 1 , 2 . ( B ) Linear Feynman graph of the lamB region that harbours the mutation in the evolved ΔmutH strain ( highlighted in grey ) . We predicted RNA secondary structure for the entire malK-lamB-malM transcription unit ( RegulonDB identifier: ECK120009315 ) and its mutated counterpart using RNAfold ( Lorenz et al . , 2011 ) . The malK-lamB-malM operon contains a repetitive extragenic palindromic ( REP ) element downstream of lamB , which prevents premature degradation of the lamB cistron following cleavage from malM . Resolution of this REP element as part of regulated degradation was previously shown to require RhlB ( Khemici and Carpousis , 2004 ) . However , comparison of predicted minimum free energy structures between wild type and mutant malK-lamB-malM transcripts suggested structural changes that do not interfere with REP element formation but rather lead to decreased positional entropy at the local level , as highlighted here . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 00810 . 7554/eLife . 04745 . 009Figure 3—figure supplement 1 . Relative fitness in competition experiments terminated in mid-exponential phase ( lamB ) . OE: overexpression; EP: empty plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 00910 . 7554/eLife . 04745 . 010Figure 3—figure supplement 2 . Relative fitness in competition experiments terminated in mid-exponential phase ( rplSsyn ) . OE: overexpression; EP: empty plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 010 For the 40k strain , where the large number of mutations precludes comprehensive analysis , we focused on two mutations in the essential ribosomal protein gene rplS—one synonymous ( rplSsyn ) and one non-synonymous ( rplSnonsyn , Supplementary file 2 ) . Both mutations are present in the 40k but not in the 20k strain . As demonstrated for rplA and rpsT in Salmonella typhimurium ( Lind et al . , 2010 ) , synonymous mutations in ribosomal protein genes can strongly compromise fitness , indicating the presence of selective constraints unrelated to amino acid composition . Moreover , fitness costs of mutations in rplA/rpsT correlated with changes in predicted mRNA free energy , albeit weakly ( Lind et al . , 2010 ) . We therefore performed competitions between the REL606 ancestor and strains carrying either the rplSsyn or rplSnonsyn mutation . These competitions revealed a deleterious effect of rplSsyn , buffered by DBRH overexpression ( Figure 3A ) . In contrast to lamB , growth phase-specific buffering patterns in rplSsyn did not echo observations in the 40k strain ( Figure 3—figure supplement 2; Figure 1—figure supplement 1 ) , consistent with the presence of multiple fitness-relevant mutations in the latter . These results establish that RNA chaperones buffer individual deleterious mutations although the mechanisms of buffering remain , at this point , unresolved . We consider likely mechanisms of buffering and appropriate experimental follow-ups in the ‘Discussion’ . Buffering of mutations that affect RNA structure and folding may be mechanistically diverse , rather than limited to a DBRH model where helicase activity catalyses the local rupture of helices and enables structural remodelling . To explore mechanistic diversity in buffering , we considered the cold shock protein CspA . By binding with low specificity to single-stranded RNA , CspA can prevent the formation of unwanted secondary structure and thereby narrow the RNA folding landscape ( Jiang and Hou , 1997 ) —a mechanism of action reminiscent of the ubiquitous protein chaperone DnaK ( Hsp70 ) , which cycles on and off nascent polypeptide chains to allow ordered stepwise folding ( Hartl et al . , 2011 ) . We found that overexpression of CspA is associated with fitness gains in both low-fitness genotypes but not in the corresponding ancestral strains ( Figure 4 ) . By contrast , overexpression of a mutant version of CspA with severely reduced nucleic acid-binding activity ( Hilier et al . , 1998 ) did not confer fitness benefits upon overexpression ( Figure 4 ) . Buffering occurs although CspA levels are relatively low compared with overexpressed DBRHs ( ∼fourfold and ∼twofold reduced relative abundance compared with CsdA and RhlB/SrmB , respectively , Figure 5 ) , likely because CspA is subject to negative autoregulation ( Bae et al . , 1997 ) . 10 . 7554/eLife . 04745 . 011Figure 4 . Effects of CspA overexpression on relative fitness . Relative fitness of REL606- and MG1655-derived strains overexpressing CspA compared with strains of the same genotype carrying the empty control plasmid . F20L: competitions in the 40k and evolved ΔmutH backgrounds , respectively , where plasmids carried a mutated version of the cspA gene yielding a protein with compromised nucleic acid binding ability ( Hilier et al . , 1998 ) . Bar heights and error bars are as described in Figure 1 . *p < 0 . 01 , p < 0 . 05 ( one-sample t-test ) . Additional results for competitions terminated in mid-exponential phase ( after 2 hr ) and competitions involving the lamB mutant in the evolved ΔmutH strain and the rplSsyn mutant in the 40k strain are shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 01110 . 7554/eLife . 04745 . 012Figure 4—figure supplement 1 . Relative fitness in competition experiments terminated in mid-exponential phase ( cspA ) . OE: overexpression; EP: empty plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 01210 . 7554/eLife . 04745 . 013Figure 5 . Relative chaperone abundances . ( A ) Representative Western blot for evolved ΔmutH strains overexpressing one of the focal RNA chaperones . Molecular weights ( from nucleotide sequence ) : CspA , 7 . 403 kD; RhlB , 47 . 126 kD; SrmB , 49 . 914 kD; CsdA , 70 . 546 kD . ( B ) Representative Coomassie-stained SDS-PAGE gel . ( C ) Relative chaperone levels are defined as the ratio of Western blot intensity to Coomassie intensity ( see ‘Materials and methods’ ) . The lowest ratio detected across triplicate experiments in all strains was set to one . Comparing these ratios between strains overexpressing different RNA chaperones gives a semi-quantitative indication of relative chaperone abundances . For example , CsdA levels in CsdA-overexpressing cells are ∼fourfold higher than CspA levels in CspA-overexpressing cells . Note that this metric does not allow conclusions about the absolute fraction of total protein that is occupied by each chaperone in the different strains . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 013 These results show that buffering can be mediated by RNA chaperones that interact with their substrates in mechanistically distinct ways . Interestingly , cold shock proteins and DBRHs have been suggested to work together , with helicases opening double-stranded structures and cold shock proteins binding to the single-stranded end products , their combined activities preventing the formation of unfavourable structures ( Hunger et al . , 2006 ) . Overexpression of different RNA chaperones , including the proteins considered here , might therefore benefit identical RNA target species . Elucidating in greater detail how different chaperones interact with wild-type and mutant RNAs will be critical to elucidate the molecular basis of buffering and how it relates to altered RNA secondary and tertiary structure . Here , we demonstrate that buffering by RNA chaperones occurs at the organismal level , establish that helicase and nucleic acid-binding activity are required for buffering by DBRHs and CspA , respectively , and identify individual mutations that are amenable to buffering . These mutations , located in lamB and rplS , constitute valuable assets to establish the precise molecular mechanism ( s ) of buffering in the future . Although this study was not designed to resolve molecular mechanism , it is nonetheless useful , if primarily to guide future work , to contemplate potential causes of fitness loss , and how buffering might occur . In particular , we wanted to know whether mutations in lamB and rplS stand out amongst other mutations in their predicted effect on RNA structure . Several metrics that quantify mutational impact on RNA secondary structure do indeed suggest comparatively severe effects for rplSsyn ( e . g . , the correlation of base pairing probabilities , Supplementary file 2 ) . Similarly , of all the mutations in the evolved ΔmutH strain , the lamB mutation is predicted to have the most severe repercussions for local RNA structure as measured by maximum local base pair distance ( dmax , Supplementary file 3 , Figure 3B ) . However , these correlates should be interpreted with caution . Existing measures of RNA structural change , however accurate , will only indicate how disruptive a given mutation is to the RNA structure but not whether the resulting defect can be rescued by chaperone activity , or whether it matters at the organismal level . Consequently , we would not necessarily anticipate a robust relationship between fitness loss , buffering , and indicators of RNA structural change . More mundanely , we cannot quantify the reliability of any ( structural ) predictor without a larger set of experimentally characterized mutations that are both deleterious and amenable to buffering . In short , while suggestive , these findings do not conclusively implicate RNA structure as the vehicle for fitness loss . Provided deleterious effects arise at the level of RNA structure , RNA chaperone activity might be beneficial through stabilizing ( CspA ) , destabilizing , or remodelling ( DBRHs ) affected structures in the focal transcript or through changing how these transcripts interact with other RNAs or RNA-binding proteins ( Pan and Russell , 2010 ) . Considering lamB , one possibility is that an increase in local stability caused by the mutation ( Figure 3B ) introduces a translational roadblock that is resolved by RNA chaperone activity . Alternatively , one might envisage a more complex scenario , where the non-synonymous lamB mutation gives rise to a dominant negative protein product and RNA chaperone overexpression ameliorates fitness defects by facilitating degradation of the mRNA , thus reducing levels of mutant LamB protein . Both of these mechanisms are speculative , and endorsing either one ( or a third or fourth option ) would be premature . Rather , competing mechanistic hypotheses will have to be confirmed or debunked through targeted follow-up experiments . For example , a logical first step to eliminate one of the hypotheses above would be to measure LamB levels ( predicted to increase and decrease upon chaperone overexpression , respectively ) . In designing insightful follow-up studies , a few issues deserve wider consideration . First , buffering might be direct ( involving interactions between the RNA chaperones and the mutant RNA ) or indirect ( involving interactions between the RNA chaperone and other components of the cell , which in turn lead to buffering ) . In addition to dissecting specific chaperone–RNA interactions , it will therefore be important to establish system-level effects of RNA chaperone overexpression . Second , both synonymous and non-synonymous mutations can affect RNA structure so that non-synonymous mutations ( like the one found in lamB ) should not be discarded a priori as unlikely candidates for buffering by RNA chaperones; deleterious consequences might arise at the RNA level even though the amino acid change is selectively neutral . Conversely , synonymous mutations can affect translation kinetics and protein folding ( Plotkin and Kudla , 2011 ) and therefore have fitness repercussions at the protein level . As a corollary , some mutations might be amenable to buffering by both protein and RNA chaperones . For example , the former might rescue misfolded proteins , whereas the latter removes translational roadblocks that predispose to misfolding . Elucidating to what extent protein and RNA chaperones have orthogonal buffering capacities will therefore be an important future objective . Third , as is the case for protein chaperones , buffering by RNA chaperones is almost certain to occur through a range of mechanisms , so that identifying general principles of buffering in the face of mechanistic plurality will be a key challenge . Our findings should provide a strong impetus to meet this challenge and embark on further investigations with the ultimate aim to unravel the ramifications of mechanistically diverse chaperoning activity for RNA biogenesis , evolution , and evolvability . Encouragingly , recent advances in probing RNA secondary structures and RNA–protein interactions at high throughput have rendered the transcriptome-wide dissection of RNA chaperone-mediated buffering a realistic prospect for the not too distant future . The strains used here were derived either from the E . coli K12 MG1655 strain by laboratory evolution and P1 transduction and/or transformation , or from the REL606 strain and its descendants in the LTEE ( Sniegowski et al . , 1997 ) . All strains used are listed in Supplementary file 4 , and their relationships and experimental derivation are illustrated in Figures 6 , 7 . Sequences of cspA , rhlB , srmB , and csdA inserted into pCA24N::Cam were obtained from the ASKA collection ( http://www . shigen . nig . ac . jp/ecoli/strain/ ) . For strain construction and subsequent experiments , bacteria were grown in LB at 37°C . To distinguish competitors during competition assays , cells were plated onto TA solid medium ( Lenski , 1991 ) . 10 . 7554/eLife . 04745 . 014Figure 6 . Relationship between REL606-derived strains . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 01410 . 7554/eLife . 04745 . 015Figure 7 . Relationship between MG1655-derived strains . *Resequencing of the MG1655 laboratory strain revealed a single difference to the NC000913 reference genome , an intergenic dinucleotide insertion at position 4296380 ( AC → ACGC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04745 . 015 Mutations in the DEAD domain of E . coli DBRHs abolish or severely reduce helicase activity , as demonstrated for the E . coli DBRHs RhlB ( Vanzo et al . , 1998 ) , DbpA ( Elles and Uhlenbeck , 2008 ) , and CsdA ( Turner et al . , 2007 ) . Here , we used DBRH mutants in which the central glutamic acid residue has been recoded to yield lysine , a change known to abolish RhlB ATPase activity , which is required for helicase activity ( Vanzo et al . , 1998 ) . Mutations in the nucleic acid-binding domain of CspA were previously evaluated for their impact on both nucleic acid binding and protein stability ( Hilier et al . , 1998 ) . The F20L mutation was found to only weakly affect protein stability but strongly reduce nucleic acid binding ( Hilier et al . , 1998 ) and was therefore chosen for our study . Plasmids carrying the mutated genes were constructed by and purchased from DNA2 . 0 . We performed pairwise competition experiments to estimate the relative fitness of two competing E . coli strains as previously described ( Lenski , 1991 ) . Briefly , the two competitors were grown separately , mixed at an initial ratio of 1:1 and diluted 100-fold in the competition environment ( LB supplemented with the relevant antibiotics for plasmid maintenance ) . Initial and final densities ( after either 24 or 2 hr ) were estimated by diluting and spreading the cells on indicator TA ( tetrazolium and arabinose ) plates , which allow the competitors to be distinguished through an arabinose-utilization marker , which is neutral under the conditions utilized ( Lenski , 1988 ) . Relative fitness was calculated asw=ln ( Af/Ai ) /ln ( Bf/Bi ) , where A and B are the densities of the two competitors , and i and f represent initial and final densities , respectively . We used one-sample t-tests to evaluate whether mean relative fitness differed from the null expectation of one . DNA was extracted from the MG1655 ancestor and the evolved ΔmutH strain using the standard phenol–chloroform extraction procedure followed by ethanol precipitation . Libraries ( 100bp paired end reads ) were prepared using the TruSeq DNA PCR-Free LT Sample Prep Kit ( Illumina , San Diego , CA ) , with median insert size of 487 bp ( ancestor ) and 495 bp ( evolved ΔmutH ) , for subsequent sequencing on the Illumina HiSeq 2000 platform using the TruSeq v3 reagent kit . Sequencing yielded 7365202 and 6504588 read pairs for the ancestral and evolved strains , respectively . The genomes of the REL606-derived 20k ( REL8602A ) and 40k ( REL10953 ) clones were previously sequenced on the Illumina Genome Analyzer platform using a single lane of single-end 36bp reads per genome . To identify genomic differences in the MG1655-derived strains , we closely followed the approach adopted by Tenaillon et al . ( 2012 ) . To detect single-nucleotide polymorphisms ( SNPs ) and short indels , paired end reads were first aligned to the reference genome ( NC000913 ) using bwa mem ( version 0 . 7 . 6a , arXiv:1303 . 3997 ) . Subsequently , duplicate reads were removed using samtools ( version 0 . 1 . 19 ) , and only those with non-zero mapping quality and no suboptimal alignments were considered further . SNPs and indels were then called using the mpileup function in samtools , requiring a minimum base quality of 30 . Mutations were validated by visual inspection of read mappings in the Integrated Genome Viewer . Non-reference alleles had to be present at a frequency of p > 0 . 75 to be considered bona fide mutations . To ensure that we did not miss pertinent SNPs/indels in non-unique regions present in structural RNAs , we repeated the above procedure without filters on mapping quality and including suboptimal alignments . We identified no additional candidate mutations . Screening for larger deletions and duplications was performed by computing per-base coverage using the genomeCoverageBed function in bedtools ( version 2 . 17 . 0 ) ( Quinlan and Hall , 2010 ) , subsequent smoothing across 250bp windows , normalizing by GC content , and applying a rolling filter to control for long-range effects , as previously described ( Tenaillon et al . , 2012 ) . This approach confirmed the mutH deletion in the mutator strain . We found no other large deletions . Similarly , no duplications were detected after visually inspecting neighbouring regions with unusual differences in coverage ( >1 . 5-fold ) . All candidate regions were found to be due to local drop-offs in neighbouring regions rather than excess coverage in focal windows . For REL606-derived strains , candidate genomic differences compared with the ancestral genome were identified using the PALOMA ( Vallenet et al . , 2012 ) and BRESEQ ( Barrick et al . , 2009 ) pipelines . Sequencing data for the MG1655-derived strains and the 20k strain have been deposited in the European Nucleotide Archive ( accession no . PRJEB7107 ) . Sequencing data for the REL606-derived 40k clone was previously deposited in the National Center for Biotechnology Information Sequence Read Archive ( accession no . SRR1536189 ) . All point mutations detected in the evolved ΔmutH strain that were located in a gene of known function ( i . e . , not located in a y-gene ) were individually introduced into the ancestral MG1655 background using a recombineering approach ( Sawitzke et al . , 2013 ) . Briefly , we designed single-stranded oligonucleotides , ∼70 nt in length , that were complementary to the region of interest and carried the desired point mutation ( Supplementary file 5 ) . They were transformed by electroporation into the TB56 ( Ara+ ) and TB62 ( Ara− ) strains supplemented with pSIM6 . In both TB56 and TB62 ( kindly provided by Tobias Bergmiller ) , the native mutS promoter has been replaced with an ara promoter . In the presence of 0 . 2% arabinose , the strains have wild-type MutS levels , whereas in the presence of 0 . 2% glucose , MutS expression is repressed . Growing strains in LB medium supplemented with 0 . 2% glucose therefore increases the likelihood that oligo-born mutations are fixed due to impaired mismatch repair . Following electroporation , cells were grown overnight at 32°C on LB agar plates supplemented with 0 . 2% arabinose , and the presence of mutations was confirmed by sequencing target regions from individual colonies on an ABI PRISM 310 Genetic Analyzer using the Big Dye Terminator 1 . 1 Cycle Sequencing kit ( Life Technologies , Carlsbad , CA ) . The primers used for sequencing are listed in Supplementary file 5 . Exponentially growing evolved ΔmutH strains overexpressing one of the four RNA chaperone proteins ( CspA , RhlB , SrmB or CsdA ) were pelleted by centrifugation and resuspended in UTCDTT buffer containing 8 M urea , 2 M thiourea , 4% CHAPS and 10 mM DTT supplemented with a mixture of protease inhibitors containing aprotinin bestatin , leupeptin , pepstatin A , E-64 and AEBSFxHCL , EDTA-free ( Life Technologies ) . Cells were incubated for 2 hr at room temperature followed by a 20-min centrifugation step at 12 , 000×g . Protein concentrations were determined using the Bradford assay ( Bradford , 1976 ) . For each sample , 15 μg of total protein extract was loaded onto an SDS-PAGE gel with a 6% stacking and 20% resolving gel . 6X-His tagged CspA , RhlB , SrmB , and CsdA were detected by Western blotting using a mouse monoclonal anti-6X His tag antibody ( Abcam , United Kingdom ) followed by a goat anti-mouse polyclonal antibody conjugated to HRP ( Abcam ) . Proteins were visualized on autoradiographic film using the Amersham ECL Advance chemiluminescence detection system ( GE Healthcare Life Sciences , United Kingdom ) . We used ImageJ ( Collins , 2007 ) to quantify the intensity of each chaperone band on the Western blot and normalized this intensity by the amount of total protein loaded into each lane ( detected by Coomassie staining of the SDS-PAGE gel and subsequent quantification with ImageJ ) . This normalized abundance allows comparing relative chaperone levels across experiments ( Figure 5C ) .	Stretches of DNA known as genes contain the instructions to make the proteins and RNA molecules that are essential for life . The DNA sequence of the gene is first copied to make a strand of RNA , which may subsequently be ‘translated’ to make a protein . To carry out their tasks , proteins and many RNA molecules must fold into specific three-dimensional structures . Since folding can easily get derailed , proteins known as chaperones assist with this process . Mutations sometimes occur in the DNA that reduce the ability of the proteins or RNA molecules to fold correctly . Previously , scientists had found that some chaperones help incorrectly folded proteins adopt ‘normal’ shapes and thus mask the harmful effects of mutations . However , it was not known whether the chaperones that fold RNA similarly suppress harmful mutations . To address this question Rudan et al . studied the effects of several RNA chaperones in Escherichia coli bacteria that had been grown in the laboratory as part of long-term evolution experiments . During this time , they had accumulated mutations that reduced the fitness of later generations in comparison with their ancestors . Rudan et al . then found that increasing the levels of certain RNA chaperones—in particular , a group called DEAD box RNA helicases—in the evolved bacteria improved their fitness . This strongly suggests that RNA chaperones , like protein chaperones , can suppress harmful mutations . Compromised versions of the same RNA chaperones , which were unable to dismantle folded RNA structures , did not show any improvements in fitness , demonstrating that the capacity to unfold and refold RNA is critical . Rudan et al . suggest that different types of chaperones are likely to alleviate RNA mutations using different mechanisms . A future challenge will therefore be to work out how these mechanisms work together to mask different mutations and allow them to persist through evolution , their harmful effects rendered invisible to the forces of natural selection .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	RNA chaperones buffer deleterious mutations in E. coli
Antibody class switching is a feature of the adaptive immune system which enables diversification of the effector properties of antibodies . Even though class switching is essential for mounting a protective response to pathogens , the in vivo patterns and lineage characteristics of antibody class switching have remained uncharacterized in living humans . Here we comprehensively measured the landscape of antibody class switching in human adult twins using antibody repertoire sequencing . The map identifies how antibodies of every class are created and delineates a two-tiered hierarchy of class switch pathways . Using somatic hypermutations as a molecular clock , we discovered that closely related B cells often switch to the same class , but lose coherence as somatic mutations accumulate . Such correlations between closely related cells exist when purified B cells class switch in vitro , suggesting that class switch recombination is directed toward specific isotypes by a cell-autonomous imprinted state . The human immune system’s antibody repertoire provides broad protection against pathogen infection . The variable regions of antibodies have been the subject of intense study due to their central role in determining the amazing breadth of molecular recognition in the antibody repertoire . However , the constant regions of antibodies also display quite dynamic behavior through the phenomenon of class switching , which is also known as isotype switching . Different classes of antibodies with distinct Fc domains mediate specialized effector functions , including activation of complement , phagocytosis , cytotoxicity , and release of inflammatory mediators ( Kindt et al . , 2007 ) . The diversification of antibody functionality via class switching is essential for mounting a protective response to different pathogens . Conversely , dysregulation of antibody class switching has been implicated in autoimmune diseases , including allergic hypersensitivity ( Sugai et al . , 2003 ) , rheumatoid arthritis ( Humby et al . , 2009 ) , systemic lupus erythematosus ( Bubier et al . , 2009; Mietzner et al . , 2008 ) , IgG4-related disease ( Stone et al . , 2012 ) , and hyperimmunoglobulin E syndrome ( Minegishi , 2009 ) . Class switching occurs during germinal center maturation and is linked to cell division and somatic hypermutation ( Hodgkin et al . , 1996; Liu et al . , 1996; Tangye et al . , 2002 ) . After antigen encounter , IgM+ and IgD+ naïve B cells can switch to expression of activated classes IgG , IgA , and IgE via genomic recombination of the immunoglobulin heavy chain constant region locus . Much of current knowledge about the mechanisms of class switching is derived from the analysis of B cells induced to undergo class switch recombination ( CSR ) in vitro . However , the patterns of antibody class switching in the natural setting within a living organism have remained largely uncharacterized . How switch recombination is directed to distinct classes in individual cells is a longstanding question ( Esser and Radbruch , 1990 ) . Cytokine signals , such as CD40 ligand , IL-4 , IFNγ , and TGFβ , induce CSR and can direct switching toward specific classes in vitro ( Stavnezer , 1996 ) . These signals likely originate from cognate Th cells and dendritic cells in vivo . Cytokine stimulation induces transcription and splicing of 'germline' transcripts from the switch region of the particular IGHC locus that is participating in CSR ( Lorenz et al . , 1995; Stavnezer-Nordgren and Sirlin , 1986 ) . These switch regions accumulate histone modifications that are associated with open chromatin conformations and high DNA accessibility ( Jeevan-Raj et al . , 2011; Wang et al . , 2009 ) . Together , these experiments suggest a model in which epigenetic control of switch region accessibility directs CSR toward specific classes ( Alt et al . , 1986; Stavnezer-Nordgren and Sirlin , 1986; Vaidyanathan and Chaudhuri , 2015 ) . In this study , we mapped the landscape of human antibody class switching using high-throughput immune repertoire sequencing ( Boyd et al . , 2009; Weinstein et al . , 2009 ) . This method has previously yielded insights into how the immune system responds to pathogen challenge and vaccination ( Jackson et al . , 2014a; Jiang et al . , 2013; Parameswaran et al . , 2013; Racanelli et al . , 2011; Vollmers et al . , 2013; Wang et al . , 2014 ) and changes with age ( Jiang et al . , 2011 , 2013; Wang et al . , 2014 ) . We developed an approach for reconstructing clonal histories of antibody lineages , including class switching events . We used this method to measure antibody class switching within clonal lineages across the entire repertoire in vivo in a cohort of healthy human twins . This comprehensive map identifies how antibodies of every class are created . Our analysis of class switching events within clonal lineages uncovered signatures of the cellular decision processes that direct CSR toward specific isotypes . To investigate human antibody class switching in vivo , we conducted immune repertoire sequencing of immunoglobulin heavy chain ( IGH ) genes of 22 healthy young adult human twins , including 9 pairs of identical twins and 2 pairs of fraternal twins . Sequencing libraries were prepared of a ~430–480 bp fragment of the IGH gene using total RNA from peripheral blood B cells drawn from each subject . Libraries were sequenced with 300 bp paired-end reads on the Illumina Miseq platform . Individual RNA molecules were labeled with unique molecular barcodes during library preparation , enabling highly accurate measurement of genetic diversity by using a consensus read approach to correct PCR and sequencing errors ( Figure 1—figure supplement 1; Vollmers et al . , 2013 ) . On average , ~261 , 000 raw reads were obtained from each individual at each time point , representing ~154 , 000 unique sequences ( Figure 1—figure supplement 2A ) . Molecular barcodes were used to enumerate unique sequences and identify distinct clones . Sequencing reads covered ~100 bp of the constant region , making it possible to determine antibody class and resolve subclasses ( IgG1 , IgG2 , IgG3 , IgG4 , IgA1 , and IgA2 ) with high accuracy . Across individuals , the most abundant class was IgM ( 75% ) , followed by IgG1 ( 10% ) and IgA1 ( 8% ) ( Figure 1—figure supplement 4A and Table 1 ) . For 14 of 22 subjects , the measurement was repeated 28 days after the initial sample as a biological replicate ( Bio . Rep . ) . To test the robustness of these measurements , we used the Jensen-Shannon distance as a measure of similarity between distributions of antibody class abundance . As expected , the class distributions are essentially the same across replicates in nearly every subject ( Figure 1—figure supplement 4A and B ) . Furthermore , V gene usage is highly similar across biological replicates ( Figure 1—figure supplement 7 ) . 10 . 7554/eLife . 16578 . 003Table 1 . Number of unique sequences of each class analyzed in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 003ClassSequences ( Sample ) Sequences ( Bio . Rep . ) IgM 2 , 423 , 262 1 , 899 , 952 IgD 70 , 169 60 , 510 IgG3 16 , 981 15 , 625 IgG1 117 , 025 143 , 053 IgA1 276 , 189 231 , 477 IgG2 213 , 574 176 , 484 IgG4 6 , 751 9 , 672 IgE 278 262 IgA2 63 , 374 62 , 251 After activation by specific antigen , naïve B cells proliferate and undergo somatic hypermutation and class switching . This process gives rise to a clonally related lineage composed of antibody IGH sequences of distinct classes that also differ in the variable region due to accumulation of somatic mutations . To study class switching after B cell activation , we developed an approach for reconstructing the clonal history of antibody lineages by using the information contained in the accumulation of somatic mutations as a molecular clock , much as ribosomal 16S sequences are used to study the evolutionary relationships of life on earth . We identified sequences belonging to the same clonal lineage as those sharing a variable ( V ) and joining ( J ) gene combination , CDR3 length , and ≥95% sequence identity in both the CDR3 and the rest of the variable region with at least one other member of the lineage ( Figure 1—figure supplement 5 , Figure 2—figure supplement 7 , and Materials and methods ) . We reconstructed a minimum evolution tree for each clonal lineage by conducting a multiple sequence alignment of all sequences in the lineage and then identifying a minimum spanning tree which includes all the sequences and minimizes the total number of mutations across the tree . CSR occurs through a genomic rearrangement of the IGH constant region locus that brings the gene segment encoding the new constant region closer to the VDJ locus . Gene segments between the old and new constant regions are looped out and deleted ( Iwasato et al . , 1990; Schwedler et al . , 1990; Yoshida et al . , 1990 ) . Therefore , class switches are irreversible and must proceed from upstream classes to downstream classes , according to the order of the IGH constant region loci on the chromosome , which is shown in Figure 1—figure supplement 3 . This provides a constraint on ancestry that we incorporated into our algorithm for clonal history reconstruction: for a sequence belonging to a given class , only sequences of upstream classes can be ancestors . The lineage trees are rooted on the germline sequence of the V and J gene combination shared by the lineage . As expected , somatic mutations accumulate as one moves from the germline sequence toward the leaves of the trees ( Figure 1—figure supplement 6A ) . We note that PCR and sequencing errors rarely give rise to sequences having different classes and therefore contribute minimally to error in measuring class switching . Because these errors terminate branches of the tree , they also do not affect our analysis of mutation accumulation and correlations in class switching patterns . Unlike previous approaches ( Barak et al . , 2008 ) , our tree reconstruction approach enables direct measurement of class switching events in clonal lineages which are supported by observed sequences without the need to infer mutations or ancestral isotype states . The inference process is challenging and indirect due to the complex mutational spectrum of somatic hypermutation . Examples of reconstructed clonal histories of activated B cell lineages from one subject are displayed in Figure 1 . The ~154 , 000 unique sequences from each subject belonged to ~34 , 000 distinct clonal lineages on average ( Figure 1—figure supplement 2B ) . On average across all subjects , each sequence displays 99 . 2% identity in the VDJ variable region to its parent ( Figure 1—figure supplement 6B ) , suggesting that the repertoire has been sampled deeply enough to enable accurate , high-resolution reconstruction of lineage history , since most pairs of sequences are separated by at most a few sites of hypermutation . 10 . 7554/eLife . 16578 . 004Figure 1 . Reconstructed clonal histories of B cell lineages . Examples of reconstructed clonal histories of antibody lineages in the repertoire of Subject 1A . All lineages with ≥6 sequences , comprising 64% of unique sequences in the repertoire , are shown in the upper left . Four examples among these lineages are also shown . Circles indicate unique IGH sequences colored by class . Edges indicate the minimum evolution tree that spans the clonal lineage and are labeled with the number of substitutions separating the sequences . The tree is rooted on the germline V and J gene sequence , indicated by the small black circle . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00410 . 7554/eLife . 16578 . 005Figure 1—figure supplement 1 . Schematic of immune repertoire sequencing strategy and data processing . Reads having identical molecular barcodes are combined to yield a consensus read , which corresponds to a single mRNA molecule whose sequence has been corrected for errors arising during PCR and sequencing . Next , molecules having the same VDJC sequence are combined and the molecular abundance of each unique sequence is counted . In our analysis , we considered only the unique sequences and ignored molecular abundance counts . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00510 . 7554/eLife . 16578 . 006Figure 1—figure supplement 2 . Number of unique sequences ( A ) and clonal lineages ( B ) identified in each subject . Twins are indicated by subject identifiers having the same number , but different letters ( e . g . 1A and 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00610 . 7554/eLife . 16578 . 007Figure 1—figure supplement 3 . Schematic of human immunoglobulin heavy chain ( IGH ) locus . Constant region loci are indicated by colored rectangles , with labels and colors corresponding to class , as in Figure 1 . Recombination signal sequences are indicated by black diamonds . Because the intervening DNA is looped out and excised during class switch recombination , class switching can only proceed from left to right . This region is located on chromosome 14 . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00710 . 7554/eLife . 16578 . 008Figure 1—figure supplement 4 . Abundance of antibody classes . ( A ) Fraction of IGH sequences belonging to each class for each subject in sample and biological replicate . Median across subjects is indicated by red line . ( B ) Differences between distributions of antibody classes ( measured by Jensen-Shannon distance ) . Lane 1 compares biological replicates for individual subjects . Lanes 2 and 3 compare pairs of subjects ( identical twins or unrelated individuals ) . Median across comparisons indicated by red line . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00810 . 7554/eLife . 16578 . 009Figure 1—figure supplement 5 . Determination of sequence identity cutoff for clonal lineages . Distributions of sequence identity within groups of sequences sharing the same V and J genes and CDR3 length ( a 'group' ) from the same repertoire are shown . For each sequence , we calculated the sequence identity with the most similar sequence in its group ( its 'nearest neighbor' ) . Plot displays CDR3 length and identity to the nearest neighbor for all sequences in our data set . Color indicates the number of sequences per bin . This plot reveals two groups of sequences: ( 1 ) sequences for which the nearest neighbor has >95% identity , implying that it belongs to a clonal lineage with the nearest neighbor; and ( 2 ) sequences for which the nearest neighbor has 40–80% identity , suggesting that it does not belong to a clonal lineage . This indicates that by using a cutoff of 95% sequence identity in the CDR3 , one can stringently identify sequences belonging to the same clonal lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 00910 . 7554/eLife . 16578 . 010Figure 1—figure supplement 6 . Features of reconstructed antibody lineages . ( A ) Somatic mutations accumulate in reconstructed antibody lineages . For every sequence , identity to the germline V gene is plotted against its depth , defined as the number of edges to the root of the tree ( the germline sequence ) . Color indicates number of sequences per bin . ( B ) Distributions of sequence identity between variable region sequences of parent-child pairs in reconstructed antibody lineages . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01010 . 7554/eLife . 16578 . 011Figure 1—figure supplement 7 . V gene usage is similar in biological replicates . Fraction of sequences mapping to each V gene in the sample and biological replicate libraries is displayed . Squared Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 011 To characterize the landscape of antibody class switching in living humans , we measured probabilities of class switching across the entire repertoire . We devised an algorithm that traverses the reconstructed tree for each lineage and counts switches between classes from ancestor to child . We then calculated the relative frequency of switching between every pair of classes . In total ~142 , 000 class switch events were observed and contributed to this data set . To characterize the accuracy of this approach , we note that we detected ~35 , 000 pairs of sequences sharing identical VDJ sequences but having different classes , as indicated by differing constant region sequences . Since every molecule has a unique barcode associated with it , we are able to prove that these sequences are not due to PCR recombination artifacts ( Figure 2—figure supplement 1A ) . These sequence pairs arose from CSR without intervening hypermutation events and enable analysis of class switching rates on a subset of the data without the need for lineage identification or tree construction . When we conducted our analysis using only these sequences , we found that the patterns of class switching correlate extremely well with the landscape measured across the entire repertoire , showing that the full lineage tree approach faithfully measures class switching patterns ( Figure 2—figure supplement 1B and C; Table 2 ) . We further confirmed that the patterns of class switching measured using only sequences that inherited all of the germline mutations from their immediate ancestor are highly similar to those measured using the full lineage tree approach ( Figure 2—figure supplement 2 ) , indicating that artifacts arising from imperfect sampling of ancestral sequences have not distorted our measurement . In addition , patterns of class switching measured using only sequences supported by at least three sequencing reads are highly similar to those measured using the full lineage tree approach , suggesting that PCR and sequencing error have not substantially distorted our measurement ( Figure 2—figure supplement 3 ) . Finally , we confirmed that patterns of class switching could not simply be explained by random switching in proportion to class abundance . After shuffling the classes of parent-child sequence pairs , the hierarchical class switching patterns that we observed vanished ( Figure 2—figure supplement 4 ) . 10 . 7554/eLife . 16578 . 012Table 2 . Counts of pairs of sequences sharing identical VDJ sequences , but different constant region sequences . Data from all subjects including both original and biological replicate samples are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 012IgM/IgDIgG3IgG1IgA1IgG2IgG4IgEIgA2IgG3 744 IgG1 6440 2234 IgA1 6374 338 4676 IgG2 3210 1325 5673 2291 IgG4 50 46 129 3 61 IgE 2 0 2 4 0 0 IgA2 1619 30 427 2581 1630 2 0 To confirm that we measured the antibody repertoire with sufficient depth to accurately characterize the class switching landscape , we performed rarefaction analysis using data from five subjects . This analysis revealed that the landscape asymptotes rapidly to the observed one as read sampling depth is increased ( Figure 2—figure supplement 5 ) . Since class switch intermediates that are missing from the data could contribute to error in our measurement , we performed additional rarefaction analysis to show that the fraction of class switches that occur via an intermediate also saturates as read sampling depth is increased ( Figure 2—figure supplement 6 ) . As further validation , we examined ~1500 sequences that were detected in both biological replicates of the same subject and found that in 99 . 9% of cases the class of the ancestor sequence was the same in both samples , indicating that the presence or absence of switch intermediates is reliably detected . Our measurement of class switching patterns uncovered a hierarchy of pathways leading to the production of antibodies of specific classes , which we have summarized as a state transition diagram showing the relative rates of all possible switches ( Figure 2A , Figure 2—figure supplement 8A , Table 3 ) . The dominant class switch pathway leads from IgM/IgD to IgG1 or IgA1 . Specifically , IgM switched most commonly to IgG1 , IgA1 , and IgG2 , which together account for ~85% of switches from IgM . Direct switches from IgM to downstream classes ( IgG4 , IgE , or IgA2 ) were rare ( ~14% ) . Instead , downstream classes are predominantly produced via indirect switches ( Figure 2B ) , most often through IgG1 or IgA1 in a secondary hierarchy of pathways . For example , IgG1 frequently switched to IgA1 or IgG2 , which together account for ~92% of switches from IgG1 , but rarely switched directly to IgG4 or IgA2 . Most IgA2 was produced by subsequent switches from IgA1 or IgG2 ( ~65% ) , instead of directly from IgG1 . We also saw that IgG3 is more likely to switch to IgG1/2 ( ~87% ) rather than IgA1/2 , suggesting that IgG3 lies along a pathway for specific generation of IgG antibodies . These results delineate the class switch pathways that give rise to specific antibody classes . The class switching landscape and the penetrance of direct and indirect switches were highly reproducible across the biological replicates , which were separated by 28 days ( Figure 2C , Figure 2—figure supplement 8B , and Figure 2—figure supplement 9 ) , confirming the robustness of our measurements and suggesting that the landscape is a temporally invariant feature of the healthy human immune system . 10 . 7554/eLife . 16578 . 013Figure 2 . Landscape of human antibody class switching . ( A ) State transition diagram of class switching . Classes are indicated as circles and possible switches as arrows . The radius of each circle indicates the relative abundance of the labeled class . The width of each arrow indicates the relative frequency of the switch ( also reported in Table 3 ) . Rare classes IgG4 and IgE have been omitted for clarity and are shown in Figure 2—figure supplement 8A . ( B ) Penetrance of direct switches from IgM/IgD . For each class , the fraction of sequences created by direct switching from IgM is shown ( mean ± s . d . across n = 22 subjects for Sample and n = 14 subjects for Bio . Rep . ) . ( C ) Rates of CSR . The rate constant of each switch path was estimated by fitting an exponential probability distribution to the distribution of the number of somatic mutations accumulated prior to CSR ( Figure 2—figure supplement 11 ) . Distributions of rate constants for switch paths from IgM/IgD to activated classes ( gray ) and from an activated class to another activated class ( white ) having ≥500 examplesin both Sample and Bio . Rep . repertoires are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01310 . 7554/eLife . 16578 . 014Figure 2—source data 1 . Counts of class switch events . Number of events observed for each possible switch from the class indicated by the row to the class indicated by the column . Data from D0 and D28 are provided separately . These data were used to calculate the switching rates depicted in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01410 . 7554/eLife . 16578 . 015Figure 2—figure supplement 1 . Patterns of class switching measured using sequences with identical VDJ sequences but different constant regions are highly similar to those measured using the full lineage reconstruction approach . ( A ) Origin of pairs of sequences having identical VDJ sequences but different constant region classes . PCR recombination artifacts ( PCR chimeras ) were detected by comparing the unique barcodes from each sequencing read . Specifically , a pair of sequences was identified as originating from PCR chimera if at least one V-region barcode was shared between the pair of sequences , accounting for ~5% of sequence pairs . ( B ) Landscape of antibody class switching measured using only pairs of sequences having identical VDJ sequences but different constant region classes , which did not originate from PCR chimeras . Top panels show the relative frequency of class switch events from the class indicated by the column to the class indicated by the row . Middle panels show the destination probability , which is the probability that a given sequence of class indicated by the column switches to the class indicated to the row . Bottom panels show the arrival probability , which is the probability that a given sequence of class indicated by the row arose via direct switch from the class indicated by the column . Sample ( left ) and biological replicate ( right ) are shown . ( C ) Comparison between the landscape of antibody class switching measured using only pairs of sequences having identical VDJ sequences but different constant region classes and the landscape measured using the full lineage reconstruction approach . The values that define the landscape ( relative switch frequencies , destination probabilities , and arrival probabilities ) are plotted against the values obtained using all parent-child sequence pairs . Squared Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01510 . 7554/eLife . 16578 . 016Figure 2—figure supplement 2 . Patterns of class switching measured using sequences inheriting all germline mutations from parent are highly similar to those measured using the full lineage reconstruction approach . ( A ) Landscape of antibody class switching measured using only sequences inheriting all germline mutations from parent . Top panels show the relative frequency of class switch events from the class indicated by the column to the class indicated by the row . Middle panels show the destination probability , which is the probability that a given sequence of class indicated by the column switches to the class indicated to the row . Bottom panels show the arrival probability , which is the probability that a given sequence of class indicated by the row arose via direct switch from the class indicated by the column . Sample ( left ) and biological replicate ( right ) are shown . ( B ) Comparison between the landscape of antibody class switching measured using only sequences inheriting all germline mutations from parent and the landscape measured using the full lineage reconstruction approach . The values that define the landscape ( relative switch frequencies , destination probabilities , and arrival probabilities ) are plotted against the values obtained using all parent-child sequence pairs . Squared Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01610 . 7554/eLife . 16578 . 017Figure 2—figure supplement 3 . Patterns of class switching measured using sequences supported by consensus reads are highly similar to those measured using the full lineage reconstruction approach . ( A ) Landscape of antibody class switching measured using only sequences supported by consensus reads formed from ≥3 sequencing reads . Top panels show the relative frequency of class switch events from the class indicated by the column to the class indicated by the row . Middle panels show the destination probability , which is the probability that a given sequence of class indicated by the column switches to the class indicated to the row . Bottom panels show the arrival probability , which is the probability that a given sequence of class indicated by the row arose via direct switch from the class indicated by the column . Sample ( left ) and biological replicate ( right ) are shown . ( B ) Comparison between the landscape of antibody class switching measured using only sequences supported by consensus reads and the landscape measured using the full lineage reconstruction approach . The values that define the landscape ( relative switch frequencies , destination probabilities , and arrival probabilities ) are plotted against the values obtained using all parent-child sequence pairs . Squared Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01710 . 7554/eLife . 16578 . 018Figure 2—figure supplement 4 . Landscape of class switching cannot be explained by random switching in proportion to the abundance of antibody classes . ( A ) Landscape of antibody class switching measured after shuffling parent-child pairs of sequences . Top panel shows the relative frequency of class switch events from the class indicated by the column to the class indicated by the row . Middle panel shows the destination probability , which is the probability that a given sequence of class indicated by the column switches to the class indicated to the row . Bottom panel shows the arrival probability , which is the probability that a given sequence of class indicated by the row arose via direct switch from the class indicated by the column . ( B ) Comparison of the landscapes of antibody class switching before and after shuffling parent-child pairs of sequences . The values that define the landscapes ( relative switch frequencies , destination probabilities , and arrival probabilities ) are plotted against each other . Squared Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01810 . 7554/eLife . 16578 . 019Figure 2—figure supplement 5 . Landscape of class switching saturates with respect to sequencing depth . Rarefaction analysis of class switching landscapes of five subjects . Sequencing reads were sampled to varying depth , and the class switching landscape was measured in each case , the values of the relative switch frequency are plotted . The relative switch frequency is obtained by dividing the number of switches for a given transition by the maximum number of switches observed for any transition . For each subject , 5 replicate subsamples were performed at each depth , and the values obtained in these replicates are indicated by points , while the line connects medians of the replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 01910 . 7554/eLife . 16578 . 020Figure 2—figure supplement 6 . Rarefaction analysis indicates that switch intermediates are robustly detected . Sequencing reads were subsampled to varying depth for the five subjects shown in Figure 2—figure supplement 5 with five replicate subsamplings at each depth . Data from all five subjects was pooled and used to calculate the fraction of switches from A to C indicated by the title of each panel that were direct ( A -> C ) and indirect ( A -> B -> C ) . Median across replicates is indicated by the red line . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02010 . 7554/eLife . 16578 . 021Figure 2—figure supplement 7 . Class switching landscape is not sensitive to the lineage clustering cutoff parameter . Clustering to identify clonal lineages of antibodies was performed on all repertoires from D0 with varying values of the clustering cutoff parameter ranging from 0 . 80 to 0 . 95 . The class switching landscape was then calculated . In this calculation , we included only lineages having ≤2500 sequences in every parameter setting to ensure computational tractability . The landscape in each case is plotted against the landscape measured when the cutoff is 0 . 95 . Squared Pearson correlation is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02110 . 7554/eLife . 16578 . 022Figure 2—figure supplement 8 . Landscape of class switching in humans . ( A ) Class switch state transition diagram including the rare classes IgG4 and IgE . Classes are indicated as pies and possible switches are indicated as arrows . Radius of each pie indicates the relative abundance of the class . The width of each arrow indicates the relative frequency of the switch ( also reported in Table 2 ) . ( B ) Heatmaps showing the class switch landscape as an average across subjects . Top panel shows the relative frequency of class switch events from the class indicated by the column to the class indicated by the row . Middle panel shows the destination probability , which is the probability that a given sequence of class indicated by the column switches to the class indicated to the row . Bottom panel shows the arrival probability , which is the probability that a given sequence of class indicated by the row arose via direct switch from the class indicated by the column . Sample ( left ) and biological replicate ( right ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02210 . 7554/eLife . 16578 . 023Figure 2—figure supplement 9 . Comparisons of class switch landscapes across individuals . ( A and B ) Differences between class switch landscapes ( measured by Jensen-Shannon distance ) . Distances were calculated between the vectors representing ( A ) relative switch frequency or ( B ) destination probability . Lane 1 compares the two biological replicates for each subject . Lanes 2 and 3 compare pairs of subjects ( identical twins or unrelated individuals ) . Median is indicated by red line . ( C ) Comparison of class switching landscapes of identical twins and unrelated pairs of subjects . Destination probabilities of identical twin pairs ( red ) and all possible pairs of unrelated subjects ( blue ) are plotted against one another . Intraclass correlation coefficient ( ICC ) for twins and unrelated pairs was calculated using bootstrap resampling of pairs of subjects ( 1000 replicates ) and reported in the legend ( 5th to 95th percentile range ) . ( D ) Comparison of the relative switch frequency of all possible class switches across subjects . Each point indicates the relative frequency of the switch indicated on the x-axis for an individual subject . Median is indicated by red line . Relative frequency of all switches is similar across all subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02310 . 7554/eLife . 16578 . 024Figure 2—figure supplement 10 . Class switching landscapes of individual subjects . ( A ) Relative switch frequency , ( B ) destination probability , and ( C ) arrival probability are shown for each subject . Twins are shown on the same row and zygosity is indicated by row label . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02410 . 7554/eLife . 16578 . 025Figure 2—figure supplement 11 . Measurement of rates of class switching . ( A ) Motifs analyzed to characterize the rate of CSR . ( B ) Distributions of mutations accumulated prior to class switching . Switches from IgM/IgD to activated classes ( IgG , IgA , IgE ) are plotted separately from switches between activated classes , as indicated by color . The p value of Kolmogorov-Smirnov test , two-sample comparing these two distributions is shown . ( C ) Cumulative probability of class switching as mutations accumulate . Origin and destination of class switch are indicated by color . ( D ) Rate constants of class switching along all switch paths where we observed >250 direct switches . Exponential distributions were fitted to the distributions of the number of mutations accumulated prior to class switching ( see examples in panel E ) and the rate constant was extracted . ( E ) Examples of exponential distributions ( modified to have an additional parameter for non-zero y-intercept [CDF ( x ) = 1 – exp ( -ax ) + b] ) fitted to the empirical distributions of the number of mutations accumulated prior to class switching . Fit was performed using the curve_fit function in the scipy . optimize module in Python , which implements the Levenberg-Marquardt nonlinear least squares algorithm . Rate constants of the fitted exponential distributions are shown in panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02510 . 7554/eLife . 16578 . 026Table 3 . Number of parent-child pairs having each possible pair of classes . Data from all subjects and both biological replicates are included . Total number of parent-child pairs is 3 , 304 , 346 . Number in parentheses indicates the relative frequency of each class switch . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 026ParentChild IgM/IgD IgG3 IgG1 IgA1 IgG2 IgG4 IgE IgA2 IgM/IgD 1 , 495 , 250 IgG3 2 , 530 ( 0 . 09 ) 48 , 357 IgG1 19 , 881 ( 0 . 74 ) 4 , 726 ( 0 . 18 ) 645 , 591 IgA1 26 , 915 ( 1 . 00 ) 935 ( 0 . 03 ) 17 , 480 ( 0 . 65 ) 493 , 999 IgG2 12 , 695 ( 0 . 47 ) 3 , 178 ( 0 . 12 ) 15 , 663 ( 0 . 58 ) 8 , 028 ( 0 . 30 ) 375 , 284 IgG4 312 ( 0 . 01 ) 157 ( 0 . 006 ) 342 ( 0 . 01 ) 57 ( 0 . 002 ) 542 ( 0 . 02 ) 16 , 091 IgE 10 ( 0 . 0004 ) 0 ( 0 ) 16 ( 0 . 0006 ) 5 ( 0 . 0002 ) 1 ( 0 . 00004 ) 3 ( 0 . 0001 ) 419 IgA2 7 , 344 ( 0 . 27 ) 139 ( 0 . 005 ) 2 , 934 ( 0 . 11 ) 12 , 176 ( 0 . 45 ) 6 , 455 ( 0 . 24 ) 17 ( 0 . 0006 ) 0 ( 0 ) 86 , 814 Next , we examined how the landscape of class switching varies between individuals . The landscapes of individual subjects are broadly similar ( Figure 2—figure supplements 9 and 10 ) , and the dominant usage of several major switch pathways is conserved across all subjects ( Figure 2—figure supplement 9D ) . This similarity is also apparent when measured by the Jensen-Shannon distance , which reveals that the magnitude of variation between subjects is similar to variation between biological replicates of the same subject ( Figure 2—figure supplement 9A and B ) . We conclude that healthy young adults share a broadly conserved landscape of antibody class switching . We also asked whether class switching landscapes were more similar among identical twins compared to unrelated individuals . We found that the class switching patterns of identical twins are no better correlated than pairs of unrelated individuals ( Figure 2—figure supplement 9A–C ) , suggesting that the regulation of CSR involves substantial environmental or stochastic influences , as has been found in many other parameters of the immune system ( Brodin et al . , 2015 ) . Despite progress in dissecting the molecular mechanisms of CSR , many fundamental characteristics of class switching under physiological conditions remain unresolved . For example , the tempo of class switching within activated B cell lineages has not been measured . Motivated by this , we used somatic mutations as a molecular clock to measure the rate of CSR between naïve and activated classes within clonal lineages . We searched the clonal lineage trees for motifs consisting of multiple sequences sharing the same class that accumulated mutations prior to a class switch event ( Figure 2—figure supplement 11A ) . We asked how many somatic mutations accumulate prior to CSR and whether different antibody classes tend to accumulate different numbers of mutations before class switching . We found that naïve classes ( IgM or IgD ) accumulated significantly more mutations before undergoing CSR to activated classes ( IgG , IgA , or IgE ) , in comparison with CSR between activated classes . Among IgM/IgD sequences , the average number of mutations accumulated in the variable region prior to CSR is 4 . 1 ± 6 . 7 ( mean ± s . d . ) . In contrast , only 2 . 5 ± 4 . 3 mutations accumulate in sequences of activated classes prior to further CSR ( Figure 2—figure supplement 11B and C; p = 1 . 8 × 10–45 and 1 . 3 × 10–12 for Samples and Bio . Rep . respectively; Mann-Whitney U test , two-sided ) . We found that the distributions of mutations accumulated prior to CSR were well fit by an exponential distribution , suggesting that CSR is a memory-less process with respect to somatic mutation and allowing us to estimate rates of CSR ( Figure 2—figure supplement 11E ) . We found that the rate constants of class switching from IgM/IgD to activated classes were ~0 . 37 mutation-1 ( Figure 2C ) . By contrast , the rate constants of switching between activated classes were ~0 . 60 mutation-1 ( p < 1 . 5 × 10–4; Mann-Whitney U test , two-sided ) . These patterns were robustly observed across biological replicates ( Figure 2—figure supplement 11D ) . Because class switching can be initiated from mutated IgM memory cells for which we may have failed to detect naïve progenitors , our measurements of the number of mutations accumulated in IgM sequences prior to class switching are likely an underestimate , supporting an argument for differences in the rate of CSR between naïve and activated classes . Thus , our results suggest that activated B cells which have already undergone class switching tend to rapidly undergo further class switching , as measured by the clock of somatic hypermutation . The isotype composition of the antibody repertoire is ultimately determined by class switch decisions made by individual B cells , which belong to clonal lineages . However , nothing is known about how class switch fates vary among cells within a clonal lineage . To address this , we traced the descent of individual B cells , using somatic mutations as both lineage markers and a molecular clock , and examined the class switch fates of clonally related cells . We asked whether closely related cells , as measured by the number of somatic mutations accumulated in their sequences since their divergence from a common progenitor , exhibit more concordant class switch fates than more distantly related cells , as well as unrelated cells ( from distinct clonal lineages ) . To find pairs of related cells , we searched the clonal lineage trees for motifs consisting of a pair of sequences that ( 1 ) shared a common progenitor , ( 2 ) were the same class , and ( 3 ) each had class-switched progeny ( Figure 3—figure supplement 1 ) . We further required that each sequence inherit all of the somatic mutations present in its ancestor , yielding ~40 , 000 pairs of sequences for this analysis . We binned these sequence pairs by their mutational distance from the common progenitor , revealing that the pairs spanned a broad spectrum of relatedness ( Figure 3—figure supplement 2 ) . For each level of relatedness ( bin ) , we calculated the probability that both sequences in a pair switched to the same class . To quantify the strength of concordance , we used Yule’s Q , which measures the agreement between pairs of sequences . Yule’s Q ranges from -1 to 1 , with 1 indicating that both sequences always switched to the same class ( perfect agreement ) , -1 indicating that the two sequences always switched to different classes ( perfect disagreement ) , and 0 indicating no correlation between the fates of sequences in a pair . We discovered that closely related cells made highly concordant class switch decisions ( Figure 3A ) . The most closely related cells , separated by ≤2 mutations from their common progenitor , had a very significant tendency to switch to the same class , in contrast to unrelated pairs of cells which were obtained by shuffling ( p values ranging from 6 × 10–5 to 1 . 1 × 10–186; Figure 3—figure supplement 3 ) . When we examined pairs of cells that were less closely related ( as measured by mutations from their common progenitor ) , we found that the concordance in class switch fates dissipated as somatic mutations accumulated , becoming indistinguishable from unrelated cells after ~10 somatic mutations ( Figure 3B and Figure 3—figure supplement 4 ) . These findings were corroborated by examining the probability distributions of class switch fate conditioned upon the fate of a closely related cell: closely related cells exhibit probability distributions that are strongly biased toward the same class switch fate , and the bias dissipates as somatic mutations accumulate ( Figure 3—figure supplement 5 ) . Importantly , we found that mutational distances ( i . e . branch lengths ) are not associated with particular switching events ( Figure 3—figure supplement 6 ) , and that mutational distances among related sequences are not correlated ( Figure 3—figure supplement 7 ) , indicating that correlations in switch fate are not due to differential sampling of lineages . These findings demonstrate that class switch decisions are coordinated within clonal lineages of B cells in living humans and that this coordination dissipates at large genealogical distances within a lineage . 10 . 7554/eLife . 16578 . 027Figure 3 . Class switch fates of closely related sequences are correlated and lose coherence as somatic mutations accumulate . ( A ) Concordance between the class switch fates of closely related sequences having ≤2 substitutions from their common progenitor , as measured using Yule’s Q . Distinct switch paths are indicated on the x-axis . Bars show standard deviation of the concordance Q betweenpairs of unrelated sequences , which were obtained by shuffling ( 1000 replicates ) . ( B ) Concordance between the class switch fates of pairs of sequences plotted by their relatedness , as measured by number of mutations from their common progenitor . For comparison , red shading indicates the probability density of concordance between unrelated sequence pairs obtained by shuffling ( 1000 replicates ) . To account for variation due to sampling statistics , the number of pairs at each level of relatedness was preserved during shuffling . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02710 . 7554/eLife . 16578 . 028Figure 3—figure supplement 1 . Motif analyzed to characterize the class switch fates of clonally related cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02810 . 7554/eLife . 16578 . 029Figure 3—figure supplement 2 . Relatedness of the pairs of cells used to characterize the class switch fates of clonally related cells in vivo . Distributions of relatedness between pairs of related sequences measured by the maximum number of mutations among the two sequences to their common progenitor ( most recent common ancestor [MRCA] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 02910 . 7554/eLife . 16578 . 030Figure 3—figure supplement 3 . Estimation of significance of correlations between class switch fates of related sequences . Plots show examples of distributions of Yule’s Q for unrelated sequence pairs obtained by shuffling with 1000 replicates , together with fitted Gaussian distribution and observed value of Yule’s Q for the most closely related cells ( 2 or fewer mutations from common progenitor ) for comparison . We calculated the exact one-sided p value of the observed value of Yule’s Q as the probability of the Gaussian random variable taking a value greater than or equal to the observed value of Q . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03010 . 7554/eLife . 16578 . 031Figure 3—figure supplement 4 . Concordance between the class switch fates of related sequences plotted against relatedness as measured by number of mutations from common progenitor for all common switch paths . Column header indicates class of the related sequences . Title of each panel indicates the downstream switch destination . Observed data are shown in red . Gray lines indicate concordance between unrelated sequence pairs obtained by shuffling ( 1000 replicates ) while preserving the number of pairs having each degree of relatedness to account for variation due to sampling statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03110 . 7554/eLife . 16578 . 032Figure 3—figure supplement 5 . Closely related B cells often switch to the same class . Probability distributions of class switch fate conditioned upon the fate of closely related cell . These plots represent the probability distribution of class switch fate of 'Related cell 2' , given the fate of 'Related cell 1' . Column header indicates class of the related sequences . X-axis label of each row indicates the class switch fate of 'Related cell 1' . Color indicates the relatedness of the cells , measured by the number of mutations from common progenitor . Number of sequence pairs contributing to each panel N is shown . Right panel shows original samples and left panel shows biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03210 . 7554/eLife . 16578 . 033Figure 3—figure supplement 6 . Branch lengths are not associated with particular switching events . Fraction of switches from naïve classes ( IgM/IgD ) to each possible destination class is shown as a function of mutational distance ( bins in increments of 2 mutations ) from ( A ) common progenitor ( most recent common ancestor [MRCA] ) to parent cell , and ( B ) parent cell to switched progeny . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03310 . 7554/eLife . 16578 . 034Figure 3—figure supplement 7 . Mutational distances among related cells , common progenitors , and switched progeny are not correlated . Number of mutations from ( A ) common progenitor ( most recent common ancestor [MRCA] ) to cell 1 and cell 2 , ( B ) cell 1 to switched progeny and cell 2 to switched progeny , and ( C ) common progenitor to cell and cell to switched progeny . The mutational distances compared in each panel are indicated by the colored edges in the motif on the right ( x-axis , orange; y-axis , blue ) . Color indicates number of pairs of related sequences in each bin . Squared Pearson correlation values are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 034 We reasoned that the coordination of class switch decisions among closely related cells could arise if CSR is directed to specific isotypes by cytokine signals originating from cognate cells in spatially localized niches , which sister cells at some point co-occupy and therefore are directed synchronously toward the same class switch fate . Alternatively , sister cells might share an imprinted state , which is transmitted from a common progenitor and directs CSR toward specific isotypes in a cell-autonomous fashion . To discriminate between these models and test whether cellular interactions are necessary to generate correlations between class switch fates of sibling cells , we measured the class switch behavior of purified primary human B cells stimulated with cytokines in culture . We purified CD19+ IgM+ B cells from whole blood ( Figure 4—figure supplement 1A ) and cultured them in the presence of multimeric CD40 ligand ( CD40L ) , IL-4 , and IL-10 for 8 days , then prepared sequencing libraries of the IGH locus . The purified B cells were 99 . 6% CD19+ ( Figure 4—figure supplement 1B ) , and sequencing of the IgM+ cells used to initiate the culture revealed that 97% of the sequences were IgM/IgD ( Figure 4A ) . During culture , the cells proliferated and underwent class switching to IgG1 , IgG2 , IgG3 , and IgA1 ( Figure 4B ) . After reconstructing the histories of clonal lineages , we identified ~1900 pairs of IgM/IgD sequences sharing a common progenitor that subsequently underwent class switching . 10 . 7554/eLife . 16578 . 035Figure 4 . Class switch fates of closely related sequences are correlated in purified B cells induced to class switch in vitro . ( A ) Class composition of CD19+ IgM+ cells used to initiate cell culture as measured by sequencing of the IGH locus . ( B ) Class composition of cells after culture for 8 days in the presence of multimeric CD40L , IL-4 and IL-10 ( mean of three replicates ) . ( C ) Concordance between class switch fates of closely related sequences having ≤2 mutations from their common progenitor measured using Yule’s Q . Distinct switch paths are indicated on the x-axis . Bars show standard deviation of the concordance Q for unrelated pairs of sequences , which were obtained by shuffling ( 1000 replicates ) . Results of three replicate experiments are indicated by color . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03510 . 7554/eLife . 16578 . 036Figure 4—figure supplement 1 . Purification of CD19+ IgM+ B cells for in vitro culture . ( A ) Fluorescence activated cell sorting ( FACS ) procedure for purifying CD19+ IgM+ B cells . Purified B cell populations obtained using RosetteSep were stained with CD19-PE , CD20-AF647 , IgM-BV421 , passed through a 40 um filter , and sorted on a Sony SH800 instrument ( top ) . Fluorescence intensity cutoffs for CD19+ and IgM+ were set based on an unstained control ( bottom ) . Analysis of CD20-AF647 fluorescence confirmed that 99 . 9% of CD19+ IgM+ cells were CD20+ . Abbreviations: PE , phycoerythrin; AF647 , AlexaFluor 647; BV421 , Brilliant Violet 421 . ( B ) Analysis of sorted CD19+ IgM+ populations immediately after sort to confirm purity . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 03610 . 7554/eLife . 16578 . 037Figure 4—figure supplement 2 . Relatedness of the pairs of cells used to characterize the class switch fates of clonally related cells in vitro . Distributions of relatedness between pairs of sister sequences , as measured by the maximum number of mutations among the two sequences to their common progenitor ( most recent common ancestor [MRCA] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 037 In vitro , the concordance between the class switch fates of the most closely related cells was as evident as in vivo ( Figure 4C ) . Closely related sequences separated from a common progenitor by ≤2 substitutions exhibited a highly significant tendency to switch to the same class , in comparison with unrelated sequences ( p values ranging from 5 × 10–4 to 5 × 10–28 ) . Since sequences having more somatic mutations were exceedingly rare ( Figure 4—figure supplement 2 ) , it was not possible to examine the dissipation of concordance as somatic mutations accumulated . Importantly , no pairs of sequences sharing a common progenitor that subsequently underwent class switching were detected in the IgM+ B cell populations that initiated the culture , confirming that imperfect IgM+ B cell purification cannot account for the class switches detected after culturing . These results demonstrate that coordination of class switch decisions in clonal lineages of B cells is not dependent upon the in vivo environment and interactions with cognate cells , but rather seems to be an autonomous property of purified B cells . This finding suggests that CSR is directed toward specific isotypes by an imprinted state , which is transmitted from a common progenitor to sister cells . Deep sequencing of the immune repertoire offers unprecedented views into the human immune system . In this study , we set out to use antibody repertoire sequencing to investigate the nature of antibody class switching in healthy humans . The mechanisms of CSR and each individual's history of immune activation leave indelible imprints on the diversity of immunoglobulin sequences . We have exploited these signatures to characterize patterns of class switching in the natural setting in living humans and dissect the cellular processes that govern CSR . Our findings provide a comprehensive map of the patterns of antibody class switching in humans . The data reveal that pathways of CSR are organized into two tiers: ( 1 ) naïve classes ( IgM/IgD ) switch predominantly to proximal classes such as IgG3 , IgG1 , or IgA1 , and ( 2 ) proximal classes may subsequently switch to distal classes , such as IgG2 , IgG4 , or IgA2 . IgG1 and IgA1 seem to be central intermediates linking naïve to distal classes . This pattern is evident from ( 1 ) the high probability of switching from naïve to proximal classes together with the low probability of switching directly from naïve to distal classes , and ( 2 ) observations of frequent sequential switches via IgG3 , IgG1 , and IgA1 intermediates . This hierarchy mirrors the linear geometry of the IGH constant region loci on chromosome 14 , suggesting that chromosomal structure or topology influences CSR . Consistent with this , previous studies which showed that mouse B cells stimulated in vitro with CD40L and IL-4 switched to IgG1 with high frequency after three divisions , while switching to the downstream classes IgE and IgG2a increased after five or six divisions ( Hasbold et al . , 1998; Hodgkin et al . , 1996 ) . A previous study on human populations reported increasing levels of point mutation in progressively further downstream IgG subclasses , supporting sequential class switching ( Jackson et al . , 2014b ) . This study also suggested the existence of a preferential switch pathway from IgG2 to IgA2 based on evidence of stronger antigen-driven selection in IgA2 than IgA1 sequences , which is consistent with our results . Our examination of clonal histories of B cell lineages with mixed isotypes provides more direct evidence for sequential class switching and the existence of dominant class switch pathways . While our lineage reconstruction approach circumvents the difficulties associated with ancestral inference and probabilistic models of class switching , one limitation is that measurements of switch rates can be affected by undersampling of ancestors , especially for switches between rare classes , such as IgE and IgG4 . We have estimated this undersampling rate by using a Chapman estimator to gauge the total number of each isotype in circulation compared to the number we detected , and found results ranging from 0 . 1% to 5% depending on the subject , which does not affect any of the conclusions of this work . Prior to this work a limited number of class transitions had been observed in human samples . Our data identifies ten transitions which to our knowledge have not been previously identified ( Table 4 ) . In conjunction with other published results , it appears that , with the possible exception of IgE , any transition which is permitted to happen by the geometry of the immunoglobulin locus is observed in healthy human samples . This finding supports the view that the machinery underlying CSR is intrinsically stochastic , and that biological regulation enforces probabilistic preferences , rather than strict rules in switch behavior . 10 . 7554/eLife . 16578 . 038Table 4 . Summary of class switch recombination events that have been observed in human cells . Switches that have previously been observed are indicated as 'Known' and the literature references are provided . All of the previous studies demonstrated the existence of switch events by sequencing recombination junctions or switch circles . 'Novel' indicates switches which have not previously been reported that we observed in our dataset of ~35 , 000 pairs of sequences sharing identical VDJ sequences , but having different constant region genes . 'Not detected' indicates switches that we did not observe in this dataset of identical sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 038Source classDestination class IgM/IgD IgG3 IgG1 IgA1 IgG2 IgG4 IgE IgG3 Known ( Fujieda et al . , 1995; Malisan et al . , 1996 ) IgG1 Known ( Fujieda et al . , 1995; Malisan et al . , 1996 ) Novel IgA1 Known ( Jabara et al . , 1993; Zan et al . , 1998 ) Known ( Lin et al . , 2014 ) Known ( Zan et al . , 1998 ) IgG2 Known ( Malisan et al . , 1996 ) Novel Novel Novel IgG4 Known ( Fujieda et al . , 1995; Jabara et al . , 1993 ) Novel Novel Novel Novel IgE Known ( Jabara et al . , 1993; Xiong et al . , 2012 ) Not detected Known ( Xiong et al . , 2012 ) Novel Not detected Known ( Jabara et al . , 1993 ) IgA2 Known ( He et al . , 2007; Lin et al . , 2014 ) Known ( Lin et al . , 2014 ) Known ( Lin et al . , 2014 ) Known ( He et al . , 2007; Lin et al . , 2014 ) Known ( Lin et al . , 2014 ) Novel Not detected The broad contours of the class switching landscape appear to be conserved across individuals , but there is variation between individuals that likely reflects differences in the history of immune activation and environmental exposure . Importantly , identical twins did not exhibit identical class switching landscapes , indicating that class switching is driven largely by non-heritable factors , which likely include exposure to pathogens or other microbes . Previous studies of identical twins have suggested that genetic background controls features of the antibody repertoire , such as IGHV , IGHD , and IGHJ gene use , and CDR3 length ( Wang et al . , 2015 ) . On the other hand , studies examining other components of the immune system have indicated that non-heritable factors dominantly influence most features of serological and cellular responses , including serum protein abundances and cell populations ( Brodin et al . , 2015 ) . Our findings suggest that variation between human in the class composition of the antibody repertoire is predominantly driven by the ability of the immune system to adapt to environmental stimuli , rather than genetic predisposition . Unique landscapes of antibody class switching in identical twins likely arise from the unpredictable stimulation of B cell clones and different exposure to many microbes over the course of a lifetime . Our measurement of the conserved class switching landscape of healthy , young adult humans provides a reference for comparison against individuals with altered immune states , such as autoimmunity or chronic infection . Our work demonstrates how somatic mutations can be exploited as a molecular clock to reconstruct the genealogies of cells and characterize the dynamics of cell state . We have uncovered strong correlations between the class switch fates of closely related B cells which have undergone maturation in the natural context in living humans . These correlations appear to decay on a timescale of ~10 somatic mutations . Such correlations between closely related cells are also generated during in vitro culture of purified B cell populations in the presence of cytokines that induce class switching , demonstrating that the correlations are an autonomous property of purified B cells and that the in vivo environment is not necessary to create them . Although cytokine signals driving CSR in vivo likely originate from cognate T helper cells and dendritic cells in localized intercellular niches , our experiments show that correlations in class switch fate between sibling cells cannot simply be attributed to exposure to common signals due to co-occupancy of the same niche . This mode of regulation contrasts with stem cell maintenance and differentiation in the mammalian lung , which are regulated by signaling from parent/progenitor cells to daughter cells in localized niches ( Pardo-Saganta et al . , 2015 ) . We note that the correlations that we observed in vitro were not quite as strong as those seen in vivo , suggesting that cellular interactions in the natural context might enhance sibling correlations in class switch fates . We note that in our culture experiments the starting populations were heterogeneous and included both naïve and memory B cells , leaving the lineage characteristics of class switching within these compartments to be examined in future work . Our data suggest a model where CSR is directed toward specific classes by a transient epigenetic state , which is transmitted from parent cells to daughter cells and relaxes on a timescale of ~10 somatic mutations . Consistent with this , directed CSR is thought to be regulated via cytokine-activated transcription at specific IGHC loci , which targets the region for modification by activation-induced cytidine deaminase ( AID ) . Germline transcription of IGHC genes is associated with histone modifications that increase DNA accessibility ( Jeevan-Raj et al . , 2011; Wang et al . , 2009 ) , and germline IGHC transcripts can form RNA-DNA hybrids with genomic DNA , exposing ssDNA to AID attack ( Reaban and Griffin , 1990; Reaban et al . , 1994; Yu et al . , 2003 ) . Using single-cell transcriptomics , we have found that single B cells stimulated to class switch in vitro often predominantly express germline transcripts from a single IGHC locus ( F . Horns , unpublished data ) . We propose that inheritance during mitosis of germline transcripts and chromatin state in the IGHC locus , including perhaps histone modifications influencing DNA accessibility , is a mechanism that generates correlations in the class switch fates of sister cells . Importantly , epigenetic inheritance of active and repressed chromatin state during mitosis has been demonstrated ( Cavalli and Paro , 1998; Grewal and Klar , 1996 ) . Our measurements suggest that the timescale on which the relative accessibility of IGHC loci persists is ~10 somatic mutations . Calibration of the mutational clock should allow recovery of information about epigenetic state and phenotypic dynamics in units of time and cellular generations . Together with recent studies of mammalian ( Spencer et al . , 2009 ) and bacterial cells in culture ( Hormoz et al . , 2015 ) , our work suggests that phenotypic correlations between sister cells due to shared inheritance are widespread . We predict that such correlations often will be detected when genealogical relationships between individual cells can be resolved . We propose that inheritance of epigenetic state provides a mechanism for orchestrating cellular behavior without the need for signaling . All study participants gave informed consent and protocols were approved by the Stanford Institutional Review Board . Twenty-two human twins aged 18–28 , including 11 males and 11 females , were recruited in 2010 . All subjects were apparently healthy and showed no signs of disease . Twin zygosity was determined by short tandem repeat analysis with 18 loci . Monozygosity was assigned when all loci and the gender-determining marker were identical . Blood was drawn from each subject by venipuncture . Peripheral-blood mononuclear cells ( PBMCs ) were isolated using a Ficoll gradient and frozen in 10% ( vol/vol ) DMSO/40% ( vol/vol ) fetal bovine serum ( FBS ) following Stanford Human Immune Monitoring Center protocols . After cells were thawed , total RNA was extracted using the Qiagen AllPrep kit ( Valencia , CA ) . Subjects were vaccinated with the 2010 seasonal trivalent inactivated influenza vaccine immediately after the sample was drawn . Biological replicates were drawn 28 days later . Biological replicates were indistinguishable from the original samples with respect to antibody class and V gene usage ( Figure 1—figure supplements 4B and 7 ) , as expected given that the most pronounced immune response occurs 7 days after vaccination ( Wrammert et al . , 2008 ) . Sequencing libraries were prepared using 500 ng of total RNA as input following the protocol described in ( Vollmers et al . , 2013 ) . Briefly , primer annealing to a pooled set of ten isotype-specific IGH constant region primers that contain 8 or 12 random nts ( Table 5 ) was carried out at 72°C for 3 min then immediately placed on ice for 2 min . First-strand cDNA synthesis was performed using Superscript III reverse transcriptase ( Life Technologies , Carlsbad , CA ) according to manufacturer's instructions . Second-strand cDNA synthesis was done using Phusion HiFi DNA polymerase ( Thermo Scientific , Waltham , MA ) and a pool of six IGH variable region primers that contain 8 random nts ( Table 6 ) ( 98°C for 4 min , 52°C for 1 min , 72°C for 5 min ) . Double-stranded cDNA was purified twice using Ampure XP beads ( Beckman Coulter , Indianapolis , IN ) at a 1:1 ratio , then amplified using Platinum HiFi enzyme ( Life Technologies ) and primers containing Illumina adapters and dual indexes . PCR products were purified once using Ampure XP beads at a 1:1 ratio then pooled for multiplexed sequencing . 10 . 7554/eLife . 16578 . 039Table 5 . IGH constant region primers . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 039NamePrimer ( 5’ to 3’ ) IgA_08N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNGGGGAAGAAGCCCTGGAC IgA_12N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNNNNGGGGAAGAAGCCCTGGAC IgG_08N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNGGGAAGTAGTCCTTGACCA IgG_12N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNNNNGGGAAGTAGTCCTTGACCA IgM_long_8N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNGAAGGAAGTCCTGTGCGAG IgM_long_12N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNNNNGAAGGAAGTCCTGTGCGAG IgE_long_8N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNAAGTAGCCCGTGGCCAGG IgE_long_12N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNNNNAAGTAGCCCGTGGCCAGG IgD_long_8N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNTGGGTGGTACCCAGTTATCAA IgD_long_12N TGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNNNNTGGGTGGTACCCAGTTATCAA 10 . 7554/eLife . 16578 . 040Table 6 . IGH variable region primers . DOI: http://dx . doi . org/10 . 7554/eLife . 16578 . 040NamePrimer ( 5’ to 3’ ) Primer1_1_70 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNSCAGCTGGTGCAGTCTGG Primer1/3/5_70 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNGTGCAGCTGGTGGAGTCTG Primer2 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNTCACCTTGAAGGAGTCTGG Primer4_1 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNTGCAGCTGCAGGAGTCG Primer4_2 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNGTGCAGCTACAGCAGTGG Primer6 ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNGTACAGCTGCAGCAGTCA High-throughput sequencing was performed on the Illumina MiSeq platform ( Illumina , San Diego , CA ) with 300 bp paired end reads . Reads were passed through a pipeline to construct consensus sequences from reads containing the same 16 nt random barcode similar to ( Vollmers et al . , 2013 ) . Base quality scores in the consensus were calculated from the error probabilities associated with bases in the raw reads . Sequences were annotated with V and J germline gene usage and CDR3 length using IgBlast ( Ye et al . , 2013 ) . Classes were determined using BLASTN against a custom database of IGH constant region fragments . We have deposited the sequencing reads in the Sequence Read Archive ( accession number PRJNA324281 ) . Preprocessed sequence data and custom analysis scripts are available for download ( doi:10 . 5061/dryad . bv989 ) . Sequences belonging to the same clonal B cell lineage were identified using clustering as follows . Sequences sharing the same V-J combination and CDR3 length were grouped . Within each group , clusters were found by performing single linkage clustering with a cutoff of 95% sequence identity across both the CDR3 and the rest of the variable region . Sequence identity was computed from ungapped pairwise alignments by counting mismatches . Stringent quality filtering was implemented by assuming mismatches at positions at which the base in either aligned sequence had Q ≤ 5 . To choose the optimal cutoff for identifying sequences in clonal lineages , we examined the distributions of pairwise sequence identity within groups of CDR3 sequences sharing the same V and J gene combination and CDR3 length ( Figure 1—figure supplement 5 ) . By plotting the identity of each sequence to the most similar sequence in its group ( its 'nearest neighbor' ) , we saw that sequences separate into two groups: sequences with a highly similar nearest neighbor ( >90% identity ) and sequences that are substantially dissimilar to the nearest neighbor ( 40–80% identity ) . This pattern suggests that the first group consists of sequences that belong to a clonal lineage , while the second group consists of singleton sequences . Thus using a stringent cutoff of 95% sequence identity in the CDR3 ensures that the identified lineages contain sequences that are clonally related . In order to show that the results are not sensitive to the particular choice of a cutoff value we repeated subsequent analyses with cutoffs varying from 80% to 95% and obtained nearly identical results ( Figure 2—figure supplement 7 ) . Rarefaction of preprocessed sequences , which represent molecules of IGH mRNA , was performed by selecting a fraction of sequences uniformly at random from the sequences prior to lineage clustering using a custom script written in Python . The clonal history of each lineage was reconstructed using a custom algorithm . An ungapped multiple alignment of sequences in each lineage was performed by aligning the anchor sequences that mark the start and end of the CDR3 . The concatenated sequences of the V and J germline genes were then added to this alignment by performing a profile-profile alignment using MUSCLE with options '-maxiters 2 –diags' which introduces a gap corresponding to the D gene and untemplated nucleotides . A pairwise distance matrix was constructed by counting the number of substitutions required to transform each aligned sequence into the others . This matrix defines a weighted graph of possible ancestor-child relationships . A constraint on ancestry based on antibody class was then applied by pruning edges that violate the geometry of the IGH constant region locus . The minimum evolution tree was identified by using Edmonds’ algorithm to find a minimum spanning tree on this directed graph ( Edmonds , 1967 ) . The tree was rooted on the germline sequence . Class switching events were identified by traversing the minimum evolution tree for each lineage and searching for ancestor-child pairs of sequences having different classes . The probabilities that define the class switching landscape were calculated as follows . Relative switch frequency is equal to the number of switches observed from A to B divided by the total number of switches . Destination probability , which describes the probability that a cell switching from the ancestral class will choose the downstream class as the destination , is equal to the number of class switches from class A to class B divided by the total number of switches exiting class A . Similarly , arrival probability , which describes the probability that a cell of a given downstream class originated from the ancestral class , is equal to the number of class switches from class A to class B by the total number of switches entering class B . To generate shuffled ancestor-child pairs , we performed sampling without replacement on the list of ancestor classes to assign an ancestor class to each child class . To quantify correlations between twins , we developed a generalization of the intraclass correlation coefficient ( Shrout and Fleiss , 1979 ) for multidimensional data . We treat each measurement as an n-dimensional vector , where n is the number of probabilities measured per twin . To calculate the between- and within-target variance , we computed the distance from the mean vector using the squared Euclidean norm . Intraclass correlation is then equal to ( B – W ) / ( B + W ) , where B is the between-target variance and W is the within-target variance . To calculate intraclass correlation for twins , the twin pairs are treated as the within-target pairs . For unrelated individuals , all possible pairs of unrelated individuals are treated as the within-target pairs . To quantify the amount of hypermutation between class switching events , we searched in the clonal lineage trees for motifs consisting of multiple parent-child pairs sharing the same class prior to a class switch event . We filtered for motifs in which each child sequence inherited all of the somatic mutations relative to the V and J germline genes that were present in its parent , yielding ~74 , 000 switches for this analysis . The number of mutations accumulated before CSR was calculated by summing the number of substitutions separating parent-child pairs prior to the class switching event . Fitting of an exponential distribution to the empirical distribution of the number of mutations prior to class switching was performed using Python and SciPy ( Oliphant , 2007 ) in the IPython environment ( Perez and Granger , 2007 ) . To explore the inheritance of class switching fates , we searched in the trees for motifs in which two clonally related sequences sharing the same class ( 1 ) descended from a common progenitor of the same class and ( 2 ) subsequently switched to different classes . We further filtered for motifs where each descendant sequence inherited all of the somatic mutations relative to the V and J germline genes present in its ancestor . We binned these sequence pairs by relatedness measured by the number of substitutions separating the sequences from the common progenitor , using the maximum number among the two sequences . We then examined the downstream class to which each sequence switched . To quantify concordance , we calculated Yule's Q from the odds ratio describing whether both sequences switched to the same downstream class . Focusing on a single downstream class , let a be the number of cases where both sequence 1 and sequence 2 switched to this class , d be the number of cases where both sequence 1 and sequence 2 did not switch to this class , and b and c be the number of cases where sequence 1 switched to this class , but sequence 2 did not , and vice versa , respectively . Then the odds ratio OR is ( ad ) / ( bc ) and Yule’s Q is ( OR – 1 ) / ( OR + 1 ) . We also examined the conditional probabilities describing the class switch fate of one sequence given the class switch fate of the other sequence . We obtained whole blood drawn from volunteers at the Stanford Blood Center and prepared enriched B cell fractions using the RosetteSep kit ( StemCell Technologies , Cambridge , MA ) according to manufacturer’s instructions . We sorted CD19+ IgM+ cells and cultured them at 5 × 105 cells/ml for 5 days at 37 C and 5% CO2 in RPMI 1640 with L-glutamine ( ThermoFisher ) supplemented with 10% fetal bovine serum , 10 mM HEPES pH 7 . 4 , 0 . 1 mM non-essential amino acid ( Sigma-Aldrich , St . Louis , MO ) , 1 mM sodium pyruvate , 100 µ/ml penicillin , 100 µg/ml streptomycin ( ThermoFisher ) , 40 µg/ml apo-transferrin , 500 ng/µl multimeric CD40 ligand ( Miltenyi Biotec , San Diego , CA ) , 200 ng/ml IL-4 ( Sigma-Aldrich ) , and 200 ng/ml IL-10 ( Sigma-Aldrich ) . We extracted RNA from the cells using the RNeasy Micro Kit ( Qiagen ) according to manufacturer’s instructions , but omitting the DNase digestion step . We then prepared sequencing libraries using 24 . 5 ng of total RNA as input as described above , except that PCR products were purified using Ampure XP beads at a 0 . 65:1 ratio instead of a 1:1 ratio before pooling for multiplexed sequencing . We processed the sequences , reconstructed the clonal lineage histories , and measured correlations between the class switch fates of related cells as described above .	The human immune system comprises cells and processes that protect the body against infection and disease . B cells are immune cells that once activated produce antibodies , or proteins that help identify and neutralize infectious microbes and diseased host cells . Antibodies fall into one of ten different classes , and each class has a different , specialized role . Certain antibody classes are responsible for eradicating viruses , while others recruit and help activate additional cells of the immune system . B cells multiply quickly once they are activated . During this proliferation process , dividing B cells can switch from making one class of antibody to another . As such , a single activated B cell can yield a group of related B cells that produce distinct classes of antibodies . Although much has been learned about antibody class switching and its role in generating a diverse set of antibodies , the process of creating different antibody classes in humans remains unknown . Horns , Vollmers et al . now reveal how antibodies of every class are created in living humans . By developing a way to reconstruct the B cell proliferation process and thereby trace the lineage of individual B cells , the occurrence of class switching events could be measured and mapped . This approach revealed that most antibodies are produced via a single dominant pathway that involves first switching through one of two antibody classes . Horns , Vollmers et al . also determined that closely related B cells , which were recently born through division of a common ancestor , often switched to the same class . The shared fate is likely explained by the existence of similar conditions inside each cell , which are inherited during cell division and direct switching toward a particular class . All together , these new findings lay a foundation for developing techniques to direct antibody class switching in ways that support the immune system . Future work will aim to understand the conditions inside a cell that direct switching toward a particular class of antibody .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "immunology", "and", "inflammation" ]	2016	Lineage tracing of human B cells reveals the in vivo landscape of human antibody class switching
How genetic changes are linked to morphological novelties and developmental constraints remains elusive . Here , we investigate genetic apparatuses that distinguish fish fins from tetrapod limbs by analyzing transcriptomes and open-chromatin regions ( OCRs ) . Specifically , we compared mouse forelimb buds with the pectoral fin buds of an elasmobranch , the brown-banded bamboo shark ( Chiloscyllium punctatum ) . A transcriptomic comparison with an accurate orthology map revealed both a mass heterochrony and hourglass-shaped conservation of gene expression between fins and limbs . Furthermore , open-chromatin analysis suggested that access to conserved regulatory sequences is transiently increased during mid-stage limb development . During this stage , stage-specific and tissue-specific OCRs were also enriched . Together , early and late stages of fin/limb development are more permissive to mutations than middle stages , which may have contributed to major morphological changes during the fin-to-limb evolution . We hypothesize that the middle stages are constrained by regulatory complexity that results from dynamic and tissue-specific transcriptional controls . The genetic mechanism of morphological diversity among multicellular organisms is of central interest in evolutionary biology . In particular , our understanding of how morphological novelties are linked to the emergence of their respective genetic apparatuses is limited ( Rebeiz and Tsiantis , 2017 ) . In addition , it is still unclear to what extent internal constraints , such as pleiotropy , affect evolvability ( Wagner and Zhang , 2011 ) . The fin-to-limb transition is a classic , yet still influential , case study that contributes to our understanding of morphological evolution . In general , tetrapod limbs are composed of three modules , the stylopod , zeugopod , and autopod , which are ordered proximally to distally ( Figure 1A ) . In contrast , fish fins are often subdivided into different anatomical modules along the anterior−posterior axis—the propterygium , mesopterygium , and metapterygium ( Figure 1A ) . Although it is still controversial how this different skeletal arrangement compares with the archetypal tetrapod limb , the autopod ( wrist and digits ) seems to be the most apparent morphological novelty during the fin-to-limb transition ( Clack , 2009 ) . Despite intensive comparative studies of developmental gene regulation , genetic machinery that differs between fins and limbs remains elusive . Instead , several studies revealed that autopod-specific regulation of Hoxa13 and Hoxd10−13 , which control autopod formation , is also conserved in non-tetrapod vertebrates ( Davis et al . , 2007; Freitas et al . , 2007; Schneider et al . , 2011 ) , except that the expression domains of Hoxa13 and Hoxa11 are mutually exclusive in mouse and chick limbs while overlapping in examined fish fin buds ( note that axolotl limbs also exhibit such fish-like overlap of these expression domains; Ahn and Ho , 2008; Metscher et al . , 2005; Sakamoto et al . , 2009; Woltering et al . , 2019 ) . Although several gene regulatory differences have been proposed to explain the anatomical difference between fins and limbs , these proposals have been exclusively focused on Hox genes ( Kherdjemil et al . , 2016; Nakamura et al . , 2016; Sheth et al . , 2012; Woltering et al . , 2014 ) . Therefore , a genome-wide systematic study is required to identify the genetic differences between fish fins and tetrapod limbs . There have been several difficulties that limit genetic comparisons between tetrapods and non-tetrapod vertebrates . For example , whereas zebrafish and medaka are ideal models for molecular studies , their rapid evolutionary speed and a teleost-specific whole-genome duplication hinder comparative analyses with tetrapods at both the morphological and genetic levels ( Ravi and Venkatesh , 2008 ) . This obstacle can be circumvented by using more slowly evolving species such as spotted gar , coelacanths , and elephantfish ( also known as elephant shark , a cartilaginous fish that is not a true shark ) with their genome sequences that have not experienced recent lineage-specific genome duplications and thus facilitate the tracing of the evolution of gene regulation ( Amemiya et al . , 2013; Braasch et al . , 2016 ) . However , the major disadvantage of these slowly evolving species is the inaccessibility of developing embryos . In contrast , although the eggs of sharks and rays ( other slowly evolving species; Hara et al . , 2018 ) are often more accessible , their genomic sequence information has not been available until recently . As a solution for these problems , this study used embryos of the brownbanded bamboo shark ( referred to hereafter as the bamboo shark ) , because a usable genome assembly was recently published for this species ( Hara et al . , 2018 ) . Importantly , its non-coding sequences seem to be more comparable with those of tetrapods than with teleosts ( Hara et al . , 2018 ) . In addition , this species is common in aquariums and has a detailed developmental staging table , providing an opportunity to study embryogenesis ( Onimaru et al . , 2018 ) . These unique circumstances of the bamboo shark enabled a comprehensive study to identify the genetic differences between fins and limbs . In this study , to identify genetic differences between fins and limbs , we performed RNA sequencing ( RNA-seq ) analyses of developing bamboo shark fins and mouse limbs . Along with this transcriptomic comparison , we also generated an accurate orthology map between the bamboo shark and mouse . In addition , we applied an assay for transposase‐accessible chromatin with high‐throughput and chromatin accessibility analysis ( ATAC-seq; Buenrostro et al . , 2013 ) across a time series of mouse limb buds , which generated a high-quality data set the showing dynamics of open-chromatin regions ( OCRs; putative enhancers ) during limb development . We also analyzed the evolutionary conservation of sequences in these OCRs to gain insights into the gene regulatory changes during the fin-to-limb transition . To compare the temporal dynamics of gene expression between bamboo shark fin and mouse limb development , we obtained RNA-seq data from a time series of growing fin and limb buds with three replicates ( Figure 1B; Supplementary file 1 for the details of RNA-seq ) . We selected limb buds from embryonic day ( E ) 9 . 5 to E12 . 5 mice because this is the period during which the major segments of the tetrapod limb—the stylopod , zeugopod , and autopod—become apparent . In particular , the presumptive autopod domain , which is a distinct structure in the tetrapod limb , is visually recognizable from E11 . 5 . For the bamboo shark , we selected developing fin stages from as wide a time period as possible ( Figure 1B ) . To perform fine-scale molecular-level comparison , we annotated its coding genes using BLASTP against several vertebrates ( listed in the Materials and methods ) and our custom algorithm . As a result , 16443 unique genes from 63898 redundant coding transcripts were annotated as orthologous to known genes of vertebrates , among which 13 , 005 genes were uniquely orthologous to mouse genes ( Table 1 for details of the transcriptome assembly; Figure 1—figure supplement 1—3 , Supplementary files 2 and 3 for gene annotations and Supplementary data for sequence information ) . The number of detected orthologs is reasonable when compared with other studies ( e . g . Hao et al . , 2020 ) . The quality of the ortholog assignment , which was assessed by examining Hox and Fgf genes , showed that our custom algorithm is more accurate than other methods ( Figure 1C; see Materials and Methods and Supplementary file 4 for details ) . Using this assembly for the bamboo shark and RefSeq genes for mice , the means and standard errors of the transcripts per million ( TPM ) values were calculated from three replicates ( see Figure 1—figure supplement 4 for other normalization methods and Supplementary files 5 and 6 for the full list of TPM values ) . In addition , for most of the analyses , TPMs were scaled by setting the highest TPM in each gene of each species to ‘1’ ( which we refer to as the Max one method ) to capture temporal dynamics rather than absolute transcript amounts . Compared to using intact TPMs and other scaling methods , Max one is relatively sensitive to interspecific differences in dynamically regulated gene expression ( see Materials and methods and Figure 1—figure supplements 5 and 6 for details ) . With this transcriptome data set and gene annotation , we first validated our data by analyzing the expression profiles of Hoxa and Hoxd genes . In mouse limb development , Hoxa and Hoxd genes undergo two phases of global regulation ( Deschamps and Duboule , 2017 ) . During the first phase , Hoxd genes are regulated by an enhancer group located 3ʹ of the entire HoxD cluster , and the Hoxd genes are sequentially upregulated from 3ʹ to 5ʹ . The outcome of this first phase helps to establish the arm and the forearm . During the second phase , enhancers located 5ʹ of the HoxD cluster start to activate expression of Hoxd10 to Hoxd13 in the presumptive autopod region ( Hoxa genes are regulated in a similar manner; Deschamps and Duboule , 2017 ) . As expected , we detected the two phases of Hoxd gene regulation in mouse limb transcriptomes; the expression levels of Hoxd1 to Hoxd8 were highest at E9 . 5 ( the first phase regulation ) , and Hoxd11 to Hoxd13 were gradually upregulated later ( the second phase regulation; Figure 1D ) . Interestingly , the expression levels of Hoxd9 and Hoxd10 were highest at E10 . 5 , which probably represents the transitional state between the first and second global regulation ( Andrey et al . , 2013 ) . A similar profile was observed for Hoxa genes ( Figure 1D ) . As with mouse limb buds , we found similar phasic regulation of Hoxa and Hoxd genes in the bamboo shark fin transcriptome ( Figure 1D ) , suggesting that these transcriptomic data cover comparable developmental stages between the two species at least with respect to Hox gene regulation . The overall similarity in the temporal dynamics of Hox gene expression between the mouse limb bud and the bamboo shark fin bud is an expected result because the second phase of Hoxd gene regulation has been found to be conserved in the fins of many fish ( Ahn and Ho , 2008; Davis et al . , 2007; Freitas et al . , 2007; Schneider et al . , 2011; Tulenko et al . , 2017 ) . However , there are several differences that are worth noting . For example , in mouse limb buds , Hoxd11 and Hoxd12 expression was highest at E11 . 5 , followed by further upregulation of Hoxd13 at E12 . 5 ( Figure 1D ) . In contrast , in bamboo shark fin buds , these three genes reached their peak expression simultaneously at [stage ( st ) ]31 ( Figure 1D ) . This led us to investigate further whether the quantitative collinearity of 5ʹ Hoxd genes , where the expression of Hoxd13 is much higher than that of its neighboring Hoxd genes , whose transcription levels decrease with increasing distance from Hoxd13 ( Montavon et al . , 2008 ) , is conserved in the bamboo shark fin buds . First , as a confirmation of the previous observation , we also found quantitative collinearity of Hoxd genes in our transcriptome data of mouse limb buds at E12 . 5 ( Figure 1—figure supplement 7 ) . However , the bamboo shark fin buds exhibited no clear relationship between the genomic loci and the expression levels of Hoxd genes at either st31 or st32 ( Figure 1—figure supplement 7 ) : Hoxd12 expression was highest among its neighbors . Hoxd9 showed the second highest expression , followed by Hoxd10 and Hoxd11 , which had roughly identical levels of transcripts . Hoxd13 expression was lowest among these 5ʹ members . Given that quantitative collinearity is considered to be a consequence of the characteristic global regulation of the HoxD cluster in the mouse limb bud ( Montavon et al . , 2008 ) , this result suggests that the bamboo shark fin bud may have a different mechanism for Hoxd gene regulation . Interestingly , a recent study also showed that the presumptive autopod domains of chick limb buds express nearly a same amount of Hoxd13 and Hoxd12 transcripts ( Yakushiji-Kaminatsui et al . , 2018 ) , suggesting that quantitative collinearity is not a universal feature of fins and limbs , rather varies among species . Taken together , although the overall temporal dynamics of Hox gene expression are conserved between the mouse limb bud and the bamboo shark fin bud , some differences in the regulation of Hox genes may exist between species . To investigate to what extent our bulk transcriptome data captured the processes of cellular differentiation , we also analyzed genes related to chrondrogenesis and myogenesis . As a result , we found that the chondrogenic pathway was at least partially conserved between bambooshark fin buds and mouse limb buds; the expression level of Sox9 and Runx3 ( key transcription factors of chondrogenesis; Fowler and Larsson , 2020 ) increased relatively early , and that of Acan ( a cartilage-specific proteoglycan; Fowler and Larsson , 2020 ) was upregulated later ( Figure 1—figure supplement 8 ) . In contrast , although Nog is known to be expressed in cartilaginous condensations in mouse limb buds ( Brunet et al . , 1998 ) , we did not detect a Nog ortholog in either the fin transcriptome or the genome assembly of the bamboo shark . As for myogenesis , our transcriptome data captured both conserved and divergent myogenetic regulation: Pax3 ( a marker of myogenic precursor cells ) was downregulated over developmental time , and the MyoD gene family ( Myog , Myod1 , Myf5 ) took turns for further differentiation ( Chal and Pourquié , 2017 ) . In contrast , whereas mouse limb buds showed upregulation of three myosin genes ( Myh3 , Myh7 , Myh8 ) at E12 . 5 , we detected the upregulation of only Myh7 in bamboo shark fin buds . Again , we did not find Myh3 and Myh8 in either the transcriptome or the genome assembly of the bamboo shark . These results suggest that our transcriptome data , even though based on bulk sampling of RNA , can reveal conserved and diverged cellular differentiation processes . Next , to find differences in gene regulation between the two species , we performed a gene-by-gene comparison of expression dynamics with hierarchical clustering ( Figure 2A ) . To find potential candidate genes that contribute to the different morphologies between fins and limbs , we annotated genes with mouse mutant phenotypes ( see Supplementary file 7 for the full list of genes , expression data , and annotation ) . The result showed that 6701 genes were significantly expressed in only one of these species ( ‘Fin-specific’ and ‘Limb-specific’ in Figure 2A; 3284 and 3417 genes , respectively ) . While the fin-specific gene group consisted of many uncharacterized genes , it included ones that are known to control only fish fin development ( Fischer et al . , 2003; Zhang et al . , 2010 ) , such as And1 ( TRINITY_DN62789_c1_g1_i3 in Supplementary data; ortholog of a coelacanth gene , XP_015216565 ) and Fgf24 ( TRINITY_DN92536_c7_g1_i2 in Supplementary data; ortholog of a coelacanth gene , XP_006012032 ) . In the limb-specific gene group , several interesting genes were listed that exhibit abnormal phenotype in the mouse limb ( e . g . Bmp2 , Ihh , and Megf8 ) . However , the number of these species-specific genes is probably unreliable and overestimated because these groups also contain genes for which their orthology was not assigned correctly . We also detected 1884 genes that were upregulated during late stages of fin/limb development for both species , including genes that are well known to be expressed later during fin/limb development , such as the autopod-related transcription factors Hoxd13 and Hoxa13 and differentiation markers Col2a1 and Mef2c ( ‘Conserved , late1 and Conserved , late2’ in Figure 2A ) . Intriguingly , 5388 genes exhibited heterochronic expression profiles; their expression levels were highest during the late stages of mouse limb bud development but were relatively stable expression throughout fin development ( ‘Heterochronic1’; 3178 genes ) or decreased during the late stages of fin development ( ‘Heterochronic2’; 2223 genes; see Supplementary file 7 for the full list of genes and annotations ) . For validation , we examined the spatio-temporal expression pattern of three heterochronic genes that exhibit limb abnormality in mouse mutants , Aldh1a2 from Heterochronic1 and , Hand2 and Vcan from Heterochronic2 . Aldh1a2 is upregulated in the interdigital web of mouse limb buds from E11 . 5 ( Figure 2—figure supplement 1A ) and known to positively regulate interdigital cell death ( Kuss et al . , 2009 ) . On the other hand , in bamboo shark fin buds , Aldh1a2 expression was initially uniform and was later restricted to the distal edge of fin buds ( Figure 2—figure supplement 1A ) . Hand2 and Vcan transcripts were upregulated in mouse forelimb buds at E12 . 5 and downregulated in bamboo shark fin buds at st32 ( Figure 2B , C ) . Thus , the temporal transcriptomic profiles were consistent with spatial expression patterns . For a comparison , we found relatively few genes that were downregulated over time in the mouse limb bud but were upregulated in the shark fin . There was a total of 241 such genes , but only 43 of them displayed a clear heterochrony ( yellow empty box in Figure 2—figure supplement 1B and Supplementary file 8 for the list of the genes ) . Of those , Fgf8 is particularly interesting as FGF8 plays a crucial role as a growth signal from the apical ectodermal ridge ( AER ) in mouse and chick limb buds ( Lewandoski et al . , 2000 ) . As shown in Figure 2—figure supplement 1C , Fgf8 expression was high during the early stages of limb buds and was gradually downregulated at later stages . In contrast , in bamboo shark fin buds , Fgf8 was expressed very weakly ( around 0 . 1 TPM ) at st . 27 and st . 27 . 5 and was upregulated at later stages . Indeed , this late upregulation of Fgf8 was also reported in the apical fin fold ( roughly equivalent to the AER ) of zebrafish pectoral fin buds ( Nomura et al . , 2006 ) . In the zebrafish pectoral fin bud , Fgf16 and Fgf24 are upregulated earlier than Fgf8 ( Draper et al . , 2003; Nomura et al . , 2006 ) . In addition , Fgf4 , Fgf9 , and Fgf17 are expressed in the AER and have a redundant function in the mouse limb bud ( Mariani et al . , 2008 ) . Therefore , we also examined these other Fgf genes and found that moderate expression of Fgf9 , Fgf16 , and Fgf24 were detected in the early stages of bamboo shark fin buds ( Figure 2—figure supplement 1C ) . Although we cannot infer the ancestral state of the expression pattern , the overlapping functions of these genes may have allowed subfunctionalization of the signaling molecules of the AER during vertebrate divergence . In sum , we detected mass heterochronic shifts in gene expression between bamboo shark fin buds and mouse forelimb buds . In particular , a mechanism to maintain upregulation of the expression of genes involved in early fin development may have been either gained in the tetrapod lineages or lost in the cartilaginous fish lineages . In tetrapod limbs , SHH controls growth and asymmetric gene expression along the anterior-posterior axis . Although previous studies have repeatedly implied a relatively delayed onset of Shh expression or a short signal duration in developing fins of several elasmobranch species ( Dahn et al . , 2007; Sakamoto et al . , 2009; Yonei-Tamura et al . , 2008 ) , there has not been solid evidence to support such a delay due to the lack of systematic gene expression analysis and the poor staging system of these species . Because the heterochronic genes identified above include basic SHH target genes , such as Ptch1 and Gli1 , we reexamined the expression dynamics of Shh and its target genes in mouse limb and bamboo shark fin buds . Because HOX genes are the upstream factors relative to Shh transcription ( Zeller et al . , 2009 ) , we used them as a potential reference for developmental time . We first found that Shh transcription was present by the earliest stages examined in both bamboo shark fin and mouse limb buds , and it peaked when the transcription level of Hoxd9 and Hoxd10 was highest , suggesting that there was no apparent heterochrony in Shh transcription timing at least between these two species ( red rectangles in Figure 3A and B ) . In contrast , SHH target genes , such as Ptch1/2 , Gli1 , Gremlin , and Hand2 ( Vokes et al . , 2008 ) , did show a relatively extended period of expression in mouse limb buds as compared with their expression in bamboo shark fin buds . Namely , whereas the expression peak of SHH target genes was concurrent with that of Shh in the bamboo shark fin bud , these SHH target genes were highly expressed in E11 . 5 limb buds , which is one day later than the Shh expression peak ( yellow rectangles in Figure 3A and B; see Figure 3—figure supplement 1 for intact TPM values ) . This timing difference is also apparent when comparing the expression peak of Hoxd11 and Hoxd12 , which was concurrent with that of SHH target genes in mouse limb buds , but came after downregulation of SHH target genes in bamboo shark fin buds ( green rectangles in Figure 3A and B ) . To confirm this observation , we performed whole-mount in situ hybridization for Ptch1 and Hoxd12 in mouse limb buds and bamboo shark fin buds . As previously reported ( Lewis et al . , 2001; Zákány et al . , 2004 ) , mouse limb buds showed a clear expansion of the expression domain of Ptch1 ( upper panel in Figure 3C ) from E10 . 5 to E11 . 5 , which is accompanied by the anterior extension of the Hoxd12 expression domain ( black arrowheads in Figure 3C ) . In contrast , Ptch1 was expressed in the posterior domain of bamboo shark fin buds at st . 29 ( white arrowheads in Figure 3D ) , but was substantially downregulated by st . 31 , whereas the Hoxd12 expression domain extended anteriorly at this stage ( black arrowheads in Figure 3D ) . These results were roughly consistent with the RNA-seq data . We cannot completely reject the possibility that this timing difference is due to the different physical time-resolution of data sampling between these species ( six time points over 20 days in the bamboo shark and four time points over 4 days in the mouse ) . However , given that this data set captured the similar expression dynamics of HoxA/D clusters between these species ( Figure 1D; also see Figure 4C ) as well as the differentiation dynamics of myocytes and chondrocytes ( Figure 1—figure supplement 8 ) , these results quite likely represent an interesting difference in the transcriptional regulation of SHH downstream genes between fins and limbs . Several studies have reported a temporally heterogeneous diversification of embryonic transcriptomes , such that the middle stages are more conserved than early or late stages ( e . g . Irie and Kuratani , 2011; Kalinka et al . , 2010; Levin et al . , 2012 ) . These observations are considered to support the notion of the developmental hourglass ( or egg timer ) , which has been proposed to explain the morphological similarity of mid-stage embryos based on developmental constraints , such as strong interactions between tissues or Hox-dependent organization of the body axis ( Duboule , 1994; Raff , 1996 ) . In addition , a previous transcriptomic analysis reported that the late stage of mammalian limb development has experienced relatively rapid evolution ( Maier et al . , 2017 ) . To examine which developmental stages of fins and limbs are conserved , we calculated the distance between the fin and limb transcriptome data . As a result , four different distance methods that we examined consistently indicated that the limb bud at E10 . 5 and the fin buds at st27 . 5–30 tended to have a relatively similar expression profile ( Figure 4A for a Euclidean distance measure and Figure 4—figure supplement 1 for other types of distance measures ) . In addition , the transcriptomic profile of all the stages of examined fin buds showed the highest similarity to that of E10 . 5 limb bud ( Figure 4B ) . Therefore , the mid-stages of limb and fin buds tend to be conserved over 400 million years of evolution . To find factors that underlie the mid-stage conservation , we analyzed Hox genes , which were proposed to be responsible for the developmental hourglass ( Duboule , 1994 ) . We found that the comparison of only Hox gene expression did not reproduce the hourglass-shaped conservation ( Figure 4C ) , suggesting that other mechanisms constrain the middle stage of development . We further performed principal component analysis ( PCA ) of gene expression profiles to identify genes responsible for the hourglass-shaped conservation . The first component , PC1 , distinguished transcriptome data mostly by species differences ( Figure 4D ) . In contrast , PC2 was correlated with the temporal order of mouse limb buds ( Figure 4D ) . PC2 was also weakly correlated with the temporal order of bamboo shark fin buds except at st27 ( Figure 4D ) , but PC3 showed a clearer correlation ( Figure 4E ) . These three components were mostly sufficient to reproduce the mid-stage conservation in Figure 4A ( Figure 4—figure supplement 2A for the ratio of explained variables and 2B for the Euclidean distance measure ) . Interestingly , the plot with PC2 and PC3 roughly mirrored the hourglass-shaped conservation because the earliest and latest stages were placed more distantly than the middle stages in this representation ( Figure 4E ) . Indeed , the major loadings of PC2 consisted of the conserved late expressed genes ( C8 ) and the heterochronically regulated genes ( C9 and C12 ) identified in Figure 2A ( see Table 2 for the top 25 genes of PC2 ) . Similarly , PC3 consisted of the conserved early genes ( a part of C15 ) and the heterochronically regulated genes ( C12 and C13; see Supplementary file 9 for the loadings of PC3 and others ) , suggesting that the presence of heterochronically regulated genes may at least partly contribute to the mid-stage conservation and the distant relationship between the early/late stages of fins and limbs . These results indicate that the mass heterochronic shift in gene expression , at least in part , contributes to the long distances between early- and late-stage expression profiles ( Figure 4E ) . Because a recent report suggests that pleiotropy of genes is related to hourglass-shaped conservation ( EXPANDE Consortium et al . , 2017 ) , we counted the number of genes with stage- or tissue-specific expression . Consistent with the previous report ( EXPANDE Consortium et al . , 2017 ) , we detected a relatively low number of stage-associated genes during the middle stages of mouse forelimb and bamboo shark fin development ( Figure 4—figure supplement 2C ) . To evaluate the tissue specificity of genes , we first calculated Shannon entropy of gene expression patterns by analyzing RNA-seq data from 71 mouse tissues as released by the ENCODE project ( Davis et al . , 2018; Supplementary file 10 for the list of RNA-seq data ) . Namely , genes expressed only in a few tissues score lower with respect to entropy ( thus , these genes are more specific ) . We counted genes with 1 . 0 ≥ TPM and 0 . 65 ≤ entropy and , again , found that the number of tissue-associated genes was relatively low at E10 . 5 ( Figure 3F ) . Together , these results indicate an inverse correlation between the hourglass-shaped conservation and the number of tissue- and stage-specific genes . Next , we systematically identified putative gene regulatory sequences involved in mouse limb development and sought a possible cause for the hourglass-shaped conservation in gene regulatory sequences . To this end , we applied ATAC-seq , which detects OCRs ( putative active regulatory sequences ) , to time-series of forelimb buds at E9 . 5–E12 . 5 with three replicates . First , as a positive control , we found that ATAC-seq peaks that were determined by MACS2 peak caller covered 10 of 11 known limb enhancers of the HoxA cluster ( Figure 5A and Figure 5—figure supplement 1 ) , suggesting a high coverage of true regulatory sequences . Consistently , our ATAC-seq data showed relatively high scores for a quality control index , fraction of reads in peaks ( FRiP ) , as compared with data downloaded from the ENCODE project ( Davis et al . , 2018; Figure 5B ) . Next , to examine evolutionary conservation , we performed BLASTN ( Camacho et al . , 2009 ) for the sequences in the ATAC-seq peaks against several vertebrate genomes . Reinforcing the result of the transcriptome analysis , we found that evolutionarily conserved sequences were most accessible at E10 . 5 ( Figure 5C ) . To confirm this result , we also used a different alignment algorithm , LAST ( Kiełbasa et al . , 2011 ) with the bamboo shark and the alligator ( Green et al . , 2014 ) genomes . Alignment results for both analyses consistently indicated that the OCRs of E10 . 5 forelimb bud more frequently contained conserved sequences relative to those of other time points ( Figure 5D; see Figure 5—figure supplement 2A and B for the absolute counts of conserved sequences ) . Therefore , activation of conserved gene regulatory sequences may be one of the proximate causes for the hourglass-shaped conservation of fin and limb transcriptome data . To further characterize the ATAC-seq peaks , we next performed a clustering analysis . Using one of the three replicates for each stage , we collected the summits of peaks and the surrounding 1400 bp and carried out hierarchical clustering , which resulted in eight clusters ( C1–C8; Figure 6A ) that consisted of broad ( C1 and C2 ) , E11 . 5/E12 . 5-specific ( C3 and C4 ) , stable ( C5 and C6 ) , E10 . 5-specific ( C7 ) , and E9 . 5-specific ( C8 ) peaks . The overall clustering pattern was reproducible by other combinations of replicates if its FRiP was ≥0 . 20 ( Figure 6—figure supplement 1 ) . Consistent with the above conservation analysis , E10 . 5-specific peaks frequently overlapped conserved sequences ( Figure 5—figure supplement 2C and D ) . To characterize the regulatory features of the clusters , we performed motif analysis in each cluster using HOMER ( Heinz et al . , 2010 ) . First , it was convincing that stable peaks ( C5 and C6 ) were enriched for the CTCF binding motif both in de novo motif discovery ( Figure 6A ) and known motif enrichment analysis ( Figure 6—figure supplement 2 ) , which is a major regulator of three-dimensional genomic structure . This result was consistent whether random genomic regions or other peak regions were used for the background ( Figure 6—figure supplement 3 ) . In addition , E11 . 5/E12 . 5-specific peak C3 was enriched for the HOX13 motif ( Figure 6A ) , which was consistent with the increase in the expression of 5ʹ Hox genes ( Figure 1D ) . C4 was also enriched for motifs similar to those of C3 , but the HOX13 motif was detected only in known motif enrichment analysis ( compare Figure 6—figure supplements 2 and 3 ) . The enrichment of the HOX9 motif in E10 . 5-specific peaks ( C7 ) was also consistent with our RNA-seq data , in which Hoxd9 and Hoxa9 expression levels peaked at E10 . 5 ( Figure 1D ) . Interestingly , in E10 . 5-specific peaks ( C7 ) , the LHX1-binding motif was ranked at the top of the motif enrichment list the closely related transcription factors Lhx2 , Lhx9 , and Lmx1b are required to mediate a signaling feedback loop between ectoderm and mesenchyme in limb development ( Tzchori et al . , 2009 ) . C8 was enriched for motifs similar to those in C7 ( e . g . COUP-TFII ) , but the top-ranked transcription factor in the de novo motif discovery analysis was VSX2 , which has a very similar binding sequence to the LHX motif ( Figure 6—figure supplement 4 ) . The LHX motif was top-ranked in C8 for the known motif enrichment analysis ( Figure 6—figure supplement 2 ) . For a better understanding of the dynamics of transcription factor motifs , we counted the average number of the above detected motifs within the OCRs of each stage , which revealed a transitional increase in LHX and HOX9 motifs at E10 . 5 and a gradual increase in the motifs detected in C3 over the developmental stages ( Figure 6—figure supplement 3 ) . In addition , Gene Ontology ( GO ) analysis for the peaks in each cluster revealed that the constitutively accessible peaks ( C5 , C6 ) were closely located to genes annotated with ‘cellular components’ ( Supplementary file 11 ) . Interestingly , the dynamically regulated peaks ( C3 , C4 , C7 , C8 ) were associated with genes with ‘developmental process’ , ‘multicellular organism development’ , and ‘anatomical structure morphogenesis’ ( Supplementary file 11 ) , suggesting that these dynamic OCRs regulate developmental genes . Together , these results suggest that there are E10 . 5-specific transient OCRs that exhibit several characteristics including their evolutionary conservation , the presence of LHX and HOX9 motifs and a close relation with developmental genes . To confirm the results from the above clustering analysis , we also determined the genomic regions that showed a statistically significant increase or decrease in the ATAC-seq signal within a day by using all replicates . As a result , ATAC-seq signals were most increased during the transition from E9 . 5 to E10 . 5 in the mouse limb bud . From E10 . 5 to E11 . 5 , the total number of decreased and increased signals was highest , indicating that the OCR landscape was most dynamically changing at E10 . 5 ( Figure 6B ) . In contrast , relatively few significant changes were observed from E11 . 5 to E12 . 5 . Thus , in contrast to the transcriptome analysis , stage-specific gene regulatory sequences are likely to be most accessible at E10 . 5 . Moreover , by comparing the peaks of each cluster identified above with ATAC-seq peaks of other cells and tissues released by the ENCODE project ( Davis et al . , 2018; Supplementary file 10 for the full list of cells and tissues ) , we discovered that the C7 cluster ( E10 . 5-specific peaks ) contained more peaks that did not overlap with those of other cells and tissues . Again , in contrast to the transcriptome analysis , the data suggest that gene regulatory sequences that are accessible only at E10 . 5 tend to be limb-specific ( Figure 6C ) . Taken together , these analyses revealed a unique regulatory landscape of forelimb buds at E10 . 5 , which is enriched for evolutionarily conserved stage-specific and tissue-specific OCRs . In this work , we applied transcriptomics and chromatin accessibility analysis to systematically study genetic changes that differentiate fins from limbs . Because of the slow sequence evolution and the embryo availability of the bamboo shark , we were able to compare transcriptional regulation of genes with high accuracy and found both heterochronic shifts and hourglass-shaped conservation of transcriptional regulation between fin and limb development . Here , we discuss the interpretations , limitations , and implications of these results . Our time-series transcriptome data indicated that a remarkable number of genes that exhibit the highest expression during the late stages of mouse limb bud development are decreased during the late stages of bamboo shark fin development ( Figure 2 ) . The simplest hypothesis for this mass heterochronic shift is that the later stages of limb development gained expression of one or a few upstream transcription factor ( s ) or signaling molecules that collectively regulate this group of genes . Interestingly , we also observed relatively extensive expression of the downstream targets of the SHH signaling pathway in mouse limb buds , as compared with bamboo shark fin buds ( Figure 3 ) . Because SHH-independent regulation of its target genes through the GLI3-HOX complex was previously reported ( Chen et al . , 2004 ) , the mismatch between the peak expression of Shh and its target genes may be caused by such SHH-independent regulatory mechanisms that are absent in bamboo shark fin development . Given that direct and genetic interactions of GLI3 and HOX have a significant impact on autopod formation , the emergence of this interaction may be a key component of the mass heterochronic shift and the acquisition of autopod-related developmental regulation in the tetrapod lineages . However , because we compared only two species , it is equally possible that the late stages of shark fin development lost this SHH-independent gene regulation . Alternatively , given that the evolutionary distance between these two species is >400 million years , it is also possible that every one of these genes independently shifted their expression to the later stages of limb development or to the early stages of shark fin development . Further taxon sampling and functional analyses will reveal the relation between the mass heterochronic shift and the emergence of the autopod . Related to the potential changes in regulation of SHH target genes , by analyzing catshark fin buds , we previously proposed that the expression domains of genes that are positively regulated by SHH might have expanded anteriorly during the fin-to-limb transition ( Onimaru et al . , 2015 ) . We speculated that this expression changes may be linked to the loss of pro- and mesopterygial elements . Recently , this hypothesis was partially supported by another group who compared the gene expression pattern of lungfish and cichlid fin development ( Woltering et al . , 2020 ) , where lungfish fin buds seem to exhibit an intermediate condition between non-sarcopterygian fish fins and tetrapod limbs in terms of gene expression distribution along the anterior-posterior axis . This group particularly emphasized that the absence and presence of the dynamics of the anterior expansion of Hoxd13 expression correlate with the difference between the metapterygial morphologies of lungfish and tetrapods ( also see Johanson et al . , 2007 for a conflicting report ) . However , the significance of changes in Hoxd13 expression remains unclear because of the following two reasons: ( a ) Hoxd13 expression pattern seems to quite vary among species—the anterior expansion of Hoxd13 expression has been observed in the fin buds of the little skate , the small-spotted catshark , and Polyodon ( Davis et al . , 2007; Freitas et al . , 2007; Nakamura et al . , 2015 ) , while not in those of zebrafish and cichlids ( Ahn and Ho , 2008; Woltering et al . , 2020 ) and ( b ) in fish fins , the expression domain of Hoxa13 , whose function is mostly redundant with Hoxd13 , commonly spans from anterior to posterior regions in fish fin buds like as tetrapod limbs ( Davis et al . , 2007; Freitas et al . , 2007; Nakamura et al . , 2016 ) . Therefore , while changes in Hoxd13 expression domain are likely to contribute to some degree of anatomical diversity , their impact is questionable in the context of the fin-to-limb transition . Nevertheless , our previous study and Woltering et al . commonly suggest that the anterior expansion of gene expression domains is likely associated with the substantial anatomical changes during the fin-to-limb transition . As discussed above , we speculate that the mass heterochronic shifts that we observed in the present study may be related to the gain of SHH-independent regulation of its target genes . Therefore , whether the anterior expansion of SHH-target gene expression is related to the mass heterochronic shifts will be one of the interesting questions to address in the future . We observed that gene expression profiles are most highly conserved between bamboo shark fin buds at st . 27 . 5–30 and mouse forelimb buds at E10 . 5 ( Figure 4 ) . Consistent with this result , our chromatin accessibility analysis reveals that OCRs at E10 . 5 tend to contain evolutionarily conserved sequences ( Figure 5 ) . Whereas transcriptomic conservation during the middle of embryonic development has been reported by many groups using different species ( e . g . Irie and Kuratani , 2011; Kalinka et al . , 2010 ) , analysis of regulatory sequence conservation during embryonic development has been either incomplete or controversial . For example , by analyzing histone acetylation marks on several developing organs in mouse embryos , Nord et al . , 2013 proposed regulatory sequences active at E11 . 5 are exposed by the highest evolutionary constraint . However , they used stem cell lines as the substitutes for organs at early stages . Another study showed that genes expressed at the segmentation stage of zebrafish embryos tended to be surrounded by highly conserved non-coding sequences ( Piasecka et al . , 2013 ) . Although their results are in line with our present study as discussed below , they did not show that these highly conserved non-coding sequences were indeed active at the segmentation stage . In addition to these studies , there is a conflicting observation that early , instead of middle , embryonic stages tend to be regulated by conserved OCRs ( Uesaka et al . , 2019 ) . Therefore , our present study is the first to convincingly show a clear correlation of conservation status between transcriptomic data and OCRs . Our results suggest that evolutionary constraints on the gene regulatory apparatus are present during the middle stage of fin and limb development . What drives the hourglass-shaped conservation is still under debate . Interestingly , we found that stage- and tissue-specific OCRs were enriched in this conserved period , during which a relatively low number of stage- and tissue-specific genes were expressed ( Figure 6 ) . These quite contrasting observations imply that the mid-stage limb development is enriched for pleiotropic genes controlled by multiple tissue-specific enhancers , including limb-specific ones , rather than by constitutive promoters that often regulate housekeeping genes . Therefore , we speculate that , at least in the case of limb development , complex regulatory sequences that execute spatiotemporally specific transcriptional controls over pleiotropic genes constrain the evolvability of this particular period of morphogenesis , probably due to the vulnerability of complex regulation to genetic mutations . In conclusion , the present study provides insights for the evolutionary origin of gene regulation that differentiates fins from limbs . In particular , comparative transcriptional analyses prompted us to hypothesize that mass heterochronic shifts of gene expression may have occurred during the fin-to-limb evolution . In addition , both transcriptome and open chromatin data point to an evolutionary constraint during mid-stage limb development , likely owing to gene regulatory complexity . Although these hypotheses require further taxon sampling and experimental tests , this study opens up new prospects for understanding not only the genetic basis of the fin-to-limb transition but also the general nature of morphological evolution . Animal experiments were conducted in accordance with the guidelines approved by the Institutional Animal Care and Use Committee ( IACUC ) , RIKEN Kobe Branch , and experiments involving mice were approved by IACUC ( K2017-ER032 ) . The eggs of brownbanded bamboo shark ( C . punctatum ) were kindly provided by Osaka Aquarium Kaiyukan and were incubated at 25°C in artificial seawater ( MARINE ART Hi , Tomita Pharmaceutical Co . , Ltd . ) and staged according to the published staging table ( Onimaru et al . , 2018 ) . For mouse embryos , C52BL/6 timed-pregnant females were supplied by the animal facility of Kobe RIKEN , LARGE and sacrificed at different days after 9 . 5–12 . 5 days of gestation . For RNA-seq , fin buds and limb buds were dissected in cold seawater and phosphate-buffered saline ( PBS ) , respectively , and stored at −80°C . For in situ hybridization , embryos were fixed overnight in 4% paraformaldehyde in PBS , dehydrated in a graded methanol series , and stored in 100% methanol at −20°C . We sampled mouse forelimb buds at E9 . 5 , E10 . 5 , E11 . 5 and E12 . 5 and bamboo shark pectoral fin buds at st27 , st27 . 5 , st29 , st30 , st31 , and st32 and pooled several individual samples by stage to obtain enough RNA for each time point . We considered this pooled sample to represent one biological replicate ( other replicates were generated using different individuals ) . Total RNAs from these samples were extracted with the RNeasy Micro and Mini plus kit ( QIAGEN , Cat . No . 74034 and 74134 ) and PicoPure RNA Isolation Kit ( ThermoFisher , Cat . No . KIT0214 ) . Genomic DNA was removed with gDNA Eliminator columns included with this kit . For quality control , the Agilent 2100 Bioanalyzer system and Agilent RNA 6000 Nano Kit ( Agilent , Cat . No . 5067–1511 ) were used to measure the RNA integrity number for each sample . Using 237 ng of each of the RNA samples , strand-specific single-end RNA-seq libraries were prepared with the TruSeq Stranded mRNA LT Sample Prep Kit ( Illumina , Cat . No . RS-122–2101 and/or RS-122–2102 ) . For purification , we applied 1 . 8× ( after end repair ) and 1 . 0× ( after adapter ligation and PCR ) volumes of Agencourt AMPure XP ( Beckman Coulter , Cat . No . A63880 ) . The optimal number of PCR cycles for library amplification was determined by a preliminary quantitative PCR using KAPA HiFi HotStart Real-Time Library Amplification Kit ( KAPA , Cat . No . KK2702 ) and was estimated to be 11 cycles for mouse limb buds and 10 cycles for bamboo shark fin buds . The quality of the libraries was checked by Agilent 4200 TapeStation High Sensitivity D1000 . The libraries were sequenced after on-board cluster generation for 80 cycles using 1× HiSeq Rapid SBS Kit v2 ( Illumina , Cat . No . FC-402–4022 ) and HiSeq SR Rapid Cluster Kit v2 ( Illumina , Cat . No . GD-402–4002 ) on a HiSeq 1500 ( Illumina ) operated by HiSeq Control Software v2 . 2 . 58 ( Run type: SR80 bp ) . The output was processed with Illumina RTA v1 . 18 . 64 for base-calling and with bcl2fastq v1 . 8 . 4 for de-multiplexing . Quality control of the obtained fastq files for individual libraries was performed with FASTQC v0 . 11 . 5 . RNA-seq was performed with three biological replicates for each stage . We used the NCBI RefSeq mouse proteins ( GRCm38 . p5; only curated proteins were used ) and two bamboo shark gene lists: a genome sequence-based gene model ( Hara et al . , 2018 ) and transcripts assembled from RNA-seq in this study ( see below ) for orthology assignment . The amino acid sequences of the published gene model of the bamboo shark are available from https://doi . org/10 . 6084/m9 . fig ( Supplementary file 1 ) . For the transcriptome assembly , the short reads from the bamboo shark RNA-seq data were trimmed and filtered with Trim Galore ! ( https://www . bioinformatics . babraham . ac . uk/projects/trim_galore/ ) and assembled using Trinity v2 . 4 . 0 ( Grabherr et al . , 2011; options: --SS_lib_type RF --normalize_max_read_cov 200 min_kmer_cov 2 ) . Protein coding sequences were predicted with a program that finds coding regions , TransDecoder v3 . 0 . 1 ( Haas et al . , 2013 ) , according to the guide in TransDecoder ( Supplementary file 2 and 3 ) . Using these coding gene lists as queries , orthologous pairs were assigned as illustrated in Figure 1—figure supplement 1 . The idea behind this algorithm is the ‘gar bridge’ ( Braasch et al . , 2016 ) , an empirical observation that a comparison including intermediate and slowly evolving animals yields a better resolution for identifying homologous sequences than a direct comparison between two species . First , BLASTP v2 . 7 . 1 was performed between mouse and bamboo shark genes reciprocally , and also against the coding genes of the elephantfish ( or elephant shark; Callorhinchus_milii-6 . 1 . 3 ) , spotted gar ( LepOcu1 ) , coelacanth ( LatCha1 ) , chicken ( GRCg6a ) , alligator ( ASM28112v4 ) , and human ( GRCh38 . p12; options: -outfmt 6 -evalue 1e-30 -window 0 ) . Then , the BLASTP results of bamboo shark queries against the animals listed above ( except for the elephantfish ) were concatenated , and the best hit across species ( cross-species best hit ) was identified for each of the bamboo shark genes . If there was no cross-species best hit , then the best hit among the elephantfish genes was retrieved , which may include cartilaginous fish-specific genes . Subsequently , orthologous pairs between mouse and bamboo shark genes were assigned by checking if a cross-species best hit from the bamboo shark BLASTP results also had a best hit in the BLASTP result of mouse genes against the corresponding animal ( species-wise best hit; Supplementary files 4 , 5 , 6 ) . For quality control , the orthology of Fgf family members was independently determined by generating molecular phylogenetic trees ( Figure 1—figure supplements 2 and 3 ) . Amino acid sequences were aligned with an alignment tool , MAFFT v7 . 419–1 ( Katoh et al . , 2002; options: --localpair --maxiterate 1000 ) and trimmed with trimAL v1 . 2 ( Capella-Gutiérrez et al . , 2009; options: -gt 0 . 9 -cons 60 ) . Then , maximum-likelihood trees were constructed with RaxML v8 . 2 . 12 ( Stamatakis , 2014; options: -x 12345 p 12345 m PROTGAMMAWAG -f a -# 100 ) . The orthology of Hox genes was confirmed based on genome synteny . These independently confirmed orthologous pairs were compared with the results of the above orthology assignment algorithm . For a comparison , we also used the results from a reciprocal best hit algorithm , proteinOrtho v6 . 0 . 4 ( Lechner et al . , 2011 ) and the previously generated orthology groups ( Hara et al . , 2018; Figure 1B ) . The trimmed RNA-seq short reads were aligned to the transcript contigs for the bamboo shark and curated RefSeq genes ( GRCm38 . p5 ) for the mouse using RSEM v1 . 3 . 0 ( Li and Dewey , 2011 ) and Perl scripts ( align_and_estimate_abundance . pl and abundance_estimates_to_matrix . pl ) in the Trinity package . TPM ( transcripts per million ) , but not TMM ( trimmed mean of M-values ) , was used for all analyses , because we found some artificial biases in TMM values ( see Figure 1—figure supplement 4 ) . TPM values from the splicing variants of a single gene were summed up to generate a single value per gene . Then , the means and standard errors of TPM values from three replicates were used for the downstream analyses . Genes with a maximum TPM <1 . 0 were considered not expressed . For clustering and distance measures , TPM values were scaled so that the maximum value of each gene of each species was set to ‘1’ ( Max 1 ) . Whereas this scaling method loses information with respect to the absolute value of the TPMs , it has a substantial advantage when comparisons are being made between evolutionarily distant species . Indeed , previous comparative transcriptome studies have scaled gene expression values in different ways . Among those approaches , the use of Z-scores ( standardization ) and log transformations are relatively common strategies ( e . g . Kalinka et al . , 2010; Leiboff and Hake , 2019; Levin et al . , 2016 ) . Some researchers have used the intact RPKM ( reads per kilobase per million ) values to compare closely related species ( Wang et al . , 2013 ) , but , because the RPKM is known to be inconsistent between samples even within a species ( Wagner et al . , 2012 ) . Scaled transcriptional values are commonly used for clustering analyses and visualization of transcriptomic data from different samples within a single species . In this case , scaling is mainly aimed at flattening the dynamic range of transcription levels among genes . For inter-specific comparisons , scaling is also useful for being simultaneously sensitive to differentially regulated genes and also insensitive to conserved housekeeping genes . Here , we examine the effect of several scaling methods and the use of intact TPM values . We define the four relevant scaling methods as follows:Mg , s , t=xg , s , tmax{xg , s , t:t=1 . . Ts}Zg , s , t=xg , s , t-x¯g , sσg , sUg , s , t=xg , s , t{xg , s , t:t=1 . . Ts}Lg , s , t=log10xg , s , t+1where xg , s , t is the intact TPM of gene g , species s , and time point t; Ts is the total number of time points in species s; Mg , s , t , Zg , s , t , Ug , s , t and Lg , s , t are scaled values that we refer to as the Max 1 , Z-score , Unit vector and Log10 methods , respectively; and x¯g , s and σg , s are the mean and standard deviation , respectively , of {xg , s , 1 . . . xg , s , Ts} . First , we take a simple example to develop some intuition as to how these calculations transform TPM values . Let us assume that we compare two species [ ( species 1 and species 2 ) ] , and each species has two genes ( gene 1 and gene 2 ) and three developmental time points ( t1 , t2 , and t3; Figure 1—figure supplement 5A ) . Gene 1 is a constitutively active gene ( i . e . a housekeeping gene ) , and gene 2 is differentially regulated between species . In this example , we want to identify t2 as the most conserved time point because gene two is expressed in both species at this time point . In addition , we want to ignore the subtle expression differences of gene one within and between species . As seen in Figure 1—figure supplement 5A , scaling by the Max 1 , Unit vector , and Log10 methods effectively conserves the expression dynamics of gene two while suppressing the expression noise of gene 1 . In contrast , Z-score scaling amplifies the expression dynamics of both genes to the same degree , which suggests that the Z-score method is sensitive to noise . Calculation of the Euclidean distances for each time point between species 1 and 2 ( ‘Distance’ in Figure 1—figure supplement 5A ) shows that although all scaling methods and the use of intact TPMs indicate that t2 is the most similar time point , Max one creates a greater contrast between conserved and non-conserved time points than the other methods . Therefore , Max one is likely to be able to sensitively detect inter-specific differences . We also examined a subset of our real transcriptomic data from mouse limb buds and bamboo shark fin buds . As an example , we chose three housekeeping genes conserved in most vertebrates , Psmd5 , Mrpl21 , and Polr1b—these genes are listed both in a housekeeping gene list https://www . tau . ac . il/~elieis/HKG/HK_genes . txt ( Eisenberg and Levanon , 2013 ) and in the BUSCO data set , a gene list used to assess the completeness of genome assemblies ( Simão et al . , 2015 ) . As shown in Figure 1—figure supplement 5B and C , the TPM values of these genes were stable throughout developmental time in both species , suggesting that these genes also play a role in the maintenance of basic cellular function in bamboo shark fin development . However , the TPM values of Mrpl21 and Polr1b in mouse limb buds were roughly twice as high as those in bamboo shark fin buds . One explanation for this finding is that the expression of housekeeping genes is low in the bamboo shark because the relatively low temperature of the environment in which it lives slows its metabolic activity . We note , however , that there are many technical uncertainties when directly interpreting TPM values , particularly when comparing distantly related species . For example , differences in DNA sequences of transcripts ( such as variations in GC content ) between species probably affects the efficiency of library preparation and sequencing . The TPM values are also likely to be biased because of the incompleteness of the reference transcriptome sequence that we used for the bamboo shark ( e . g . some genes lack 3ʹ untranlated regions ) . Therefore , the dynamics of TPM values extracted by scaling methods rather than absolute TPM values are likely to contain more biologically relevant information . Of the scaling methods , Max 1 , Unit vector , and Log10 conserved the stable expression profile of the housekeeping genes , whereas the Z-score method amplified the subtle variation in TPM values as seen in the above simple example ( Figure 1—figure supplement 5B ) . In particular , the Max one and Unit vector methods transformed the TPM values into relatively comparable values between the two species ( compare Figure 1—figure supplement 5B with C ) . For a comparison , we also examined three genes that are heterochronically regulated between bamboo shark fin buds and mouse limb buds ( Figure 1—figure supplement 6A and B ) . In this case , all of the scaling methods seemed to conserve the temporal dynamics of gene expression . To obtain an objective measure , we calculated the ratio of the interspecific Euclidean distance of the three housekeeping genes to that of the three heterochronic genes with different scaling methods ( Figure 1—figure supplement 6C and D ) . Namely , the Euclidean distance of expression values was close to zero if we used only housekeeping genes ( left panel of Figure 1—figure supplement 6C ) , but it was larger when comparing heterochronic genes ( right panel of Figure 1—figure supplement 6C ) . As a result , the Max1 method resulted in the highest ratio ( Figure 1—figure supplement 6D ) , suggesting that Max1 is most sensitive to interspecific differences in dynamically regulated genes . The scaled values of each orthologous pair were concatenated as a 10-dimensional vector ( consisting of four stages for mouse limb buds and six stages for bamboo shark fin buds ) , and all gene expression vectors were dimensionally reduced with UMAP ( hyper parameters: a = 10 , b = 1 . 8 ) followed by hierarchical clustering ( hyper parameters: method = 'ward' , metric = 'euclidean'; the code is available at https://github . com/koonimaru/easy_heatmapper; copy archived at https://archive . softwareheritage . org/swh:1:dir:b1b8edece650ac9e8a7458354aaf69e74f437092;origin=https://github . com/koonimaru/easy_heatmapper;visit=swh:1:snp:a69e903d0efcde99cb203ec86832c5e5c56a43e5;anchor=swh:1:rev:ba1fde133621a52390b82b4c9f73711a56f252b8/ ) . To find genes that have an opposite trend in their expression relative to 'Heterochronic2' , a Pearson correlation coefficient ( PCC ) for TPM values and developmental stages was calculated for each gene for each species , and genes with PCC > 0 . 5 for bamboo shark fin buds and PCC < −0 . 5 for mouse limb buds were listed ( Figure 2—figure supplement 1B and Supplementary file 8 ) . For the distance measurements , four different distance methods were calculated: Euclidean distance ( ∑ui-vi2 ) , correlation distance ( 1-u-u¯v-v¯u-u¯2v-v¯2 ) , Shannon distance ( -12∑uilogui+vi2ui+vilogui+vi2vi ) , standardized Euclidean distance ( ∑ui-vi2/Vi ) , where u and v are gene expression vectors of two samples and Vi is the variance computed over all the values of gene i . For PCA analysis , we used the PCA module in a python package , scikit-learn ( https://scikit-learn . org/stable/ ) . For the stage-associated gene analysis in Figure 3—figure supplement 1B and C , we first calculated the z-score of each gene at each stage as uk , i-u¯iσi , where uk , i is the TPM value of gene i at stage k , u¯i is a mean of TPM over all the stages , and σi is the standard deviation of the TPM . Genes with TPM ≥ 1 . 0 and the absolute Z-score ≥1 . 0 were counted as stage-associated genes . For the tissue-associated gene analysis , the entropy of each gene was calculated using RNA-seq data of 71 tissues downloaded from the ENCODE web site ( https://www . encodeproject . org/; see Supplementary file 10 for all list ) . Entropy was calculated as follows:pk , i=TPMk , i∑kTPMk , iHi=-∑kpk , ilogpk , iwhere TPMk , i is the TPM value of gene i in tissue k , pk , i is a probability distribution and Hi is entropy . Genes with TPM ( of mouse limb buds ) ≥ 1 . 0 and 0 . 65 ≤ entropy were counted as tissue-associated genes . To clone DNA sequences for RNA probes , we used primers that were based on the nucleotide sequences in the ENSEMBL database ( https://www . ensembl . org ) for mouse genes and in the transcriptome assembly ( Supplementary file 3 ) ; bamboo shark Hand2 ( Chipun0004250/g4250 . t1/ TRINITY_DN85524_c0_g1_i1 ) , 5′-ACCAGCTACATTGCCTACCTCATGGAC-3′ and 5′-CACTTGTTGAACGGAAGTGCACAAGTC-3′; bamboo shark Vcan ( Chipun0003941/g3941 . t1/ TRINITY_DN95522_c0_g1_i8 ) , 5′-AGCTTGGGAAGATGCAGAGAAGGAATG-3′ and 5′-AGAGCAGCTTCACAATGCAGTCTCTGG-3′; bamboo shark Hoxd12 ( Chipun0005654/g5654 . t1/TRINITY_DN85970_c0_g1_i1 ) , 5′-GCCAGTATGCAACAGATCCTCTGATGG-3′ and 5′-CTAATGACCTGTTGTACTTACATTCTC-3′; bamboo shark Ptch1 ( Chipun0003320/g3320 . t1/TRINITY_DN92499_c0_g1_i3 ) , 5′-TTCAGCCAGATTGCAGATTACATCAACC-3′ and 5′-TTCTCTGTGTTTCACATTCAACGTCCTG-3′; bamboo shark Aldh1a2 ( Chipun0010503/g10503 . t1/TRINITY_DN81423_c0_g1_i1 ) , 5′-TTGAACTTGTACTAAGTGGTATCGCTG-3′ and 5′-AGGATGTGAACATTAGGCTGACCTCAC-3′; mouse Hand2 ( ENSMUST00000040104 . 4 ) , 5′-ACCAAACTCTCCAAGATCAAGACACTG-3′ and 5′-TTGAATACTTACAATGTTTACACCTTC-3′; mouse Vcan ( ENSMUST00000109546 . 8 ) , 5′-TGCAAAGATGGTTTCATTCAGCGACAC-3′ and 5′-ACACGTGCAGAGACCTGCAAGATGCTG-3′; mouse Hoxd12 ( ENSMUST00000109546 . 8 ) , 5′-TGCAAAGATGGTTTCATTCAGCGACAC-3′ and 5′-ACACGTGCAGAGACCTGCAAGATGCTG-3′; mouse Aldh1a2 ( ENSMUST00000034723 . 5 ) , 5′-ACCGTGTTCTCCAACGTCACTGATGAC-3′ and 5′-TCTGTCAGTAACAGTATGGAGAGCTTG-3′; mouse Ptch1 ( ENSMUST00000192155 . 5 ) , 5′-GGGAAGGCAGTTCATTGTTACTGTAACTG-3′ and 5′-TGTAATACGACTCACTATAGGTCAGAAGCTGCCACACACAGGCATGAAGC-3′ . Note that although we also tried bamboo shark Shh expression analysis using several RNA probes , we did not obtain specific signals . Fixed embryos were processed for in situ hybridization as described ( Westerfield , 2000 ) with slight modifications . Briefly , embryos were re-hydrated with 50% MeOH in PBST ( 0 . 01% Tween 20 in PBS ) and with PBST for 5–30 min each at room temperature ( RT ) . Then , embryos were treated with 20 μg/ml proteinase K ( Roche ) in PBST ( 5 s for mouse E11 . 5 and E12 . 5 embryos , 5 min for st . 27 and st . 29 bamboo shark embryos , 10 min for st . 31 and st . 32 bamboo shark embryos ) . After the proteinase treatment , embryos were fixed in 4% paraformaldehyde/PBS for 1 hr , followed by one or two washes with PBST for 5–10 min each . Optionally , if embryos had some pigmentation , they were immersed in 2% H2O2 until they became white . Then , embryos were incubated for 1 hr in preheated hybridization buffer ( 50 ml formaldehyde; 25 ml 20× SSC , pH 5 . 0; 100 μl 50 mg/ml yeast torula RNA; 100 μl 50 mg/ml heparin; 1 ml 0 . 5 M EDTA; 2 . 5 ml 10% Tween 20; 5 g dextran sulfate; and DEPC-treated MilliQ water to a final volume of 100 ml ) at 68°C . Subsequently , embryos were incubated with fresh hybridization buffer containing 0 . 25–4 μl/ml of RNA probes at 68°C overnight . Embryos were washed twice with preheated Wash buffer 1 ( 50 ml formaldehyde; 25 ml 20× SSC , pH 5 . 0; 2 . 5 ml 10% Tween 20; and DEPC-treated MilliQ water to a final volume of 100 ml ) for 1 hr each at 68°C; once with preheated Wash buffer 2 , which consisted of equal volumes of Wash buffer 1 and 2× SSCT ( 10 ml 20× SSC , pH 7 . 0; 1 ml 10% Tween 20; and MilliQ water to a final volume of 100 ml ) , for 10 min at 68°C; once with preheated 2× SSCT at 68°C for 10 min; and once with TBST at room temperature for 10 min . Embryos were then incubated with a blocking buffer ( 20 μl/ml 10% bovine serum albumin , 20 μl/ml heat-inactivated fetal bovine serum in TBST ) for 1 hr at room temperature , followed by incubation with 1/4000 anti-digoxigenin ( Roche ) in fresh blocking buffer at 4°C overnight . Embryos were washed four times with TBST for 10–20 min each and were incubated at 4°C overnight . Finally , embryos were incubated with NTMT ( 200 μl 5 M NaCl; 1 ml 1 M Tris-HCl , pH 9 . 8; 500 μl 1 M MgCl2; 100 μl 10% Tween 20; and MilliQ water to a final volume of 10 ml ) for 20 min and then with 15 μg/ml nitro-blue tetrazolium chloride ( NBT ) and 175 μg/ml 5-bromo-4-chloro-3-indolyphosphate p-toluidine salt ( BCIP ) in NTMT for 10 min to 2 hr until signals appeared . Pictures were taken with an Olympus microscope . For bamboo shark embryos , experiments were performed for at least two biological replicates . Mouse forelimb buds at E9 . 5 , E10 . 5 , E11 . 5 , and E12 . 5 were dissected , and samples from several individuals were pooled by stage to obtain enough cells . We considered this pooled sample to represent a biological replicate ( other replicates were generated using different individuals ) . To obtain single-cell suspensions , pooled samples were treated with collagenase for 10 min at room temperature . The tissues were then dissociated into single-cell suspensions by pipetting the mixture and passing it through a 40 μm mesh filter ( Funakoshi , Cat . No . HT-AMS-14002 ) ; the cell suspension was frozen in CryoStor medium ( STEMCELL Technologies , Cat . No . ST07930 ) with Mr . Frosty ( Thermo Scientific , Cat . No . 5100–0001 ) at −80°C overnight , according to Milani et al . , 2016 . An ATAC-seq library was prepared as described ( Buenrostro et al . , 2013 ) with some minor modifications . For library preparation , stored cells were thawed in a 38°C water bath and centrifuged at 500 g for 5 min at 4°C , which was followed by a wash using 50 μl of cold PBS and a second centrifugation at 500 g for 5 min at 4°C . Ten thousand cells per sample were collected , without distinguishing dead cells , and were lysed using 50 μl of cold lysis buffer ( 10 mM Tris-HCl , pH 7 . 4; 10 mM NaCl; 3 mM MgCl2; and 0 . 1% IGEPAL CA-630 ) . Immediately after lysis , cells were spun at 1000 g for 10 min at 4°C , and the supernatant was discarded . For the transposition reaction , cells were re-suspended in the transposase reaction mix ( 25 μl 2× TD buffer , 2 . 5 μl Tn5 transposase [in the Nextera DNA Sample Preparation Kit , Illumina , Cat . No . FC-121–1031] , and 22 . 5 μl nuclease-free water ) and incubated for 30 min at 37°C . The reaction mix was purified using DNA Clean and Concentrator-5 ( Zymo Research , Cat . No . D4014 ) by adding 350 μl of DNA-binding buffer and eluting in a volume of 10 μl . After a five-cycle pre-PCR amplification , the optimal number of PCR cycles was determined by a preliminary PCR using KAPA HiFi HotStart Real-Time Library Amplification Kit and was estimated to be four cycles . The PCR products were purified using 1 . 8× volumes of Agencourt AMPure XP . As a control , 50 ng of mouse genomic DNA was also transposed following the standard procedure of the Nextera DNA Sample Preparation Kit . Sequencing with HiSeq X was outsourced to Macrogen , Inc , which was carried out with HiSeq Control Software 3 . 3 . 76 ( Run type: PE151bp ) . The output was processed with Illumina RTA 2 . 7 . 6 for base-calling and with bcl2fastq 2 . 15 . 0 for de-multiplexing . Quality control of the obtained fastq files for individual libraries was performed with FASTQC v0 . 11 . 5 . ATAC-seq was performed with three biological replicates for each stage . The short-read data from ATAC-seq were trimmed and filtered with Trim-Galore ! ( v0 . 5 . 0; options: --paired --phred33 -e 0 . 1 -q 30 ) . We also removed reads that originated from mitochondrial genome contamination by mapping reads to the mouse mitochondrial genome using bowtie2 v2 . 3 . 4 . 1 ( Langmead and Salzberg , 2012 ) . The rest of the reads were mapped onto the mouse genome ( mm10 ) using bwa v0 . 7 . 17 with the ‘mem’ option ( Li and Durbin , 2010 ) . Among the mapped reads , we removed reads with length >320 bp to reduce noise . The rest of the reads were further down-sampled to around 83 . 2 million reads to equalize the sequence depth of every sample . Peak calls were done with MACS2 v2 . 1 . 1 ( Zhang et al . , 2008; options: --nomodel --shift −100 --extsize 200 f BAMPE -g mm -B -q 0 . 01; the genomic reads were used as a control for all samples ) . For FRiP score calculation , a module , ‘countReadsPerBin . CountReadsPerBin’ in deepTools v3 . 2 . 1 ( Ramírez et al . , 2016 ) , was used to count reads in peaks , and these read counts were then divided by the total number of reads . To evaluate reproducibility among the replicates , we first divided the mouse genome into 500 bp bins . Then , the ATAC-seq peaks were re-distributed into these bins with bedtools ( Quinlan and Hall , 2010; options: intersect -F 0 . 4 f 0 . 4 -e -wo ) . Peaks of >500 bp were subdivided into 500-bp-long regions , and those of <500 bp were extended to fit within the closest 500 bp window . Subsequently , these peaks were converted into one-hot vectors , in which ‘1’ means that a 500-bp-long genomic region harbors an ATAC-seq peak . Genomic regions that lacked ATAC-seq peaks in all data were omitted . Using these one-hot vectors , Euclidean distances between the ATAC-seq data were calculated ( Figure 5—figure supplement 1A ) . For the conservation analysis , the significant variation in the length of ATAC-seq peaks complicated this evaluation . To deal with such variation , we the ATAC-seq peaks were re-distributed into 100 bp bins with bedtools ( Quinlan and Hall , 2010; options: intersect -F 0 . 4 f 0 . 4 -e -wo ) as described above . The sequences in these peaks were retrieved with BLASTN v2 . 7 . 1 against the genomes of 16 vertebrate species listed in Supplementary file 10 ( BLASTN options: -task dc-megablast -max_target_seqs 1 ) . The blast hits that scored ≥40 were considered as conserved sequences . In this way , the final figures shown in Figure 5C represent the fraction of the total conserved sequence length in the peaks of each stage rather than the number of conserved peaks . For confirmation , we also used a different alignment algorithm , LAST v961 ( Kiełbasa et al . , 2011 ) to find conserved sequences . To generate mouse genome databases for LAST , we first masked repeat sequences with N and split the genome file into multiple files , each of which contained a single chromosome sequence . Then , databases were generated using lastdb ( options: -cR01 ) . Alignments with the bamboo shark genome ( Cpunctatum_v1 . 0; https://transcriptome . riken . jp/squalomix/resources/01 . GCA_003427335 . 1_Cpunctatum_v1 . 0_genomic . rn . fna . gz ) and the alligator genome ( ASM28112v3 ) were carried out by lastal ( options: -a1 -m100 ) . Only a unique best alignment was selected using last-split . These alignment results were converted into the bed format , and regions that overlapped with the ATAC-seq peaks that were subdivided into 100 bp bins were counted . For the clustering analysis , we converted the alignment files of the ATAC-seq reads into mapped reads in bins per million ( BPM ) coverage values with 200 bp resolution using bamCoverage in deepTools v3 . 2 . 1 ( Ramírez et al . , 2016; options: -of bedgraph --normalizeUsing BPM --effectiveGenomeSize 2652783500 -e -bs 200 ) . Then , BPMs at the summits of ATAC-seq peaks and an additional 600 bp to the left and to the right of each summit ( 1400 bp in total ) were collected and clustered by t-SNE ( https://github . com/DmitryUlyanov/Multicore-TSNE; hyper parameters: perplexity = 30 . 0 , n_iter = 5000 ) followed by hierarchical clustering ( hyper parameters: method = ‘ward’ , metric = ‘euclidean’ ) . Enriched motifs were detected using a Perl script , findMotifsGenome . pl in HOMER v4 . 10 . 4 ( Heinz et al . , 2010; options: -size 100 -mask ) . To count the number of motif occurrences , ‘-find’ option of findMotifsGenome . pl was used , and sequences that scored ≥75% of the highest motif score were counted . For GO analysis , annotatePeaks . pl in HOMER was used . For the tissue-specificity analysis , we downloaded several aligned and unaligned reads of ATAC-seq experiments on 25 different tissues from the ENCODE web site ( https://www . encodeproject . org/; see Supplementary file 10 for a complete list ) , and peaks were called as described above . Then , peaks that did not overlap with other tissues/cells were detected using bedtools . RNA-seq and ATAC-seq data sets generated during the current study are available in the Gene Expression Omnibus ( GEO ) repository under accession number GSE136445 . Other sequence data and raw data are available in the figshare ( DOI: 10 . 6084/m9 . figshare . 9928541 ) . Code for clustering analysis is available at https://github . com/koonimaru/easy_heatmapper . Materials related to this paper are available upon request from the corresponding authors .	Animals come in all shapes and sizes . This diversity arose through genetic mutations during evolution , but it is unclear exactly how these variations led to the formation of new shapes . There is increasing evidence to suggest that not all shapes are possible and that variability between animals is limited by a phenomenon known as “developmental constraint” . These limitations direct parts of the body towards a specific shape as they develop in the embryo . Therefore , understanding the mechanisms underlying these developmental constraints could help explain how different body shapes evolved . The limbs of humans and other mammals evolved from the fins of fish , and this transition is often used to study the role developmental constraints play in evolution . This is an ideal model as there is already a detailed fossil record mapping this evolutionary event , and data pinpointing some of the genes involved in the development of limbs and fins . But this data is incomplete , and a full comparison between the genes activated in the fin and the limb during embryonic development had not been achieved . This is because most fish used for research have undergone recent genetic changes , making it hard to spot which genetic differences are linked to the evolution of the limb . To overcome this barrier , Onimaru et al . compared genetic data from the developing limbs of mice to the developing fins of the brown-banded bamboo shark , which evolves much slower than other fish . This revealed that although many genes commonly played a role in the development of the fin and the limb in the embryo , the activity of these shared genes was not the same . For example , genes that switched on in the late stages of limb development , switched off in the late stages of fin development . But in the middle of development , those differences were relatively small and both species activated very similar sets of genes . Many of these genes were pleiotropic , which means they have important roles in other tissues and therefore mutate less often . This suggests that the mid-stage of limb development is under the strongest level of constraint . Darwin’s theory of natural selection explains that mutations drive evolution . But the theory cannot predict what kinds of new body shapes new mutations will produce . Understanding how the activity levels of different genes affect development could help to fill this knowledge gap . This has potential medical applications , for example , understanding why some genetic changes cause more serious problems than others . This work suggests that mutations in genes that are active during the mid-stage of limb development may have the most serious impact .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2021	Developmental hourglass and heterochronic shifts in fin and limb development
During awake consciousness , the brain intrinsically maintains a dynamical state in which it can coordinate complex responses to sensory input . How the brain reaches this state spontaneously is not known . General anesthesia provides a unique opportunity to examine how the human brain recovers its functional capabilities after profound unconsciousness . We used intracranial electrocorticography and scalp EEG in humans to track neural dynamics during emergence from propofol general anesthesia . We identify a distinct transient brain state that occurs immediately prior to recovery of behavioral responsiveness . This state is characterized by large , spatially distributed , slow sensory-evoked potentials that resemble the K-complexes that are hallmarks of stage two sleep . However , the ongoing spontaneous dynamics in this transitional state differ from sleep . These results identify an asymmetry in the neurophysiology of induction and emergence , as the emerging brain can enter a state with a sleep-like sensory blockade before regaining responsivity to arousing stimuli . During emergence from general anesthesia , the brain transitions out of the unconscious state and recovers its ability to process complex sensory information and coordinate behavioral responses . General anesthesia causes a profound disruption of information processing across large-scale cortical ( Alkire et al . , 2008; Liu et al . , 2012; Hudetz , 2012; Lewis et al . , 2012; Dehaene et al . , 2014; Sarasso et al . , 2014 ) and thalamocortical networks ( Alkire et al . , 2000; Ching et al . , 2010; Mhuircheartaigh et al . , 2010; Ní Mhuircheartaigh et al . , 2013; Vijayan et al . , 2013; Akeju et al . , 2014; Baker et al . , 2014; Flores et al . , 2017 ) , and suppression of arousal systems ( Lydic and Baghdoyan , 2005; Franks , 2008; Brown et al . , 2011 ) , which then spontaneously reverts as the anesthetic drug clears . While several studies have characterized the transition in electrophysiological dynamics occurring at loss of consciousness during anesthetic inductions ( Gugino et al . , 2001; Feshchenko et al . , 2004; Ferrarelli et al . , 2010; Supp et al . , 2011; Lewis et al . , 2012 ) , less is known about the dynamics that occur during emergence from general anesthesia and how they support recovery of a behaviorally responsive state ( Purdon et al . , 2013; Hudson et al . , 2014; Mukamel et al . , 2014 ) . Electrophysiological evidence shows that many anesthetic-induced neurophysiological dynamics undergo relatively symmetric transitions: shifts in spectral power , spatial correlations , phase-amplitude coupling , and spike coherence that are observed during anesthetic induction gradually reverse as drug concentrations are lowered ( Breshears et al . , 2010; Lee et al . , 2010; Purdon et al . , 2013; Mukamel et al . , 2014; Vizuete et al . , 2014 ) . However , it is also clear that the process of emerging from anesthesia is not identical to anesthetic induction . Emergence occurs at lower anesthetic doses than induction ( Friedman et al . , 2010 ) , and this hysteresis suggests that state-dependent processes also shape the transitions in and out of anesthesia . Behaviorally , some patients experience delirium , a transient state of agitation and confusion which can arise during emergence from anesthesia ( O'Brien , 2002 ) , suggesting that distinct neural mechanisms may underlie emergence . EEG and local field potential recordings have suggested that the process of emergence may involve stepping through discrete dynamical states ( Lee et al . , 2011; Hudson et al . , 2014 ) . Electrophysiological studies in rodents show that propofol-induced coherent alpha and delta oscillations , which appear to mediate the functional disruption of thalamus and cortex during anesthesia , recover in a spatiotemporal sequence during emergence that is different from induction ( Flores et al . , 2017 ) . These observations are consistent with a history-dependent process , in which the current brain state influences the process by which the next brain state is reached . However , a neurophysiological mechanism or network dynamic that is engaged selectively during emergence , rather than induction , is not known . In addition to shifts in spontaneous neurophysiological dynamics , sensory processing is also strongly affected by induction and emergence from general anesthesia . Sensory-evoked potentials ( event-related potentials , ERPs ) index specific phases of cognitive information processing and can provide diagnostic measures of unconscious patients ( Boly et al . , 2011; King et al . , 2013 ) . Several studies of ERPs during anesthesia have shown that disruption of higher level cognitive processing is reflected by a reduction in amplitude of the mismatch negativity ( MMN ) , potentials evoked by unexpected sensory input . The MMN declines in amplitude during induction of anesthesia ( Simpson et al . , 2002; Heinke et al . , 2004 ) , whereas lower-level responses such as the auditory steady-state response persist during sedation , and are abolished at deep anesthetic levels ( Plourde and Picton , 1990 ) . Cortical responses to direct stimulation using TMS are more spatially constrained and less complex during propofol-induced unconsciousness ( Sarasso et al . , 2015 ) , consistent with fragmentation of large-scale brain network activity during propofol anesthesia ( Lewis et al . , 2012 ) . The propagation of sensory information through thalamocortical circuits is thus differentially affected at increasing doses of anesthesia , with higher-level , longer-latency responses extinguished at low drug levels and then further suppression of short-latency evoked activity at high drug levels . At the deepest levels of anesthesia , when brain activity enters a state of ‘burst suppression’ alternating between periods of isoelectric silence ( suppressions ) and periods of high-amplitude activity ( bursts ) , sensory stimuli can trigger the onset of a burst ( Hartikainen et al . , 1995; Kroeger and Amzica , 2007 ) . It is therefore clear that external sensory input can still influence cortical activity during profound anesthesia . However , evoked responses during burst suppression are qualitatively different than those observed during normal sensory processing , as they typically manifest as a large-amplitude burst containing the spectral dynamics of the pre-bursting state ( Lewis et al . , 2013 ) , rather than the distinct ERP waveform with classical components related to specific phases of cognitive information processing seen in the waking state . Sensory input during burst suppression thus appears to drive nonspecific cortical activity rather than effective processing of sensory information . The neural dynamics supporting the brain’s ability to spontaneously recover wakeful consciousness , regain sensory perception and resume complex cortical responses following the profound disruption caused by general anesthesia are not well understood . Late components of the ERP continue to be disrupted even after patients have recovered consciousness and early components have returned to baseline ( Plourde and Picton , 1991; Koelsch et al . , 2006 ) , suggesting that emergence represents a graded and prolonged return to the normal awake state rather than a simple reversal of anesthesia induction . It is still unclear what ongoing brain dynamics contribute to altered sensory processing during emergence from anesthesia . Here , we use two independent datasets − intracranial recordings from patients emerging from anesthesia after surgery , and high-density EEG recordings from a study of emergence in healthy volunteers under controlled laboratory conditions − to provide a multiscale analysis of neural dynamics during emergence from anesthesia . By defining the trajectory of changes in ongoing neural dynamics and sensory evoked responses during the process of emergence , we identify a new transitional brain state that occurs just before emergence from anesthesia . This state is marked by stimulus-evoked cortical down states that resemble the K-complexes which are hallmarks of stage two non-rapid eye movement ( N2 ) sleep . However , its spontaneous dynamics qualitatively differ from sleep . We show that this state occurs primarily in the minutes prior to awakening , identifying a novel transitional brain state that is selective to anesthetic emergence . During emergence from general anesthesia , we observed that in a subset of trials , auditory stimuli elicited a large ( >100 µV ) and slow ( duration >1 s ) evoked potential ( Figure 1b , c ) across many electrodes . We developed an automatic detection algorithm to identify these events , which we termed large potentials ( LP ) . LPs were defined as events of >400 µV amplitude lasting >400 ms ( see Materials and methods for additional details ) . We chose these thresholds to conservatively detect only large events while ignoring small or ambiguous LP-like events . 16% of electrodes ( n = 1095 electrodes ) exhibited at least five events using this detector . This number included electrodes from every patient , as at least two electrodes with ≥ 5 LPs were detected in each emergence session . To characterize the relationship between these events and the auditory stimulation , we analyzed all trials on which an LP occurred within two seconds of stimulus onset . The mean stimulus-triggered event on each electrode ( Figure 1d ) had a median peak amplitude of 236 µV ( quartile range ( QR ) : 183–295 ) , a value that was lower than the detection threshold due to averaging together events with slightly different peak times . The peak of the mean stimulus-triggered event occurred 1 . 01 s ( QR:0 . 7–1 . 38 ) after stimulus onset , and lasted 0 . 28 s ( full-width at half max; QR: 0 . 05–0 . 47 ) , a waveform that was far slower and larger in amplitude than typical auditory-evoked responses in the awake state . The LPs thus rank among the largest electrophysiological signals observed in human cortex , indicating synchronization of electrical signaling among a substantial fraction of the local neuronal population . The average stimulus-aligned waveform across patients can be temporally blurred due to differences in timing across subjects and electrode locations . To more precisely assess the amplitude and waveform of these events , we selected the electrode with the most events in each subject , and analyzed the mean waveform of all detected events aligned to their peak . The peak-aligned events on these electrodes were larger ( median amplitude = 550 µV ) and had an asymmetric morphology ( Figure 1c , d , e ) , with a sharper onset than offset ( mean rise = 165 ms , mean fall = 285 ms , 95% confidence interval ( CI ) for difference=[84 156] ms , bootstrap; p=0 . 0002 , Wilcoxon signed-rank test ) and large post-peak rebound . Aligning to stimulus onset thus confirmed these events were auditory-evoked , whereas analyses aligning to the peak demonstrated that the waveform of the events was large and asymmetric , with substantial variability in exact time-to-peak . The large , slow , and asymmetric waveform of the LPs resembles K-complexes ( KCs ) , a characteristic electrophysiological graphoelement that occurs spontaneously or following sensory stimulation during stage two non-rapid eye movement ( NREM ) sleep ( Loomis et al . , 1938; Colrain , 2005; Halász , 2016 ) . The KC corresponds to a cortical DOWN state ( Cash et al . , 2009 ) , in which local neuronal firing is suppressed . To test whether LPs mark a similar cortical dynamic , we analyzed high-frequency power in the LFP , which is correlated with local spike rates ( Ray and Maunsell , 2011 ) , during all detected LPs . We selected the electrode with the most LPs in each subject and computed the peak-triggered power , and found that LPs correlated with a strong reduction in broadband gamma-range ( 40–100 Hz ) power ( −1 . 29 dB , CI=[−0 . 4–2 . 5] , bootstrap; p=0 . 04 , Wilcoxon signed-rank test , Figure 1f , g ) , suggesting they too represent a DOWN state with suppression of neural activity . This peak-locked analysis included both stimulus-evoked and spontaneous events . A substantial proportion ( 28% ) of detected LPs appeared to occur spontaneously , as they were not preceded by an experimental stimulus within 2 s , although other auditory input present in the clinical environment may have contributed to their generation . When the spectral analysis was instead performed relative to the onset of the auditory stimulus , including only trials where LPs appeared within 2 s of a stimulus , we found that this decrease in high-frequency power reached a minimum at 1 . 3 s post-stimulus , suggesting that the auditory-evoked potentials were also associated with prolonged suppression of neuronal activity . This slow timecourse is also similar to the timing of auditory-evoked KCs during sleep ( Colrain et al . , 1999 ) . Intracranial recordings provide precise , millimetre-scale spatial resolution , enabling mapping of the cortical sources of LPs . We measured the amplitude of the mean stimulus-evoked response across all electrodes , on trials that evoked an LP in at least one electrode . We aligned mean responses to stimulus onset , to allow consistent comparisons across channels that could exhibit different peak times . Most subjects exhibited LPs on multiple electrodes , with amplitude of the evoked potential varying widely across regions ( Figure 2a ) . However , many electrodes exhibited no sign of an LP despite showing ongoing local electrophysiological activity , indicating that these were not global cortical events . The percentage of grid and strip electrodes with at least five detected LPs was highest in frontal and temporal cortex ( 39% of frontal electrodes , 36% of temporal electrodes , Figure 2b ) , which had significantly higher proportions than the mean rate ( 26% , CI=[24 29] , p<0 . 05 , Bonferroni corrected binomial test ) . Fewer parietal electrodes exhibited detectable LPs ( 11% ) . We also found that LPs were recorded on 35 of the 129 depth electrodes placed in gray matter ( 27% ) , including on deep contacts placed in hippocampus . The peak timing and morphology of the evoked potential varied across space within individuals ( Figure 2c ) . Overall , our intracranial recordings suggest that LPs were restricted to a specific frontotemporal network of cortical regions rather than a globally coherent slow wave . To determine the timecourse of the stimulus-evoked LPs , we computed sliding window measures of the mean evoked response over time , including all trials , on the electrode for each subject that exhibited the most LPs . The LPs were primarily observed after propofol was turned off but before the patient exhibited signs of recovery ( Figure 3a , b ) . This effect was seen in the mean amplitude of the ERP over time: the normalized ERP amplitude increased across subjects in the ~300 s prior to the first behavioral sign of emergence , and subsided again thereafter ( Figure 3c ) . In the eight patients who had recordings in both induction and emergence , we analyzed the mean ERP amplitude relative to behavioral state changes and found that the LPs occurred predominantly during emergence , particularly in the pre-return of consciousness ( pre-ROC ) period , and not during induction ( Figure 3d ) . Since our induction used a gradual infusion ( Figure 1a ) , patients were guaranteed to pass through a plasma concentration level during induction that matched their level at emergence , demonstrating this transient state was selective to the process of emergence rather than only a particular dosage level . To test what ongoing dynamics accompanied this transitional LP state , we analyzed spectral content within each epoch . We found that the dynamics during emergence were substantially different from induction , exhibiting significantly greater low-frequency ( <2 Hz ) and alpha power even after awakening ( Figure 3e ) . Comparing the three minutes immediately after behaviorally defined loss of consciousness ( LOC ) during induction , and the three minutes immediately prior to return of consciousness ( ROC ) , a smaller but otherwise similar power difference was evident ( Figure 3f ) . While the intracranial recordings suggested asymmetry between induction and emergence , due to time constraints in the operating room we were not able to measure intracranially over prolonged periods . To test these dynamics in a more controlled setting and in a population of healthy subjects , we next analyzed scalp EEG data recorded during a stepped infusion of propofol in healthy volunteers ( Purdon et al . , 2013; Mukamel et al . , 2014 ) , during presentation of auditory stimuli that were click trains , words , and the subject's own name . This stepped infusion protocol induced slow changes in propofol concentration and behavioral responses ( Figure 4a , b ) . The steady-state auditory evoked potential to the auditory click train stimuli also declined , quantified as the induced power at 40 Hz , corresponding to the click frequency ( Figure 4c ) . To confirm this decrease was selective to the auditory-evoked band rather than broadband , we also analyzed power at a 'control' frequency of 22 Hz ( i . e . , not the stimulus frequency ) and found no change . We next examined the traces and found that large evoked potentials were clearly visible during emergence after large-amplitude slow oscillations subside ( Figure 4d , e , f ) . To apply the same LP detector , we focused on a frontal EEG electrode , as frontal electrodes had high LP rates in our intracranial data ( Figure 2b ) and did not exhibit the large auditory-evoked potentials of the temporal electrodes . We detected stimulus-evoked LPs ( peak >7 s . d . , Figure 4 ) during emergence from propofol anesthesia in 4 out of 10 subjects , despite the lower spatial resolution of the scalp recordings ( Figure 4e , f , g ) . If the detection threshold was lowered ( peak >5 s . d . ) , we could also observe brief traces of similar events in the induction period in 3 of the 10 subjects . However , these periods were brief and infrequent ( Figure 4e ) , suggesting that this brain state occurs primarily ( but not exclusively ) during emergence ( Figure 4—figure supplement 1 ) . While we detected these events in a frontal electrode , LP events were observed broadly across the scalp ( Figure 4—figure supplement 1 ) , consistent with the widespread spatial profiles we observed in the intracranial data . These results in healthy volunteers confirmed that the LPs were not related to epileptic events in the patients . Furthermore , they show that LPs occurred primarily during emergence ( Figure 4h , i , fig . supp . 1 ) even in these experiments with a prolonged induction period , lasting more than twice as long as the emergence period . In these subjects , the LPs were also found to be stimulus-selective: they occurred preferentially in response to the sound of words and names , and did not occur following click-train stimuli ( Figure 4h , Figure 4—figure supplement 1 ) . In contrast , no such stimulus selectivity was observed in the intracranial patients , as each stimulus type could elicit LPs ( in the channel with the most events in each subject , LPs occurred within 2 s of 21% of word stimuli; 20% of sounds stimuli; 22% of click-train stimuli ) . A key difference between these two datasets was the relative frequency of the name and word stimulus categories , which were infrequent ( 20% names/words , 80% clicks ) in the scalp data but were evenly distributed in the intracranial data ( 30% words , 40% clicks , 30% sounds ) . The increased saliency of an infrequent stimulus may thus increase the probability of an LP , similar to reports for KCs during sleep ( Colrain et al . , 1999 ) . We observed LPs for a prolonged period that could extend after the initial ROC in the scalp EEG dataset ( Figure 4h , i , Figure 4—figure supplement 1 ) , whereas LPs were only present before ROC in the intracranial dataset ( Figure 3d , Figure 4—figure supplement 1 ) . This difference likely reflects the differences in arousal state across these two datasets: in the intracranial study , the drug was completely shut off and patients emerged rapidly as drug levels monotonically decreased . In contrast , in the scalp EEG dataset , the propofol levels were lowered in a gradual , stepped fashion ( Figure 4a ) , leading to a prolonged emergence period over tens of minutes . These large LPs therefore may be present not only in the minutes prior to any sign of ROC , but may continue through emergence until a relatively heightened arousal state is reached . Although the LPs shared some properties with the spontaneous KCs that occur during N2 sleep , the propofol emergence period could be expected to also exhibit significant differences from natural sleep . To test the similarity between events during sleep and during emergence , we obtained intracranial recordings during sleep from a subset of the patients ( n = 3 patients ) . To compare the LPs detected during propofol with the spontaneous KCs during sleep , we first verified that the automatic detection algorithm could identify events in the sleep datasets . We found that 64% of manually identified KCs were also flagged by the automatic detector , suggesting this approach could be used to quantitatively compare the two phenomena within this patient cohort ( although the high number of misses , 36% , suggests it should not be employed as a KC detector for more general purposes ) . The LPs recorded during emergence and the sleep KCs shared an overall profile of large ( >100 µV ) and slow ( ~1 s ) waveforms ( Figure 5a ) . No significant difference in median amplitude was seen between the propofol and sleep datasets ( grouped median in sleep = 644 µV ( CI=[630 673] ) , propofol = 647 µV ( CI=[623 665] ) , bootstrap; p=0 . 59 , Wilcoxon rank-sum test ) , and the distributions of event amplitudes shared a high degree of overlap ( Figure 5b ) . The spatial distribution of sleep KCs also appeared very similar to that seen during emergence from propofol anesthesia ( Figure 5c , d ) . To test this spatial similarity , we computed the mean event waveform across all electrodes , triggered on the peak of each event detected in a single electrode with a large number of events in both the sleep and propofol recordings . We found that the mean event amplitude across electrodes was significantly correlated when comparing sleep and propofol ( R = 0 . 92 , 0 . 62 , 0 . 71 , for the three patients , Figure 5c , Figure 5—figure supplement 1 ) , meaning that electrodes with large KCs were likely to also exhibit large LPs during emergence . Overall , the shared waveform , spatial distribution , and timing of these events suggest that the LPs observed during propofol emergence may engage the circuit mechanisms that generate KCs during natural sleep . Given the resemblance of the LPs to KCs , we next tested whether ongoing spectral dynamics within the LP period resembled N2 sleep . We computed the power spectrum during a manually selected period exhibiting LPs during emergence , and compared these to segments of sleep recordings manually identified as N2 sleep . We found substantial differences in these spectra , with propofol emergence exhibiting more power across a broad frequency range of 10 to 40 Hz throughout all recorded cortical regions ( Figure 6 , median difference = 5 . 6 dB , CI=[3 . 9 6 . 1] , bootstrap; p<0 . 001 in each subject , Wilcoxon signed-rank test ) . In addition , the sleep spectra exhibited clear spindle power ( 10–14 Hz ) peaks across cortical regions , whereas the emergence spectra exhibited either no peak or a spatially restricted frontal alpha ( ~10 Hz ) peak characteristic of deep propofol anesthesia ( Figure 6c ) . These results demonstrate that while some common neurophysiological events can be observed in stage two sleep and in this transient emergence period , emergence is a distinct brain state that is not identical to sleep . Using both intracranial ECoG and scalp EEG recordings , we found that emergence from general anesthesia is accompanied by a transient state in which auditory stimuli can evoke large potentials ( LPs ) corresponding to all-or-none cortical suppressions lasting several hundred milliseconds . LPs strongly resemble the K-complexes observed in N2 sleep , although the neural dynamics of emergence from general anesthesia nevertheless represent a distinct state . This state appeared primarily during emergence and foreshadowed the return of behavioral responsiveness , suggesting it represents a distinct brain state through which patients transition as they recover consciousness . Our data indicate that the brain’s response to propofol is hysteretic , such that the current state is determined not only by the drug concentration but also by the recent history of the brain’s activation . The brain state we observed appears to be distinct from the sedated state experienced by patients during slow anesthetic induction , as it was exclusively observed during emergence and not induction of general anesthesia in the intracranial recordings . A small number of LPs were detected during induction of anesthesia in a subset of subjects over the course of extremely long ( >1 hr ) inductions in the scalp EEG dataset , but these were rare and vastly outnumbered by the more frequent LPs occurring during emergence . In addition , a previous intracranial study of slow ( ~1 hr ) inductions of propofol general anesthesia did not report analogous events ( Nourski et al . , 2017 ) , suggesting this phenomenon is primarily a signature of emergence . We also found that this transitional state is not identical to sleep: comparing neural dynamics during sleep and emergence from general anesthesia in the same subjects identified substantial differences in the power spectrum . The frontal alpha rhythm characteristic of propofol anesthesia is still present during emergence ( Feshchenko et al . , 2004; Murphy et al . , 2011; Purdon et al . , 2013 ) , but not during N2 sleep , indicating these are distinct brain states . Spontaneous alpha rhythms during propofol are thought to be generated by increased inhibitory tone in thalamic circuits , causing an intrinsic ~10 Hz dynamic to emerge ( Ching et al . , 2010 ) . These alpha rhythms are still present during the LPs , suggesting the thalamus may be exhibiting an altered excitatory/inhibitory balance as compared to sleep . However , despite the difference in spontaneous dynamics , the LP events themselves share many common properties with sleep , exhibiting highly similar waveforms and spatial profiles . In addition , LPs occurred at higher rates in response to more salient stimuli . A similar effect has also been found in sleep , as salient stimuli ( such as rare stimuli or the subject’s own name ) produce larger KC peaks during sleep ( Colrain et al . , 1999; Perrin et al . , 1999 ) . These events may therefore reflect an analogous effect of arousing stimuli in sleep and emergence , which could conceivably be related to some similarity in circuit state , such as ongoing tonic vs . bursting dynamics in thalamus . The common morphology of the LPs we observe during emergence and the KCs characteristic of sleep suggest that similar circuit mechanisms are engaged by auditory stimuli despite differences in the ongoing spontaneous dynamics . There is evidence that neuromodulatory arousal systems mediate emergence from general anesthesia , distinct from induction . Disruption of orexinergic signaling increases the time required for emergence from anesthesia , but does not change the dose-response sensitivity for induction ( Kelz et al . , 2008 ) . Coherent alpha ( 8–12 Hz ) and delta ( 1–4 Hz ) oscillations develop rapidly and pervasively across medial prefrontal cortex and thalamus at loss of consciousness induced by propofol , and likely mediate the functional disruption of these areas , contributing to the state of unconsciousness ( Flores et al . , 2017 ) . During emergence , these oscillations dissipate in a sequence distinct from induction , beginning with superficial cortical layers and medial and intralaminar thalamic nuclei , following known cortical and thalamic projection patterns for dopaminergic and cholinergic signaling ( Flores et al . , 2017 ) . Neuromodulatory activity during emergence could therefore create unique cortical and thalamic circuit states that enable LP responses to sensory stimulation . Given the similarity between LPs under anesthesia and sleep K-complexes , similar mechanisms might also play a role in modulating levels of arousal during sleep . The LPs we observe are qualitatively different from the ongoing slow oscillations that occur during deep anesthesia ( Steriade et al . , 1993; Breshears et al . , 2010; Murphy et al . , 2011; Lewis et al . , 2012 ) . LPs occur after the ongoing slow oscillation has largely subsided and reflect an isolated cortical DOWN state elicited by auditory stimulation , rather than a rhythmic cortical dynamic . However , the occurrence of LPs increases power in the same low-frequency bands of the spectrum that are occupied by the slow oscillation . Future studies may therefore need to take care that their analyses differentiate between these two distinct states , as increased low-frequency power may indicate isolated LP occurrence and foreshadow awakening , and will be important to distinguish from the slow oscillations of deep anesthesia . While the LPs were strikingly large , they may have been obscured in previous studies due to the brief and transient nature of the state in which they occur . In addition , we observed substantial heterogeneity across patients in terms of the frequency and timing of the LPs . In the intracranial data this heterogeneity may be partially explained by variation in electrode location , the duration and complexity of the surgery , and dosage of clinical medications administered to each patient . In the scalp EEG data , however , drug levels were controlled and no surgery was performed , yet heterogeneity across subjects was still present . This heterogeneity is also consistent with clinical observations , as patients are much more variable in how long they take to emerge than they are in induction . Following anesthetic emergence , patients exhibit variable levels of arousal , with some patients taking hours to return to alertness ( Larsen et al . , 2000 ) . While animal studies have reported stereotyped transitions between states ( Hudson et al . , 2014 ) , possibly due to increased experimental control and genetic similarity between individual animals , human studies have suggested that individuals may exhibit different trajectories during emergence from anesthesia ( Hight et al . , 2014 ) , and undergo different transitions between distinct , potentially sleep-like states ( Chander et al . , 2014 ) . This variability may reflect individual differences in arousal regulatory circuits or even in drug diffusion rates across the brain . It may also be that some patients pass through the transient stage too quickly for it to be identified using our analyses . Another possibility is individual physiological differences , such as receptor density or vascular properties , could modulate the relative rate of drug clearance in cortex and subcortex , and that only some individuals may experience this state . However , since these events were detected in all the intracranial patients we studied , it may be that the transient state occurs in most patients but is more challenging to detect in scalp EEG due to blurring of signals measured at the scalp . In addition , the healthy volunteers received a smaller total amount of propofol than the clinical patients , and may therefore have been more likely to emerge too rapidly to detect this brief state . The precise circuit mechanisms that generate the LP phenomenon are not clear , and will be challenging to identify with certainty using data from human subjects . However , we suggest that the sleep K-complex may share some mechanistic parallels with the LPs observed here . The KC is an isolated cortical DOWN state ( Cash et al . , 2009 ) and is likely to also involve thalamic circuits ( Jahnke et al . , 2012; Mak-McCully et al . , 2014 ) . While previous animal studies have identified spontaneous KCs during maintenance of ketamine-xylazine anesthesia ( Amzica and Steriade , 1998a ) , these occurred as part of an ongoing slow oscillation rather than the isolated auditory-evoked events seen here and during N2 sleep . Moreover , those events were not selective to anesthetic emergence , suggesting they represent a different phenomenon . Stimulus-evoked potentials in animal studies have primarily reported stimulus-evoked responses with a faster timecourse than the LPs reported here ( Amzica and Steriade , 1998b ) , perhaps reflecting a different phenomenon during relatively stable states of anesthesia in most animal studies , compared with dynamic changes during awakening . Future animal studies should therefore track the gradual process of emergence to identify the mechanisms of the isolated LPs identified here . One possible mechanism is that increased thalamic activation leads to strong stimulation of the thalamic reticular nucleus ( TRN ) , leading to a thalamic and subsequent cortical suppression . This theory would be consistent with animal studies that have induced slow waves through stimulation of TRN and suppression of thalamocortical neurons ( Lewis et al . , 2015 ) , and with human imaging studies demonstrating that emergence is associated with increased activity in subcortical arousal structures such as thalamus ( Långsjö et al . , 2012 ) . Alternatively , it may be that an inhibitory shift in the excitatory/inhibitory balance in cortex leads to a local profound suppression in response to sensory input , generating a local LP that can then spread across cortex or through corticothalamic projections . Future studies could explore these theories further through causal manipulations of cortical and thalamic activity during a gradual emergence process . These future investigations could also address some limitations of the current study . As intracranial electrodes are placed solely based on clinical need , we did not obtain whole-brain coverage , and had no thalamic recordings . Animal studies could investigate more systematically the spatial profile of the observed LPs . In addition , due to the nature of our experiment taking place in the operating room , we were constrained in timing and could not record throughout a continuous induction , maintenance , and emergence . In addition , the induction and emergence recordings were not counterbalanced in time due to the ordering of implant and explant procedures . They could potentially exhibit small changes in electrode position and signal-to-noise-ratio . While our data suggest no major difference in recording quality that could explain the striking LPs we observe , and we observe the LPs in scalp EEG as well despite opposite temporal ordering , more subtle phenomena could depend on differences in the recordings across these sessions . Highly controlled volunteer studies , as in our scalp data , will therefore be useful counterparts to any future intracranial investigations of these phenomena . Finally , our patient sample was small due to the rare nature of these recordings , and therefore we could not examine how the heterogeneity of LP dynamics might relate to emergence time or other clinical outcomes . Gathering datasets in larger patient cohorts would be very valuable for investigating how these dynamics can inform patient monitoring and predict functional outcomes . In particular , the LP events could potentially be used to monitor depth of anesthesia or predict when a patient will emerge , or they may be found to relate to emergence-related clinical outcomes , such as delirium . Future clinical studies would be highly beneficial for investigating these questions . In summary , we identified a transient brain state that occurs asymmetrically during emergence from general anesthesia . While deep states of anesthesia have been well characterized and exhibit stereotyped electrophysiological signatures , tracking transitions between states demonstrates the existence of transient and heterogeneous dynamics that occur selectively in the minutes before emergence . This state engages similar sensory-evoked circuit dynamics as in sleep , suggesting the brain may sometimes experience a sleep-like sensory blockade before recovering from general anesthesia . Written informed consent was obtained from all patients and experimental procedures were approved by the Massachusetts General Hospital/Brigham and Women’s Hospital Institutional Review Board . The enrolled patients had medically intractable epilepsy and underwent surgery to implant intracranial electrodes for clinical monitoring purposes ( Figure 1—figure supplement 1 ) . The location and number of electrodes implanted was determined by clinical criteria without regard to this study . Recordings were performed in the hospital operating room as the patients emerged from propofol general anesthesia . Recordings began after surgery was completed and while the clinical infusion of propofol was still running , and continued throughout the period after the infusion was stopped and patients emerged from anesthesia , until patients had to be disconnected for transport outside of the operating room . No seizures were recorded in these data . Patients received the typical clinical regimen of medication throughout the surgery ( including paralytics and analgesics ) , and in most cases the maintenance infusion also included remifentanil . We acquired intracranial recordings from 15 patients during emergence . Data from two patients were excluded due to poor recording quality , and data from one patient was excluded due to failure of the auditory stimulus equipment . A second emergence recording was acquired from one patient with electrodes implanted in different locations , and this recording was treated as another subject in the analysis ( total analyzed = 13 sessions , drawn from 12 individuals , five female , mean age 34 . 5 years , range 21–48 years ) . In four sessions , patients had only depth electrodes , and in nine sessions they had both subdural grid/strip and depth electrodes . Eight of these patients were also studied during gradual induction of general anesthesia when they returned 1–3 weeks later to undergo electrode removal surgery . In the induction recording , propofol was infused gradually using STANPUMP software with a target plasma concentration rising linearly over 10 min to a maximum of 6 µg/mL ( Schnider et al . , 1999 ) . Auditory stimuli were presented every 3 . 5–4 . 5 s with uniform temporal jittering ( 11 sessions ) or every 6 s ( two sessions ) using EPrime software and air-tube earphones to avoid stimulus-related artifacts in the electrophysiology data . Stimuli consisted of either a click train with a frequency of 40 Hz in one ear and 84 Hz in the other , lasting 2 s; a non-verbal sound ( e . g . door closing , alarm ) ; or a spoken word . Stimulus types were pseudorandomized throughout the presentation . Words and sounds were of neutral or negative affect; these distinctions were not analyzed in detail here . During the induction of anesthesia prior to the start of the surgery , patients listened to the stimuli and were asked to press a button to indicate whether the stimulus was a word . During emergence , stimulus presentation began near the time that the propofol infusion was stopped , and continued until the patient became responsive . The total presentation duration was 20 min , and if the patient had not yet emerged at that time then the presentation was restarted . Only two patients began performing the task at emergence . Due to this behavioral observation , clinical staff also periodically ( approximately every ~1–2 min ) asked subjects to open their eyes . Return of behavioral responsiveness was marked manually using two definitions: the first spontaneous movement observed by research staff ( labeled ‘First movement’ ) , and the time at which patients began responding to verbal requests to open their eyes or move their hands ( labeled ‘First response’ , defined as return of consciousness ( ROC ) ) . In 8 of these patients , the same behavioral task was used during induction of anesthesia 1–3 weeks later when patients returned for surgery to remove the intracranial electrodes . The task began 4 min prior to the start of the gradual propofol infusion and continued for 4 min after the target plasma level reached its maximum level . During anesthetic induction and emergence , intracranial recordings were acquired from depth and/or subdural grid and strip electrodes , with placement selected solely by clinical staff for clinical purposes . Recordings were acquired with an XLTEK acquisition system at a 2000 Hz sampling rate . Bad electrodes were manually identified and excluded from further analysis . Depth and strip electrodes were re-referenced to a bipolar montage in which an adjacent contact was subtracted from each channel . Grid electrodes were referenced to a Laplacian montage by subtracting the mean of the immediately neighboring electrodes . Data were detrended , lowpass filtered below 200 Hz , downsampled to 500 Hz , and highpass filtered above 0 . 16 Hz . The automated detector was designed to conservatively select events , missing some events but also reducing false positives . Since occasional large artifacts interfered with event detection , automatic detection of spontaneous events was restricted to the longest manually identified continuous segment with acceptable recording quality . All other timepoints were excluded from the automatic detection window . This approach was chosen due to the nature of the intracranial recording: we began recording as soon as possible , but clinical interaction with the patient at the beginning and end of the experiment , as well as connecting and disconnecting electrodes , led to very large artifacts at these timepoints whereas we obtained a long , stable recording during the emergence process . The median duration of this segment across patients was 650 s ( inter-quartile range: 580–1410 s ) . This long segment typically still included some periods with noise , which were rejected automatically in further analyses as described below . Data were first filtered between 0 . 2–4 Hz . All positive and negative peaks with an amplitude of at least 400 µV were identified . The duration of this peak , defined as the amount of time spent over a threshold of 40 µV , was required to be at least 400 ms . Peaks with amplitude greater than 1200 µV were discarded as artifact , and events occurring within 500 ms of a previous event were discarded . All events within a single electrode were required to have the same polarity , selected as whichever polarity was most frequent across all automatically detected events , since the referencing montage allowed potentials to be either negative- or positive-going depending on local polarity and electrode positioning . Trials with a range ( peak-trough difference ) exceeding 1500 µV were discarded as artifact . Event trials were defined as those trials with an automatically detected event occurring within 2 s of stimulus onset . The mean of all event trials was computed for each electrode that had at least five event trials . Because different electrodes had different polarities , the sign for negative-going electrodes was flipped . The median and quartiles were then computed across the pool of all electrode waveforms ( restricted to electrodes with at least five event trials ) and all 12 patients . Analyses of individual waveforms ( e . g . Figure 2c ) selected the electrode with the most detected events in each patient . Rise times and fall times were computed on the mean waveform for each selected electrode , by calculating the amount of time it took to rise and fall from a threshold of 200 µV to the peak of the mean event waveform . Bootstrapped 95% confidence intervals were calculated by resampling across subjects with replacement 1000 times , and reporting the 2 . 5th and 97 . 5th percentile of the resulting distribution . Spectrograms were computed using the electrode with the most LPs in each subject . Triggered spectrograms were computed relative to the peak of the LP waveform selected by the automatic detector . Spectral analysis was performed using multi-taper estimation ( Chronux , http://chronux . org , [Bokil et al . , 2010] ) . The analysis used three tapers and a sliding window of 200 ms duration every 50 ms . Spectrograms were normalized within frequencies to the mean power at that frequency between [−2–1] s prior to the peak . Broadband gamma power was computed by taking the mean power between 40 and 100 Hz , relative to the mean gamma power in the [−2–1] s window . Statistical analyses of gamma power were performed on the mean gamma power in the 300 milliseconds post-peak using the Wilcoxon signed-rank test . Spectra for ongoing spontaneous dynamics ( Figure 3 ) used six 30 s epochs within a continuous 3 min time window , using 19 tapers . Spectra were downsampled by a factor of 4 for display . Statistical comparison between time windows was performed by a hierarchical bootstrap resampling procedure: ( 1 ) resample subjects; ( 2 ) resample epochs within subjects; ( 3 ) compute the mean spectrum for each time window on the resampled time windows; ( 4 ) calculate the difference between the two spectra . This procedure was repeated 1000 times to obtain 1000 bootstrap estimates of the difference in the spectra; differences outside the [2 . 5 , 97 . 5] percentile for more contiguous frequency points than the spectral resolution of the multitaper estimate were labelled as significant and marked in red . One subject was excluded from the post-ROC vs . awake baseline comparison because electrode quality became too poor ( s . d . >500 µV ) after the patient emerged due to motion artifacts . Shaded error bars in the plot were computed in Chronux using jackknife estimation . Because the automatic detector imposes an artificial threshold on amplitude for events , the spatial analysis was performed on the stimulus-evoked potential over all electrodes . This analysis included only trials that were identified as generating an LP on at least one electrode , and excluded any trials with amplitude above 1500 µV as noise . The peak amplitude of the mean evoked potential in each electrode was plotted in color on a 3D reconstruction of the cortical surface generated using Freesurfer ( Fischl , 2012 ) and with grid and strip electrode coordinates registered to the surface of the brain ( Dykstra et al . , 2012 ) . To categorize the spatial location of electrodes , the nearest anatomical label from the Freesurfer automatic subcortical segmentation or cortical parcellation ( Destrieux et al . , 2010 ) was assigned . Electrodes identified as being in white matter and electrodes in regions with fewer than five contacts ( e . g . , putamen , occipital cortex ) were excluded from the spatial analysis . Statistical testing of which of the nine regions had significantly high proportions of electrodes with >5 LPs was performed with a binomial test , comparing each region to the mean across regions , with a Bonferroni correction for multiple comparisons across regions . Displayed grid timecourses are lowpass filtered below 30 Hz and downsampled to 100 Hz for display . Sliding window plots over induction and emergence were calculated by averaging all trials within a window of 60 s sliding every 15 s . For z-score analysis ( Figure 3c ) , the peak amplitude of the ERP was normalized to the standard deviation of the 1 . 5 s pre-stimulus , across each 60 s window , and the plots display the resulting z-scores . Calculations were only included when at least eight stimuli occurred within the window . When analyzing mean evoked amplitude across time windows ( Figure 3d ) , a 3 min period for each window was defined , and the mean evoked response was computed . The mean amplitude in the 0 . 5–1 . 5 post-stimulus window was then computed for each subject . As before , one subject was excluded from the post-ROC condition because electrodes began to be disconnected and recording quality was not usable . Recordings of natural sleep were obtained for three of the intracranial recording patients during their hospital stay ( after the emergence recording and prior to the induction recording ) . An experienced neurophysiologist ( G . P . ) scored the sleep data and manually labelled the onset and offset times of a subset of clearly visible KCs in the intracranial recordings for initial validation of the approach . Sleep data was acquired on a clinical system with a sampling rate of either 500 or 512 Hz . To match the propofol recording , the same reference electrodes were used for each electrode as in the emergence dataset , and then all electrodes were filtered between 0 . 16 Hz and 200 Hz . Any electrodes where the same reference electrodes were not available in both datasets were excluded . For analysis of median peak amplitude in individual events , electrodes with at least four events in each dataset were included . The histogram reflects all detected events on these electrodes , whereas the statistical test drew the same number of events from both the sleep and the propofol datasets for each subject . For bootstrap confidence interval estimation , data across subjects were pooled due to the small number of patients , and the bootstrap drew from datapoints pooled across the three patients . For comparison , within-subject statistics are also presented . To compare the spatial distribution of events across both datasets , event times were selected from a single electrode with at least 20 events in both datasets , and then the peak-triggered waveform across all electrodes was computed using these selected times . The mean value of the peak-triggered waveform between 100 ms pre-peak and 100 ms post-peak was calculated , and this mean event value was then compared across electrodes . Spectra were compared by identifying four 30 s windows of clean recordings with high LP rates in the emergence dataset , and randomly selecting four 30 s consecutive windows of N2 in the sleep dataset . Spectra were computed using Chronux with 19 tapers , downsampled by a factor of 4 for display , and error bars were computed with the jackknife method at p<0 . 05 . Scalp EEG analysis used data that was previously published ( Purdon et al . , 2013; Mukamel et al . , 2014 ) with full details provided in those publications . Briefly , healthy volunteers underwent monitoring with 64-channel EEG during a slow infusion of propofol , targeting a stepped increase from 0 to 5 µg/mL plasma concentration over one hour , and then a stepped decrease until the subjects recovered consciousness . Stimuli consisted of click trains ( 2 s duration ) , words , or the subject's own name , with stimulus type pseudorandomized throughout the experiment . 80% of the stimuli were click trains , 10% were words , and 10% were names . The LP analysis used a single frontal EEG electrode . For each stimulus presentation , we subtracted the mean and divided by the standard deviation during the 2 s pre-stimulus period . We then computed the maximum stimulus-evoked amplitude during the 1 s following stimulus presentation , and averaged these over 1 min windows .	General anesthesia is essential to modern medicine . It allows physicians to temporarily keep people in an unconscious state . When infusions of the anesthetic drug stop , patients gradually recover consciousness and awaken , a process called emergence . Previous studies using recordings of electrical activity in the brain have documented spontaneous changes during anesthesia . In addition , the way the brain responds to sounds or other stimulation is altered . How the brain switches between the anesthetized and awake states is not well understood . Studying the changes that happen during emergence may help scientists learn how the brain awakens after anesthesia . A key question is whether the changes that occur during emergence are the reverse of what happens when someone is anesthetized , or whether it is a completely different process . Knowing this could help clinicians monitoring patients under anesthesia , and help scientists understand more about how the brain transitions into the awake state . Now , Lewis et al . show that people go through a sleep-like state right before awakening from anesthesia-induced unconsciousness . In the experiments , recordings were made of the electrical activity in the brains of people emerging from anesthesia . One set of recordings was taken in people with epilepsy , who had electrodes implanted in their brains as part of their treatment . Similar recordings of brain electrical activity during emergence were also made on healthy volunteers using electrodes placed on their scalps . In both groups of people , Lewis et al . documented large changes in electrical activity in the brain’s response to sound in the minutes before emergence . These patterns of electrical activity during emergence were similar to those seen in patients during a normal stage of sleep ( stage 2 ) . Patients who were about to wake up from general anesthesia had suppressed brain activity in response to sounds , such as their name . Moreover , this sleep-like state happened only during emergence , indicating it is a distinct process from going under anesthesia . The experiments also suggest that the brain may use a common process to wake up after sleep or anesthesia . More studies may help scientists understand this process and how to better care for patients who need anesthesia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	A transient cortical state with sleep-like sensory responses precedes emergence from general anesthesia in humans
Most motile bacteria sense and respond to their environment through a transmembrane chemoreceptor array whose structure and function have been well-studied , but many species also contain an additional cluster of chemoreceptors in their cytoplasm . Although the cytoplasmic cluster is essential for normal chemotaxis in some organisms , its structure and function remain unknown . Here we use electron cryotomography to image the cytoplasmic chemoreceptor cluster in Rhodobacter sphaeroides and Vibrio cholerae . We show that just like transmembrane arrays , cytoplasmic clusters contain trimers-of-receptor-dimers organized in 12-nm hexagonal arrays . In contrast to transmembrane arrays , however , cytoplasmic clusters comprise two CheA/CheW baseplates sandwiching two opposed receptor arrays . We further show that cytoplasmic fragments of normally transmembrane E . coli chemoreceptors form similar sandwiched structures in the presence of molecular crowding agents . Together these results suggest that the 12-nm hexagonal architecture is fundamentally important and that sandwiching and crowding can replace the stabilizing effect of the membrane . Bacteria sense and respond to their environment through a chemotactic system that translates ligand binding ( stimulus ) into preferential flagellar rotation ( response ) , leading to either running in a favorable direction or tumbling to find a more favorable direction ( Hazelbauer et al . , 2008 ) . Briefly , ligands bind to the periplasmic domains of the methyl-accepting chemotaxis proteins ( MCPs ) , either directly ( Milburn et al . , 1991; Englert et al . , 2010 ) or via periplasmic binding proteins ( Tam and Saier , 1993 ) . This results in conformational changes that traverse the length of the MCP , through one or more HAMP ( histidine kinases , adenyl cyclases , MCPs , and some phosphatases ) domains and the cytoplasmic coiled-coil signaling domain to either activate or inactivate a histidine kinase , CheA , bound at the distal tip ( Wadhams and Armitage , 2004; Kentner and Sourjik , 2006; Hazelbauer et al . , 2008 ) . The bacterial chemotactic system is best understood in Escherichia coli , where system components are located in a single operon ( Silverman and Simon , 1976 ) . In E . coli , CheA is a 5-domain ( P1-P5 ) protein that functions as a homodimer . When activated by the MCP in unfavorable environments , CheA undergoes autophosphorylation and then transfers the phosphoryl group to one of two response regulators . One of these , CheY , binds the flagellar motor when phosphorylated , switching the motor’s direction from clockwise to counter-clockwise and thus increasing the frequency of tumbling to find a more favorable direction ( Turner et al . , 2000 ) . The other response regulator , CheB , is a methylesterase whose activity is stimulated by phosphorylation . Its activity is opposed by the constitutively active methyltransferase CheR and the balance of these two activities determines the methylation state of specific glutamate residues in the MCPs . These methylations confer adaptation on the system , modulating its response based on recent environmental conditions ( Kleene et al . , 1979; Toews et al . , 1979; Lupas and Stock , 1989 ) . In E . coli , MCPs associate into trimers of dimers that span the inner membrane ( Kim et al . , 1999; Studdert and Parkinson , 2004 ) and complex with CheA and the coupling protein CheW ( Figure 1A ) . This results in hexagonally packed trimers-of-receptor-dimers surrounding a ring of alternating CheA P5 domains and CheW . Neighboring rings are linked together by CheA P3 dimerization domains to form extended hexagonal lattices – the so-called transmembrane chemoreceptor arrays ( Briegel et al . , 2012; Liu et al . , 2012 ) . The architecture of these membrane-bound chemoreceptor arrays is not specific to E . coli , indeed it is universal among bacteria ( Briegel et al . , 2009 ) . The architecture of these arrays is likely key to the high sensitivity , wide dynamic range , cooperativity , and feedback control of this system ( Duke and Bray , 1999; Li and Weis , 2000; Gestwicki and Kiessling , 2002; Sourjik and Berg , 2002 , 2004; Li and Hazelbauer , 2005; Endres and Wingreen , 2006 ) . 10 . 7554/eLife . 02151 . 003Figure 1 . Structure of membrane-bound and cytoplasmic receptor complexes . ( A ) Schematic showing the topology of receptor-trimers-of-dimers ( purple ) , CheA ( domains P1–P4 light gray/dark gray indicating domains in each CheA monomer , domain P5 orange ) , and CheW ( blue ) in membrane-bound arrays . The methylation region of each receptor dimer is indicated by a pink star . IM = inner membrane . ( B ) Top-view of the arrangement of the array components showing the interaction sites between the receptors , CheA and CheW , colored as in ( A ) . ( C ) Schematic showing the topology of receptor , CheA , and CheW complexes in cytoplasmic arrays , colored as in ( A ) . Cytoplasmic chemoreceptors assemble into two hexagonally packed arrays interacting at their presumably ligand-binding tips . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 003 Besides the single chemosensory operon encoding components of the transmembrane array , exemplified by E . coli , many bacteria contain additional chemotaxis operons , whose products are less well understood . The best-studied case is Rhodobacter sphaeroides , in which there are three operons , two of which are essential for normal chemotaxis in laboratory conditions . One of these operons , CheOp2 , encodes components of the chemotaxis array associated with the transmembrane MCPs . The other , CheOp3 , encodes components of a cytoplasmic chemotaxis cluster including CheB , CheR , CheY , CheW , two atypical CheA proteins ( CheA3 and CheA4 ) ( Porter et al . , 2008 ) , and a soluble chemoreceptor , TlpT . Each of the CheA proteins in the cluster lacks canonical domains compared to E . coli CheA: CheA3 contains only the P1 and P5 domains separated by a 794 amino acid sequence of unknown structure , whereas CheA4 lacks the P1 and P2 domains . CheA4 can transfer phosphoryl groups to P1 of CheA3 ( Porter et al . , 2008 ) . Both CheA3 and CheA4 are required for chemotaxis ( Wadhams et al . , 2003; Porter et al . , 2008 ) . A second MCP homolog lacking any transmembrane domain , TlpC , is encoded in CheOp2 . Fluorescent fusions of both TlpT and TlpC localize to a single cytoplasmic focus at mid-cell , which divides and is segregated during cell division ( Wadhams et al . , 2002 , 2003 ) . CheA3 and CheA4 , along with CheW4 and a CheR , also localize to this cytoplasmic cluster ( Wadhams et al . , 2003 ) . CheW4 and TlpT are required to form the cluster , and the positioning of the cluster depends on the ParA homolog PpfA , which is encoded next to TlpT in CheOp3 ( Thompson et al . , 2006 ) . PpfA is a chromosome-associated ParA-like ATPase and controls the localization and segregation of the cytoplasmic cluster through an interaction with the N-terminus of TlpT , which is thought to stimulate the ATPase activity of PpfA in a ParB-like manner ( Roberts et al . , 2012 ) . Sequence analysis of TlpT suggests that the 567 amino acid protein forms a classical MCP coiled-coil structure ( Alexander and Zhulin , 2007 ) . The 580 amino acid TlpC lacks significant homology to the canonical MCP structure and lacks a recognizable HAMP domain or methylation region . Cytoplasmic chemoreceptors are not unique to R . sphaeroides . The genomes of many bacterial species encode MCP homologs lacking transmembrane domains ( Ulrich and Zhulin , 2010; Wuichet and Zhulin , 2010 ) . However , the organization and precise function of these cytoplasmic chemoreceptor clusters remain unknown . In the case of R . sphaeroides , it has been speculated that cytoplasmic MCPs modulate chemotactic response based on the current metabolic state of the cell ( Armitage and Schmitt , 1997; Porter et al . , 2008 ) , while the inputs of cytoplasmic receptors in other organisms such as Vibrio cholerae are less clear . In the case of E . coli transmembrane arrays , the membrane is thought to be important for proper assembly and function ( Miller and Falke , 2004; Draheim et al . , 2006; Amin and Hazelbauer , 2012 ) . We were therefore interested in how cytoplasmic receptors cluster in the absence of organizing membrane . We used tomography of freeze-substituted , plastic-embedded sections; immunoelectron microscopy; electron cryotomography ( ECT ) of both intact cells and cryosections ( Gan and Jensen , 2012 ) ; and correlated cryogenic fluorescence light microscopy and ECT ( cryo-FLM/ECT ) to characterize the structure of cytoplasmic chemoreceptor clusters in two bacteria: R . sphaeroides and V . cholerae . We also report the ability of normally transmembrane chemoreceptors from E . coli to form cytoplasmic-like arrays in the absence of a membrane . We initially tried to image cytoplasmic chemoreceptor arrays in intact R . sphaeroides by ECT . Unfortunately , we found that the size and cytoplasmic density of this species limits the achievable resolution in tomograms of intact cells , and we could not resolve the cytoplasmic clusters . To circumvent this problem , we employed two approaches that allowed us to view sections of the otherwise thick samples: cryosectioning vitreously frozen cells and room temperature sectioning of high-pressure frozen , freeze-substituted cells . Tomographic reconstructions of cryosections of R . sphaeroides cells ( containing the chemotaxis protein TlpC tagged with GFP to aid in future identification ) revealed curved , double-layered structures that might correspond to cytoplasmic clusters ( Figure 2A , B ) , but they were small and difficult to differentiate from the complex environment . To gain confidence in our identification , we used a strain in which some cytoplasmic chemotaxis components are overexpressed . This strain exhibited chemotaxis in swim plate assays , indicating that the limited overexpression did not significantly impair chemotactic function . By fluorescence light microscopy ( FLM ) , short cells contained single fluorescent foci and long cells two foci , as expected ( Wadhams et al . , 2002 ) . In tomograms of cryosections of these cells , we observed similar structures to those seen in WT cells but with greater length , appearing in cross-section as a curved , double-layered sandwich approximately 27 nm wide ( measured from density peak-to-peak; Figure 2C–E ) . 10 . 7554/eLife . 02151 . 004Figure 2 . Tomography of R . sphaeroides cryosections reveals cytoplasmic clusters . Tomographic slices of cryosections through R . sphaeroides cells expressing TlpC-GFP at wild-type ( A and B ) or overexpressed ( C–E ) levels . ( A ) Tomographic slice of a cell expressing TlpC-GFP , revealing a potential cytoplasmic chemoreceptor array ( black arrows ) , pseudo-colored red in ( B ) . ( C–E ) Cryosections of cells overexpressing cytoplasmic chemoreceptor components , including TlpC-GFP , contain similar structures to those observed under WT expression conditions: two curved layers approximately 27 nm apart ( brackets ) , with perpendicular striations between them ( black arrowheads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 004 It was difficult to identify the location of the structure in the cryosections , so we turned to a technique that would allow us to place the structure in the context of the whole cell . In tomograms of high-pressure frozen , freeze-substituted cells , we again observed extended , double-layered structures ( Figure 3 ) . Cross-sections through the structures revealed two curved layers with perpendicular striations connecting them . The positions of the structures in the cells were consistent with FLM observations ( Thompson et al . , 2006 ) , strengthening our confidence in the identification of the cytoplasmic array . We did occasionally see multiple clusters at these positions , consistent with fluorescence results in cells overexpressing these proteins . 10 . 7554/eLife . 02151 . 005Figure 3 . Tomography of cytoplasmic clusters in freeze-substituted R . sphaeroides . Dual-axis tomographic reconstruction of high-pressure frozen , freeze-substituted cells reveals cytoplasmic clusters in most cells . Clusters appear as partially circular structures , typically located near storage granules . ( A ) Reconstruction of a longitudinally sectioned cell showing a single cluster . Inset: high-magnification detail of the cluster in a single , 6 . 7-nm tomographic slice . Scale bar 50 nm . ( B ) Cell containing two cytoplasmic clusters located next to storage granules . Inset: high-magnification detail of one cluster . ( C ) Model ( blue ) superimposed over the tomogram shown in B , highlighting the clusters . Scale bars 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 005 To check whether the structures we observed contained the cytoplasmic chemotaxis protein TlpC , we performed immunogold labeling of the GFP tag in our sections . In this technique , gold-conjugated anti-GFP antibodies recognize and bind epitopes on the surface of the section , which is subsequently imaged by tomography . We consistently observed gold labeling of the curved structures described above , and tomography of the volume beneath the label confirmed the structural details seen in cryosections ( n = 17 cells , example shown in Figure 4 ) , indicating that these structures are indeed cytoplasmic chemoreceptor arrays . 10 . 7554/eLife . 02151 . 006Figure 4 . Identification of cytoplasmic clusters in R . sphaeroides by tomography of immunolabeled , negatively stained sections . Sections were immunolabeled with antibodies against GFP and gold-conjugated secondary antibodies . ( A–C ) Three examples of immunolabeled cells ( left ) and corresponding tomographic slices ( right ) . Clusters ( black arrowheads ) were typically labeled by 1–3 gold particles ( red arrows ) , depending on the amount of antigen present on the section surface . The heavily labeled cluster in ( C ) is likely oriented en face and near the section surface , allowing for access to a larger number of GFP antigens . Scale bars 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 006 To corroborate this identification , we used correlated cryo-FLM/ECT . Since whole R . sphaeroides cells are too thick and dense to image by ECT with high resolution , we flattened them by gentle lysis . However , unlike transmembrane receptors that remain well-organized following cell lysis ( Briegel et al . , 2009 , 2012 , 2013 ) , fluorescent foci were not observed after lysis , indicating disruption of the cytoplasmic receptor cluster . We reasoned that in the absence of membrane , the density of the cytoplasm might play an important role in stabilizing cytoplasmic clusters . We therefore added molecular crowding agents , either polyethylene glycol ( PEG-8000 ) or polyvinylpirrolidone ( PVP ) , to the lysis protocol and used this approach to image an R . sphaeroides strain overexpressing the cytoplasmic chemoreceptor TlpT fused to YFP . We observed preservation of TlpT-YFP signals by FLM after lysis ( Figure 5A–C ) . We then froze lysed cells on an EM grid and localized a single TlpT-YFP signal by cryo-FLM , then imaged the indicated location at high resolution by ECT . We observed that the fluorescence signal corresponded to the structure previously identified ( Figure 5D–F ) , further verifying that these are indeed cytoplasmic chemoreceptor arrays containing TlpT . 10 . 7554/eLife . 02151 . 007Figure 5 . Correlative cryo-FLM/ECT of R . spheroides cytoplasmic arrays . ( A–C ) Molecular crowding preserves cytoplasmic chemoreceptor clusters during lysis . Overlay of phase contrast and TlpT-YFP fluorescence images of R . sphaeroides cells of strain JPA1558OE before ( A ) and after lysis either in the absence ( B ) or presence ( C ) of 10% PVP . ( D–F ) Correlative cryo-FLM/ECT identifies a cytoplasmic array . ( D ) Overlay of fluorescent signal from TlpT-YFP ( yellow ) and a low magnification cryo-EM image of the same lysed cell . ( E ) Overlay of TlpT-YFP fluorescent signal ( yellow ) and a cross-section of a reconstructed tomogram corresponding to the same region , identifying the structure of a cytoplasmic chemoreceptor array . ( F ) An enlarged view of the cytoplasmic array reveals a partial ring consisting of two layers of chemoreceptors . Scale bar 100 nm . ( G–J ) Receptor packing is identical in R . sphaeroides cytoplasmic and transmembrane arrays . ( G ) Top view through one of the two layers of the array imaged in D–F and corresponding power spectrum ( H ) reveal a hexagonal receptor arrangement with center-to-center spacing of 12 nm . Scale bar 25 nm; power spectrum not to scale . ( I ) Top view through a membrane-bound chemoreceptor array and corresponding power spectrum ( J ) reveal hexagonal packing identical to that of the cytoplasmic array shown in ( G–H ) . Scale bar 25 nm; power spectrum not to scale . ( K–L ) Additional examples of cytoplasmic chemoreceptor arrays ( brackets with arrowheads ) in tomograms of R . sphaeroides cells lysed with molecular crowding agents added . Distance between baseplates is approximately 30 nm . Scale bars 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 007 We compared the structure of the cytoplasmic chemoreceptor cluster identified by correlated cryo-FLM/ECT to that of a transmembrane receptor cluster in the same cell . Interestingly , the arrangement and packing of the receptors was identical in both cases: a hexagonal lattice with center-to-center spacing of 12 nm when viewed from above ( Figure 5G–J ) . In the transmembrane array , this receptor lattice is associated with a single baseplate , whereas in the cytoplasmic array , there are two , one on either side , approximately 30 nm apart . Unfortunately , due to the tight curvature , we could only confirm the hexagonal order of one of the two layers of the cytoplasmic array . As discussed above , due to the substantial density and thickness of R . sphaeroides cells , we were unable to identify cytoplasmic chemoreceptor arrays in intact cells . To obtain higher resolution information from intact cells , we turned to V . cholerae , which is also predicted to contain cytoplasmic chemoreceptors and flattens significantly during freezing for cryo-EM , yielding a thinner sample . By ECT of whole V . cholerae cells , we observed cytoplasmic chemoreceptor clusters with the same architecture as those in R . sphaeroides , though lacking curvature ( Figure 6 ) . The improved resolution in the thinner cells allowed us to clearly distinguish two hexagonal arrays of trimers-of-receptor-dimers sandwiched between two baseplates , 35 nm apart ( Figure 6 ) . The kinase-binding regions of the receptors were highly ordered , with decreasing order toward the center , as previously observed for transmembrane chemoreceptor arrays . 10 . 7554/eLife . 02151 . 008Figure 6 . In vivo architecture of the V . cholerae cytoplasmic array . Left: side view of a membrane-bound chemoreceptor array ( MA ) and a cytoplasmic chemoreceptor array ( CA ) . The cytoplasmic array is composed of two parallel CheA/W baseplates approximately 35 nm apart . The chemoreceptors are sandwiched between the two baseplates and are hexagonally packed with a 12 nm center-to-center spacing . Right: top views of the receptor packing close to the CheA/CheW baseplate on either side and corresponding power spectra ( top insets ) , as well as sixfold symmetrized subvolume averages ( bottom insets ) reveal that the hexagonal arrangement of the receptors is identical to that of the membrane bound array described previously ( Briegel et al . , 2009 ) . Scale bars 50 nm . CA , Cytoplasmic chemoreceptor array; MA , membrane-bound chemoreceptor array; IM , inner membrane; OM , outer membrane . Power spectra not to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 008 We observed a striking similarity between the cytoplasmic arrays in R . sphaeroides and V . cholerae and an in vitro chemoreceptor preparation from E . coli . We purified a cytoplasmic fragment of the aspartate-sensing Tar receptor , as well as CheA and CheW , from E . coli ( Montefusco et al . , 2007; Fowler et al . , 2010 ) and assembled complexes in vitro using these components , with CheA and CheW present in excess . The mixture of purified proteins also contained the molecular crowding agents PEG-8000 and trehalose to simulate cytoplasmic conditions . Once formed , the complexes activated the kinase as effectively as when assembled on vesicles ( Fowler et al . , 2010 ) . Using ECT , we observed extended arrays with identical architecture to that of the in vivo cytoplasmic clusters described above: two baseplates , approximately 31 nm apart , flank two hexagonal lattices of chemoreceptor trimers with 12 nm center-to-center spacing ( Figure 7 ) . The receptors from the two sides interact at their methylation tips . Again , as we observed for transmembrane chemoreceptors , the kinase-binding tips were well ordered , with decreasing order towards the middle of the sandwich . 10 . 7554/eLife . 02151 . 009Figure 7 . E . coli Tar chemoreceptors lacking transmembrane regions can form cytoplasmic-like arrays in the presence of CheA , CheW , and a molecular crowding agent . Tomographic slice showing cytoplasmic fragments of Tar forming extended arrays in the presence of CheA , CheW , and molecular crowding agents . These arrays closely resemble the cytoplasmic chemoreceptor arrays seen in V . cholerae ( Figure 6 ) . Side-view ( black arrowheads ) reveals two flat , parallel CheA/W baseplates spaced approximately 31 nm apart . Top views of the chemoreceptors close to the baseplates ( white arrows ) reveal a well-ordered , hexagonal arrangement with a center-to-center spacing of 12 nm . Enlarged subvolume average ( inset ) confirms that the packing is identical to that of in vivo arrays . Scale bar 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 009 The most striking difference between the cytoplasmic arrays in R . sphaeroides and V . cholerae was the curvature observed in the former . In all cases observed , cytoplasmic V . cholerae arrays and arrays formed in vitro from E . coli cytoplasmic receptor fragments were flat , while those of R . sphaeroides exhibited kinks and regions of high curvature , even forming full rings in some cases . We wondered if this curvature may be induced by interactions between the fluorophores rather than being a property of the arrays themselves , as tags are known to introduce such artifacts ( e . g . , Swulius and Jensen , 2012 ) . To test this , we performed tomography of high-pressure frozen , freeze-substituted cells overexpressing untagged components of the cytoplasmic cluster . We observed structures identical to those observed in the tagged strains ( Figure 8A ) , indicating that curvature is not an artifact induced by attached fluorophores . 10 . 7554/eLife . 02151 . 010Figure 8 . R . sphaeroides cytoplasmic arrays are inherently curved . ( A ) Tomographic slice of a high-pressure frozen , freeze-substituted cell overexpressing native , untagged components of the cytoplasmic chemoreceptor cluster . Arrowheads indicate curved array . ( B ) Tomographic slice of a similarly treated cell lacking the partitioning protein PpfA . Arrowheads indicate curved cytoplasmic array . Scale bars 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 010 Cytoplasmic arrays in R . sphaeroides are segregated by a ParA homolog , PpfA , which associates with the N-terminus of TlpT and with the chromosomal DNA . To test whether this interaction imposes curvature on the cytoplasmic array , we similarly imaged cells lacking PpfA . Again , we observed structures with similar curvature , indicating that this is an intrinsic property of the array , not affected by the segregation machinery ( Figure 8B ) . Over half of all genomes sequenced of motile bacteria have more than one putative chemosensory pathway and many of these encode putative soluble chemoreceptors ( Wuichet and Zhulin , 2010 ) . In this study , we describe the structure of cytoplasmic chemoreceptor arrays in two distantly related bacterial species , R . sphaeroides and V . cholerae . Notably , cytoplasmic arrays in both species display a hexagonal arrangement of receptor-trimers-of-dimers , with 12 nm center-to-center spacing that is identical to that of transmembrane chemoreceptor arrays , suggesting a fundamental utility for this architecture . Whereas the transmembrane array is embedded in the membrane with a CheA/W baseplate at the receptors’ membrane-distal tips and ligand-binding domains in the periplasm , the cytoplasmic arrays assemble as a sandwich , with two CheA/W baseplates flanking two sheets of MCPs that interact in the middle , presumably at their ligand-binding domains ( Figure 1C ) . No membrane is associated with these arrays . The identity of the outer baseplates as CheA/W is supported by our observation of the same structure in in vitro preparations from E . coli , where only three proteins are present: receptor fragments , CheA , and CheW . The similarity of the array organization in membrane-bound and cytoplasmic clusters suggests a common organizational theme , in which either the membrane or the crowded environment of the cytoplasm provides necessary stability to the conserved architecture of a 12 nm hexagonal array of receptor-trimers-of-dimers . Interestingly , while transmembrane arrays remain intact upon cell lysis , presumably due to the stabilizing effect of the membrane , cytoplasmic arrays fall apart unless molecular crowding agents are added . This suggests that cellular crowding contributes to the stability of the array and may be an important factor in assembly of these , and likely other , cellular structures ( Ellis , 2001; Zhou et al . , 2008; Zhou , 2013 ) . We observe the same requirement for molecular crowding ( mimicking the cellular environment ) in our in vitro preparation from E . coli . This supports the idea that chemoreceptor interactions can be enhanced by membrane binding ( Shrout et al . , 2003 ) , but in the absence of a membrane , this can be achieved by molecular crowding and sandwiching ( Fowler et al . , 2010 ) . The hexagonal order of the R . sphaeroides and V . cholerae cytoplasmic arrays , with 12 nm center-to-center spacing , is identical to the arrangement of transmembrane chemoreceptor arrays in other bacteria ( Briegel et al . , 2009 ) , as well as to arrays formed from cytoplasmic fragments of E . coli receptors . This indicates that the atypical R . sphaeroides CheA homologs ( CheA3 and CheA4 ) , despite their sequence anomalies , assemble in a similar fashion to canonical CheA proteins , including the three CheA homologs found in V . cholerae ( Ulrich and Zhulin , 2010 ) . From what is known in other organisms , R . sphaeroides CheA3 and CheA4 likely intercalate into CheA/W rings with their P5 domains ( Briegel et al . , 2012; Liu et al . , 2012; Li et al . , 2013; Natale et al . , 2013 ) . This packing would position both atypical CheA homologs in close proximity to each other , thereby facilitating phosphotransfer from the P4 domain of CheA4 to the P1 domain of CheA3 . The most striking difference between cytoplasmic chemoreceptor arrays in R . sphaeroides and V . cholerae is their curvature . In our observations , all cytoplasmic V . cholerae arrays and arrays formed in vitro from E . coli cytoplasmic receptor fragments were flat , while those of R . sphaeroides exhibited kinks and regions of high curvature , even forming full rings in some cases . This indicates that curvature may not affect performance unduly , with functioning arrays found in linear , concave , and convex conformations . However , there may be an upper limit of curvature tolerance . The discontinuities we observe in cytoplasmic arrays may result from ‘breaking’ of the CheA/W baseplate on one side under conditions of extreme curvature . In transmembrane arrays , ligand-binding domains are located in the periplasm . In cytoplasmic arrays , we assume that these domains are located in the middle of the sandwich . This configuration is unlikely to inhibit access to signaling molecules as the receptor arrangement is much more porous than the outer membrane , which extracellular ligands need to cross in order to bind transmembrane arrays . The signaling implications of this arrangement remain unclear , but it is intriguing to think that ligand binding in the middle of the sandwich could result in CheA activation in both baseplates , facilitating amplification . Strains used in this study are listed in Table 1 . R . sphaeroides strains were cultured in succinate medium ( Sistrom , 1960 ) supplemented with 25 µg/ml kanamycin and 30 µM IPTG ( to induce expression of plasmid-borne FliA from pIND [Ind et al . , 2009] ) at 30°C with shaking . V . cholerae strains were cultured in LB media at 37°C with shaking . V . cholerae cells were then mixed 1:1 with another bacterial species grown in Ca-HEPES buffer ( 25 mM HEPES , 2 mM CaCl2 at pH 7 . 6 ) and grown for an additional 16 hr at 30°C with shaking . 10 . 7554/eLife . 02151 . 011Table 1 . Strains used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 02151 . 011StrainRelevant genotypeSourceJPA543WS8N TlpC-GFPWadhams et al . , 2002JPA543OETlpC-GFP pIND-FliAThis studyJPA1558OETlpT-YFP pIND-FliAThis studyWS8NOEpIND-FliAThis studyJPA1330OEΔppfA TlpC-GFP pIND-FliAThis studyMKW1383N16961 ΔctxAB::kanMatthew Waldor Cells were immobilized on agarose pads and imaged using a Nikon Eclipse 90i microscope ( Nikon Instruments Inc . , Melville , NY ) using a 100× oil immersion lens and images were recorded on a Coolsnap HQ2 camera ( Photometrics , Tuscon , AZ ) operated using the Metamorph software ( Molecular Devices , Chicago , IL ) . Vitreous cryosections were prepared as previously described ( Ladinsky et al . , 2006; Ladinsky , 2010 ) . Briefly , R . sphaeroides cells were pelleted and resuspended in ∼100 μl of 50% dextran in SUX buffer then rapidly-frozen in a BalTec HPM-010 high-pressure freezer ( Leica Microsystems , Vienna ) using dome-shaped brass planchettes , then stored in liquid nitrogen . Once opened under liquid nitrogen , the sample-containing planchette was placed in a FC6/UC6 cryoultramicrotome equipped with a model M micromanipulator ( Leica Microsystems ) . Blockfaces were trimmed with a diamond trimming knife and sections were cut with a 25° Cryo-Platform knife ( Diatome-US ) at −145°C . Ribbons of sections ( 130 nm ) were transferred to carbon-coated , 200-mesh copper grids ( Electron Microscopy Sciences ) and stored in liquid nitrogen . 20 µl cell culture was mixed with pelleted 100 µl colloidal gold solution , BSA treated to avoid aggregation ( Iancu et al . , 2007 ) . 3 µl of this cell-gold mixture was then applied to R2/2 copper Quantifoil grids ( Quantifoil Micro Tools ) . After blotting away excess liquid using a Vitrobot ( FEI ) , the sample was plunge-frozen in a liquid ethane-propane mixture ( Iancu et al . , 2007; Tivol et al . , 2008 ) . Images were collected using either an FEI G2 ( FEI Company , Hillsboro , OR ) 300 kV field emission gun electron microscope or an FEI TITAN Krios ( FEI Company , Hillsboro , OR ) 300 kV field emission gun with an image corrector for lens aberration correction . Both microscopes were equipped with Gatan image filters ( Gatan , Pleasanton , CA ) and K2 Summit counting electron detector cameras ( Gatan , Pleasanton , CA ) . Data were collected using the UCSFtomo software ( Zheng et al . , 2007 ) using cumulative electron doses of ∼160 e/A2 or less for each individual tilt-series . The images were CTF corrected , aligned , and reconstructed using weighted back projection using the IMOD software package ( Kremer et al . , 1996 ) . SIRT reconstructions were calculated using TOMO3D ( Agulleiro and Fernandez , 2011 ) . Subvolume averaging and symmetrizing were done using PEET ( Nicastro et al . , 2006 ) . R . sphaeroides cells were grown at 30°C to stationary phase in SUX buffer , centrifuged and the pellets resuspended in SUX buffer +10% Ficoll ( 70 kD; Sigma ) . The cells were centrifuged gently and the supernatant was removed . Packed cells were placed in aluminum or brass planchettes ( #39201 , Ted Pella , Inc . ) and high-pressure frozen as described above . Frozen samples in closed planchettes were transferred under liquid nitrogen to cryogenic vials ( Nalge Nunc International , Rochester NY ) containing 2% OsO4 , 0 . 05% uranyl acetate in acetone . Vials were placed in an AFS freeze-substitution machine ( Leica Microsystems ) and processed at −90°C for 72 hr , warmed over 10 hr to −20°C and further processed at that temperature for 24 hr . The samples were brought to room temperature , rinsed four times with acetone , removed from the planchettes and infiltrated with Epon-Araldite epoxy resin ( Electron Microscopy Sciences , Hatfield PA ) . Samples were placed in embedding molds and polymerized at 60°C for 24 hr . Embedded pellets of R . sphaeroides cells were serially sectioned with an EM-UC6 ultramicrotome ( Leica Microsystems ) using a diamond knife ( Diatome-US , Hatfield PA ) . Thick ( 350 nm ) sections were placed on Formvar-coated , copper-rhodium 1-mm slot grids ( Electron Microscopy Sciences ) and stained with 3% aqueous uranyl acetate and lead citrate . Colloidal gold particles ( 10 nm ) were placed on both surfaces of the grid to serve as fiducial markers for image alignment . Grids were placed in a model 2040 dual-axis tomography holder ( Fischione Instruments , Export PA ) and imaged with a Tecnai TF30ST-FEG transmission electron microscope ( FEI ) at 300 keV . Dual-axis tilt-series were acquired automatically using the SerialEM software package ( Mastronarde , 2005 ) . Samples were tilted ± 60° and imaged at 1° intervals about orthogonal axes . Images were recorded digitally with an UltraScan 994 1000XP camera ( Gatan , Inc . Pleasanton CA ) . Tomographic data was subsequently processed , analyzed , and modeled using the IMOD software package ( Kremer et al . , 1996 ) . R . sphaeroides cells were cultured to stationary phase as described above . TlpC-GFP signal was verified by fluorescence microscopy . The cells were pelleted , the supernatant was removed and the pellet was resuspended in 10% paraformaldehyde , 5% sucrose in SUX buffer . The cells were pelleted again and the supernatant was replaced with fresh fixative without disturbing the pellet . The cells were fixed overnight at 4°C . The samples were brought to room temperature , the fixative was removed and the pellets were infiltrated into 2 . 1M sucrose in PBS over the course of 1 day , with the sucrose solution changed at 1-hr intervals . The pellets were transferred to aluminum sectioning stubs and rapidly frozen in liquid nitrogen . Semi-thin ( 90 nm ) cryosections were cut at −110°C with a FC6/UC6 cryoultramicrotome ( Leica Microsystems ) using a Cryo-Immuno diamond knife ( Diatome-US ) . Cryosections were picked up in a drop of 2 . 3M sucrose in PBS and transferred to Formvar-coated , carbon-coated , glow-discharged 100-mesh copper-rhodium grids . Sections were incubated for 30’ with 10% calf serum in PBS to block nonspecific antibody binding sites , then labeled with a monoclonal antibody against GFP ( Rockland Immunochemicals , Inc . Gilbertsville PA ) diluted in 5% calf serum in PBS , followed by a colloidal gold ( 15 nm ) conjugated anti-mouse secondary antibody ( BBI International , Grand Forks ND ) . After labeling , the sections were negatively stained with 1% uranyl acetate and stabilized with 1% methylcellulose . Immunolabeled sections were imaged with a Tecnai T12 electron microscope at 120 keV . Tomographic tilt-series were acquired and analyzed as described above . Mid-log phase cells of strain JPA1558 were collected by centrifugation and resuspended in crowding buffer ( 20% sucrose , 7% PEG-8000 , 4% trehalose , 25 mM Tris–HCl , pH 7 . 0 ) with 2 mg/ml lysozyme , then incubated for 15 min at room temperature . Following addition of 2 mM MgCl2 , 250 µM CaCl2 , and 1 mg/ml DNase I , cells were incubated for 30 min at 37°C , collected by centrifugation and resuspended in crowding buffer for imaging . Identical results were obtained with a different crowding agent , replacing the PEG-8000 and trehalose with 10% PVP and treating the cells as before . Cells of strain JPA1558 were lysed as described above and plunge-frozen on copper EM finder grids ( Quantifoil Micro Tools ) , loaded into a cryo-FLM stage ( FEI company ) and imaged on a Nikon 90Ti inverted microscope ( Nikon Instruments Inc . , Melville , NY ) using a 60x ELWD air objective lens and a Neo 5 . 5 sCMOS camera ( Andor Technology , South Windsor , USA ) using the NIS Elements software ( Nikon Instruments Inc . , Melville , NY ) . The grid was then transferred to the cryo-EM and tilt series were recorded from the same cells imaged by FLM previously . The grid was kept below −150°C at all times during the imaging process . Cytoplasmic fragments of the Tar receptor ( CF4Q ) , CheW , CheA , and CheY were expressed and purified as previously described ( Fowler et al . , 2010 ) . The Tar fragments contained methylation-mimicking glutamine residues at the four major sites of receptor methylation . Protein purity was assessed with SDS-PAGE analysis , and protein concentrations were determined using a BCA assay ( Thermo Fisher Scientific ) . PEG 8000 ( Fluka ) and D- ( + ) -trehalose ( Sigma–Aldrich ) were prepared as 40% ( wt/vol ) stock solutions in deionized water and passed through a 0 . 22-μm syringe filter prior to use . A modified kinase buffer ( 50 mM potassium phosphate , 50 mM KCl , 5 mM MgCl2 , pH 7 . 5 ) was used for sample preparation . Formation and characterization of kinase-active ternary complexes followed published methods ( Fowler et al . , 2010; Mudiyanselage et al . , 2013 ) , further specified as follows . PEG-mediated CF4Q complexes were prepared by incubating 50 μM CF4Q , 20 μM CheW , and 12 μM CheA with final concentrations of 7 . 5% wt/vol PEG 8000 and 4% wt/vol trehalose . CF4Q was added last to minimize CF-promoted aggregation ( Montefusco et al . , 2007 ) and samples were incubated overnight at 25°C in a circulating water bath and subjected to an enzyme-coupled assay and gel-based cosedimentation assay to check for phosphorylation activity and ternary complex formation . Kinase activity of PEG-mediated CF4Q complexes was similar to that observed for complexes assembled via binding to vesicles .	Many bacteria swim through water by rotating tiny hair-like structures called flagella . In E . coli , if all the flagella on the surface of a bacterium rotate in a counterclockwise fashion , then it will swim in a particular direction , but if the flagella all rotate in an clockwise fashion , then the bacterium will stop swimming and start to tumble . Bacteria use a combination of swimming and tumbling in order to move towards or away from certain chemicals . For example , a bacterium is able to move towards a source of nutrients because it is constantly evaluating its environment and will swim forward for longer periods of time when it recognizes the concentration of the nutrient is increasing . And if it senses that the nutrient concentration is decreasing , it will tumble in an effort to move in a different direction . Many bacteria , such as E . coli , rely on proteins in their cell membrane called chemoreceptors to sense specific chemicals and then send signals that tell the flagella how to rotate . These transmembrane receptors and their role in chemotaxis—that is , movement towards or away from specific chemicals in the environment—have been widely studied . However , other bacteria also have chemoreceptors in the cytoplasm inside the bacterial cell , and much less is known about these . Now , Briegel et al . have examined the cytoplasmic chemoreceptors of two unrelated bacteria , R . sphaeroides and V . cholera , and found that the cytoplasmic chemoreceptors arrange themselves in hexagonal arrays , similar to the way that transmembrane chemoreceptors are arranged . However , the cytoplasmic chemoreceptors arrange themselves in a two-layer sandwich-like structure , whereas the transmembrane chemoreceptors are arranged in just one layer . The next step is to understand how chemical binding causes these arrays to send their signals to the motor . A complete understanding of this signaling system may ultimately allow scientists to re-engineer it to draw bacteria to targets of medical or environmental interest , such as cancer cells or contaminated soils .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2014	Structure of bacterial cytoplasmic chemoreceptor arrays and implications for chemotactic signaling
Plexins are cell surface receptors that bind semaphorins and transduce signals for regulating neuronal axon guidance and other processes . Plexin signaling depends on their cytoplasmic GTPase activating protein ( GAP ) domain , which specifically inactivates the Ras homolog Rap through an ill-defined non-canonical catalytic mechanism . The plexin GAP is activated by semaphorin-induced dimerization , the structural basis for which remained unknown . Here we present the crystal structures of the active dimer of zebrafish PlexinC1 cytoplasmic region in the apo state and in complex with Rap . The structures show that the dimerization induces a large-scale conformational change in plexin , which opens the GAP active site to allow Rap binding . Plexin stabilizes the switch II region of Rap in an unprecedented conformation , bringing Gln63 in Rap into the active site for catalyzing GTP hydrolysis . The structures also explain the unique Rap-specificity of plexins . Mutational analyses support that these mechanisms underlie plexin activation and signaling . Plexins are a large group of type I transmembrane proteins that serve as the major receptors for semaphorins ( Yazdani and Terman , 2006; Tran et al . , 2007 ) . Plexin-mediated semaphorin signaling controls neuronal axon guidance as well other essential processes such as angiogenesis and immune responses ( Sakurai et al . , 2012; Takamatsu and Kumanogoh , 2012 ) . Aberrant plexin/semaphorin signaling has been implicated in numerous pathologies including neurological disorders and cancer ( Yaron and Zheng , 2007; Tamagnone , 2012; Gu and Giraudo , 2013 ) . Plexins all possess a large multi-domain extracellular region , a single transmembrane helix and a multi-domain cytoplasmic region . Binding of semaphorin to the extracellular region of plexin triggers activation of the cytoplasmic region , which relays the signal to downstream pathways . The plexin cytoplasmic region contains a juxtamembrane segment , a RhoGTPase binding domain ( RBD ) and a GTPase activating protein ( GAP ) domain ( Rohm et al . , 2000; Hu et al . , 2001; He et al . , 2009; Tong et al . , 2009; Bell et al . , 2011 ) . The juxtamembrane segment has been suggested to regulate plexin signaling by interacting with the GAP domain or mediating oligomerization ( He et al . , 2009; Bell et al . , 2011 ) . Binding of RhoGTPases such as Rac1 and RND1 to the RBD facilitates plexin activation ( Vikis et al . , 2000; Driessens et al . , 2001; Zanata et al . , 2002; Turner et al . , 2004; Tong et al . , 2007 ) , the mechanism of which is not well understood ( Bell et al . , 2011; Wang et al . , 2012 ) . The GAP domain in plexin shows structural homology to RasGAPs such as p120GAP , and contains a functionally essential arginine residue corresponding to the catalytic ‘arginine finger’ in RasGAPs ( Rohm et al . , 2000; Oinuma et al . , 2004; He et al . , 2009 ) . Plexins have been reported previously to be GAPs for the Ras homologs R-Ras and M-Ras ( Oinuma et al . , 2004; Saito et al . , 2009 ) . Our recent study , however , has demonstrated that the plexin GAP does not act directly on R-Ras or M-Ras ( Wang et al . , 2012 ) . Instead it is active specifically to the Ras homolog Rap , and this RapGAP activity is critical for plexin signaling . GTP-bound active Rap is a key activator of integrin for promoting cell–matrix adhesion ( Gloerich and Bos , 2011 ) . Conversion of Rap into the GDP-bound inactive form by the plexin GAP likely contributes to plexin-mediated repulsive axon guidance and other cell morphological changes through causing inactivation of integrin and weakening cell–matrix adhesion ( Wang et al . , 2012 ) . RasGAPs such as p120GAP and neurofibromin facilitate GTP hydrolysis of Ras , R-Ras and M-Ras by providing the conserved arginine finger to stabilize the leaving γ-phosphate group ( Li et al . , 1997; Scheffzek et al . , 1997; Scheffzek et al . , 1998; Quilliam et al . , 1999; Ohba et al . , 2000; Bos et al . , 2007 ) . Concomitantly , a conserved glutamine in the GTPases ( Gln61 in Ras ) coordinates the nucleophilic water for hydrolysis . Many other GAP/small GTPase pairs use similar mechanisms to catalyze GTP hydrolysis ( Bos et al . , 2007 ) . Rap is distinct from Ras/R-Ras/M-Ras in that it has a threonine at position 61 , which lacks the ability to coordinate the catalytic water . Canonical RapGAPs are structurally unrelated to RasGAPs and catalyze Rap GTP hydrolysis by providing an asparagine residue ( referred to as the ‘Asn thumb’ ) to fulfill the water coordination role of Gln61 in Ras ( Scrima et al . , 2008 ) . SynGAP ( Synaptic GAP ) , and three GAP1 family members Rasal ( Ras-GTPase-activating-like protein ) , CAPRI ( Ca2+-promoted Ras inactivator ) and GAP1IP4BP ( tetrakisphosphate binding protein ) are dual-specificity GAPs , active to both Ras and Rap . Plexins and these dual-specificity GAPs share the RasGAP fold that contains the arginine finger but lack a conserved Asn thumb ( Kupzig et al . , 2006; Pena et al . , 2008 ) . They facilitate GTP hydrolysis for Rap through a distinct , poorly understood mechanism . A recent study has suggested that Gln63 in Rap plays a role analogous to Gln61 in Ras in the non-canonical catalysis of the dual-specificity GAPs ( Sot et al . , 2010 ) . Mutating Gln63 in Rap abolishes GTP hydrolysis catalyzed by both the dual-specificity GAPs and plexins ( Sot et al . , 2010; Wang et al . , 2012 ) . Plexin signaling is critically dependent on the on/off switch of the RapGAP activity under the control of semaphorin ( Wang et al . , 2012 ) . Our previous structural analyses have suggested that the plexin GAP is autoinhibited by adopting a closed conformation that sequesters the active site ( He et al . , 2009 ) . A pre-formed inhibitory dimer of plexin may also be involved in suppressing the GAP activity prior to semaphorin binding ( Antipenko et al . , 2003; Tong et al . , 2007; Nogi et al . , 2010 ) . Semaphorins are dimeric molecules and have been suggested to induce dimerization or oligomerization of plexin for triggering downstream signaling ( Klostermann et al . , 1998; Koppel and Raper , 1998; Driessens et al . , 2001; Perrot et al . , 2002; Antipenko et al . , 2003; Love et al . , 2003 ) . A model of plexin activation involving oligomerization mediated by the RBD/RhoGTPase interaction has been proposed ( Bell et al . , 2011 ) , but existence of this oligomeric structure in solution or on the cell surface has not been established ( Siebold and Jones , 2013 ) . Recent structural studies have demonstrated how dimeric semaphorin brings two copies of the plexin extracellular region into proximity ( Janssen et al . , 2010; Liu et al . , 2010; Nogi et al . , 2010; Janssen et al . , 2012 ) . We have shown that the purified plexin cytoplasmic region displays low RapGAP activity , which can be activated dramatically by fusing it to the coiled-coil dimerization motif of GCN4 ( general control non-repressed 4 ) ( Wang et al . , 2012 ) . These observations collectively support that semaphorin-induced formation of an active dimer of plexin is the major mechanism for activation of the GAP domain and intracellular signaling . In this study we sought to understand how the plexin RapGAP is activated by induced-dimerization and facilitates GTP hydrolysis specifically for Rap . We systematically screened various coiled-coil dimer fusions of the plexin cytoplasmic region for optimal activation of the GAP . These experiments led to crystallization and structure determination of the active dimer of zebrafish PlexinC1 . In addition , we employed a novel protein ligation system to covalently link the plexin cytoplasmic region and Rap , which stabilized their weak interaction and allowed us to crystallize and determine the structure of a PlexinC1/Rap complex . The structures and the associated mutational analyses together reveal the basis for the dimerization-induced activation , the non-canonical catalysis and the unique specificity of plexin for Rap . Our previous study has shown that the RapGAP activity of the cytoplasmic region of plexins ( plexinscyto ) can be activated by fusing it to the coiled-coil motif of GCN4 through a flexible linker of various lengths ( Wang et al . , 2012 ) . Our extensive crystallization trials of these coiled-coil induced dimers of plexinscyto all failed , presumably due to the flexibility of the linker . We therefore removed the linker and directly fused the coiled-coil with the juxtamembrane helix ( the N-terminal helix in the juxtamembrane segment ) of plexincyto ( Figure 1 ) . Assuming the coiled-coil motif and the juxtamembrane helix of plexin merge into a continuous helix , varying the relative register between them by adding or removing residues at the junction can result in dramatically different relative orientations between the two plexin monomers in the dimer . Without knowing the ideal arrangement of the two monomers for active dimer formation , we systematically tested fusing plexincyto to each of the seven unique positions on the heptad repeat of the coiled-coil ( Figure 1A ) . We used mouse PlexinA1cyto for the screening experiments , because it displayed the highest level of activation by induced dimerization in our previous study ( Wang et al . , 2012 ) . We chose Ala1272 , located near the N-terminus of the juxtamembrane helix in PlexinA1cyto , as the reference for naming the fusion constructs . These constructs are referred to as CC ( x ) PlexinA1cyto , in which ‘x’ indicates the position of Ala1272 in PlexinA1 on the heptad repeat ( Figure 1A ) . 10 . 7554/eLife . 01279 . 003Figure 1 . Activation of the plexin GAP by coiled-coil fusion . ( A ) Design of the coiled-coil fusions of mouse PlexinA1cyto and zebrafish PlexinC1cyto . The juxtamembrane segment sequences from mouse PlexinA1cyto and zebrafish PlexinC1cyto are aligned . The constructs are named CC ( x ) Plexincyto , where ‘x’ ( in red ) is the position of Ala1272 in PlexinA1 or Gln553 in PlexinC1 on the heptad repeat . The ‘a’ and ‘d’ positions in the GCN4 coiled-coil are highlighted gray . Residues at the active dimer interface are highlighted pink . ( B ) Diagram of the CC ( x ) Plexincyto constructs . ( C ) GAP activity of mouse CC ( x ) PlexinA1cyto . Activity of monomeric PlexinA1cyto is too low to be measured reliably . The fold increase of kcat/KM is calculated relative to CC ( c ) PlexinA1cyto , which is the least active among the dimers but approximately 10-fold more active than the monomer . ( D ) GAP activity of zebrafish CC ( x ) PlexinC1cyto . In both ( C ) and ( D ) , error bars represent standard error of the kcat/KM . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 003 GAP activity assays showed that all these dimer constructs are substantially more active than the monomer ( Figure 1C ) . Remarkably , CC ( a ) PlexinA1cyto , CC ( d ) PlexinA1cyto and CC ( g ) PlexinA1cyto , which confer in general similar inter-monomer orientations , achieve much higher activation levels than CC ( b ) PlexinA1cyto , CC ( c ) PlexinA1cyto , CC ( e ) PlexinA1cyto and CC ( f ) PlexinA1cyto . We also tested four zebrafish CC ( x ) PlexinC1cyto constructs , which showed the same trend of activation levels ( Figure 1A , D ) . These results further support the notion that a specific association mode between the two plexin monomers is required for the optimal dimerization-induced activation ( Wang et al . , 2012 ) . We screened for crystals of those highly active dimer constructs , and obtained crystals of zebrafish CC ( a ) PlexinC1cyto and determined the structure at 3 . 3 Å resolution ( Table 1 ) . 10 . 7554/eLife . 01279 . 004Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 004Data collectionCrystalCC ( a ) PlexinC1cytoPlexinC1cyto/Rap1BSpace groupP212121P1Cell dimensions a , b , c ( Å ) 53 . 22 , 146 . 10 , 209 . 5876 . 28 , 84 . 73 , 138 . 75 α , β , γ ( ° ) 90 , 90 , 9091 . 09 , 95 . 15 , 90 . 32Resolution ( Å ) 50 . 0–3 . 30 ( 3 . 36–3 . 30 ) 50 . 0–3 . 30 ( 3 . 36–3 . 30 ) Rsym11 . 1 ( 86 . 8 ) 5 . 2 ( 48 . 7 ) I/σ19 . 7 ( 1 . 4 ) 18 . 6 ( 1 . 6 ) Completeness ( % ) 95 . 8 ( 79 . 5 ) 91 . 0 ( 89 . 2 ) Redundancy10 . 4 ( 4 . 3 ) 1 . 9 ( 1 . 9 ) RefinementResolution ( Å ) 3 . 303 . 30No . reflections21 , 08747 , 207Completeness ( % ) 83 . 32†90 . 23Rwork/Rfree ( % ) 22 . 6/28 . 224 . 3/30 . 0No . atoms888822 , 228 Protein887122 , 086 Ligand/ion0132 Water1710B-factors Protein98 . 9143 . 5 Ligand/ion–128 . 2 Water49 . 389 . 7R . m . s deviations Bond lengths ( Å ) 0 . 0050 . 004 Bond angles ( ° ) 0 . 850 . 70Ramanchandran plot Favored ( % ) 91 . 793 . 1 Allowed ( % ) 8 . 16 . 7 Disallowed ( % ) 0 . 20 . 2Highest resolution shell is shown in parenthesis . †The data were corrected for anisotropy in HKL2000 . This treatment eliminated many weak reflections and reduced the completeness of the data used for refinement compared to the completeness reported for data collection . Our attempts to co-crystallize Rap with various plexinscyto also failed , likely due to their weak interaction ( Wang et al . , 2012 ) . To stabilize the interaction , we covalently linked plexinscyto and Rap1B in vitro by using a protein ligation system based on the transpeptidase activity of sortase from Staphylococcus aureus ( Figure 2; see details in ‘Materials and methods’ ) ( Popp et al . , 2009 ) . We used the GAP activity assay to characterize the ligated complex of zebrafish PlexinC1cyto and human Rap1B connected by a 24-residue flexible linker and the sortase-recognition motif . The ligated complex catalyzes GTP hydrolysis at much higher rates than the two individual proteins mixed at the same concentrations ( Figure 2C ) , indicating enhanced formation of the catalytically competent plexin/Rap complex when the two proteins are tethered . 10 . 7554/eLife . 01279 . 005Figure 2 . Sortase-mediated ligation and characterization of the ligated PlexinC1cyto/Rap1B complex . ( A ) Scheme of the sortase-mediated ligation . ( B ) Representative gels of purified PlexinC1cyto , Rap1B and the ligated PlexinC1cyto/Rap1B complex with the 24-residue linker and the ‘LPETGG’ sortase recognition motif . ( C ) Comparison of the GTP hydrolysis activity between the ligated complex and the individual PlexinC1cyto and Rap1B proteins mixed at the same concentrations . The hydrolysis rates are averages of three replicates . Error bars represent standard deviation of the mean . ( D ) Analytical ultracentrifugation showing AlFx-induced dimerization of the ligated PlexinC1cyto/Rap1B complex . In the absence of AlFx ( the left panel ) , the majority of the complex behaves as a monomer with a sedimentation coefficient of 4 . 5 S . In the presence of AlFx ( the right panel ) , a dimeric species ( sedimentation coefficient of 6 . 7 S ) appears and becomes more abundant at higher protein concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 005 As mentioned above , the dimerization induces the active conformation of plexincyto , which enhances Rap binding and GTP hydrolysis . Conversely , stabilization of the active conformation of plexin by Rap binding is expected to facilitate formation of the plexin dimer . Due to basal GAP activity of plexincyto , Rap in the ligated plexincyto/Rap complex is GDP-bound and cannot stably bind or induce dimerization of plexincyto . We therefore used the γ-phosphate analog aluminum fluoride ( AlFx , x = 3 or 4 ) ( Vetter and Wittinghofer , 2001 ) to induce formation of the transition state complex between Rap ( GDP ) and plexin . Our analytical ultracentrifugation experiments showed that while the ligated PlexinC1cyto/Rap1B complex itself did not dimerize , it dimerized robustly in the presence of AlFx ( Figure 2D ) . We crystallized this complex with AlFx and determined the structure to 3 . 3 Å resolution ( Table 1 ) . In the CC ( a ) PlexinC1cyto structure , the two plexin monomers in the asymmetric unit form a symmetric side-by-side dimer ( Figure 3A ) . The two juxtamembrane helices are oriented approximately in parallel , extending well beyond the main body of the proteins and integrating into the C-termini of the coiled-coil moiety . On the plasma membrane , this configuration of the plexin dimer orients the active sites of the two GAP domains toward the membrane surface and leaves sufficient space for binding of the membrane anchored Rap substrate , as observed in the PlexinC1cyto/Rap1B complex structure ( Figure 3A , B ) . 10 . 7554/eLife . 01279 . 006Figure 3 . Overall structures of the zebrafish CC ( a ) PlexinC1cyto active dimer and the PlexinC1cyto/Rap1B complex . ( A ) Structure of the CC ( a ) PlexinC1cyto dimer . ( B ) Structure of the PlexinC1cyto/Rap1B complex . One of the two active dimers of plexin with Rap1B bound in the asymmetric unit is shown . In both ( A ) and ( B ) , domains from one plexin monomer in the dimer are colored and labeled . The other monomer is shown in white in ( A ) and gray in ( B ) . ( C ) Comparison of the active dimers in the structures of CC ( a ) PlexinC1cyto and the PlexinC1cyto/Rap1B complex . The coiled-coil moiety is omitted for clarity . The color schemes are the same as in ( A ) and ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 006 The asymmetric unit of the PlexinC1cyto/Rap1B complex structure contains four protomers of the complex , which are virtually identical to one another . The four PlexinC1 molecules form two pairs of dimers , consistent with the dimerization observed in solution . The conformation of PlexinC1 and its mode of dimerization are highly similar in the two structures ( Figure 3C ) , supporting that they represent the active state of plexin and are not artifacts induced by the fusion constructs . The coiled-coil moiety in the CC ( a ) PlexinC1cyto structure is nearly identical to the isolated coiled-coil structures reported previously ( O’Shea et al . , 1991 ) . Comparison of the active dimers in the two structures suggests that there is a small geometric incompatibility between the coiled-coil and the plexin dimer , as the N-terminal portion of the juxtamembrane helix ( residues 553–566 ) seems to bend slightly near its junction with the coiled-coil ( Figure 3C ) . This portion of the juxtamembrane helix does not mediate any inter-molecular interactions and likely has some flexibility . The flexibility can further compensate for deletion or insertion of one residue at the junction between the coiled-coil and the juxtamembrane helix , allowing several constructs to induce the active dimer and achieve similarly high activation levels ( Figure 1 ) . More deletions or insertions at the junction probably cannot be accommodated without severe distortion of the juxtamembrane helix , explaining the much lower activation levels of those constructs ( Figure 1 ) . We will refer to the CC ( a ) PlexinC1cyto structure for the following discussion on the active dimer unless otherwise stated , because the dimer interface in this structure is better resolved in the electron density map . The dimer interface is formed by the juxtamembrane helix and one side of the GAP domain , burying a total of ∼3200 Å2 surface area ( Figure 4 ) . The RBDs in the two monomers are far away from each other and not involved in dimer formation . The center of the dimer interface is a 4-helix bundle structure comprised of the C-terminal portion of the juxtamembrane helix ( residues 567–584 ) and the N-terminal portion of helix 11 in the GAP domain ( residues 929–943 ) from each monomer ( Figure 4 ) . The core of the 4-helix bundle is dominated by hydrophobic interactions , involving residues Ile571 , Ile575 , Phe579 and Leu582 from the juxtamembrane helix and Met933 , Ile936 and Leu939 from helix 11 ( Figure 4B ) . The core interface is supported by peripheral electrostatic interactions mediated by Arg572 , Arg576 and Asp581 from the juxtamembrane helix and Glu770 , Glu932 , and Lys937 from the GAP domain ( Figure 4A ) . 10 . 7554/eLife . 01279 . 007Figure 4 . The dimer interface in the CC ( a ) PlexinC1cyto structure . ( A ) Periphery of the dimer interface . The coiled-coil moiety is not shown . ( B ) Hydrophobic core of the dimer interface . Residue labels for one monomer are omitted for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 007 A loop-helix segment ( residues 1038–1058 ) between helix 15 and 17 in the GAP domain of each monomer wraps around the C-terminal portion of the 4-helix bundle . The interactions involve Leu1045 , Lys1047 , Leu1054 , Leu1055 , and Lys1058 in the loop-helix segment and Phe579 , Gln583 , Thr584 and Leu939 from the 4-helix bundle ( Figure 4A ) . We call the loop-helix element ‘the activation segment’ in plexin since it plays a major role in regulating the GAP activity ( see the next section for details ) , functionally resembling the well known activation segment in protein kinases ( Huse and Kuriyan , 2002 ) . Comparisons of the dimer structure with previously determined structures of plexinscyto reveal several substantial conformational differences . The most striking difference is in the juxtamembrane helix ( Figure 5A ) . Except in one of the PlexinB1 structures where it is disordered ( Bell et al . , 2011 ) , the juxtamembrane helix in all other previous structures adopts a kinked conformation , with both the N- and C-terminal halves interacting with the GAP domain ( He et al . , 2009; Tong et al . , 2009; Wang et al . , 2012 ) . In the active dimer structure , the last two turns in the juxtamembrane helix ( residues 585–591 , corresponding to residues 1282–1288 in mouse PlexinA3 ) convert to an extended loop . This loop and the following segment use Asp588 , Leu589 , Asp591 and Val593 to make a distinct set of intra-molecular interactions with the GAP domain ( Figure 5B ) . The remaining N-terminal helical portion ( residues 553–584 ) adopts a straight conformation and rotates by ∼90° in relation to the inactive structures ( Figure 5A ) to mediate the formation of the 4-helix bundle at the center of the dimer interface ( Figure 4B ) . Helix 11 undergoes a small tilt to accommodate the juxtamembrane helix from the dimer partner , and the top part ( residues 929–934 ) adopts a 310 helix like conformation to pack against the hydrophobic core of the 4-helix bundle ( Figure 5C ) . 10 . 7554/eLife . 01279 . 008Figure 5 . Dimerization-induced conformational changes of the juxtamembrane helix and helix 11 . ( A ) Conformational change of the juxtamembrane helix . One monomer in the PlexinC1cyto active dimer is superimposed onto the monomeric PlexinA3cyto structure ( PDB code: 3IG3 ) . The GAP domain and RBD of PlexinA3cyto are shown in the surface representation . ( B ) Intra-molecular interactions made by the extended portion of the juxtamembrane segment in the CC ( a ) PlexinC1cyto structure . ( C ) Conformational change of helix 11 . The structure superimposition is the same as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 008 The conformational changes in the juxtamembrane helix and helix 11 are coupled to changes in the activation segment . In all the previously reported structures of plexinscyto , the highly conserved helical portion of the activation segment adopts essentially the same ‘closed’ conformation ( Figure 6A ) ( He et al . , 2009; Tong et al . , 2009; Bell et al . , 2011; Wang et al . , 2012 ) . An asparagine residue in the helix ( Asn1774 in mouse PlexinA3 ) is invariably hydrogen bonded with a conserved aspartate ( Asp1758 in PlexinA3 ) in helix 15 . A proline residue ( Pro1772 in PlexinA3 ) at the N-terminus of the helix acts as a lid that covers the asparagine and blocks its access to the incoming Rap substrate . Docking Rap to PlexinA3 based on the PlexinC1/Rap complex structure results in a number of clashes between Rap and the activation segment ( Figure 6A ) . The proline ‘lid’ ( Pro1772 ) sterically clashes with Tyr40 in Rap , while the carbonyl oxygen on the sidechain of Asn1774 makes an unfavorable contact with the sidechain of Asp38 in Rap . The loop portion of the activation segment appears to be rather flexible , as it displays high B-factors in PlexinA3 ( PDB ID: 3IG3 ) and the PlexinB1/Rac1 complex ( PDB ID: 3SU8 ) and is partially disordered in apo-PlexinB1 ( PDB ID: 3HM6 ) and the PlexinA1/Rac1 complex ( PDB ID: 3RYT ) . The loop likely samples many conformations , some of which may impose additional hindrance on Rap binding . 10 . 7554/eLife . 01279 . 009Figure 6 . Dimerization-induced opening of the activation segment . ( A ) Docking of Rap to the inactive PlexinA3cyto structure ( PDB code: 3IG3 ) . The docking is based on a superimposition between PlexinA3 and PlexinC1 in the PlexinC1/Rap complex structure ( see ‘Materials and methods’ for details ) . Red dashed line: hydrogen bond . Red arrows: steric clashes and unfavorable interactions . ( B ) Sigma-A weighted simulated annealing omit map of the activation segment in CC ( a ) PlexinC1cyto . The map was calculated using the model with residues 1050–1056 in one of PlexinC1 molecules removed . The map was contoured at 3σ , with the final model shown . ( C ) Comparison of the activation segment in the structures of CC ( a ) PlexinC1cyto and PlexinA3cyto . Conformational differences important for GAP activation are highlighted by black arrows . ( D ) Comparison of the activation segment in the structures of CC ( a ) PlexinC1cyto and the PlexinC1cyto/Rap complex . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 009 In contrast , the activation segment in the active dimer adopts an open conformation and shifts away from the GAP active site ( Figure 6B , C ) . This shift appears to be induced by the interactions between the activation segment and the 4-helix bundle in the dimer interface ( Figure 6C ) . The outward shift pulls Asn1052 ( Asn1774 in PlexinA3 ) away from Asp1036 ( Asp1758 in PlexinA3 ) , precluding hydrogen bond formation . Pro1050 in the dimer structure also moves outward compared to Pro1772 in PlexinA3 ( Figure 6C ) . The activation segment in the structure of the PlexinC1/Rap complex adopts a similar open conformation ( Figure 6D ) . Therefore , a major mechanism in the dimerization-induced activation of plexin appears to be the outward shift of the activation segment , which opens the otherwise obstructed active site to allow Rap binding and catalysis of GTP hydrolysis . While this conformational change in the plexin GAP domain seems small , it is known that interactions between small GTPases and their regulators or effectors can be strongly influenced by subtle changes at the binding interface ( Nassar et al . , 1996; Snyder et al . , 2002 ) . The activation segment in the PlexinC1/Rap complex is slightly more closed than that in the coiled-coil-induced PlexinC1 dimer ( Figure 6D ) , indicating that the active dimer promotes a conformation that is more open than required for accommodating Rap . Binding of Rap induces a slight closure of the active site for optimal interactions and catalysis of GTP hydrolysis . The RBD and the subdomain composed of the first three and the last two helices in the GAP domain show conformational variations among all the structures of plexins . Given the fact that they are not involved in the dimer interface or Rap binding , the variations of these structural elements likely reflect their intrinsic flexibility and are not relevant to the activation mechanism . We performed extensive mutational analyses to test the activation mechanism revealed by the dimer structure . Arg576 , Asp581 , Asp588 , Val593 and Met933 are involved in the dimer interface or intra-molecular interactions that stabilize the new conformation of the juxtamembrane segment ( Figures 4 and 5B ) . In the inactive monomer structures , residues at these positions are surface exposed and do not make any interactions . We made the R576E , D581K , D588K , V593E and M933E single mutations in CC ( a ) PlexinC1cyto . The GAP assay showed that R576E , D581K , D588K and M933E strongly impaired dimerization-induced activation ( Figure 7A ) . The deleterious effect of V593E on GAP activation is weaker but clearly observable at a lower plexin concentration ( Figure 7A , C ) . To test the coupling between the dimerization and the conformation of the activation segment , we designed the Q583A , T584A , L1045A , K1047A and L1054A mutations to disrupt the interactions between the activation segment and the juxtamembrane helix from the dimer partner ( Figure 6C ) . The GAP assay showed that while K1047A modestly decreased dimerization-induced GAP activation , L1054A , L1045A and T584A greatly reduced the activation ( Figure 7B , C ) . 10 . 7554/eLife . 01279 . 010Figure 7 . Mutational analyses of the dimerization-driven activation mechanism . ( A–C ) Mutational analyses of the activation mechanism using the GAP activity assay . Residues mutated in ( A ) are involved in stabilizing the active dimer , whereas residues in ( B ) couple the dimer formation to the opening of the activation segment . In ( A ) and ( B ) the concentration of plexin is 2 µM . In ( C ) , the concentration of plexin is 0 . 25 µM . The Rap concentration is 120 µM for all the assays . Data shown are representative of three replicates . ( D ) Mutational analyses using the COS7 cell collapse assay . The results for the wild type and the arginine-finger mutant ( R1407/1408A ) are shown as positive and negative controls , respectively . Error bars represent standard error of the mean from three independent experiments . At least 150 cells were counted for each sample in each experiment . Statistical significance between wild type and each mutant is determined by two-tailed Student’s t-test ( p<0 . 05; p<0 . 01; p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 010 We further examined the activation mechanism by using a functional assay , which assesses the ability of plexin to induce COS7 cell collapse upon semaphorin stimulation ( Takahashi et al . , 1999 ) . Since the ligand for zebrafish PlexinC1 was not available , mouse PlexinA3 and its ligand Sema3F were used in these assays ( He et al . , 2009 ) . The K1273E , E1278R , E1285R and M1290E mutations of mouse PlexinA3 , corresponding to R576E , D581K , D588K and V593E of zebrafish PlexinC1 respectively , all significantly impaired plexin-mediated COS7 cell collapse ( Figure 7D ) . A previous study identified a large panel of mutations that abolished PlexinB1-mediated COS7 cell collapse ( Bell et al . , 2011 ) . These mutations were designed to test the model of plexin activation by Rac1-induced oligomerization . The results are also consistent with the activation mechanism shown here , as most of the mutated resides are conserved in zebrafish PlexinC1 and are involved in formation of the active dimer . Some mutations of highly conserved residues in the dimer interface have been identified in cancer patients , including R2040W in PlexinB1 ( Gui et al . , 2011 ) ( corresponding to Lys1058 in zebrafish PlexinC1 ) and R1680Q/W in PlexinA2 ( Cancer Genome Atlas Research Network , 2012 ) ( corresponding to Lys937 in zebrafish PlexinC1 ) . Both of these mutations likely prevent formation of the active dimer of plexins , consistent with the tumor suppressor function of plexins suggested by previous studies ( Gu and Giraudo , 2013 ) . The species mismatch of the plexin/Rap complex does not affect their interaction , since the human Rap1B construct contains only three residues non-identical to their counterparts in zebrafish Rap1B , which are all located far from the plexin/Rap interface ( Figure 8A , middle panel ) . The linker between PlexinC1cyto and Rap1B is not visible in the electron density map , suggesting that it is flexible as designed and does not impose restraints on the plexin/Rap interaction . A superimposition of Rap and Ras in the PlexinC1cyto/Rap and p120GAP/Ras complexes shows that the overall binding modes of the two with their respective GAPs are similar ( Figure 8A , left panel ) ( Scheffzek et al . , 1997 ) . The GAP domain in PlexinC1 and Switches I ( residues 30–38 ) and II ( residues 59–67 ) in Rap constitute the majority of the binding interface , whereas the RBD in plexin is not involved and its role in GAP regulation remains unclear ( Figure 8 ) . The core of the interface is composed of several hydrophobic residues , which are surrounded by numerous charge–charge interactions at the periphery . Most of the Rap-binding residues are conserved among the plexin family members , suggesting that they all interact with Rap in the same mode ( Figure 8B , C ) . The presumed arginine finger ( Arg711 ) in PlexinC1 superimposes well with the arginine finger ( Arg789 ) in p120GAP , playing the same role in catalysis by interacting with the AlFx and GDP in the active site ( Figure 8A ) . While the bound AlFx is not clearly resolved in the relatively low-resolution map , the shape of the density suggests that it is the trigonal AlF3 , the same as in the p120GAP/Ras structure . We therefore modeled it AlF3 in the structure . The second conserved arginine ( Arg1001 ) in PlexinC1 is equivalent to Arg903 in p120GAP , which stabilizes the position of the arginine finger ( Figure 8A ) . The functional importance of these two arginine residues in plexin has been demonstrated by previous mutational studies ( Rohm et al . , 2000; Oinuma et al . , 2004; He et al . , 2009; Wang et al . , 2012 ) . 10 . 7554/eLife . 01279 . 011Figure 8 . Overall view of the interface between zebrafish PlexinC1cyto and human Rap1B in the complex structure . ( A ) The PlexinC1cyto/Rap interface and its comparison with that in the p120GAP/Ras complex structure . The middle panel shows the overall structure of the PlexinC1cyto/Rap complex , with the three residues ( 48 , 105 and 140 ) different between human and zebrafish Rap1B highlighted . The left panel shows a superimposition of the active sites in the PlexinC1cyto/Rap1B and p120GAP/Ras ( PDB ID: 1WQ1 ) structures . The superimposition is based on Rap1B and Ras . The right panel shows a Sigma-A weighted simulated annealing omit map of Switch II in Rap , calculated using the model with residues 60–66 in one of the Rap1B molecules removed . The map is contoured at 3σ , with the final model of the structure shown . ( B ) Rap-binding surface on PlexinC1cyto . Residues in PlexinC1 within 4 Å distance of the bound Rap1B molecule are colored green . ( C ) Sequence conservation projection on the molecular surface of PlexinC1cyto . The conservation scores were calculated based on an alignment of zebrafish PlexinC1 and all the plexins from mouse ( Plexin A1 , A2 , A3 , A4 , B1 , B2 , B3 , C1 and D1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 011 Switch I of Rap makes numerous interactions with the activation segment in PlexinC1 ( Figure 9A ) . As mentioned above , the activation segment in the PlexinC1/Rap complex structure adopts the open conformation similar to that in the CC ( a ) PlexinC1 dimer structure . Pro1050 at the N-terminus of the helical portion of the activation segment packs against Tyr40 in Rap . Asn1052 forms two hydrogen bonds with the carboxyl group of Asp38 and the backbone amide of Ser39 in Rap . Lys1053 apparently makes electrostatic interactions with Asp38 in Rap and Asp1036 in plexin . Gln1032 in helix 15 also contributes to Switch I binding through forming three hydrogen bonds . GAP activity assays showed that while the P1050A mutation caused a modest activity decrease , the Q1032E , N1052E and K1053A mutations largely abolished the activity ( Figure 9B ) . COS7 cell collapse assays showed that both the Q1754E and K1775A mutations of mouse PlexinA3 , equivalent to zebrafish PlexinC1 Q1032E and K1053A respectively , greatly impaired the cell collapse activity ( Figure 9C ) . Mutations of Pro2032 in PlexinB1 and Lys1809 in PlexinB3 , equivalent to Pro1050 and Lys1053 in zebrafish PlexinC1 respectively , have also been found in cancer patients ( Cancer Genome Atlas Research Network , 2012; Seshagiri et al . , 2012 ) . 10 . 7554/eLife . 01279 . 012Figure 9 . Interaction between the activation segment in PlexinC1 and Switch I in Rap . ( A ) Interface between the activation segment and Switch I . Polar interactions and potential hydrogen bonds are indicated by dashed lines . ( B ) GAP activity assays for mutations at the activation segment/Switch I interface . Monomeric PlexinC1cyto was used in these assays . The plots are representatives of three replicates . ( C ) COS7 cell collapse assays for mutations at the activation segment/Switch I interface . Q1754E , and K1775A of mouse PlexinA3 correspond to Q1032E and K1053A of zebrafish PlexinC1 , respectively . The data analysis and presentation are the same as in Figure 7D . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 012 Switch II of Rap in the complex structure adopts an unprecedented conformation that is markedly different from both the p120GAP/Ras and the RapGAP/Rap complexes ( Figure 10 ) ( Scheffzek et al . , 1997; Scrima et al . , 2008 ) . Residues 60–63 in Switch II form a tight hairpin-like turn , which brings Gln63 close to AlF3 ( therefore named the Gln63-in conformation ) . The Gln63 sidechain is placed in a nearly identical position in the active site as Gln61 in the p120GAP/Ras complex ( Figure 10B ) . This comparison strongly supports that Rap Gln63 indeed fulfills the catalytic role of Gln61 in Ras , that is stabilizing the nucleophilic water ( Sot et al . , 2010; Wang et al . , 2012 ) . Consistently , mutation of Gln63 in Rap has been shown to abolish GTP hydrolysis catalyzed by both plexin and the dual-specificity GAPs ( Sot et al . , 2010; Wang et al . , 2012 ) . The segment following Gln63 ( residues 64–67 ) adopts an extended conformation , allowing it to span the distance between Gln63 in the active site and the helix following Switch II . In contrast , the corresponding segments in the p120GAP/Ras and the Rap/RapGAP complexes adopt helical structures , holding residue 63 away from the active site ( Figure 10B , C ) . 10 . 7554/eLife . 01279 . 013Figure 10 . The Gln63-in conformation of Switch II in the PlexinC1cyto/Rap1B complex . ( A ) Sequence alignment of human Rap1B , R-Ras , M-Ras and Ras . Black circles denote residues in Rap1B that are involved in binding PlexinC1cyto . Gln63 in Rap1B and Gln61 in Ras are highlighted by blue arrows . ( B ) Comparison of Switch II in the PlexinC1/Rap and the p120GAP/Ras ( PDB ID: 1WQ1 ) complexes . The nucleophilic H2O is not included in the PlexinC1cyto/Rap1B structure due to low resolution of the density map . ( C ) Comparison of Switch II in the PlexinC1/Rap and the RapGAP/Rap ( PDB ID: 3BRW ) complexes . ( D ) Specific interactions between PlexinC1 and Switch II in Rap1B . Polar interactions and potential hydrogen bonds are indicated by dashed lines . ( E ) Interaction between Pro611 in PlexinC1 and Thr65 in Rap1B . ( F ) GAP activity assays for mutations at the plexin/Switch II interface . Monomeric PlexinC1cyto was used in these assays . The plots are representatives of three replicates . ( G ) COS7 cell collapse assays for mutations at the plexin/Switch II interface . The data analysis and presentation are the same as in Figure 7D . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 013 The Gln63-in conformation of Switch II is stabilized by numerous specific interactions between PlexinC1 and Rap . The side chains of Arg1001 , Asn1005 and Asn1009 in helices 13 and 14 of PlexinC1 form a network of hydrogen bonds with the backbone of Switch II ( Figure 10D ) . Pro611 in the second helix of the juxtamembrane segment makes van der Waals interactions with Thr65 in Switch II ( Figure 10E ) . Mutation of either Asn1005 or Asn1009 dramatically decreased the GAP activity ( Figure 10F ) . The N1728E mutation in mouse PlexinA3 ( equivalent to N1005E of zebrafish PlexinC1 ) also abolished the cell collapse activity ( Figure 10G ) . Mutating Pro611 to glycine , which eliminates its interaction with Thr65 in Switch II , decreased the GAP activity ( Figure 10F ) . Conversely , the wild-type PlexinC1 showed decreased activity towards the Rap T65A mutant ( Figure 10F ) . The Switch II-interacting residues are highly conserved among the plexin family members , suggesting they all use the same mechanism to stabilize the Gln63-in conformation . The dual-specificity GAPs do not share some of the Switch II-interacting residues with plexin ( Figure 11 ) . For example , Asn1005 in PlexinC1 is replaced by a proline in the dual-specificity GAPs ( Pro585 in SynGAP ) ( Pena et al . , 2008 ) , lacking the ability to stabilize the Gln63-in conformation of Rap through hydrogen bonds . This loss may be compensated by the extra domains outside of the GAP domain in the dual-specificity GAPs , which have been shown to be required for their RapGAP activity but not for the RasGAP activity ( Kupzig et al . , 2006; Pena et al . , 2008; Kupzig et al . , 2009; Sot et al . , 2010 ) . It has been suggested that the extra domains contribute to the catalysis by stabilizing a certain conformation of Switch II ( Kupzig et al . , 2006; Kupzig et al . , 2009 ) . 10 . 7554/eLife . 01279 . 014Figure 11 . Comparison of the Switch II-interacting region between plexin , RasGAPs and dual-specific GAPs . ( A ) Packing interactions made by Phe64 in Rap1B with PlexinC1 and Tyr64 in Ras with p120GAP . The PlexinC1ctyo/Rap1B and the p120GAP/Ras structures are superimposed by using Rap1B and Ras as references . ( B ) Sequence alignment of the major Switch II-interacting segment in plexins , RasGAPs and dual-specificity GAPs . The black circles highlight the three residues ( Arg1001 , Asn1005 and Asn1009 ) in zebrafish PlexinC1 that make critical interactions with Switch II of Rap . zf: zebrafish; m: mouse; h: human; r: rat . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 014 RasGAPs such as p120GAP and neurofibromin also contain a proline at the position of Asn1005 in PlexinC1 . GAP1m , the only GAP1 family member that is active toward Ras but not Rap , has a valine at this position ( Figure 11B ) . Proline-to-valine mutants of the dual-specificity GAP1 family members ( Rasal , CAPRI and GAP1IP4BP ) remain active toward Ras , but lose activity toward Rap ( Kupzig et al . , 2009 ) . The superimposition of the p120GAP/Ras and PlexinC1/Rap structures suggests the basis for how this residue determines the substrate specificity of these GAPs ( Figure 11A ) . Pro907 in p120GAP contributes to Ras binding by stacking against Tyr64 in Switch II of Ras . A valine residue at the position of Pro907 ( Val515 in GAP1m in Figure 11B ) appears to be readily accommodated in this Ras binding mode ( Figure 11A ) . Assuming Rap adopts the same Gln63-in conformation when it binds the dual-specificity GAPs , a proline residue at this position in the GAPs is compatible with the interaction . However , the Gln63-in conformation of Rap places Phe64 much closer to the proline residue ( Figure 11A ) . Replacing the proline with a bulkier valine residue likely cause steric clashes with Phe64 in Rap , leading to loss of the RapGAP activity . In addition to Switch II , the PlexinC1/Rap interface involves several other residues in Rap that diverge from Ras/R-Ras/M-Ras . Residue 31 in Rap and Ras is a key residue for determining the binding specificity for downstream effectors of these two closely related small GTPases ( Nassar et al . , 1996 ) . Our PlexinC1cyto/Rap1B structure suggests that residue 31 is also a determinant for the specificity between the plexin GAP and Rap . Rap possesses a lysine at this position , which is replaced by a negatively charged residue ( aspartate or glutamate ) in Ras/R-Ras/M-Ras ( Figure 10A ) . Lys31 and Asp33 in Rap form a charge–charge pair and are buried by the activation segment in PlexinC1 ( Figure 12B ) . We made a Rap ( K31E ) mutant to render it more similar to Ras/R-Ras/M-Ras . This mutation is predicted to destabilize the PlexinC1/Rap interaction , since it closely places two buried negative charges . The GAP assay indeed showed that PlexinC1 failed to catalyze GTP hydrolysis for the K31E mutant ( Figure 12C ) . 10 . 7554/eLife . 01279 . 015Figure 12 . Additional specificity determinants in the PlexinC1/Rap1B complex . ( A ) Potential interaction between Lys666 in PlexinC1 and Asp95 in Rap . The side chain of Lys666 in PlexinC1 is not built in the final model due to weak electron density . It is modeled to show its potential interaction with Asp95 in Rap . ( B ) Burial of Lys31 in Rap1B at the PlexinC1/Rap1B interface . ( C ) GAP activity assays for the specificity determinants . Monomeric PlexinC1cyto was used in these assays . The plots are representatives of three replicates . ( D ) COS7 cell collapse assays for the R1360D mutant of mouse PlexinA3 ( equivalent to K666D of zebrafish PlexinC1 ) . The data analysis and presentation are the same as in Figure 7D . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 015 A potential salt-bridge between Asp95 in Rap1B and Lys666 in PlexinC1 may also contribute to their interaction and specificity ( Figure 12A ) . Consistent with this notion , Rap2 has a proline residue at position 95 and is less responsive to the plexin GAP ( Wang et al . , 2012 ) . The corresponding residues in Ras , R-Ras and M-Ras are glutamine , lysine and arginine respectively ( Figure 10A ) . Mutating Rap Asp95 to lysine , as in R-Ras , substantially decreased the rate of PlexinC1-catalyzed GTP hydrolysis ( Figure 12C ) . Likewise , PlexinC1 ( K666D ) displayed lower GAP activity than the wild-type PlexinC1 ( Figure 12C ) . The PlexinC1 ( K666D ) and Rap ( D95K ) charge-swapped pair only slightly restored the GTP hydrolysis activity ( Figure 12C ) , which may be due to disruption of the electrostatic complementarity at the plexin/Rap interface by the mutations . We also tested the importance of this interaction in the cell-based assay , which showed that the equivalent mutation of mouse PlexinA3 ( R1360D ) impaired the cell collapse activity ( Figure 12D ) . The same residue in human PlexinA1 ( Arg1384 ) has been found mutated to cysteine in cancer patients ( Seshagiri et al . , 2012 ) . These analyses together with the unique plexin/Switch II interface support the notion that plexins have evolved to recognize residues in Rap that have diverged from other Ras family members , leading to loss of activity toward Ras/R-Ras/M-Ras . This study together with the previous structures of the plexin extracellular regions establishes a framework for understanding plexin regulation ( Figure 13; Video 1 ) ( Janssen et al . , 2010; Liu et al . , 2010; Nogi et al . , 2010; Janssen et al . , 2012 ) . Semaphorin binding to the plexin extracellular region induces formation of the active dimer of the cytoplasmic region , which triggers its GAP activity to inactivate Rap through the non-canonical catalytic mechanism for signal transduction . Conformational changes similar to those undertaken by the plexin GAP domain upon the dimerization may serve as on/off switches for other related GAPs such as CAPRI , which is also activated by dimerization ( Dai et al . , 2011 ) . In addition to activation of the GAP , the dimerization-induced structural rearrangements may underlie the activation state-selective binding of plexins by signal transducers such as FARP2 ( FERM , RhoGEF and pleckstrin homology protein 2 ) and MICAL ( molecule interacting with CasL ) ( Toyofuku et al . , 2005; Schmidt et al . , 2008 ) . The structures of the several extracellular membrane-proximal domains and the transmembrane helix of plexins have not been determined . Our data suggest that , upon semaphorin-induced dimerization , these domains are arranged precisely to ensure the proper juxtaposition of the juxtamembrane helix for inducing the active dimer of the cytoplasmic domain ( Figure 13 ) . Future work on these domains in the active dimeric state will fill in the missing links , leading to a complete structural model of semaphorin-activated plexin . 10 . 7554/eLife . 01279 . 016Figure 13 . Schematic model for the activation of the plexin RapGAP by semaphorin-induced dimerization . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 01610 . 7554/eLife . 01279 . 017Video 1 . Dimerization-induced activation of plexincyto and binding of Rap to the GAP active site . The video is based on the crystal structures of inactive monomeric PlexinA3cyto ( PDB ID: 3IG3 ) , CC ( a ) PlexinC1cyto and the PlexinC1cyto/Rap1B complex . It is rendered for illustrating the dimerization-induced structural rearrangements and the binding mode between plexin and Rap . The actual order of the events and conformational transition trajectories likely do not follow those in the video . DOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 017 Clarifying the substrate specificity for the plexin GAP is essential for understanding plexin signaling . The results shown here and in our previous study ( Wang et al . , 2012 ) together demonstrate that while sharing the same domain fold with RasGAPs and dual-specificity GAPs , plexins are a unique group that are active to Rap , but not to Ras/R-Ras/M-Ras . Our analysis of the plexin/Rap complex structure reveals residues in both plexin and Rap that contribute to this specificity . P120GAP has been shown to bind GTP-bound Rap strongly but fail to catalyze its GTP hydrolysis , making Rap an effective inhibitor of the GAP activity of p120GAP to Ras/R-Ras/M-Ras ( Frech et al . , 1990; Hata et al . , 1990; Yatani et al . , 1991 ) . The apparent GAP activity of plexins towards R-Ras and M-Ras reported previously may be caused indirectly by inactivation of Rap and alleviation of its inhibition on p120GAP . The induced Gln63-in conformation of Rap seen in the PlexinC1/Rap complex structure likely represents the general mechanism by which plexins and the dual-specificity GAPs facilitate GTP hydrolysis for Rap . This conformation is stabilized by specific interactions made by several conserved residues in plexin . The dual-specificity GAPs achieve this through different mechanisms that likely involve the extra domains , the precise basis for which awaits structural studies of these GAPs in complex with Rap . The human Rap1B construct ( residues 2–167 ) in a modified pET28 vector ( Novagen , Darmstadt , Germany ) that encodes a N-terminal His6-tag and a recognition site for the human rhinovirus C3 protease has been described previously ( Wang et al . , 2012 ) . The Rap1B constructs ( 2–166 ) containing a C-terminal flexible linker followed by a sortase recognition motif ( one letter-code sequence: LPETGG ) were generated by PCR and subcloned into the same vector . Seven versions of the linker were generated: 0-residue ( containing the LPETGG motif only ) , 11-residue ( sequence: GGSGGSGSGSS ) , 14-residue ( sequence: SGGSGSGSSGGSGS ) , 16-residue ( sequence: GGSGGSGSGSSGGSGS ) , 21-residue ( sequence: GGSGGSGSGSSGGSGSGGGSG ) , 24-residue ( sequence: SGGSGSGSSGGSGSGGGSGSGSSG ) and 26-residue ( sequence: GGSGGSGSGSSGGSGSGGGSGSGSSG ) . The vector encodes a glycine residue at the second position from the N-terminus , which becomes the N-terminal residue after removal of the methionine residue encoded by the start codon during protein expression . An N-terminal glycine on the Rap1B protein would hinder the sortase-mediated ligation with Plexin ( see below ) ( Popp et al . , 2009 ) . To avoid this problem , the vector was mutated to replace the glycine residue with an aspartate using a Quickchange reaction ( Stratagene , La Jolla , CA ) . The Rap1B proteins were expressed in the bacteria strain BL21 ( DE3 ) and purified as described previously ( Wang et al . , 2012 ) . The coding region for the zebrafish PlexinC1cyto ( residues 552–1153 ) with a N-terminal di-glycine tag was synthesized ( GenScript , Piscataway , NJ ) based on the gene bank entry XM_685667 . 4 . The region encoding residues 552–1147 was subcloned into another modified pET28 vector containing a N-terminal tandem His6–SUMO tag ( Wang et al . , 2012 ) . The GCN4 coiled-coil motif was fused to the N-terminus of the PlexinC1cyto ( residues 553–1153 ) without the di-glycine motif by PCR . The fusion was subcloned into the same modified pET28 vector . Quikchange ( Stratagene ) was used to alter the residues at the junction between the coiled-coil and PlexinC1 . The coiled-coil fusion constructs of mouse PlexinA1cyto were cloned by using similar procedures . The protein was expressed in the bacteria strain ArcticExpress ( Stratagene ) and purified as described previously ( He et al . , 2009; Wang et al . , 2012 ) . The His6-SUMO-tag was removed by treatment with the SUMO-specific protease Ulp1 . For the construct containing the di-glycine encoding sequence , the Ulp1 treatment yielded the PlexinC1 protein with a N-terminal GG-tag . All mutants of Rap and plexins were generated by Quickchange reactions ( Stratagene ) , and expressed and purified as the respective wild-type proteins . Ligation of the N-terminal His6/C-terminal LPETGG-tagged Rap1B and the N-terminal GG-tagged PlexinC1 was catalyzed by the transpeptidase activity of sortase from Staphylococcus aureus ( plasmid provide by Dr Hidde Ploegh ) ( Popp et al . , 2009 ) . Sortase with a N-terminal His6-tag was expressed and purified by using Ni-NTA chromatography . Sortase first cleaves the peptide bond between the threonine and first glycine within the LPETGG motif in Rap1B . In the second step , the GG-tagged PlexinC1 is added to the threonine to regenerate a native peptide bond between the two proteins . The reaction mix contained Rap1B , PlexinC1 and sortase at 450 , 69 and 25 μM respectively . Reactions were performed at room temperature for 3 hr with simultaneous dialysis to remove the di-glycine by-product . The dialysis buffer contained 20 mM Tris pH 8 , 150 mM NaCl , 10% glycerol , 2 mM MgCl2 , 2 mM DTT , 10 mM CaCl2 . The ligated PlexinC1/Rap1B complex was purified by Ni-NTA , ion exchange and gel filtration chromatographic steps . The N-terminal His6-tag was removed by treatment with the human rhinovirus C3 protease . The GAP assay was performed by coupling release of inorganic phosphate during GTP hydrolysis to the purine nucleoside phosphorylase-catalyzed conversion of 2-amino-6-mercapto-7-methylpurine ribonucleoside to ribose-1-phosphate , which can be monitored photometrically at the wavelength of 360 nm ( Webb and Hunter , 1992 ) . For analyzing various structure-based mutations of CC ( a ) PlexinC1cyto , the single turnover GAP assay was used ( Wang et al . , 2012 ) . The concentration of plexin in the assays shown in Figure 7C was 0 . 25 µM . In the assays shown in Figure 7A , B , the concentration of plexin was 2 µM . The concentration of Rap1B ( GTP ) was 120 µM . In the assays for analyzing various mutants of the PlexinC1cyto monomer and Rap1B , the concentrations of PlexinC1 and Rap1B ( GTP ) were 5 μM and 60 μM respectively . For determining the activation levels of the CC ( x ) Plexinscyto constructs , the initial reaction rate V0 was measured at different Rap ( GTP ) concentrations ( [S] ) ( Table 2 ) . Fitting the data to the Michaelis–Menten equation ( V0 = ( Vmax[S] ) / ( KM + [S] ) ) suggested that the Rap ( GTP ) concentrations used ( 25–150 µM ) were far below KM ( >1 mM ) . For plexin constructs exhibiting low GAP activity , V0 was determined by linear fitting of the initial period of the reaction ( 5–8 min ) when less then 10% of Rap ( GTP ) had been hydrolyzed . After subtraction of the baseline rate from reaction without plexin , the kcat/KM value of each construct was estimated by fitting the data to the equation V0 = ( kcat/KM ) [E][S] ( when [S] << KM ) , where [E] is the total plexin concentration . For plexin constructs with high GAP activity , single turn-over reaction curves measured at different Rap ( GTP ) concentrations were baseline-subtracted and simultaneously fitted to the single exponential equation: A ( t ) = ( Amax − Amin ) ( 1 − exp ( −kt ) ) + Amin , where k = ( kcat/KM ) [E] . In the fitting , k was treated as a global parameter . The plexin and Rap concentrations and the analysis methods used are listed in Table 2 . 10 . 7554/eLife . 01279 . 018Table 2 . Protein concentrations and fitting methods used for determining kcat/KM of plexinsDOI: http://dx . doi . org/10 . 7554/eLife . 01279 . 018Plexin constructPlexin concentration ( µM ) Rap-GTP concentrations ( µM ) Data fitting method†CC ( a ) PlexinA1cyto1 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialCC ( b ) PlexinA1cyto5 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0LinearCC ( c ) PlexinA1cyto5 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0LinearCC ( d ) PlexinA1cyto1 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialCC ( e ) PlexinA1cyto5 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialCC ( f ) PlexinA1cyto5 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0LinearCC ( g ) PlexinA1cyto1 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialMonomer PlexinC1cyto2 . 025 . 0 , 50 . 0 , 75 . 0 , 100 . 0 , 150 . 0LinearCC ( a ) PlexinC1cyto2 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialCC ( b ) PlexinC1cyto2 . 025 . 0 , 50 . 0 , 75 . 0 , 100 . 0 , 150 . 0LinearCC ( d ) PlexinC1cyto2 . 050 . 0 , 75 . 0 , 100 . 0 , 150 . 0Single-exponentialCC ( g ) PlexinC1cyto2 . 025 . 0 , 50 . 0 , 75 . 0 , 100 . 0 , 150 . 0Linear*The plexin concentrations were chosen in order for the reaction rates to be within the dynamic range of the assay . †Linear fitting: kcat/KM determined by fitting data to V0 = ( kcat/KM ) [E][S]; Single-exponential fitting: kcat/KM determined by fitting data to A ( t ) = ( Amax − Amin ) ( 1 − exp ( −kt ) ) + Amin , in which k= kcat/KM[E] and was fitted as a global parameter . Due to the high intrinsic activity of the ligated plexincyto/Rap complexes , all the bound GTP molecules were hydrolyzed to GDP during the purification process . To measure the GAP activity for these complexes , we used the multiple-turnover assay , in which ( NH4 ) 2SO4 at 10 mM and EDTA at 1 mM are added to promote constant exchange of GTP/GDP for Rap in the complex , allowing continuous GTP hydrolysis provided sufficient GTP is present in the assay solution ( Webb and Hunter , 1992 ) . As the ligated complex with the 24-residue linker and the LPETGG motif crystallized and was used for structure determination , the same construct was chosen for extensive activity analyses at various concentrations . Unligated PlexinC1 and Rap1B mixed at the same concentrations were subjected to the same assay for comparison . Sedimentation velocity analytical ultracentrifugation experiments were carried out using the ligated PlexinC1cyto/Rap1B complex with the 24-residue linker and the LPETGG motif . Protein samples were prepared in Centrifugation Buffer ( 10 mM Tris pH 8 , 50 mM NaCl , 2 mM TCEP , and 2 mM MgCl2 ) . Samples at 0 . 5 , 4 , and 20 μM were used for the experiments without AlFx . Samples at 0 . 5 , 4 , 8 and 20 μM were used for the experiments in the presence of 2 mM AlFx . All samples were equilibrated ∼14 hr at 4°C , then ∼400 μl of the samples were loaded into the ‘sample’ sides of dual-sectored charcoal-filled Epon centerpieces that were sandwiched between sapphire windows in a cell housing; the ‘reference’ sectors were filled with the same volume of Centrifugation Buffer . Filled cells were placed in an An50Ti rotor and equilibrated for 2 . 5 hr under vacuum in the centrifuge at 20°C prior to centrifugation . Experiments were conducted using a Beckman Optima XL-I analytical ultracentrifuge at 42 , 000 rpm at 20°C . Absorbance data at 280 nm were collected using the Beckman control software until all components had fully sedimented . Protein partial-specific volume , solvent viscosity , and density values were calculated using the program Sednterp ( Laue et al . , 1992 ) . The data were analyzed using the c ( s ) distribution in the program SEDFIT ( Schuck , 2000 ) . A regularization level of 0 . 68 was routinely employed . Time-invariant noise elements were removed from the data ( Schuck and Demeler , 1999 ) . Data-acquisition timestamp errors ( Zhao et al . , 2013 ) were examined with SEDFIT and were found to be ∼0 . 1%; we deemed this small error acceptable and did not correct the timestamps . Plots were generated with the program GUSSI ( http://biophysics . swmed . edu/MBR/software . html ) . Mouse CC ( d ) PlexinA1 , CC ( g ) PlexinA1 , zebrafish CC ( d ) PlexinC1 and CC ( a ) PlexinC1 were subjected to crystallization trials . CC ( a ) PlexinC1cyto at 8 mg/ml crystallized initially at 20°C in 0 . 1M Bicine , pH 9 . 0 , 20% PEG 6000 in sitting-drop 96-well plates . Larger crystals were grown by sitting-drop vapor diffusion at 20°C in 0 . 1M Bis-Tris propane , pH 9 . 1 , 21% PEG 6000 . Crystals were cryo-protected using the crystallization solution supplemented with 25% glycerol and flash cooled in liquid nitrogen . Diffraction data were collected at 100 K on beamline 19ID at the Advanced Photon Source ( Argonne National Laboratory ) . Data were indexed , integrated and scaled by using HKL2000 ( Otwinowski and Minor , 1997 ) . A 3 . 3 Å dataset in the P212121 space group was collected . The ‘autocorrections’ option in HKL2000 was selected to truncate and scale the anisotropic data , which was then converted to the mtz format by using the Ctruncate program in CCP4 ( Padilla and Yeates , 2003; Winn et al . , 2011 ) . The structure of the GAP domain of mouse PlexinA3 ( PDB ID: 3IG3 ) was used as the molecular replacement search model using the Phaser module in the Phenix package ( Adams et al . , 2002; Mccoy et al . , 2007 ) . Ligated complexes of human Rap1B and several plexinscyto from mouse and zebrafish each with one of the 7 versions of the linker mentioned above were all subjected to crystallization trails . The ligated complex of zebrafish PlexinC1cyto and human Rap1B with the 24-residue linker and the LPETGG-tag at 4 mg/ml crystallized initially at 20°C in 0 . 1 M HEPES pH 7 . 5 , 10% 2-propanol , 20% PEG 4K in sitting-drop 96-well plates . Larger crystals were grown by hanging-drop vapor diffusion at 20°C in 0 . 1 M HEPES pH 7 . 3 , 5% 2-propanol , 25% PEG 3350 , 3 . 6% polypropylene glycol P400 . Cryo-protection of the crystals was achieved using with the crystallization solution supplemented with 25% glycerol . Cryo-protected crystals were snap cooled in liquid nitrogen . The data collection and processing were performed in a similar manner as described for the CC ( a ) PlexinC1cyto crystal , expect that the ‘autocorrections’ option was not used . The diffract pattern extended to 3 . 3 Å and was consistent with the symmetry of the P1 space group . One protomer from the CC ( a ) PlexinC1cyto structure was used as the molecular replacement search model for plexin . The structure of Rap1B from the Rap1B/RapGAP complex ( PDB ID: 3BRW ) was used as the search model for Rap1B . Iterative model building and refinement were performed using the Phenix and Coot programs respectively ( Adams et al . , 2002; Emsley and Cowtan , 2004 ) . In the PlexinC1cyto/Rap1B structure , the linker between the C-terminus of Rap1B and the N-terminus of PlexinC1cyto is not included in the final model due to lack of discernable electron density . Assuming the complexes in the crystal are formed by the covalently linked pairs of Rap1B and PlexinC1cyto , the linker and the disordered flanking residues from the two proteins ( a total of ∼32 residues ) are sufficient for spanning the ∼45 Å distance between the two ends without imposing restraints on the plexin/Rap binding mode ( Figure 8A , middle panel ) . The structural superimpositions shown in the Figures 5 and 6 were based on helices 13 , 14 and 15 in the plexin GAP domain , because they are at the center of the GAP active site and adopt highly similar conformations in all the plexin structures . Comprehensive model validation was performed by using MolProbity ( Chen et al . , 2010 ) . Detailed statistics of data collection and refinement are listed in Table 1 . Structure figures were prepared in PyMOL ( the PyMOL Molecular Graphics System , Schrodinger ) . Sequences were aligned by using T-Coffee ( Notredame et al . , 2000 ) and rendered with ESPript ( Gouet et al . , 1999 ) . Molecular surface area was calculated using the get_area function in PyMOL . Morph frames in Video 1 were generated by using the Yale morph server ( Krebs and Gerstein , 2000 ) and rendered in PyMOL . Mutants of mouse PlexinA3 were designed based on a sequence alignment of zebrafish PlexinC1 with all mouse plexins ( Plexin A1 , A2 , A3 , A4 , B1 , B2 , B3 , C1 and D1 ) . COS7 cell collapse assays using full-length mouse PlexinA3 were performed as described previously ( He et al . , 2009 ) . Briefly , 1 × 105 COS7 cells were plated in each well of a 6-well plate one day prior to transfection . FuGENE 6 ( Promega , Madison , WI ) was used to transfect each well with PlexinA3 ( 1 µg plasmid ) and the co-receptor Neuropilin2 ( 0 . 5 µg plasmid ) following the manufacturer’s instructions . 2 days post transfection , 5 nM alkaline phosphatase-tagged Sema3F was added to each well and incubated for 25 min at 37°C . The cells were washed , fixed and heat-treated at 65°C for 1 hr to inactivate endogenous phosphatases . Cells were stained with the BCIP/NBT alkaline phosphatase substrate ( Sigma , St . Louis , MO ) , and counted using a randomized and blind method .	A key question in neurobiology is how the brain becomes wired up . How do axons—the ‘wires’ along which neural signals flow—know in which direction to grow to reach their intended targets ? A family of signalling proteins called semaphorins contribute to this process by acting as stop signals for axons that are heading in the wrong direction . The actions of semaphorins are mediated by receptors known as plexins , which are found on the membranes of axons . Plexins contain an extracellular domain that binds semaphorin , and a large domain inside the cell that can turn semaphorin binding into cellular responses . When a semaphorin protein binds to the extracellular domain of a plexin receptor , the domain inside the cell joins with the intracellular domain of a neighbouring receptor to form a dimer . This activates the intracellular domain , which turns on its ability to inactivate a molecule called Rap . The end result is that the axon stops growing and changes direction , but the molecular mechanisms through which these events occur are not well understood . Now , Wang , Pascoe et al . have worked out the structure of the dimers formed by the intracellular plexin domains , both alone and in complex with Rap . The structures reveal how the dimer drives a shape change of the intracellular domain to enable it to bind Rap , and show that Rap itself adopts a novel conformation upon binding to plexin . This conformational change in Rap catalyses the breakdown of a signalling molecule called GTP , which inactivates Rap and triggers an intracellular signalling cascade that causes the axon to collapse and change direction . Lastly , Wang , Pascoe et al . have shown that the highly specific nature of these interactions depends on particular amino-acid residues in both Rap and the plexin receptor . Further work is now required to determine whether this pattern of activation represents a general mechanism for signalling by plexin receptors , and for the inhibition of Rap .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Structural basis for activation and non-canonical catalysis of the Rap GTPase activating protein domain of plexin
Internal ribosome entry sites ( IRESs ) are powerful model systems to understand how the translation machinery can be manipulated by structured RNAs and for exploring inherent features of ribosome function . The intergenic region ( IGR ) IRESs from the Dicistroviridae family of viruses are structured RNAs that bind directly to the ribosome and initiate translation by co-opting the translation elongation cycle . These IRESs require an RNA pseudoknot that mimics a codon-anticodon interaction and contains a conformationally dynamic loop . We explored the role of this loop and found that both the length and sequence are essential for translation in different types of IGR IRESs and from diverse viruses . We found that loop 3 affects two discrete elongation factor-dependent steps in the IRES initiation mechanism . Our results show how the IRES directs multiple steps after 80S ribosome placement and highlights the often underappreciated significance of discrete conformationally dynamic elements within the context of structured RNAs . A vital step in infection by viruses is translation of the viral RNA . Many RNA viruses initiate translation using internal ribosome entry sites ( IRESs ) , which are cis-acting RNA elements that recruit the host cell’s translation machinery in a cap- and end-independent fashion ( Filbin and Kieft , 2009; Doudna and Sarnow , 2007; Plank and Kieft , 2012 ) . Most viral IRESs use a subset of the canonical initiation factor proteins to recruit and position the ribosome , but the intergenic region ( IGR ) IRESs of the Dicistroviridae family of viruses use a more streamlined mechanism ( Figure 1A ) . Specifically , the ~200 nucleotide long , compactly folded IRES RNA interacts directly with both ribosomal subunits to assemble 80S ribosomes ( Nishiyama , 2003; Costantino and Kieft , 2005; Pfingsten et al . , 2006 ) , eliminating the requirement for initiation factors ( Sarnow et al . , 2005; Jan , 2006 ) . The IRES binds between the two subunits and , akin to a tRNA , must translocate through the ribosome ( Spahn et al . , 2004; Schüler et al . , 2006 ) , the only known non-tRNA molecule to do so . In addition , an IGR IRES was recently shown to be able to facilitate translation initiation in live bacteria , although the mechanism in bacteria is very different from the mechanism in eukaryotes ( Colussi et al . , 2015 ) . Current mechanistic models for how the IGR IRESs operate in eukaryotes suggest that after the IGR IRES assembles an 80S ribosome , eukaryotic elongation factor ( eEF ) 2 catalyzes an initial pseudotranslocation event ( translocation without peptide bond formation ) which positions the first codon of the open reading frame in the A site ( Figure 1A ) ( Fernández et al . , 2014; Koh et al . , 2014; Zhu et al . , 2011 ) . This is followed by eEF1A-catalyzed delivery of the first cognate ac-tRNA to the A site and a second eEF2-driven pseudotranslocation event that vacates the A site , allowing delivery of another ac-tRNA , subsequent peptide bond formation , and assumption of the normal translation elongation cycle ( Yamamoto et al . , 2007; Sasaki and Nakashima , 1999; Jan and Sarnow , 2002; Pestova , 2003; Pestova et al . , 2004 ) . Thus , initiation by this RNA structure-driven process has evolved to use the catalytic action of two GTPase elongation factors . The IGR IRESs have been studied using ribosomes , tRNA , elongation factors , lysate , and cells from sources as diverse as yeast , human , rabbit , shrimp , and wheat germ , often employed in combinations ( representative references: Nishiyama , 2003; Costantino and Kieft , 2005; Spahn et al . , 2004; Koh et al . , 2014; Yamamoto et al . , 2007; Jan and Sarnow , 2002; Pestova , 2003; Pestova et al . , 2004; Cevallos and Sarnow , 2005; Wilson et al . , 2000; Masoumi et al . , 2003; Thompson et al . , 2001; Au et al . , 2012; Costantino et al . , 2008; Jan et al . , 2003; Muhs et al . , 2015; Kamoshita et al . , 2009; Landry et al . , 2009; Fukushi et al . , 2001; Hertz and Thompson , 2011; Deniz et al . , 2009; Jang et al . , 2009; Pfingsten et al . , 2010 , 2007 ) . The mechanism that has emerged is consistent across these systems . This reflects the streamlined IGR IRES mechanism that depends on an RNA structure that manipulates conserved features of the eukaryotic translation machinery . In addition , this feature allows the use of diverse convenient reagents to study the IGR IRESs , a characteristic we took advantage of in this study . 10 . 7554/eLife . 08146 . 003Figure 1 . Intergenic region ( IGR ) internal ribosome entry site ( IRES ) mechanism and loop 3 . ( A ) Schematic of the IGR IRES initiation factor-independent translation initiation mechanism . The IGR IRESs occupy the same binding sites as tRNAs in the ribosome . Elongation factor-catalyzed steps are shown in red type and arrows , and proposed reverse reactions are shown with gray arrows . ( B ) Secondary structure cartoon of an IGR IRES with domain III boxed and loop 3 in red . PKI in the figure denotes the pseudoknot base pairs that mimic the codon–anticodon interaction . ( C ) Cryo-electron microscopy ( cryo-EM ) reconstruction of the Taura Syndrome Virus ( TSV ) IGR IRES bound to Saccharomyces cerevisiae 80S ribosomes ( Koh et al . , 2014 ) . The TSV IRES RNA model is shown in yellow , with loop 3 in red . Density within 8 Å of the IRES model is shown , at a threshold of 2 . 5 . To the right is a close-up view of loop 3 . ( D ) Same as panel C , but of a Cricket Paralysis Virus ( CrPV ) IGR IRES bound to Kluyveromyces lactis 80S ribosomes ( Fernández et al . , 2014 ) . Density within 4 Å of the IRES model is shown , at a threshold of 2 . 5 . ( E ) Same as panel C , but of a CrPV IGR IRES bound to Oryctolagus cuniculus 80S ribosomes with eukaryotic release factor 1 ( eRF1 ) bound ( Muhs et al . , 2015 ) . Density within 5 Å of the IRES model is shown , at a threshold of 3 . 0 . ( F ) Diagram of the dual luciferase ( LUC ) reporter RNA used in all in vitro translation assays . IRES activity is determined as a ratio of Firefly LUC activity to Renilla LUC activity . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 00310 . 7554/eLife . 08146 . 004Figure 1—figure supplement 1 . IGR IRES location in viral RNA , and alignment and structure of domain III . ( A ) Diagram of the Dicistroviridae RNA genome . The IGR IRESs initiate translation of the second open reading frame . ( B ) Alignment of domain III sequences from 14 Dicistroviridae family members ( class I and II ) . The location of loop 3 is indicated in red . Conserved sequence is in bold . CrPV , Cricket Paralysis Virus; ALPV , Aphid Lethal Paralysis Virus; BQCV , Black Queen Cell Virus; DCV , Drosophila C Virus; HiPV , Himetobi P Virus; HoCV , Homalodisca coagulata Virus; PSIV , Plautia stali Intestinal Virus; RhPV , Rhopalosiphum padi Virus; TrV , Triatoma Virus; ABPV , Acute Bee Paralysis Virus; IAPV , Israeli Acute Paralysis Virus; KBV , Kashmir Bee Virus; SInV , Solenopsis invicta Virus-1; TSV , Taura Syndrome Virus . ( C ) A model of the CrPV IGR IRES from cryo-electron microscopy ( magenta ) bound to an 80S ribosome ( PDB ID: 4CUX ) ( Fernández et al . , 2014 ) overlaid with A- , P- , and E-site tRNAs ( green ) bound in a 70S ribosome in the presence of paromomycin ( PDB ID 2WDK ) ( Voorhees et al . , 2009 ) . Domain 3 of the IRES is boxed in red and loop 3 is indicated with an arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 00410 . 7554/eLife . 08146 . 005Figure 1—figure supplement 2 . Loop 3 composition and length in diverse IGR IRESs . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 005 Although IRES structural features that drive formation of the IRES–80S ribosome complex have been mapped , how the IGR IRES co-opts elongation factor function to drive pseudotranslocation through the ribosome is poorly understood . During the canonical elongation cycle tRNA translocation requires specific tRNA–ribosome interactions and conformational states ( Frank et al . , 2007; Joseph , 2003; Schmeing and Ramakrishnan , 2009; Frank and Gonzalez , 2010 ) ; it has been proposed that IGR IRESs fulfill these requirements through a strategy that involves both global and local tRNA mimicry ( Costantino et al . , 2008; Jang et al . , 2009 ) . Globally , the ribosome-bound IGR IRES occupies the spaces normally bound by tRNAs , spans all three tRNA binding sites ( Figure 1—figure supplement 1 ) ( Spahn et al . , 2004; Schüler et al . , 2006; Fernández et al . , 2014; Koh et al . , 2014; Muhs et al . , 2015 ) , interacts with tRNA-binding surfaces on the ribosome , and potentially mimics or induces a hybrid-like state ( Frank et al . , 2007; Frank and Gonzalez , 2010; Moazed and Noller , 1989 ) . Locally , the IRES mimics tRNA using a pseudoknot-containing domain ( pseudoknot I [PKI] in domain III ) that structurally mimics the mRNA-tRNA codon–anticodon interaction located just upstream of the translation start site ( Figure 1B ) ( Zhu et al . , 2011; Costantino et al . , 2008; Jan et al . , 2003 ) . Previous biochemical and structural studies show that domain III is not needed for initial subunit recruitment and 80S ribosome formation but is essential for establishing the reading frame by docking precisely in the ribosome’s decoding groove ( Nishiyama , 2003; Costantino and Kieft , 2005; Jan and Sarnow , 2002 ) . However , domain III has features that suggest additional roles . Specifically , x-ray crystal structures of domain III in both the unbound form and bound to ribosomes ( Zhu et al . , 2011; Costantino et al . , 2008 ) , and chemical probing experiments ( Jan and Sarnow , 2002; Pfingsten et al . , 2010 , 2007 ) , revealed that the single-stranded loop of RNA ( ‘loop 3’ ) that links the anticodon-like hairpin to the mRNA-like sequence is conformationally dynamic ( Figure 1B ) . Mutation or elimination of some bases in loop 3 affects IRES function , purportedly by impairing ribosome positioning , although other effects are possible ( Au et al . , 2012 ) . Cryo-electron microscopy reconstructions provide structural models for loop 3 but the electron density corresponding to this loop is generally weaker than in other parts of the IRES , not continuous , or of low resolution ( Figure 1C–E ) ( Schüler et al . , 2006; Fernández et al . , 2014; Koh et al . , 2014; Muhs et al . , 2015 ) , again suggesting conformational dynamics or structural heterogeneity . These observations are surprising , as domain III comprises an H-type pseudoknot in which the analogous loop usually forms a stable structure ( Staple and Butcher , 2005; Aalberts , 2005; Westhof and Jaeger , 1992 ) . Comparing the sequences of IGR IRESs from different species reveals conservation in terms of the length range and base composition , in particular a high adenosine content ( Figure 1—figure supplements 1 , 2 ) . Adenosine residues in pseudoknot loops often form stable tertiary contacts that are not observed in domain III ( Staple and Butcher , 2005; Aalberts , 2005 ) . These features , combined with our previous work showing that conformationally dynamic structural elements in the IGR IRES can play important roles in IRES function ( Pfingsten et al . , 2010 ) , led us to analyze the mechanistic role of loop 3 , focusing on the poorly characterized events following 80S ribosome recruitment . We discovered that conformationally dynamic loop 3 operates within the context of the highly structured IRES RNA to influence the activity of elongation factors co-opted to drive initiation . We found that both the length and sequence of loop 3 are essential for efficient translation initiation in IGR IRESs from diverse members of the Dicistroviridae family . Using the IGR IRES from Cricket Paralysis Virus ( CrPV ) , we demonstrate that loop 3 affects multiple eEF-directed steps , including both pseudotranslocation events . Our findings provide an example of how RNAs can use dynamic regions within the context of a globally stable structure to facilitate function . Because loop 3 is unlikely to interact directly with elongation factors and translocation is a process that depends on ribosome conformational dynamics , our data also suggest a hypothesis in which loop 3 affects ribosome conformations to assist in non-canonical translocation . We assessed the functional importance of loop 3 in IGR IRES-driven translation using a dual luciferase ( LUC ) reporter construct in rabbit reticulocyte lysate ( RRL ) ( Figure 1F ) . RRL was chosen as it has proven to be a consistent system for examining the activity of most IGR IRESs . First , we measured the relative translation initiation efficiencies of several IGR IRES RNAs in RRL ( Figure 2A ) . Based on this , we chose representative IRESs with differing activities , including Class I and II IGR IRESs ( from the Cripa- and Apara-virus subfamilies ) , to study the role of loop 3 . We made several mutants ( Table 1 ) : ( 1 ) we shortened loop 3 by three nucleotides , reasoning this would reduce flexibility that may be important for function ( △3 mutants ) ; ( 2 ) noting the loops’ high adenosine content , we replaced several adenosines with guanosines ( G-rich mutants ) ; ( 3 ) because sequence alignment from various IRESs suggested the presence of conserved bases in loop 3 ( Figure 1—figure supplement 1B ) ( Au et al . , 2012 ) , we replaced a single conserved adenosine with a guanosine in the highly active Israeli Acute Paralysis Virus ( IAPV ) IRES . These mutants are similar to those studied by Au et al . , ( 2012 ) , but are more aggressive in the sense that we deleted more nucleotides ( three ) and substituted more bases ( three ) . Each mutation had a substantial impact on IRES activity ( Figure 2B , C ) . Thus , loop 3 plays a functional role in IGR IRES activity , and this role is shared by diverse members of both IRES classes . 10 . 7554/eLife . 08146 . 006Figure 2 . Function of diverse wild type ( WT ) and loop 3 mutant intergenic region ( IGR ) internal ribosome entry site ( IRESs ) in rabbit reticulocyte lysate ( RRL ) . ( A ) Activity of different WT IGR IRESs . Mutant Cricket Paralysis Virus ( CrPV ) -K/O has pseudoknots III and I disrupted and is the negative control ( Jan and Sarnow , 2002; Costantino et al . , 2008 ) . ( B and C ) Function of WT IRESs ( black bars ) and loop 3 mutants ( gray bars ) . WT levels are normalized to 1 for each IRES . ( D ) Diagrams of CrPV IGR IRES domain III mutants . Mutations are boxed and X indicates deletion of a nucleotide . ( E ) Activity of CrPV loop 3 mutants in RRL . Error bars represent standard error of the mean over at least three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 00610 . 7554/eLife . 08146 . 007Figure 2—figure supplement 1 . Degradation of input reporter mRNA in RRL . Top: Input reporter mRNAs were body-labeled with 32P during transcription , purified , then incubated in RRL at 30°C for the times indicated . RNA was then recovered from the reactions , resolved on a denaturing polyacrylamide gel , and full-length RNA was quantitated by phosphorimaging . The graph indicates the average percent of each input RNA remaining as a function of time from three independent experiments , and a linear fit of these data . Error bars indicate one standard error from the mean . Bottom: An example of raw data from this experiment . Numbers indicate incubation time in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 00710 . 7554/eLife . 08146 . 008Table 1 . Activity of IGR IRESs in RRL and mutations tested . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 008VirusWT activityLoop 3 mutants testedClass I G-rich△3ConservedCrPV++++HiPV+HoCV+UUAGGGGCCG UUAGA - - - CAPSIV+Class II ABPV++++IAPV+++++GAGGUGCCAGGAAUACCAKBV++GAAGUGCCG GAAAUA - - -SInV++++TSV+Site of mutation is shown in bold italics . Site of deletion is shown as a dash . ABPV , Acute Bee Paralysis Virus; CrPV , Cricket Paralysis Virus ( CrPV ) ; HiPV , Himetobi P Virus; HoCV , Homalodisca coagulata Virus; IAPV , Israeli Acute Paralysis Virus; IGR , intergenic region; IRES , internal ribosome entry site; KBV , Kashmir Bee Virus; PSIV , Plautia stali Intestinal Virus; RRL , rabbit reticulocyte lysate; SInV , Solenopsis invicta Virus-1; TSV , Taura Syndrome Virus . Having established the conserved functional importance of loop 3 , we selected the CrPV IGR IRES as a model IRES for additional exploration because it has been widely studied biochemically and structurally , and also because it has the aforementioned characteristic of displaying a consistent mechanism of action when studied with a variety of reagents from diverse species . Several more mutants were designed to assess the importance of loop 3 ( Figure 2D , E ) . Shortening loop 3 in the CrPV IGR IRES by just one nucleotide ( △1 ) had a small effect on function while deleting two nucleotides ( △2 ) caused a significant loss of activity; this agrees with previous results ( Au et al . , 2012 ) . The △3 mutant’s activity is even more substantially reduced , matching the activity of the negative control PKI/III knockout mutant ( Jan and Sarnow , 2002; Costantino et al . , 2008 ) . Likewise , CrPV IRES mutants analogous to the aforementioned G-rich mutants and another mutant in which three conserved bases were mutated ( GGC mutant ) were substantially decreased in their abilities to initiate translation . Because these differences in measured IRES activity could be due to different amounts of input reporter mRNA or rates of mRNA degradation , we controlled for this in two ways . First , the presence of the upstream Renilla LUC ( not under IRES control ) provides an internal normalization control for small differences in the amount of RNA in the reaction . Second , we measured the rates of degradation of all reporter mRNAs in the RRL translation reaction , finding that all were equal ( Figure 2—figure supplement 1 ) . These data indicate that both loop 3 base composition and length are important for CrPV IGR IRES function , and the mutants now provide a set of tools for querying the specific mechanistic role of loop 3 . Numerous direct ribosome binding studies have shown that domain III can be completely removed or the PKI interaction abrogated without decreasing the IRES’s affinity for the ribosome ( Nishiyama , 2003; Costantino and Kieft , 2005; Jan and Sarnow , 2002 ) . This suggests that the effects we observe when loop 3 is mutated are not due to alterations in 80S ribosome binding , but rather in events downstream of initial ribosome recruitment . To test this prediction , we used radiolabeled IRES RNAs in RRL to generate IRES–ribosome complexes and resolved them by ultracentrifugation through a sucrose gradient , using an antibiotic to halt the complexes after initial formation ( Figure 3—figure supplement 1A ) . All loop 3 mutants robustly assemble 80S–ribosome complexes in RRL . Although there is some variability in the amount of 80S complexes produced in this assay , the amounts do not correlate with the translation activity levels . As a second test for ribosome binding , we measured the approximate on- and off-rates of two mutant IRESs with purified ribosomes from yeast and shrimp sources using filter binding ( Figure 3—figure supplement 1B ) . We chose yeast and shrimp ribosomes to complement the RRL and also to test a different source of ribosomes to enable their use in subsequent assays . The measured rates are the same for wild type ( WT ) and mutant IRES RNAs . Taken together , these data are consistent with the conclusion that the functional effects of mutating loop 3 cannot be accounted for by defects in initial ribosome association with the IRES . To explore events after initial ribosome binding , we used toeprinting assays to determine if the mutant IRESs are properly positioned within the decoding groove of 80S ribosomes and if they are competent to pseudotranslocate . We chose RRL to match the translation activity assays . Since rabbit and yeast ribosomes produce an identical pretranslocation ( PRE ) toeprint at the +14/15 position ( Figure 3—figure supplement 2 ) , we used yeast 80S ribosomes as a marker for the initial IRES location in the ‘pretranslocated’ state ( Figure 3A lanes 2 and 18 ) . Toeprinting of the WT CrPV IGR IRES in RRL supplemented with the elongation inhibitor cycloheximide ( CHX ) reveals that the IRES translocates twice ( +20/21 toeprint , Figure 3A lanes 3 and 19 ) as previously observed ( Wilson et al . , 2000 ) . Without CHX no strong toeprints are seen , indicating that the antibiotic traps IRES–ribosome complexes that can be observed in this assay . 10 . 7554/eLife . 08146 . 009Figure 3 . Ribosome docking , translocation , and reading frame maintenance . ( A ) Toeprinting analysis of Cricket Paralysis Virus ( CrPV ) wild-type ( WT ) internal ribosome entry site ( IRES ) and loop 3 mutants in the free ( f ) and yeast 80S ribosome-bound ( 80S ) forms , and in rabbit reticulocyte lysate ( RRL ) with or without 3 mg/ml cycloheximide ( +/- CHX ) . The +14/15 toeprint indicates the position of the edge of the pretranslocation ribosome , and the +20/21 toeprint shows the position of the edge of the 2x translocated ribosome . Gels are representative of at least six independent experiments . ( B ) Quantification of translocated toeprint bands ( +20/21/ ( ( +14/15 ) + ( +20/21 ) ) ) in RRL+CHX ( n = 6–9 ) , error bars represent standard error of the mean . ( C ) In vitro translation assay of dual luciferase reporters with +0 ( normal ) , +1 , or +2 reading frames . Error bars represent standard error of the mean of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 00910 . 7554/eLife . 08146 . 010Figure 3—figure supplement 1 . IGR IRES loop 3 mutants bind the 80S ribosomes . ( A ) Assembly of 80S ribosomes on CrPV intergenic region ( IGR ) IRES loop 3 mutants in rabbit reticulocyte lysate . Radiolabeled CrPV IRES RNAs were incubated in RRL supplemented with hygromycin B for 20 min before separation of initiation complexes on a 15–30% sucrose gradient . Free and 80S-bound IRES complexes are indicated . ( B ) Approximate on- and off-rates of IRES-ribosome binding measured by filter binding . The on-rate experiment measures the association of IRES with ribosomes or ribosomal subunits as a function of time . Pure shrimp ribosomes were used for the on-rate experiment . The off-rate experiment used unlabeled competitor IRES RNA to detect dissociation of IRES from ribosomes as a function of time . Purified yeast subunits were used for the off-rate experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01010 . 7554/eLife . 08146 . 011Figure 3—figure supplement 2 . Toeprinting of WT CrPV IGR IRES with purified 40S subunits and 40S + 60S ( 80S ) ribosomes from two sources . R indicates rabbit; Y indicates yeast . The gel lanes shown are spliced from a single gel with irrelevant lanes removed . The locations of the toeprint with 40S and 80S are identical with both yeast and rabbit subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 011 Like WT , all length mutants ( △1 , △2 , △3 ) have a pretranslocated toeprint at +14/15 when bound to pure yeast ribosomal subunits , indicating these IRESs are correctly positioned within the decoding groove of 80S complexes ( Figure 3A lanes 6 , 10 , 14 ) . However , in RRL the loop 3 length mutants retain the +14/15 toeprint both with and without CHX to a degree that is roughly inversely correlated with their translation activities , showing that pseudotranslocation is inhibited ( lanes 7 , 8 , 11 , 12 , 15 , 16 ) . A mutation that abrogates codon–anticodon base pairing in PKI does not generate a PRE toeprint at all ( Jan and Sarnow , 2002 ) ; the fact that each mutant IRES still exhibited a PRE toeprint indicates that the mutations tested here probably do not disrupt pseudoknot formation . Furthermore , the +20/21 toeprint is decreased in the △2 mutant and is completely missing in the △3 mutant . The decreases in the +20/21 toeprint are accompanied by an increase in the pretranslocated toeprint , consistent with a decrease in the ability to undergo the first two rounds of pseudotranslocation . Our experience with the toeprinting method leads us to take great care not to use toeprinting as a quantitative assay of the amount of ribosome binding , given the nature of the assay ( not at equilibrium conditions , detected indirectly by reverse transcription , etc . ) . In general , we conservatively use toeprinting as a robust way to assess the position of ribosomes that are bound , and their movements . After normalization of the signal and with analysis of many replicates , we determined the change in toeprint band intensities at the +14/15 and +20/21 positions to get a semi-quantitative measure of the percent of ribosomes that successfully perform two pseudotranslocations ( Figure 3B ) . In contrast to the measurements of 80S ribosome binding , these data show that shortening loop 3 inhibits the first two steps of pseudotranslocation in a way that correlates very well with the measured translation activity ( Figure 3B and 2E ) . Like the length mutants , the G-rich and GGC sequence mutants also form 80S complexes that are properly positioned at the +14/15 location ( Figure 3A lanes 22 and 26 ) . However , these sequence mutants match WT’s ability to generate a strong +20/21 band ( lanes 23 and 27 ) , suggesting they assemble 80S complexes that can translocate ( Figure 3B ) . To verify the results with CHX , we performed toeprinting in RRL with the translocation inhibitor hygromycin B , which binds the ribosome in a different location and has a different mechanism of action than CHX ( Borovinskaya et al . , 2008; Wilson , 2014 ) ( Figure 4A ) . The WT , G-rich , and GGC mutants pseudotranslocate once ( +17/18 toeprint ) , but the length mutants show a decreased ability to execute the first pseudotranslocation event . Taken together , these data indicate that mechanistic steps affected by loop 3 include the first pseuodotranslocation events after 80S ribosome association . 10 . 7554/eLife . 08146 . 012Figure 4 . Toeprinting with hygromycin B . ( A ) Toeprinting analysis in rabbit reticulocyte lysate ( RRL ) without or with 0 . 66 mg/mL hygromycin B ( -/+ ) . ( B ) Toeprinting analysis in RRL without or with 3 . 33 μg/mL hygromycin B ( -/+ ) added after 1 min of incubation of the internal ribosome entry site ( IRES ) in lysate . Normalized traces of the wild type ( WT ) , △3 , G-rich , and GGC IRES RNAs in RRL+ hygromycin B are shown at right . Image is from a single gel , asterisk indicates where two irrelevant lanes were removed . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01210 . 7554/eLife . 08146 . 013Figure 4—figure supplement 1 . RNase T1 probing ( single-stranded G bases ) of unbound WT , △3 , and G-rich Cricket Paralysis Virus ( CrPV ) intergenic region ( IGR ) IRES RNAs containing only domain III . Cleavage products in the denatured ( no Mg2+ ) and native ( + Mg2+ ) states were resolved next to a hydrolysis ladder ( OH ) on a sequencing gel ( left ) . Graph shows the difference in RNase T1 cleavage in the native state minus the denatured state after normalizing the total amount of radiation in each lane to the WT , no Mg2+ levels . The dashed box indicates residues that are in loop 3 . G 6204 , 6208 , and 6209 are specific to the G-rich mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 013 To identify the step at which the G-rich and GGC mutants are inhibited , we adapted the toeprinting assay to examine their effect on rounds of translocation after the two allowed by CHX . Dilute hygromycin B was added to RRL after addition of IRES RNA ( in the experiments described above , RRL was pretreated with high concentrations of hygromycin B or CHX ) . By altering the concentration of hygromycin B and the time when it was added , we were able to empirically capture the positions of ribosomes after they had loaded and started elongation . WT IRES toeprinting shows four–five rounds of translocation ( Figure 4B , lane 2 ) . As expected , △1 behaved similarity to WT while the △2 and △3 mutants did not proceed past the initial binding location ( lanes 6 and 8 ) . Surprisingly , the sequence mutants displayed toeprinting patterns similar to WT ( lanes 10 and 12 ) , although the bands generated from the first few rounds of translocation are less intense , assessed after careful normalization ( Figure 4B , right ) . Thus , the G-rich and GGC mutants can translocate at least four–five times in RRL , and the source of their reduced translation initiation activity must be more subtle than a complete failure to translocate . Although all of the mutants showed defects in translation initiation ( Figure 2E ) , the toeprinting data indicate that the reasons differ between the length and sequence mutants . The G-rich and GGC mutants do not completely block translocation while the length mutants do , indicating loop 3 has two independent roles in IGR IRES- driven translation initiation . The ability of the G-rich and GGC mutants to translocate in the toeprinting assays suggests they disrupt a different process than do the length mutants . Domain III is essential for establishing the proper reading frame , so the mutations could induce the ribosome to initiate out-of-frame . To test this , we measured translation in RRL using dual LUC constructs with one or two additional nucleotides inserted immediately before the AUG of the firefly LUC open reading frame ( +1 and +2 frames ) , which could rescue out-of-frame initiation ( Figure 3C ) . Neither alternate frame rescues IRES activity in the G-rich or GGC loop 3 mutants , indicating the G-rich and GGC mutants do not induce out-of-frame initiation . If the G-rich and GGC mutants initiate in-frame and can translocate at least four times as indicated by the toeprinting assay , why is their translation activity decreased ? It is unlikely that loop 3 acts after the IRES no longer interacts with the ribosome , the presumed situation after four translocation events . Alternatively , decreased toeprint band intensity in these mutants ( Figure 4B lanes 10 and 12 ) suggested there could be subtle changes in kinetics of the translocation events . Because toeprinting is not an ideal assay to examine this , we directly explored differences in the rate of peptide synthesis between the WT and the sequence mutants in an in vitro reconstituted translation system by quench-flow ( diagrammed in Figure 5—figure supplement 1 ) . For these experiments , we used ribosomes from yeast or shrimp eggs , reflecting "one of" the Dicistroviridae’s natural arthropod hosts , elongation factors from yeast , and tRNAs of either bacterial or yeast origin . As mentioned above , the use of convenient and high-activity heterologous systems is prevalent in IGR IRES research , and is justified because IGR IRESs appear to function identically in all tested eukaryotic systems . Where appropriate , we indicate the source of each component of the reconstituted system . Because toeprinting suggested at least four rounds of translocation on the G-rich and GGC mutants in RRL , we first assayed the rate of conversion of tripeptide to tetrapeptide on shrimp ribosomes with the coding sequence for the peptide "Phenylalanine-Valine-Lysine-Methionine" ( FVKM ) placed downstream of the IRES . Compared to WT , both the G-rich and GGC mutants displayed substantially decreased abilities to convert tripeptide to tetrapeptide , at levels that reflected their relative translation activities ( Figure 5A ) . These data suggest that the loss of translation activity in the loop 3 sequence mutants is imparted by at least one defective elongation step at or preceding tetrapeptide formation . 10 . 7554/eLife . 08146 . 014Figure 5 . Characterization of early steps in intergenic region ( IGR ) internal ribosome entry site ( IRES ) initiation in a reconstituted translation system , using purified shrimp ribosomes and yeast elongation factors . ( A ) Time course of tetrapeptide formation from tripeptide . Data are representative of two independent experiments . ( B ) [3H]Phe-tRNAPhe binding to the P site in the presence of eukaryotic elongation factor 2 ( eEF2 ) . Triplicate reads were averaged and normalized to set wild type ( WT ) equal to 1 . ( C ) Anisotropy measurements of Phe-tRNAPhe ( prf ) binding to IRES–80S ribosome complexes . For each set of experiments performed , a determination was made of the anisotropy difference ( △ ) between free ternary complex ( TC ) and TC added to the WT IRES–80S complex , and differences between TC added to other complexes and free TC were normalized to this value . Error bars represent one standard error from the mean of two–four replicates . ( D ) Translocation efficiency of ac-tRNA from the A to the P site in the △1 and △2 mutants . Data were normalized to set the anisotropy-based A site binding levels ( data from C ) to 1 , and those factors were applied to the cosedimentation-based P site binding levels ( data from B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01410 . 7554/eLife . 08146 . 015Figure 5—figure supplement 1 . Schematic overviews of experiments performed in the reconstituted system . Data from these experiments is presented in Figure 5 . Top: Tetrapeptide formation assay . Initiation complexes with the IGR IRES and Artemia salina ( shrimp ) ribosomes were first assembled before adding TCs consisting of F- , V- , and K-charged tRNAs with eIF1A-GTP ( the tRNA delivery factor ) and eEF2-GTP ( the translocase ) to form tripeptides . These complexes were then combined with [35S]Met-tRNAMet TC , quenched with strong base , and the resultant peptides were analyzed by thin layer electrophoresis . Bottom left: P site tRNA binding by cosedimentation . Bottom right: A site tRNA binding by anisotropy . Details of the assays can be found in the supporting methods section . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01510 . 7554/eLife . 08146 . 016Figure 5—figure supplement 2 . Codon- and factor-dependent tRNA binding to IRES–80S complexes . The graph depicts the amount of Phe-tRNAPhe[3H] or Arg-tRNAArg[3H] recovered after cosedimentation with shrimp 80S ribosome complexes through a sucrose cushion . The IRES RNA construct is from the WT Cricket Paralysis Virus ( CrPV ) IGR IRES , but initiates with a UUC ( Phe ) codon . The identity and source of the isolated tRNA is indicated beneath the bars . The presence or absence of factor ( s ) and IRES RNA is indicated beneath the graph . Arg-tRNAArg was used as a control for association of non-cognate tRNA . Phe-tRNAPhe from yeast or Escherichia coli stably associates with the ribosomes in an eEF1A- and eEF2-dependent manner , while non-cognate Arg-tRNAArg did not . Omitting either elongation factor greatly decreased stable binding . Error bars represent one standard deviation from the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01610 . 7554/eLife . 08146 . 017Figure 5—figure supplement 3 . Normalized anisotropy data . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01710 . 7554/eLife . 08146 . 018Figure 5—figure supplement 4 . Raw anisotropy data of controls . The anisotropy from the labeled tRNA shows a progressive increase as the mass of bound factors increases . The ‘no mRNA’ control shows the background level of tRNA+GTP+eEF1A to empty ribosomes . The presence of the IRES within the 80S ribosome causes a substantial increase in anisotropy . Error bars represent one standard error from the mean of 2–4 replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 018 The decreased peptide synthesis described above could result from inhibition of any step preceding tetrapeptide formation , including binding of the first ac-tRNA to the IRES–80S ribosome complex . To measure the efficiency of this step , we delivered [3H]Phe-tRNAPhe to WT and mutant 80S–IRES ( coding for FVKM ) shrimp ribosome complexes in the presence of eEF1A-GTP ( which forms a ternary complex , TC , with ac-tRNA ) and eEF2-GTP and collected these complexes by ultracentrifugation through a sucrose cushion ( diagrammed in Figure 5—figure supplement 1 ) . As expected , ac-tRNA delivered by eEF1A and translocated to the P site by eEF2 bound stably enough to survive this purification , whereas A-site associated ac-tRNA did not ( Figure 5—figure supplement 2 ) ( Yamamoto et al . , 2007 ) . Furthermore , ac-tRNA delivery and binding to the P site depended on a cognate codon–tRNA anticodon interaction ( Figure 5—figure supplement 2 ) . This latter control is important as it shows that the delivery and binding event we observe in this experiment depends on the presence of the IRES and the placement of the correct codon directly downstream of the IRES within the A site . Therefore , this assay measures the efficiency of completion of all three eEF-dependent steps ( Figure 1A ) . As expected , stable [3H]Phe-tRNAPhe binding was observed with WT IRES with eEF2 ( Figure 5B ) , consistent with previous reports ( Yamamoto et al . , 2007 ) . When mutants △1 , △2 , and △3 were assayed , they showed a progressive decrease in bound [3H]Phe-tRNAPhe . Interestingly , the G-rich and GGC mutants also showed decreased P-site ac-tRNA association with IRES–80S ribosome complexes at levels that mirror their relative translation activities . Therefore , mutations to loop 3 length and base composition cause decreased association of the first ac-tRNA in the P site . Because eEF2-GTP was included in the above experiment , we could not distinguish whether decreased ac-tRNA association in the P site resulted from reduced eEF2-driven pseudotranslocation of domain III from the A site to the P site , subsequent ac-tRNA delivery to the A site , or the second pseudotranslocation that moves ac-tRNA from the A site to the P site . To discriminate between these possibilities , we employed a fluorescence anisotropy experiment in which proflavin-labeled Phe-tRNAPhe [Phe-tRNAPhe ( prf ) ] TC was delivered to WT and mutant IGR IRES–80S ribosome complexes ( shrimp ribosomes ) in the absence of eEF2 ( diagrammed in Figure 5—figure supplement 1 ) . The measured anisotropy of unbound Phe-tRNAPhe ( prf ) was 0 . 205 +/- 0 . 002 ( Figure 5—figure supplement 4 ) . As expected , addition of eEF1A-GTP to the ac-tRNA resulted in an increase in measured anisotropy to 0 . 210 +/- 0 . 003 , consistent with formation of the eEF1A+GTP+Phe-tRNAPhe ( prf ) TC . Addition of empty 80S ribosomes ( lacking an mRNA or IRES , indicated as ‘no IRES’ ) resulted in only a slight increase in change in anisotropy relative to the TC alone ( Figure 5C ) . However , when a complex of CrPV IGR IRES bound to 80S ribosomes was added to the TC , we observed a much larger increase in anisotropy , to 0 . 272 +/- 0 . 006 . This change in anisotropy between TC alone and in the presence of 80S ribosomes+IRES ( 0 . 061 +/- 0 . 003 ) is consistent with delivery of ac-tRNA to the A site of the IRES–80S ribosome complex by the TC . To verify that IRES-dependent delivery of tRNA was specific for the first codon following the IRES , we delivered ac-tRNA to an IRES–80S ribosome complex in which the UUC codon for tRNAPhe was replaced by the non-cognate GCU codon ( ‘non-cognate’ , Figure 5C ) . This resulted in a smaller increase in anisotropy compared to the IRES with a cognate Phe codon , but larger than the ‘no IRES’ control . Importantly , the observation that eEF2-independent ac-tRNA binding to the ribosome requires a cognate codon is consistent with the idea that the first codon enters the A site and is queried by the ac-tRNA anticodon . This supports the idea that domain III can spontaneously move to the P site to some degree , perhaps akin to the observed ability of tRNAs to undergo slow spontaneous translocation on bacterial ribosomes ( Gavrilova et al . , 1976; Gavrilova and Spirin , 1971; Pestka , 1969; Southworth et al . , 2002; Fredrick and Noller , 2003; Moore , 2012; Robertson and Wintermeyer , 1987; Semenkov et al . , 1992 ) . The nature of the ac-tRNA’s association with the ribosome likely differs depending on whether an IRES RNA with a non-cognate or cognate codon is present; the former probably represents transient TC interaction with the tRNA in a A/T state during a decoding step , the latter likely represents full and longer-lived accommodation of the tRNA into the A/A state . The results outlined above validate the use of this assay to explore the effect of loop 3 mutations on ac-tRNA association with the IRES–ribosome complex independent of eEF2 activity . Mutants △1 , △2 , and △3 showed a progressive decrease in anisotropy ( Figure 5C ) , following the trend established by the translation initiation and pseudotranslocation data . These data indicate that these mutants have a defect in initial ac-tRNA binding; in the case of △3 , this defect is more severe than the effect of a non-cognate codon . This may be because the movement of the first codon into the A site has been compromised . ac-tRNA delivery to IRES–80S ribosome complexes with the △1 and △2 mutants was less than to WT , but equal to or greater than to the IRES with a non-cognate codon . To approximate the percentage of these A-site ac-tRNAs that successfully translocated to the P site , we normalized their P site binding levels to the A site interaction levels ( Figure 5D ) . For △1 , the percentage is ~80% while for △2 it is ~25% . When we consider these data in light of the proposed mechanism of IGR IRES-driven initiation ( Figure 1A ) , they suggest that these mutants have defects in both pseudotranslocation events and these defects become progressively worse as loop 3 is shortened . In contrast , the G-rich and GGC mutants display ac-tRNA binding similar to the WT IRES ( Figure 5C ) . Thus , the defect in these sequence mutants is restricted to the second pseudotranslocation event which moves ac-tRNA from the A site to the P site , and domain III from the P site to the E site . Taken together , the data from all mutants suggest that loop 3 has two independent functions to facilitate two elongation factor-driven steps , which depend on loop 3 length and base composition . The anisotropy data show that loop 3 is important for initial ac-tRNA association with the ribosome , but do not directly address eEF2’s role in this process . The decreased ac-tRNA association in mutant IRES–80S ribosome complexes observed in the anisotropy experiment could result from a decrease in spontaneous vacating of the A site , or from decreased TC association even if the A site is available . To address this , we used single-molecule total internal reflection fluorescence microscopy to directly visualize the colocalization of Cy5 fluorophore-labeled Phe-tRNAPhe with Cy3 fluorophore-labeled IRES–80S ribosome complexes ( from yeast ) that had been tethered ( via the IRES RNA ) to the surface of a microfluidic observation flowcell ( Figure 6—figure supplement 1 ) . This colocalization data reports on the ac-tRNA occupancy of the 80S–IRES ribosome complexes . We chose WT and △3 IRESs to study as they exhibited the most differing behaviors in the previous experiments . As expected , addition of just Phe-tRNAPhe ( Cy5 ) +GTP ( without eEFs ) to 80S–IRES ribosome complexes , followed by incubation and subsequent flushing of the flowcell to remove unbound ac-tRNA , revealed very low ac-tRNA occupancies for both WT and △3 IRESs ( Figure 6 ) . When GTP+eEF2 was included with the Phe-tRNAPhe ( Cy5 ) ( but no eEF1A ) the ac-tRNA occupancy of the IRES–80S ribosome complexes formed with WT IRES increased to 9 . 7 ± 2 . 5% , consistent with a low , but enhanced level of eEF1A-independent ac-tRNA binding . When this experiment was repeated with the △3 IRES , we observed a lower ac-tRNA occupancy ( 1 . 5 ± 1 . 1% ) compared to the WT IRES . Higher eEF1A-independent , but eEF2-dependent , ac-tRNA occupancy on WT IRES complexes compared to △3 IRES complexes suggests that the difference between these two IRESs in the anisotropy experiment ( Figure 5C ) is not due to altering eIF1A function . Rather , those data may indicate a decrease in clearing of the A site by the △3 mutant , suggesting the △3 mutant’s main defect is in the first pseudotranslocation and not in the A-site ac-tRNA binding event itself . 10 . 7554/eLife . 08146 . 019Figure 6 . Effect of eukaryotic elongation factor 2 ( eEF2 ) on colocalization of Phe-tRNAPhe ( Cy5 ) with individual 80S ribosome–internal ribosome entry site ( IRES ) complexes formed with either wild type ( WT ) ( Cy3 ) IRES or △3 ( Cy3 ) IRES . Addition of elongation factors and Phe-tRNAPhe ( Cy5 ) ( tRNA ( Cy5 ) ) to 80S ribosome–IRES complexes formed with either ( A ) WT ( Cy3 ) IRES ( black bars ) or ( B ) △3 ( Cy3 ) IRES ( gray bars ) are depicted as percent Cy3-Cy5 colocalized spots . The presence or absence of factor ( s ) is indicated beneath the graphs and error bars represent one standard deviation from the mean . Elongation factors and ribosomes are from yeast . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 01910 . 7554/eLife . 08146 . 020Figure 6—figure supplement 1 . Schematic of the single-molecule colocalization experiments . ( A ) 80S complexes ( with yeast ribosomes ) were assembled on IRES ( Cy3 ) molecules ( hybridized to a biotinylated DNA ) that were tethered to microscope slide surfaces coated with polyethylene glycol ( PEG ) -Biotin via a streptavidin bridge . In the cartoon shown here , the wild-type IRES has undergone the first pseudotranslocation event so that domain 3 sits at the P site and the Phe-tRNAPhe ( Cy5 ) is delivered to the A site by eEF1A ( tRNA from Escherichia coli ) , thereby generating a surface-tethered complex with spatially colocalized Cy3 and Cy5 spots . Sample Cy3 and Cy5 frames from an experiment in which Phe-tRNAPhe ( Cy5 ) was delivered as a ternary complex with eEF1A and GTP , to yeast 80S complexes assembled on WT Cricket Paralysis Virus ( CrPV ) IRES ( Cy3 ) , are depicted in ( B ) and ( C ) , respectively . The imaged Cy3 and Cy5 spots in these frames are false colored as green and red , respectively . ( D ) Superposition of the two frames in which regions that appear to have colocalized green and red spots , just by manual inspection , are false colored as yellow for visual clarity; the actual analysis of the extent of colocalization involves a much more rigorous mathematical treatment of the raw data using home-built codes . The panels below ( B ) – ( D ) show a representative region from the corresponding frames , magnified 6x , to demonstrate the well-resolved distribution of spots and the precision of colocalization . ( E ) Identical images as ( B–D ) , except in the presence of eEF2 , which results in higher levels of colocalization . DOI: http://dx . doi . org/10 . 7554/eLife . 08146 . 020 To examine eEF1A-dependent ac-tRNA delivery , we assembled TC with Phe-tRNAPhe ( Cy5 ) +eEF1A+GTP and delivered this to the immobilized IRES–80S complexes without eEF2 . Compared to the reactions lacking eEF1A , both IRESs show increased and similar ac-tRNA occupancies ( WT: 17 . 9 ± 4 . 8% , △3: 20 . 8 ± 5 . 4% ) . These data initially seem at odds with the anisotropy data in which eEF2-independent ac-tRNA association with 80S–WT IRES ribosome complexes is much greater than complexes with △3 . This apparent discrepancy is likely due to the fact that anisotropy data are obtained under equilibrium conditions where transient interactions are observed , whereas the single-molecule fluorescence data are collected after the flowcell is flushed and thus only show stable long-lived association . Combining the data from both experiments reveals that eEF2-independent ac-tRNA association to WT IRES–80S ribosomes is transient and is inhibited by the △3 mutation . Finally , when eEF2+GTP+TC was delivered to the tethered 80S–IRES ribosome complexes , we observed a dramatic increase in the ac-tRNA occupancy on complexes formed with the WT IRES ( 82 . 8 ± 15 . 7% ) , but not with the △3 IRES ( 26 . 6 ± 10 . 9% ) . This demonstrates that the △3 mutation inhibits the IRES–ribosome complex from using eEF2 to facilitate stable ac-tRNA delivery . Overall , our data suggest that loop 3 is important for eEF2’s ability to catalyze both pseudotranslocations , the first of which moves domain III to clear the A site for ac-tRNA binding and the second which moves the first ac-tRNA to the P site . Our toeprinting experiments performed in RRL and experiments conducted with reconstituted systems show some differences . Specifically , toeprinting with the G-rich and GGC mutants in RRL+CHX shows at least two rounds of translocation ( Figure 3A ) and at least four in RRL+ hygromycin B at low concentrations and post-treatment ( Figure 4B ) . However , in the reconstituted assays these mutants fail before two rounds of pseudotranslocation ( Figure 5B ) . We consider it unlikely that this discrepancy is due to differences in the species of ribosomes used ( purified subunits were made from yeast and shrimp sources , versus rabbit subunits in RRL ) because IGR IRESs function in diverse systems and contact highly conserved ribosome features . A more likely possibility is that the presence or effective concentrations of various components ( ribosomes , ac-tRNAs , GTP , or unidentified factors ) is different in the lysate as compared to the reconstituted system , which may alter the kinetics of the translocation reactions . In addition , the presence of antibiotics such as CHX or hygromycin B ( which we only used in RRL-based experiments ) may suppress the effects of sequence mutation to loop 3 by altering ribosome conformational dynamics ( Wilson , 2014 ) . Despite this uncertainty , taken together our data clearly identify loop 3 as important in more than one round of pseudotranslocation and also illustrate the importance of employing multiple experimental approaches . To function , IGR IRESs must have affinity for the ribosome , promote subunit joining , manipulate elongation factor action , and move through the tRNA binding sites . In this study we show that conformationally dynamic loop 3 in the tRNA-mimicking domain controls two independent , non-canonical translocation events , demonstrating how a viral RNA can carry out intricate ribosome manipulation using dynamic RNA structure . This strengthens the previously postulated idea that structured regions are important for overall IRES architecture and ribosome positioning , whereas conformationally dynamic regions help drive the IRES through the ribosome in elongation factor-dependent steps to initiate translation ( Pfingsten et al . , 2010 ) . The strategy of using a combination of conformationally flexible elements with stably structured domains is likely a strategy used by many RNAs that control dynamic cellular machines . Our data show that the length and sequence of loop 3 are both important for function . A previous study also examined the effect of loop 3 length and sequence on IGR IRES translation efficiency ( Au et al . , 2012 ) . The mutants in that complementary study showed modest defects in translation activity . However , toeprinting results showed that the position of domain III within the ribosome is similar , although differences in toeprint band intensity were sometimes observed . Overall , toeprint band intensity did not correlate well with translation activity , suggesting that something else regulates the modest defects that were identified in that study . Because we discovered mutants with more pronounced translation defects , and whose toeprint intensities did not correlate with translation activity , we could use this to more deeply dissect the specific mechanistic role of loop 3 in more depth using a battery of quantitative analyses . Our data indicate that domain III’s loop 3 is involved in the two non-canonical pseudotranslocation events following initial IGR IRES recruitment of the 80S ribosome . Although domain III was originally proposed to first bind in the P site , the most recent structural and mechanistic models , based on both additional structural information and reexamination of earlier published biochemical data , places domain III in the A site ( Figure 1A ) ( Fernández et al . , 2014; Koh et al . , 2014; Zhu et al . , 2011; Muhs et al . , 2015 ) . In this mechanistic model , initial pseudotranslocation by eEF2 is needed to clear the A site before ac-tRNA can bind the ribosome . Consistent with this , our data and other studies show that stable association of ac-tRNA with the IRES–ribosome complex depends on eEF2 ( Yamamoto et al . , 2007 ) . Additionally , eukaryotic release factor 1 ( eRF1 ) only binds in the A site of IRES–80S ribosome complexes ( and induces a change in the toeprint ) in the presence of eEF2 ( Jan et al . , 2003; Muhs et al . , 2015 ) . However , no pseudotranslocation is observed with pure WT IGR IRES–80S ribosome complexes treated with eEF2 only ( assayed by toeprinting ) ( Pestova , 2003 ) . A mechanistic model that reconciles this observation posits that eEF2 first moves domain III from the A site to the P site , but this is a transient state and without immediate ac-tRNA delivery domain III spontaneously reverse-translocates to the A site ( Fernández et al . , 2014 ) . This is validated by the toeprinting experiment demonstrating one round of translocation in high concentrations of hygromycin B ( Figure 4A ) , which has been shown to potently inhibit reverse translocation ( Borovinskaya et al . , 2008; Szaflarski et al . , 2008 ) . If this explanation is true , the transient position of domain III in the P site would preclude detection of this state by traditional biochemical approaches; possibly , the toeprinting assay itself may facilitate reverse-translocation . This mechanistic model is supported by our data and agrees with all previously published data . Assuming domain III begins in the A site , shortening loop 3 appears to inhibit movement of domain III to the P site before any ac-tRNA is bound . Given that domain III and loop 3 are positioned to interact with components of the 40S subunit head known to be involved in translocation ( ribosomal protein uS13 when domain III is in the A site , for example [Cukras et al . , 2003] ) , our data favor a mechanistic model where the loop 3 length mutants fail to efficiently execute the first pseudotranslocation event and this blocks access of ac-tRNA to the A site . This is supported by the anisotropy data with the non-cognate RNA which show an increase above background levels established by the no-IRES control . This likely indicates the transient binding of the ac-tRNA TC to the A site and subsequent rejection . In comparison , the fact that the △3 mutant yields even lower anisotropy levels than the non-cognate RNA suggests that the TC can never bind the △3 IRES–ribosome complex even transiently . This is consistent with the idea that the initial movement of domain III does not occur with this mutant , either spontaneously or with eEF2 , and domain III remains in the A site . Given that our sequence mutants ( G-rich and GGC ) inhibit the second pseudotranslocation , this interpretation makes loop 3 , despite being a short and apparently conformationally dynamic element , a key player in non-canonical translocation events that move the IGR IRES through all three tRNA binding sites . There is no obvious analogous structure to loop 3 in tRNA , raising the question of how this loop exerts its effects . One possibility is that loop 3 interacts directly with the ribosome in ways not yet clearly observed using structural methods . Recent cryoEM reconstructions of CrPV ( Fernández et al . , 2014 ) and Taura Syndrome Virus ( TSV ) ( Koh et al . , 2014 ) IGR IRESs bound to 80S ribosomes in the pretranslocated ( PRE ) state ( domain III in the A site ) at resolutions of 3 . 8 and 6 Å respectively and of CrPV–80S–eRF complexes in the post-translocation ( POST ) state ( domain III in the P site ) at 8 . 7 Å ( Muhs et al . , 2015 ) provide structural models for loop 3 . However , the local resolution for loop 3 is low in all structures , consistent with conformational dynamics ( Figure 1C–E ) . Interestingly , in the class I ( CrPV ) versus class II ( TSV ) IRESs , loop 3 spans somewhat different space when domain III is in the A site . In both structures , the 3’ ends of loop 3 terminate in the decoding center of the A site where they may interact with elements of the decoding groove . In contrast , the 5’ ends of loop 3 differ in these structural models . In CrPV the 5’ nucleotides of loop 3 wrap around the 5’ terminal nucleotides of the PKI stem in the A site . In the TSV structural model , loop 3 interacts with the apical loop of rRNA helix 24 , part of a constriction between the P and E sites . In bacterial ribosomes this constriction is essential for maintaining the P-site tRNA in its proper place to prevent slipping of the mRNA ( Schuwirth , 2005 ) , and must be remodeled by 30S subunit head swiveling for tRNA to translocate from the P to the E site ( Zhou et al . , 2013; Ratje et al . , 2010 ) . If loop 3 contacts this constriction , it could affect a known structural regulator of translocation , affecting the conformation of the ribosome in a way that favors eEF2 function . In the POST structure with eRFs , loop 3 is modeled to interact with uS7 , a key frame-maintenance and translocation regulator ( Devaraj et al . , 2009; Galkin et al . , 2007; Robert and Brakier-Gingras , 2003 ) . Interestingly , the HCV IRES is also thought to communicate with uS7 ( Fukushi et al . , 2001; Filbin et al . , 2013; Boehringer et al . , 2005 ) , pointing to this ribosomal protein as an important ‘gatekeeper’ to ribosome function that is exploited by viral IRES RNAs . Precisely what loop 3 interacts with , how and when it makes these interactions , and how these interactions affect the conformation of the IRES–ribosome complex remains to be determined , as does the question of whether loop 3 functions differently in the two classes of IGR IRESs . In addition to making contacts to the ribosome , loop 3 could also affect pseudotranslocation by altering the conformational landscape of domain III , which comprises an H-type pseudoknot . Many H-type pseudoknots use adenosines in loop 3 to make minor groove interactions with an adjacent helix . Although no minor groove interactions have been identified in domain III , most IGR IRES loop 3s have adenosine content greater than 40% ( Figure 2—figure supplement 1 ) ; this may be an important feature of loop 3 . Indeed , the G-rich and GGC mutations ( 22% and 33% adenosine , respectively ) show substantially decreased translation activity . Transient or dynamic interactions between the loop and the rest of domain III may be important for altering the conformation of the pseudoknot as it moves through the ribosome . tRNAs are known to undergo substantial conformational changes as they transit through the ribosome ( Dunkle et al . , 2011; Fei et al . , 2011 ) ; loop 3 could help domain III do the same . Alternatively , it may be important for loop 3 to remain unstructured . Indeed , structural probing of these mutants in the unbound form show decreases in loop 3 accessibility to single-stranded ribonuclease ( Figure 4—figure supplement 1 ) . The presence and importance of these changes within the ribosome are unknown , although it is tempting to speculate that a decrease in flexibility may drive the defects observed in this study . There is growing evidence that molecular mimicry is a common tool viruses use to infect their host cells; indeed , several plant viruses display tRNA mimicry in their 3’ untranslated regions ( UTRs ) to enhance viral protein translation ( Dreher , 2010; Simon and Miller , 2013 ) . Yet , molecular mimicry is not limited to structural similarity; the binding partners of these mimics must also be fooled by conformational dynamics and overall molecular interactions . Our work suggests that the flexible elements of the IGR IRES facilitate these additional aspects of mimicry that remain understudied . This discovery that IRES RNA flexibility rather than defined structure is important for function may be particularly important in the context of ribosome manipulation since the ribosome has been suggested to act as a Brownian machine that fluctuates between conformational states ( Frank and Gonzalez , 2010 ) , and thus this and other elements of the translation machinery are highly tuned to respond to and exploit the dynamics of their ligands . The pCrPV1-1 dual LUC vector was a gift from Dr Eric Jan . Reporter vectors containing WT IAPV , Homalodisca coagulata Virus ( HoCV ) , Kashmir Bee Virus ( KBV ) , Himetobi P Virus ( HiPV ) , TSV , Solenopsis invicta Virus-1 ( SInV ) , and Acute Bee Paralysis Virus ( ABPV ) IGR IRES sequences were generated by polymerase chain reaction ( PCR ) amplification of the IRES sequence ( plasmids were gifts from Dr Eric Jan and Dr Sunnie Thompson ) and subsequent ligation into a dual LUC vector ( pDBS , derived from pBluescript , a gift from Dr Les Krushel ) . Mutagenesis was employed using the QuikChange ( Agilent ) method . DNA sequences encoding the RNA for assembly assays ( ‘CrPV4’: full IRES RNA sequence including GCU start codon ) and RNase T1 probing ( ‘CrPV11’: domain III only , no start codon ) were cloned into pUC19-derived vectors with a T7 promoter and a 5’ Hammerhead ribozyme and 3’ hepatitis delta virus ( HDV ) ribozyme flanking the IRES sequence . Constructs for reconstituted functional analysis ( ‘FVKM RNAs’ ) were built by PCR from the CrPV1-1 vector using primers that contained the appropriate mutations and flanked with restriction sites for cloning into pUC19 ( without ribozymes ) . All cloned sequences including the LUC open reading frames were verified by standard sequencing methods using appropriate primers . RNAs for translation assays were in vitro transcribed from XbaI-linearized vectors using the MEGAscript Kit ( Life Technologies , Carlsbad , CA ) . RNA purification was performed by extraction with TriReagent ( Sigma , St . Louis , MO ) followed by chloroform extraction and column purification using the RNeasy Kit ( Qiagen , Germantown , MD ) ( Plank et al . , 2013 ) . RNAs for all other assays were made by in vitro transcription using T7 RNA polymerase and PCR-generated DNA templates , as described previously ( Pfingsten et al . , 2007 ) . These RNAs were purified on 10% polyacrylamide-urea denaturing slab gels , passively eluted at 4°C , then concentrated and buffer-exchanged using appropriate MWCO centrifugal ultrafiltration devices ( Millipore , Billerica , MA ) . All RNAs were assessed for quality using denaturing PAGE . RNAs not made with ribozymes were treated with rAPid alkaline phosphatase ( Roche , San Francisco , CA ) to remove the 5’ triphosphate , whereas no treatment was needed for RNAs made with a 5’ ribozyme or for synthetic primers ( IDT , Integrated DNA Technologies , Coralvile , IA ) , which have a 5’ hydroxyl . RNA was 5’ end-labeled using T4 polynucleotide kinase ( New England Biolabs , Ipswitch , MA ) and 32P-gamma-ATP ( PerkinElmer , Waltham , MA ) , then purified by denaturing gel electrophoresis , eluted , and precipitated as described previously ( Kieft et al . , 1999 ) . Pure dual LUC reporter RNAs were incubated in RRL ( Promega , Madison , WI ) supplemented with 150 mM potassium acetate ( final concentration ) and amino acids for 90 min at 30°C . LUC production was measured using the Dual Luciferase Reporter Assay System ( Promega ) and the GloMax Multi Detection plate reader . Data shown are from five independent experiments . Dual LUC reporter RNAs were body-labeled by including 1 μL of 50 μM ( 40 μCi total ) 32P-alpha-UTP during transcription ( described above ) , treated with TURBO DNase , and then desalted through G50 spin columns ( GE Healthcare , Piscataway , NJ ) . Purified RNAs were diluted in nuclease-free water to 34 , 000 cpm/μL . Equal concentrations were verified by gel electrophoresis and phosphorimaging . For each time point , 2 μL of 34 , 000 cpm/μL dual LUC RNA were added to 8 μL of RRL and incubated at 30°C . These 10 μL reactions were collected at 0 , 10 , 30 , 60 , and 90 min , and were minimally processed by adding 30 μL of nuclease-free water and 40 μL of 2X urea loading buffer . Samples were kept on ice until 50 μL were electrophoresed on an 8% denaturing polyacrylamide gel ( 1 mm gel thickness ) at 40 W for 1 hr and 45 min . The gel was wrapped in plastic and then exposed to a phosphorscreen at -20°C overnight . Phosphorscreens were imaged using a Typhoon scanner and data were analyzed in ImageQuant software by drawing equal sized boxes around the full length RNA at each time point and then normalizing data to the amount of signal in the time=0 sample for each RNA . Data were analyzed by linear regression analysis in Microsoft Excel . For unbound IRES RNAs , 0 . 5 μg of toeprint RNA was mixed with 1 . 5 μL of 10X Toeprint Buffer A ( 1X: 20 mM Tris pH 7 . 5 , 100 mM KOAc , 2 . 5 mM MgOAc2 , 2 mM dithiothreitol [DTT] , 1 mM ATP , 0 . 25 mM spermidine ) , 0 . 5 μL of RNasin Plus ( 40 U/μL , Promega ) , and nuclease-free water to a final volume of 15 μL . For ribosome-bound RNAs ( purified yeast 40S and 60S subunits or purified rabbit 40S ) , reactions were set up in the same way as above but included 8 pmol of each purified subunit . For RRL-incubated RNAs , 11 μL of RRL was pre-incubated with 1 μl of 45 mg/mL CHX or 1 μL nuclease-free water for 5 min at 37oC , and added to RNA and 10X buffer A as above . All reactions were incubated at 30°C for 5 min to allow for folding and binding . Then , 1 μL of 40 , 000 cpm/μL toeprint primer ( internal photinus ) and 24 μL of 1X Buffer A were added and incubated at 30°C for 5 min for primer annealing . Reverse transcription was performed by addition of 4 μL dNTPs ( 1 . 25 mM each ) , 1 μL 320 mM MgOAc2 , and 0 . 5 μL avian myoblastosis virus reverse transcriptase ( 25 U/μL , Promega ) to each reaction . Primer extension proceeded at 30°C for 45 min , and was quenched with 4 μL of 4M NaOH and heated at 85°C for 5 min to hydrolyze RNA . Following this , 100 μl of nuclease-free water was added to each reaction before extraction with phenol:chloroform:isoamyl alcohol ( PCIAA , 24:24:1 , ThermoFisher , Waltham , MA ) , followed by CIAA ( 24:1 ) ( ThermoFisher ) extraction , and ethanol precipitation with 3 volumes of 100% ethanol and 1/10 volume of 3M NaOAc pH 5 . 3 . Pellets were washed with 70% cold ethanol . Precipitated RNA pellets were dried and resuspended to equal counts/μL in 1X TBE + 9M urea loading buffer , and then equal volumes ( typically 10 μL ) were loaded on a 10% polyacrylamide sequencing gel ( 0 . 4 mm gel thickness ) with a sequencing ladder of the WT RNA ( made by dideoxy-NTP incorporation as previously described; Filbin et al . , 2013 ) and electrophoresed at 65 W for approximately 2 hr . Gels were dried and exposed to a phosphorscreen overnight; they were imaged on a Storm scanner ( GE Healthcare ) and analyzed in ImageQuant . ‘Percent translocated’ toeprints were calculated for each RNA in RRL with CHX treatment by quantifying the intensity of the +14/15 toeprint and the +20/21 toeprint in equal sized boxes in ImageQuant , and using these values in the equation: ( +20/21 ) / ( +14/15 + +20/21 ) . Toeprinting assays using concentrated hygromycin B were performed essentially as described above; however 1 μL of 30 mg/mL hygromycin B ( Roche ) was added to the RRL and pre-incubated for 5 min at 37°C . For toeprinting assays in the presence of dilute hygromycin B , 0 . 5 μg of each RNA was incubated for 1 min in RRL/Buffer A/RNasin mix ( as above ) at 30°C before adding 1 μL of 0 . 05 mg/mL hygromycin B ( ‘+’ ) or nuclease free water ( ‘-’ ) . Reactions were incubated at 30°C for 5 min before adding radiolabeled primer and buffer as above . Reverse transcription and gel analysis were performed as described above . Both yeast ( Saccharomyces cerevisiae ) and shrimp ( Artemia salina ) eggs were used as sources of 40S and 60S ribosomal subunits . Yeast subunits were purified from strain YAS2488 ( gift from J . Lorsch ) as described ( Acker et al . , 2007 ) . Briefly , cells were lysed using a liquid nitrogen mill , and clarified lysates were spun through 250 mM sucrose cushions under high-salt conditions to obtain clean 80S ribosomes . Subunits were separated by treatment with puromycin and resolved on 5–20% sucrose gradients . Crude shrimp egg 80S ribosomes were prepared from dried , frozen cysts as previously described ( Iwasaki and Kaziro , 1979; Thiele et al . , 1985 ) with some modifications . After the shrimp cysts were ground open , debris was removed by centrifugation at 30 , 000xg for 15 min and crude 80S ribosomes were precipitated from the supernatant by addition of 4 . 5% ( w/v ) polyethylene glycol ( PEG ) 20K according to previous methods ( Ben-Shem et al . , 2011 ) . Subunits were resolved on 10–30% sucrose gradients after puromycin treatment . eEF1A was purified from yeast according to published methods ( Thiele et al . , 1985 ) . His6-eEF2 was isolated from an overexpressing yeast strain ( TKY675; obtained from Dr Terri Kinzy ) , and purified as described ( Jørgensen et al . , 2002 ) . Rabbit subunits were purified as described ( Kieft et al . , 2001 ) . Preinitiation complexes ( Pre-ICs ) were formed by incubation of shrimp egg 40S and 60S subunits with FVKM IRES RNA constructs at 37°C for 5 min in buffer 4 ( 40 mM Tris-HCl pH 7 . 5 , 80 mM NH4Cl , 5 mM MgOAc2 , 100 mM KOAc , 3 mM β-mercaptoethanol ) . tRNAs were charged with appropriate amino acids as described ( Pan et al . , 2009 ) . Phenylalanine , valine , lysine , and 35S-methionine TCs with purified yeast eEF1A were formed as separate complexes by incubating the relevant charged tRNA ( 1 . 6 μM , based on amino acid stoichiometry ) with eEF1A ( 8 µM ) in buffer 4 supplemented with 1 mM GTP and 1 mM ATP at 37°C for 5 min . Tripeptide complexes were made by mixing Pre-ICs with 1 μM eEF2 and F , V , and K TCs at 37°C for 15 min . Using a quench-flow instrument , tetrapeptide complexes were made by mixing the tripeptide complexes with 35S-Met TC for defined time points on the millisecond scale . Reactions were quenched with 0 . 8 M KOH and peptide was released from tRNA by further incubation at 37°C for 3 hr . Samples were neutralized with acetic acid , lyophilized and suspended in water . Following centrifugation to remove particulates ( which contained no 35S ) , the supernatant was analyzed by thin layer electrophoresis as previously described ( Youngman et al . , 2004 ) . The identities of the tri- and tetrapeptides were confirmed by their comigrations with authentic samples obtained from GenScript ( Piscataway , NJ ) . A further demonstration of tetrapeptide identity was provided by matrix-assisted laser desorption/ionization ( MALDI ) mass spectrometric analysis ( Ultraflex III TOF/TOF , Bruker , Ewing , NJ ) . Phe-tRNAPhe ( prf ) was prepared as previously described ( Wintermeyer and Zachau , 1974; Betteridge et al . , 2007 ) . TC ( 0 . 1 µM , 250 µL ) was incubated with shrimp 80S or shrimp 80S–IRES complex ( 0 . 1 µM , 250 µL ) in buffer 4 for 15 min at 37°C and then kept on ice until anisotropy measurement , which was performed at 23°C . Steady-state fluorescence anisotropy was determined using a Photon Technology International ( PTI , Birmingham , NJ ) QuantaMaster fluorometer with polarizer in L-format , with excitation at 462 ± 2 nm and fluorescence emission collected at 490 ± 2 nm . Instrument-integrated monochromators were used as filters for the fluorescence emission and the excitation light . The g-factor and anisotropy value were calculated using the instrument software as described ( Lakowicz , 1999; Ameloot et al . , 2013 ) . The instrument was calibrated by using suspended nonfat dry milk aqueous solution as scatter . Experimental data were processed and analyzed by Felix software ( from PTI ) . Shrimp 80S–IRES complexes containing Phe-tRNAPhe in the P site were formed by incubation of pre-IC ( 16 pmol ) and Phe-TC ( 32 pmol ) at 37°C for 15 min in the presence of 1 μM eEF2 , in a total volume of 40 µL . The 80S–IRES complexes were isolated by ultracentrifugation at 4°C ( 540 , 000xg ) for 40 min through a 1 . 1 M sucrose cushion , with 600 pmol of pure 30S bacterial ribosome subunits added as carrier to enhance pelleting and allow facile calculation of complex recovery . The pellets were gently washed twice with buffer 4 and dissolved in 100 µL of buffer 4 for A260nm determination . Recoveries typically varied between 60% and 80% . 3H counts from the pellet were measured to determine the amount of [3H]-Phe-tRNAPhe bound to the complex . The percent A-site ( Figure 5C ) and P-site ( Figure 5B ) tRNA binding levels were each divided by the percent of A site binding for the WT , △1 , and △2 mutants , and then multiplied by 100% . This permits analysis of the percentage of A-site tRNA that was moved to the P site for each of these RNAs . WT and △3 IRES RNAs for single-molecule analysis were generated with a 5’ extension of sequence ( 5' ) -CA AAU CAA CCU AAA ACU UAC ACA- ( 3' ) such that a complementary , 3’-biotinylated DNA oligo ( ( 5' ) -TGT GTA AGT TTT AGG TTG ATT TG/3Biotin/- ( 3' ) ) could be hybridized to the IRES constructs . The biotin at the 3’ end of the DNA oligo that had been hybridized to the IRES RNAs could then be used to tether the 80S–IRES ribosome complexes to the polyethylene glycol- , biotin-polyethylene glycol- , and streptavidin-derivatized quartz surface of a microfluidic observation flowcell ( Fei et al . , 2008; Blanchard et al . , 2004; Ha et al . , 2002 ) . The 3’ end of the IRES RNAs contained one codon for Phe ( UUC ) , followed by the hepatitis delta ribozyme to generate a clean 3’ end . 2’-3’ cyclic phosphates were removed as previously described ( Kieft et al . , 1999 ) . IRES RNAs were labeled using Cy3-maleimide ( GE Healthcare ) and the 3’ DNA End-Tag Kit ( Vector Labs , Burlingame , CA ) , which added one additional dG residue harboring the Cy3 label to the 3’ end of the IRES construct . IRES ( Cy3 ) RNAs were purified from free dye by multiple phenol extractions and ethanol precipitation , or centrifugal filtration with a 10 , 000 Da MWCO ( Millipore ) . Labeling efficiencies determined by A260nm and A550nm readings were typically low , ranging from 3% to 20% . A diagram of the RNA constructs is shown in Figure 6—figure supplement 1 . Stocks of IRES ( Cy3 ) RNAs that had been hybridized to the biotinylated DNA oligo were prepared by incubating a 10-fold excess ( 50 nM ) of the 3’-biotinylated DNA oligo with either 5 nM WT IRES ( Cy3 ) or 5 nM △3 IRES ( Cy3 ) RNA ( in a reaction volume of 100 µL ) at 95°C for 2 min , slowly cooling the hybridization reactions to room temperature , transferring the hybridization reactions to ice , aliquoting , flash-freezing in liquid nitrogen , and storing the stocks at -80°C . These stocks , therefore , had 5 nM of either WT IRES ( Cy3 ) or △3 IRES ( Cy3 ) RNA . Purified Escherichia coli tRNAPhe ( Sigma ) was fluorescently labeled with Cy5-NHS ester ( GE Healthcare ) at the primary aliphatic amino group of its naturally modified acp3U47 residue , according to previously published protocols ( Fei et al . , 2010 ) . The labeling reaction was quenched with 0 . 3 M NaOAc ( pH 5 . 2 ) , phenol-chloroform extracted , ethanol precipitated , and the Cy5-labeled tRNAPhe ( tRNA ( Cy5 ) ) was separated from unlabeled tRNAPhe by hydrophobic interaction chromatography ( HIC ) using a TSK gel Phenyl-5PW column ( Tosoh Biosciences , Tokyo , Japan ) attached to an ÄKTA fast protein liquid chromatography ( FPLC ) system ( GE Healthcare ) as previously described ( Fei et al . , 2010 ) . The HIC-purified tRNAPhe ( Cy5 ) was charged with phenylalanine ( Sigma ) as described using E . coli Phe-tRNA synthetase that was overexpressed and purified as previously described ( Fei et al . , 2010 ) . The charging reaction was quenched with 0 . 3M NaOAc ( pH 5 . 2 ) , phenol-chloroform extracted , ethanol precipitated , resuspended in 10 mM ice-cold KOAc ( pH 5 ) , passed through a Micro Bio-Spin Gel Filtration spin-column ( Bio-Rad , Hercules , CA ) , aliquoted , flash-frozen in liquid nitrogen , and stored at -80°C . Charging efficiency was estimated by running an aliquot through a Phenyl-5PW column to detect the charged Cy5-Phe-tRNAPhe and uncharged Cy5-tRNAPhe , separated by HIC . The typical charging efficiency in these reactions was >90% . For each colocalization experiment , IRES–80S ribosome complexes were initially assembled using 1 . 25 nM oligo-hybridized-Cy3-IRES RNA and 100 nM each of yeast 40S and 60S subunits in 1X Eukaryotic Polymix Buffer ( EPB: 50 mM Tris-acetate at pH 7 at 25°C , 100 mM KOAc , 10 mM MgOAc2 , 0 . 5 mM spermidine , and 10 mM β-mercaptoethanol ) . In a separate reaction tube , a TC was prepared using 500 nM Phe-tRNAPhe ( Cy5 ) , 5 µM eEF1A , and 2 mM GTP in 1X EPB . Each of these two reaction tubes were incubated at 37°C for 10 min . Then , 1 µM eEF2 and 2 mM GTP were added to the IRES–80S ribosome complex to initiate the first pseudotranslocation reaction and the reaction was allowed to proceed for an additional 10 min at 37°C ( during which the reaction tube containing the TC was kept on ice ) . Subsequently , the TC was added to the IRES–80S ribosome complex ( containing eEF2 and GTP ) and the entire reaction incubated for another 10 min at 37°C . Finally , the entire reaction was diluted fivefold in 1X EPB and the diluted reaction was delivered into the polyethylene glycol- , biotin-polyethylene glycol- , and streptavidin-derivatized quartz microfluidic observation flowcell ( Blanchard et al . , 2004 ) . The 80S–IRES ribosome complex was incubated in the flowcell for 5 min and components that remained untethered to the surface of the microfluidic flowcell at the conclusion of the 5 min were washed out of the flowcell using an imaging buffer composed of 1X EPB and a protocatechuic acid/protocatechuate-3 , 4-dioxygenase based oxygen scavenging system ( Aitken et al . , 2008 ) . Cyclooctatetraene ( COT , Sigma ) and 0 . 012% v/v 3-Nitrobenzyl alcohol ( NBA , Sigma ) were included as triplet state quenchers in these experiments . Surface-tethered , Phe-tRNAPhe ( Cy5 ) -bound 80S–IRES ribosome complexes were imaged using a custom-built , prism-based total internal reflection fluorescence microscope . Cy3 and Cy5 fluorophores were excited with a 532 nm laser and a 640 nm laser , respectively , with their powers attenuated such that the laser beams measured ~8 mW when they hit the prism . Emission data were directed to the image sensor of an electron-multiplying charge-coupled device ( EMCCD ) camera that records the fluorescence emission as a ~2 min movie with a frame rate of 100 msec . Prior to striking the image sensor of the EMCCD camera , the fluorescence emission from Cy3 and Cy5 are wavelength-separated using dichroic beamsplitters such that they could be directed onto the two separate halves of the image sensor . Colocalization data were analyzed from the imaged frames , using the standard software MetaMorph , as follows: the 256 pixel x 256 pixel imaged frames were split into the green and red halves , each half being 128 pixel x 256 pixel . Spots were picked from the red frame , using automated features in MetaMorph® and designated as ‘Areas’ . The red frames were then stacked on the green frames and the ‘areas’ were transferred from the red to the green frames . Automated algorithms set thresholds to the intensities , assigned geometric coordinates to the spots , calculated the spread of each spot intensity over an average of four adjacent pixels , superimposed each Cy5 frame on the corresponding Cy3 frame and calculated the number of spots that showed significant spatial overlap . This analysis is performed on every frame of the movie captured for a given reaction condition . For the experiments designed to test the effect that the absence of eEF2 , prior to addition of the TC , had on the colocalization , the first 10-min incubation step of the IRES–80S complex with eEF2-GTP was omitted . For these experiments , after imaging the IRES–80S complexes with Phe-tRNAPhe ( Cy5 ) delivered by eEF1A , the same channel was washed three times with 1X EPB to remove all unbound components , and a fresh mix of pre-incubated eEF2-eEF1A-GTP-Phe-tRNAPhe ( Cy5 ) was delivered to the flowcell prior to a second round of imaging aimed at monitoring the rescue of colocalization by addition of eEF2 . Similarly , in experiments targeted to detect the effect of eEF1A on colocalization , eEF1A was not added to the initial reaction tube in which the TC was set up . In this case , after imaging the IRES–80S complexes with Cy5-Phe-tRNAPhe , the channel was washed with 1X EPB , and a fresh mix of pre-incubated TC containing eEF2-eEF1A-GTP-Phe-tRNAPhe ( Cy5 ) was delivered to the flowcell to detect restoration of colocalization . All experiments were performed at least in duplicate and data from at least five movies for each experiment were averaged to calculate the colocalization percentage under a given set of conditions . In 30 mM HEPES-KOH pH 7 . 5 and 10 mM MgCl2 , 1000 cpm of 5’ end-labeled CrPV3 RNAs ( IRES alone , no coding sequence ) were folded by heat-cooling . Folded RNAs were incubated at 37°C in 30 μL RRL containing 1 . 2 mg/mL hygromycin B for 20 min . All samples were diluted in 500 μl ribosome association dilution buffer ( RADB , 50 mM Tris pH 7 . 5 , 50 mM NaCl , 5 mM MgCl2 , 1 mM DTT ) and separated by 15–30% sucrose gradient density fractionation in an SW41 rotor for 3 hr at 36 , 000 rpm , 4°C . Fractions were collected on a BioComp gradient maker and fractionation system . The amount of 32P in each fraction was determined by filter binding and exposure to a phosphorscreen .	Many viruses store their genetic information in the form of strands of ribonucleic acid ( RNA ) , which contain building blocks called nucleotides . Once inside an infected cell , the virus hijacks the cellular structures that build proteins ( called ribosomes ) , which forces the cell to start making viral proteins . Many RNA viruses manipulate the cell’s ribosomes using RNA elements called Internal Ribosome Entry Sites , or IRESs . In a family of viruses called Dicistroviridae , which infect a number of insects , a section of the IRES RNA binds directly to the ribosome . Proteins called elongation factors then trigger a series of events that lead to the cell starting to make the viral proteins . By mutating the RNA of many different Dicistroviridae viruses that infect a variety of invertebrates , Ruehle et al . have now investigated how a particular loop in the structure of the IRES helps to make cells build the viral proteins . This loop is flexible , and interacts with the ribosome to enable the IRES to move through the ribosome . Mutations that shorten the loop or alter the sequence of nucleotides in the loop prevent the occurrence of two of the steps that need to occur for the cell to make viral proteins . Both of these steps depend on elongation factors . Determining how the entire IRES might change shape as it moves through the ribosome is an important next step , since the ribosome is exquisitely sensitive to the shape and motions of its binding partners .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	A dynamic RNA loop in an IRES affects multiple steps of elongation factor-mediated translation initiation
The standard reference Caenorhabditis elegans strain , N2 , has evolved marked behavioral changes in social feeding behavior since its isolation from the wild . We show that the causal , laboratory-derived mutations in two genes , npr-1 and glb-5 , confer large fitness advantages in standard laboratory conditions . Using environmental manipulations that suppress social/solitary behavior differences , we show the fitness advantages of the derived alleles remained unchanged , suggesting selection on these alleles acted through pleiotropic traits . Transcriptomics , developmental timing , and food consumption assays showed that N2 animals mature faster , produce more sperm , and consume more food than a strain containing ancestral alleles of these genes regardless of behavioral strategies . Our data suggest that the pleiotropic effects of glb-5 and npr-1 are a consequence of changes to O2 -sensing neurons that regulate both aerotaxis and energy homeostasis . Our results demonstrate how pleiotropy can lead to profound behavioral changes in a popular laboratory model . It is tempting to compare the endless forms of life and create adaptive hypotheses to explain their differences . Why are polar bears white ? Perhaps as camouflage for when they hunt . Or perhaps to make it easier to absorb heat from the sun . Both explanations make sense , but designing experiments to distinguish between these possibilities is not trivial . Further , as Gould and Lewontin critiqued , relying on adaptive evolution as the sole explanation for phenotypic change while ignoring alternative explanations such as genetic drift , adaptive constraints , or pleiotropy does not follow Darwin’s pluralistic approach ( Gould and Lewontin , 1979 ) . In the current era , inexpensive next-generation sequencing and increasingly sophisticated bioinformatics analysis enable the identification of causative mutations with signatures of selection , yet it is difficult to determine why these alleles are under selection . Indeed , the pervasive effects of pleiotropy means that signatures of selection alone are not enough , adaptive hypotheses must be tested directly . Experimental evolution offers a route to test hypothesis directly ( Fisher and Lang , 2016; Lenski , 2017; Teotónio et al . , 2017 ) . The ability to manipulate model organisms in the lab provides a greater opportunity to test adaptive hypotheses beyond arguments of plausibility and address the role of these other competing themes in the evolution of biological traits . These studies are also useful for understanding how organisms adapt to laboratory conditions . Since the fundamental work of Gregor Mendel elucidating the laws of genetic transmission , model organisms have enabled experimenters to gain fundamental insights into many biological processes . Modern research tools are facilitating the use of new and unusual species to analyze longstanding biological questions ( Alfred and Baldwin , 2015; Gladfelter , 2015; Goldstein and King , 2016; Russell et al . , 2017 ) . More and more species are reared in the laboratory as models for biological traits of interest . An issue for these approaches , particularly for comparative analysis or for those addressing evolutionary questions , is the extreme shift in environment and associated selective pressures that these populations experience . All species evolve through the process of natural selection and genetic drift; many model organisms have evolved by exposure to the novel and artificial conditions experienced in the lab ( Orozco-terWengel et al . , 2012; Duveau and Félix , 2012; Goto et al . , 2013; Kasahara et al . , 2010; Marks et al . , 2010; Stanley and Kulathinal , 2016; Yvert et al . , 2003 ) . Understanding the process of adaptation of wild populations to captivity is necessary to understand how the genetic , developmental , and neural circuits are changed in these laboratory populations . As a model for understanding laboratory adaptation in a multicellular organism , we have focused our studies on the N2 strain of Caenorhabditis elegans . N2 is the canonical reference strain used by hundreds of C . elegans labs across the world . While this strain was introduced to the genetics research community by Sydney Brenner in 1974 ( Brenner , 1974 ) , it was actually isolated by L . N . Staniland and Warwick Nicholas from mushroom compost in 1951 , spending multiple decades ( ~300–2000 generations ) in two primary growth conditions: on agar plates where bacteria was its primary food source or in liquid axenic media ( Sterken et al . , 2015 ) . A small number ( ~100 ) of new mutations that arose and fixed in the N2 strain following isolation from the wild have been identified ( McGrath et al . , 2011 ) , including a neomorphic , missense mutation in the neuropeptide receptor gene npr-1 and a recessive , 765 bp duplication in the nematode-specific globin gene glb-5 . These mutations were originally identified for their role in foraging and aerotaxis behaviors and were initially thought to represent natural genetic variants ( de Bono and Bargmann , 1998; Persson et al . , 2009 ) ( Figure 1a ) . A large body of work has found that these genes regulate the activity of the URX-RMG neuronal circuit that controls O2 responses on food ( Chang et al . , 2006; Coates and de Bono , 2002; Gray et al . , 2004; Macosko et al . , 2009; McGrath et al . , 2009; Persson et al . , 2009 ) . Animals with the ancestral alleles of npr-1 and glb-5 prefer ~10% O2 concentrations while foraging and follow O2 gradients to the border of bacterial lawns ( ~12% O2 ) and feed in groups ( called social behavior ) ; animals containing the derived alleles of these genes are less sensitive to 10–21% O2 gradients in the presence of food and feed alone ( called solitary behavior ) ( Chang and Bargmann , 2008; Chang et al . , 2006; Gray et al . , 2004 ) . We have previously proposed that the derived alleles of glb-5 and npr-1 were fixed by selection as solitary animals are more likely to be picked when propagating animals to new plates ( McGrath et al . , 2009 ) . However , aggregation behavior in the ancestral npr-1 strain appears to create local food depletion leading to a weak starvation state , which reduces reproduction and growth ( Andersen et al . , 2014 ) . Potentially , this starvation difference could be responsible for the fitness differences of the strains . Consistent with both hypotheses , a number of experimental crosses or competition experiments between parental strains that are polymorphic for npr-1 have resulted in enrichment of the derived allele of npr-1 , suggesting it confers a fitness advantage under standard lab husbandry ( Gloria-Soria and Azevedo , 2008; Noble et al . , 2017; Weber et al . , 2010 ) . In order to distinguish between these hypotheses , we performed pairwise competition experiments following a number of environmental and/or genetic manipulations . Surprisingly , our results suggest that neither hypothesis is correct . While the derived npr-1 and glb-5 alleles increase fitness of animals on agar plates , the differences in social vs . solitary behavior are not necessary for their differences in fitness . Instead , our work suggests that fitness gains are due to increases in food consumption and changes in reproductive timing , mediated by O2-sensing body cavity neurons that are also required for social feeding behaviors . Our work demonstrates that even when alleles are identified that confer fitness advantages , care must be taken in inferring the phenotypes that are responsible due to the pleiotropic actions of genetic changes . In previous reports , we have used multigenerational pairwise competition experiments to compare the relative fitness of two strains ( Figure 1b ) utilizing Droplet Digital PCR with a custom TaqMan probe to quantify the proportion of each genotype ( Evans et al . , 2017; Greene et al . , 2016; Large et al . , 2016 ) . To quantify this change , we used a generic selection model to estimate the relative fitness difference ( w ) between the two strains ( Figure 1c ) . In this context , relative fitness measures the generational change in relative abundance of each of the two strains . We also used CRISPR-enabled genome engineering to create strains with a silent mutation in the dpy-10 gene using a previously published high-efficiency guide RNA ( Figure 1d ) ( Arribere et al . , 2014 ) , which we will refer to as barcoded strains . These strains allow us to use a common Taqman probe to quantify the relative fitness of a test strain against these barcoded strains . We confirmed that the dpy-10 silent mutation had no statistically significant effect on fitness in two genetic backgrounds studied throughout this report ( Figure 1e ) . In order to test the fitness effect of the derived alleles of npr-1 and glb-5 , we utilized three previously described near isogenic lines ( NILs ) containing ancestral alleles of npr-1 ( QG1 ) , glb-5 ( CX10774 ) , or both genes ( CX12311 ) introgressed from the Hawaiian CB4856 wild strain into the standard N2 background ( Bernstein and Rockman , 2016; McGrath et al . , 2009; McGrath et al . , 2011 ) . The npr-1 introgressed region is ~110 kb in size and the glb-5 introgressed region is ~290 kb in size . For brevity , we will refer to genotype of these introgressed regions throughout this report by the ancestral/derived allele they contain ( e . g . the ancestral allele of npr-1 vs the introgressed region containing the ancestral allele of npr-1 ) . For clarity , we will refer to the NILs colloquially using the ancestral introgression ( s ) they contain instead of their opaque strain names ( i . e . N2 = N2 , CX10774 = N2glb-5 , QG1 = N2npr-1 , and CX12311 = N2glb-5 , npr-1 ) . If needed , readers can find the strain name used in each figure in the supplemental source data files . In contrast to the N2 strain , the N2glb-5 , npr-1 strain aggregates at the border of bacterial lawns where O2 levels are lowest due to the increased height of the bacterial lawn . We confirmed previous reports that both the derived alleles of npr-1 and glb-5 suppress bordering behavior to varying degrees ( Bendesky et al . , 2012; de Bono and Bargmann , 1998; McGrath et al . , 2009 ) ; npr-1 accounted for the majority of the difference with glb-5 playing a modulatory role ( Figure 2a ) . To compare the relative fitness of the four strains , we competed each strain against the barcoded N2glb-5 , npr-1 strain , transferring animals each generation by washing to minimize potential sources of investigator bias toward picking social or solitary animals ( Figure 2b ) . The N2 strain was the most fit in these conditions , with a relative fitness ( w ) of ~1 . 30 . Interestingly , the fitness effects of the glb-5 and npr-1 regions were epistatic - the derived allele of glb-5 increased the relative fitness in the derived npr-1 background but showed no effect in the ancestral allele of npr-1 . The derived npr-1 allele increased fitness in both backgrounds of glb-5 . To confirm the fitness advantage of the derived glb-5 allele in the derived npr-1 background , we also competed the N2glb-5 strain against the barcoded N2 strain ( Figure 2b ) . The estimated selective coefficient ( a common measure of the fitness difference of a beneficial allele ) of the glb-5 allele in the npr-1 derived background was s = 0 . 10 ( 0 . 06–0 . 13 95% confidence interval ) , the estimated selective coefficient of the npr-1 allele in the glb-5 ancestral background was s = 0 . 17 ( 0 . 12–0 . 23 95% confidence interval ) , and the estimated selective coefficient of the npr-1 allele in the glb-5 derived background was s = 0 . 30 ( 0 . 27–0 . 34 95% confidence interval ) . These selective coefficients are comparable to beneficial alleles identified in other organisms , such as the haplotype responsible for lactase persistence ( ~0 . 01–0 . 19 ) ( Bersaglieri et al . , 2004 ) and the sickle-cell trait ( 0 . 05–0 . 18 ) in humans ( Li , 1975 ) . While the introgressions surrounding the npr-1 and glb-5 genes are relatively small , these NIL strains carry additional polymorphisms in surrounding genes from the CB4856 strain . We also performed competition experiments using two previously published npr-1 loss-of-function alleles ( ad609 and ky13 ) ( de Bono and Bargmann , 1998 ) against the N2 barcoded strains . Both the npr-1 ( ad609 ) and npr-1 ( ky13 ) loss-of-function alleles decreased the animal’s relative fitness in an amount comparable to the ancestral allele ( Figure 2c ) . We did not perform similar experiments on the glb-5 gene . Altogether , our work suggests that the npr-1 derived allele increases fitness of animals in laboratory conditions and also suggests that the derived allele of glb-5 increases the fitness of animals in a npr-1-dependent manner . Animals with reduced function of npr-1 sense environmental O2 levels and aerotax towards their preferred O2 levels ( 10% ) in the presence of foods , which results in aggregation of animals at the borders of the lawn ( Chang et al . , 2006; Cheung et al . , 2005; Gray et al . , 2004 ) . This behavior can be suppressed by lowering environmental O2 levels to the animals preferred O2 concentrations ( Gray et al . , 2004 ) . We decided to use this environmental manipulation to test the hypothesis that the social foraging behavior was necessary for the fitness disadvantage experienced by strains containing the ancestral alleles of npr-1 and glb-5 . Our above experiments hinted that this hypothesis might be incorrect as the derived glb-5 allele reduced bordering behavior in the ancestral npr-1 background without an associated increase in fitness . We first confirmed that we could suppress the bordering behavior differences between N2glb-5 , npr-1 and N2 by reducing environmental O2 levels to 10% or 3% using a Biospherix chamber ( Figure 3a and Videos 1–4 ) . N2glb-5 , npr-1 animals did not form any social groups in the center of the lawn at the lowered O2 levels and were also indistinguishable from N2 by visual inspection . We also verified that this O2 manipulation also suppressed roaming/dwelling behavior ( Figure 3b ) . While feeding , C . elegans worms alternate between bouts of active exploration ( roaming ) and periods of inactive movement ( dwelling ) . Animals that are mutant for npr-1 show increased amounts of roaming behavior ( Stern et al . , 2017 ) . Despite the behavioral similarity of these animals at these lower O2 levels , the relative fitness differences between the N2 and N2glb-5 , npr-1 strains remained ( Figure 3c ) . To further confirm that aggregation behavior was not necessary for the fitness differences , we also performed competition experiments on uniform bacterial lawns ( UBLs ) , which are constructed so that the entire plate is covered with a thin bacterial lawn to remove the O2 gradients created by the unequal thickness of bacteria in normal lawns . UBLs have been used to suppress npr-1-dependent differences in survival in response to bacterial pathogens ( Reddy et al . , 2009 ) ; however , the UBLs were unable to suppress the fitness advantage of N2 animals ( Figure 3d ) . Animals that carry the ancestral npr-1 allele can burrow into agar when food is depleted ( de Bono and Bargmann , 1998 ) , raising the possibility that the fitness gains of N2 could be a result of the transfer process , which selects for animals on the surface of plates . While visual inspection of the two strains at 10% and 3% did not reveal any obvious differences in burrowing behavior , we also tested the role of burrowing in the fitness differences more rigorously by using modified nematode growth plates that contain agarose that prevents burrowing ( Andersen et al . , 2014 ) . The relative fitness differences between N2 and N2glb-5 , npr-1 remained unchanged ( Figure 3d ) . Finally , we tested whether differences in resistance to infection could be responsible for the differences in fitness . The E . coli bacterial strain that is used to feed C . elegans , OP50 , can also infect and kill animals , resulting in a decreased lifespan ( Garigan et al . , 2002; Gems and Riddle , 2000 ) . Both glb-5 and npr-1 have been implicated in innate immunity and survival to pathogen exposure ( Andersen et al . , 2014; Reddy et al . , 2009; Styer et al . , 2008; Zuckerman et al . , 2017 ) . However , the fitness advantage of the N2 strain compared to the N2glb-5 , npr-1 strain remained when animals were competed against each other on OP50 bacteria killed by ultraviolet radiation ( Figure 3e ) . The relative fitness on killed OP50 bacteria was slightly decreased; however , this could reflect differences in population demographics , as the killed OP50 supported less overall growth per plate . These experiments motivated us to also test the relative fitness differences of 11 other wild strains isolated from different parts of the world using strains provided by the C . elegans Natural Diversity Resource ( Cook et al . , 2017 ) . Each strain was competed against a barcoded N2glb-5 , npr-1 . Consistent with their npr-1 genotype , these wild strains all aggregated at the borders of the bacterial lawn ( Figure 4a ) , but their relative fitness differences varied wildly ( Figure 4b ) . The relative fitness of two of the strains ( CB4856 and DL238 ) was greatly reduced compared to the N2glb-5 , npr-1 strain . The relative fitness of five of the strains were comparable to the N2 . The relative fitness of the remaining four strains was statistically indistinguishable from the barcoded N2glb-5 , npr-1 . These results further support that social behavior is not the major determinant of fitness levels in laboratory conditions . To gain more insight into the phenotypes that could be responsible for the fitness increases of the N2 strain , we performed RNA sequencing to analyze the transcriptomes of bleach-synchronized N2 and N2glb-5 , npr-1 animals grown in either 10% O2 or 21% ambient O2 levels . Animals were allowed to develop to the L4 stage and harvested at identical times . We first performed Principal Component Analysis ( PCA ) analysis on differentially expressed genes to analyze how the environmental and genetic differences globally regulated the transcriptomes of the animals . If environmental O2 and the genetic background had independent effects on the transcriptomes , we expected to find two major components in the PCA analysis . However , the PCA analysis identified a single component that explained the majority of the variance ( 77 . 9% ) . The genetic and environmental perturbations had similar effects on the first component in an additive manner ( Figure 5a ) . Reducing O2 levels from 21% to 10% had similar effects on the transcription profiles as changing the background from N2glb-5 , npr-1 to N2 . Consequently , the animals that differed in both genetic background and environmental O2 levels ( N2–21% O2 vs N2glb-5 , npr-1–10% O2 ) also showed the most similar transcriptional profiles . These patterns were also seen in Hierarchical Clustering using the 1202 differentially expressed genes ( Figure 5b ) . These results suggest that the foraging behavioral differences are not responsible for the underlying transcriptomics differences between the different strains and environmental conditions . The effects of the derived npr-1 and glb-5 alleles mimics the effects of lowering environmental O2 from 21% to 10% . To further gain insight into this connection , we plotted the average transcriptional change between the strain backgrounds vs the average transcriptional change between the environmental O2 concentrations for each gene ( Figure 5c , Supplementary file 1 ) . Surprisingly , we observed a bimodal distribution of values , with a cluster of 652 genes centered at 1 . 2 log2-fold change ( Figure 5c – red circle ) . This is unexpected , as it suggests that the environmental and genetic perturbations had identical effects on transcription for all these genes . When we inspected this list of genes , we noticed a large number of genes that are known to be involved in spermatogenesis . We further investigated the developmental regulation of these 652 genes using previously published transcriptomics data isolated from hermaphrodites or males at specific developmental time points ( Boeck et al . , 2016 ) ( Figure 5d ) . The expression of the majority of these genes peaked during the L4 stage in hermaphrodites , was further enriched in L4 males , and suppressed in somatic cells isolated from L4 animals . These observations are consistent with this cluster of genes being involved in spermatogenesis , which occurs during the L4 stage ( when RNA was isolated ) in hermaphrodite animals . We reasoned that the transcriptomics data could indicate a difference in the relative timing of spermatogenesis and/or the number of sperm that are produced in each genetic background/environmental condition . L1 larval stage animals were synchronized; subsequent differences in developmental speed would result in animals in slightly different stages of L4 . To test this , we synchronized N2glb-5 , npr-1 and N2 animals , placed them in 10% or 21% environmental O2 , and identified the number animals containing mature sperm at 2 hr intervals from 48 to 56 hr . N2 animals began spermatogenesis approximately 2 hr earlier than the N2glb-5 , npr-1 animals , regardless of the environmental O2 levels ( Figure 5e ) . Hermaphrodites undergo spermatogenesis for a fixed period of time before permanently switching gametogenesis to the production of oocytes , resulting in the development of a fixed number of self-sperm that are stored in the spermathecae ( Hubbard and Greenstein , 2005 ) . To test whether these strains produced the same number of sperm , we used DAPI staining to count the number of sperm found in the spermathecae . Not only did N2 animals start spermatogenesis earlier , they also produced more sperm ( Figure 5f ) . The total fecundity of N2 hermaphrodites that do not mate with males is determined by the number of self-sperm . We confirmed that the difference in self-sperm number also resulted in a larger overall brood size ( Figure 5g ) and as expected from computational modeling ( Large et al . , 2017 ) , an increased rate of egg-laying later on in life ( Figure 5h ) . The timing of sexual maturity is an important factor in determining the fitness of animals . We also tested whether the differences in timing of spermatogenesis could lead to differences in when fertilized eggs are produced . We performed similar experiments as above and monitored the time fertilized eggs could be observed in the uterus at two-hour intervals . Again , we observed a difference in N2 and N2glb-5 , npr-1 animals at both 10% and 21% environmental O2 levels . N2 animals were observed to contain fertilized eggs approximately 1 hr earlier that N2glb-5 , npr-1 animals ( Figure 5i ) . The difference in timing of spermatogenesis and fertilization ( 2 hr vs 1 hr ) , potentially reflects the fact that N2 animals produce more sperm before switching to oogenesis . These experiments suggest that the differences in transcription between N2 and N2glb-5 , npr-1 could be caused by differences in sexual maturity . We are unable , however , to explain the differences in transcription we observed between 10% and 21% O2 as mature sperm was observed at similar times in these different environmental conditions ( Figure 5e ) . Potentially , the rate of spermatogenesis or expression levels of genes are modified by O2 levels that are not reflected in the timing of the presence of mature sperm . Life-history tradeoffs have been proposed in evolutionary theory to account for the linkage between two different traits . Assuming an individual can acquire a finite amount of energy , the investment of energy into one trait leads to consequential changes in other traits as energy resources are shunted into different directions . For example , artificial selection experiments on early fecundity in C . elegans resulted in decreased reproduction late in life ( Anderson et al . , 2011 ) . The N2 strain seems to violate this tradeoff , as it sexually matures earlier than N2glb-5 , npr-1 , but also produces more eggs later on in life . We measured the size of N2 and N2glb-5 , npr-1 animals and found that N2 animals were also larger than N2glb-5 , npr-1 animals at synchronized time points ( Figure 6a ) . These observations suggest that the assumption of a fixed energy acquisition for N2glb-5 , npr-1 and N2 might be violated . This would be consistent with Andersen et al’s observation that metabolism genes were upregulated by the derived npr-1 allele , which they proposed represented differences in food intake ( Andersen et al . , 2014 ) . It would also be consistent with the role of orthologs of npr-1 in other species . npr-1 encodes an ortholog to neuropeptide Y receptors , which are reported to regulate feeding behavior in fishes , birds , and mammals ( Ando et al . , 2001; Lecklin et al . , 2002; Matsuda , 2009 ) . To test this hypothesis , we first utilized a previously described feeding assay to measure the ability of a strain to clear E . coli OP50 bacteria from liquid S-media ( Gomez-Amaro et al . , 2015 ) . In this assay , individual wells are seeded with a defined number of bacteria and 20 worms . Each day , the optical density of each well is measured to estimate the amount of food consumed by the worms . In these conditions , N2 cleared the bacteria faster than N2glb-5 , npr-1 animals ( Figure 6b ) . While these assays supported our hypothesis , liquid media is fundamentally different from the conditions experienced on agar plates , making it difficult to generalize the results from one condition to the other . To this end , we developed a new food consumption assay on agar media in 24-well plates . In this assay , each well was seeded with a defined amount of OP50-GFP , which we found could be quantified in a linear manner using a plate reader ( Figure 6c ) . When we tested N2 and N2glb-5 , npr-1 animals in 10% or 21% environmental O2 levels , we found N2 consumed more food than N2glb-5 , npr-1 in both environmental conditions ( Figure 6d ) . Interestingly , we found animals grown in 10% O2 also consume more food than animals grown in 21% O2 . These experiments indicate that N2 animals consume more food than N2glb-5 , npr-1 . We next decided to test whether the derived allele of npr-1 could increase the fitness and feeding rate in a different genetic background . We used the CB4856 wild strain isolated from pineapple fields in Hawaii , which has relatively low relative fitness in laboratory conditions ( Figure 4b ) , taking advantage of a previously constructed NIL of npr-1 introgressed from N2 into the CB4856 background ( CX11400 ) ( Bendesky et al . , 2012 ) ( Figure 6e ) . We found that the N2 region surrounding npr-1 also conferred a fitness advantage in the CB4856 background ( Figure 6f ) . The estimated selective coefficients of the derived allele of npr-1 was higher in the CB4856 background than the N2 background ( s = 0 . 61 vs s = 0 . 30 ) , potentially due to the lower relative fitness of the CB4856 strain . The food consumption of these strains was consistent with the fitness differences ( Figure 6g ) . The derived allele of npr-1 increased food consumption in both genetic backgrounds but its effect was higher in CB4856 . Food is consumed from the environment by the periodic contraction and relaxation of the pharyngeal muscle which serves to bring material from the environment into the pharynx and filter out bacterial cells ( Fang-Yen et al . , 2009 ) . To test whether the increase in food consumption could be explained by an increase in the rate of pumping , we measured the pharyngeal pumping rate of the N2glb-5 , npr-1 , N2 , CB4856 , and CX11400 strains . The effects of the derived allele of npr-1 was epistatic with respect to the N2 or CB4856 background . The derived allele decreased the pumping rate in the CB4856 background but had no effect in the N2 background ( Figure 6h ) . The effect of the derived allele of npr-1 on pumping rate is surprising . Pumping rate is often used as a proxy for food consumption; our results indicate that increased pharyngeal pumping does not necessarily lead to increases in food consumption . We also measured a number of size parameters of the pharynx but found no obvious differences that could account for the increased food consumption ( Figure 6—figure supplement 1 ) . Potentially , the pharynx is more efficient at bringing food in from the external environment due to stronger pump strength , more efficient filtering processes or other unknown behavioral differences that contribute to food intake . We next decided to gain insight into the cellular mechanisms by which npr-1 and glb-5 increased fitness of the strains . Previous studies have shown that npr-1 and glb-5 regulate social behavior through the URX-RMG neuronal circuit ( Figure 7a ) . glb-5 tunes O2-sensititivies of the URX oxygen-sensing neuron pair through regulation of O2-sensing guanylyl cyclases , leading to changes in influx of Ca++ into the cell body ( Abergel et al . , 2016; Gross et al . , 2014; McGrath et al . , 2009; Oda et al . , 2017; Persson et al . , 2009 ) . The derived allele of npr-1 inhibits the activity of the RMG hub interneuron which suppresses aerotaxis and social behavior ( Laurent et al . , 2015; Macosko et al . , 2009 ) . The RMG neurons connect to URX and a number of other sensory neurons through gap junctions , which are necessary for foraging behaviors ( Jang et al . , 2017 ) . URX neurons also integrate O2 with internal nutrient reserves ( Witham et al . , 2016 ) . To test the role of URX in the fitness gains of the npr-1 and glb-5 derived alleles , we used the qaIs2241 integrated cassette that specifically kills the O2-sensing neurons URX , AQR and PQR ( Chang et al . , 2006 ) . We crossed this cassette into the N2npr-1 , N2glb-5 and N2glb-5 , npr-1 strains and repeated the pairwise competition experiments performed in Figure 2a using strains that now also contained the qaIs2241 cassette . In all cases , the relative fitness gains of the derived alleles were decreased by the presence of the neuronal ablation ( Figure 7b ) . These experiments suggest that the derived alleles either activate or disinhibit the URX , AQR , and or PQR neurons which leads to increases in fitness . To distinguish between these possibilities , we competed N2 and N2glb-5 , npr-1 strains with and without the qaIs2241 against each other . Strains that carried the qaIs2241 cassette were dramatically less fit than the control worms , suggesting that URX , AQR , and PQR promote fitness in laboratory conditions ( Figure 7c ) . We and others have shown that glb-5 and npr-1 are pleiotropic , regulating social behavior and food consumption . Potentially this pleiotropy arises from the ability of the URX , AQR , and PQR neurons to these biological traits . To test this , we phenotyped strains that carried the qaIs2241 cassette for social behaviors , food consumption and reproductive timing ( Figure 7d–g ) . These experiments indicated that these neurons are required for each of these three traits . Interestingly , food consumption in the qaIs2241 strains was reduced without a corresponding change in pharyngeal pumping rate , further confirming that these phenotypes could be separated from each other at a genetic and cellular level . We also decided to test whether ascaroside pheromones were necessary for the fitness differences between N2 and N2glb-5 , npr-1 . Nematodes release a number of ascaroside molecules , which are in turn sensed by a distributed neural circuit that integrates and modifies a number of behavioral and developmental phenotypes ( Butcher , 2017; Ludewig and Schroeder , 2013 ) . There are a few reasons to think that ascaroside pheromones might be involved in the fitness gains of the N2 strain . First , work by Andersen et . al indicated that population density directly impacts lifetime fecundity and adult body length differences between N2 and CB4856 strains ( Andersen et al . , 2014 ) . Second , our previous studies of C . elegans domestication to liquid cultures has found that pheromone signaling was modified by fixed genetic changes ( Large et al . , 2016; McGrath et al . , 2011 ) . Finally , the derived alleles of npr-1 and glb-5 have been shown to modify pheromone valence in a variety of contexts ( Fenk and de Bono , 2017; Jang et al . , 2012; Macosko et al . , 2009; Oda et al . , 2017 ) . To test the role of ascaroside pheromones , we followed previous publications using a genetic knockout of the daf-22 gene , which encodes a peroxisomal enzyme required for the biosynthesis of C . elegans pheromones ( Butcher et al . , 2009 ) and accumulation of lipid droplets ( Zhang et al . , 2010 ) , using CRISPR-Cas9 enabled genome editing to create a large deletion of daf-22 in the N2 strain , which was then crossed to the N2glb-5 , npr-1background . Competition experiments demonstrated that daf-22 was necessary for the fitness advantage of derived npr-1 and glb-5 alleles ( Figure 8a ) . In addition , daf-22 was necessary for the faster sexual maturity ( Figure 8b ) and increased food intake ( Figure 8c ) of the N2 strain compared to N2glb-5 , npr-1 . These data suggest that npr-1 and glb-5 reprogram pheromone responses resulting in increased sexual maturity and ability to consume food . daf-22 encodes a peroxisomal fatty acid β-oxidation enzyme . Besides its role in the biosynthesis of ascaroside pheromones , daf-22 has recently been shown to play a distinct role in ASK neurons , where it is required for the metabolization of fatty acids that stimulate the endoplasmic reticulum stress response , promoting the transcription of insulin-like peptides that regulate dauer formation and other biological processes ( Park and Paik , 2017 ) . daf-22 mutants also accumulate massive amounts of fatty acids and fatty acyl-CoAs in their intestines ( Butcher et al . , 2009; Joo et al . , 2009; Li et al . , 2016 ) , which can potentially regulate feeding behavior through homeostasis mechanisms ( Hyun et al . , 2016 ) . To determine if the differences observed in the daf-22 mutants were caused by the lack of ascaroside pheromones , we attempted to rescue these phenotypes using two concentrations of crudely purified pheromones isolated from animals grown in liquid cultures . Neither of these concentrations were able to rescue the differences in food consumption or reproductive timing ( Figure 8d , e ) . These experiments suggest that the effects of the daf-22 mutants we have observed might be independent of their role in producing ascaroside pheromones . In this report , we studied the fitness consequences of two derived alleles that arose and fixed in the N2 strain after isolation from the wild . We find that both alleles can be adaptive , with selective coefficients that are larger than many characterized beneficial alleles from other species . These results are consistent with the derived alleles spreading through the ancestral N2 populations due to positive selection . If this was true , it would suggest that the derived allele of npr-1 arose first , as the derived glb-5 allele is only beneficial in this derived genetic background . However , the demographic history and laboratory environment of how N2 was grown at the time these alleles arose is largely lost ( Sterken et al . , 2015 ) . The exact laboratory growth conditions ( liquid axenic vs . solid media ) , transfer processes ( picking vs . chunking ) and effective population sizes ( between 4 and 1000 ) used to propagate a C . elegans strain is incredibly variable . It is likely that the evolutionary forces responsible for the fixation of these alleles will remain lost to history . Nevertheless , the ability of positive selection to act upon the derived npr-1 allele can be observed in current experiments . A recent example is provided by Noble and colleagues , who created a large mapping population between 16 parental strains ( including N2 and CB4856 ) to create a large panel recombinant inbred lines ( RILs ) ( Noble et al . , 2017 ) . During the outcrossing phase of construction , the N2 allele of npr-1 spread through the population to fixation , consistent with its dominant action and the strong selective advantage of this allele . Potentially , variation in npr-1 affected allele frequencies of unlinked loci as well . For example , an excess of CB4856 haplotypes was observed in the RILs , suggesting that CB4856 haplotypes were more likely to contain beneficial alleles . Our measurements of the relative fitness of the CB4856 strain , however , creates an apparent paradox , as CB4856 was one of the least fit strains among the wild strains we tested ( Figure 4b ) . Potentially , epistatic interactions between CB4856 alleles and the derived allele of npr-1 could help resolve this; the effect of npr-1 on food intake and fitness is higher in the CB4856 background ( Figure 6f , g ) . Differences in effect size of a focal allele in different genetic backgrounds is considered evidence for the existence of epistasis ( Gibson and Dworkin , 2004 ) . Potentially , the presence of laboratory-derived alleles in mapping populations will skew not only the allele frequencies of these beneficial alleles , but also natural genetic variants that interact epistatically with them . Evolution of behavioral traits is one strategy for animals to respond to a new environment . The identification of a polymorphism in npr-1 has served as an example of how behavioral variation can arise from genetic variation . However , our work suggests that the social/solitary feeding behavioral changes of N2 are not sufficient for explaining its fitness gains . Rather , we propose that changes to food intake , sexual maturity , and fecundity are more important . One unresolved question is why wild strains do not eat as much food as the N2 strain . We believe there must be some sort of tradeoff – either energetically or developmentally – that makes the derived mutation unfavorable in their natural environments . Mechanistic understanding of the energetic forces necessary for C . elegans to bring food into their pharynx is lacking . In fact , pharyngeal pumping rates are often used as proxies for food intake , which we have shown here can be unrelated to the amount of food consumed . Potentially , the thick slurry of food in laboratory plates is completely different biophysically from the mixed bacterial species encountered on rotting material in the wild . Alternatively , differences in feeding behaviors unrelated to social/solitary behaviors might also mediate the differences in food intake . Our experiments suggest that previously described roaming/dwelling differences are also not responsible , however , additional uncharacterized behavioral differences could influence food intake . The changes to fitness and food consumption in the N2 strain appear to be mediated by the nervous system , which we propose occurs through changes in the function and/or downstream effects of the URX sensory neurons . In this paper , we have also shown that animals that lack the URX neurons consume food at lower rates . How does URX modify food consumption ? One possibility is URX regulates pharyngeal neurons extrasynaptically through neuropeptides or through chemical synapses onto the RIP interneurons , which represent the only connection between the somatic and pharyngeal nervous system . Alternatively , URX could regulate food consumption indirectly by stimulating metabolism of fatty acids . URX , along with AQR and PQR , are body cavity neurons , sending ciliated dendrites into the coelomic fluid , which serves as the circulatory system for C . elegans ( White et al . , 1986 ) . Besides sensing external O2 , URX neurons monitor fat stores , which are thought to regulate tonic Ca++ responses of the URX neurons ( Witham et al . , 2016 ) . URX , in turn , can stimulate fat loss , creating a homeostatic loop that ensures that fat mobilization only occurs when there are sufficient fat reserves and when environmental O2 is high enough to metabolize the fatty acids into energy ( Witham et al . , 2016 ) . The changes to the N2 strain could have resulted in URX triggering fatty acid metabolism at a higher rate in laboratory conditions . The access energy could be used to speed development and increase growth . Why , then , would the animals consume more food ? Fat metabolism has been shown to regulate satiety behavior in C . elegans , which could account for increases in food consumption we see in these strains ( Hyun et al . , 2016 ) . This model could also explain the effects we see in the daf-22 mutants , which we originally explored for the potential role of pheromone responses in feeding and fitness changes . daf-22 mutants accumulate large amounts of fatty acids in their bodies , which potentially inhibits the food consumption rates of these animals . However , our experiments do not preclude a role for pheromones in these fitness and food consumption changes to N2 . Potentially , our crude purifications do not capture physiologically relevant levels and ratios of the complex pheromone mixtures . Pheromones might also contribute to these differences in combination with other daf-22-dependent pathways . Primer pheromones have been shown to influence body fat metabolism in C . elegans through the ADL sensory neuron ( Hussey et al . , 2017 ) . ADL sensory neurons are regulated by pheromones in an npr-1-dependent manner ( Abergel et al . , 2016; Fenk and de Bono , 2017; Jang et al . , 2012; Jang et al . , 2017; Macosko et al . , 2009 ) . It is possible these changes , or other parts of the pheromone circuits are also necessary . Future experiments , enabled by the development of the on-plate food consumption assay , should be enlightening . Our work underscores issues with growing organisms in the laboratory for multiple generations . Despite the attempts of researchers to create fertile conditions for nematodes to grow in , we found a large difference in relative fitness between different strains of C . elegans when competed in the laboratory . Natural genetic variation and de novo variation both result in fitness differences that selection can act on . Experimenters using wild strains of nematodes must take care in designing experiments to account for this , especially in wild strains with lower initial fitness levels . We believe that the laboratory selection pressures we characterized here will generalize to other invertebrate and vertebrate animals . If so , the behaviors and physiology of these animals will also be modified over generations of growth . Our work suggests that not only will the traits that confer fitness advantages be modified , but potentially additional traits due to the pleiotropic actions of many genes , and relaxed stabilizing selection on traits in laboratory conditions . Animals were grown following standard conditions . With exceptions listed below , animals were cultivated on modified nematode growth medium ( NGM ) plates containing 2% agar seeded with 200 μl of an overnight culture of the E . coli strain OP50 in an incubator set at 20°C . Strains were grown for at least three generations without starvation before any assays were conducted . For assays manipulating the environmental O2 levels , animals were grown inside a BioSpherix C474 chamber using a BioSpherix C21 single chamber controller to control ambient O2 levels . For these assays , animals were not grown in temperature incubators , and the room temperature was typically kept ~21°C . For competition experiments on non-burrowing plates , 1 . 25% agarose and 0 . 75% agar replaced the agar concentrations of normal growth plates . To create uniform lawns , liquid cultures of OP50 bacteria were poured onto plates to cover the entire surface area of the plate and then poured off . Competition experiments were performed as previously ( Large et al . , 2016 ) . Briefly , Ten L4 stage animals from each strain were picked onto 9 cm NGM plates seeded with 300 μL of an overnight E . coli OP50 culture and incubated at room temperature for 3 days . After 5 days , animals were transferred to an identically prepared NGM plate and then subsequently transferred every 4 days for five to seven generations . For transfers , animals were washed off from the test plates using M9 buffer and collected into 1 . 5 mL centrifuge tube . The animals were mixed by inversion and allowed to stand for approximately one minute to settle adult animals . 50 uL of the supernatant containing ~1000–2000 L1-L2 animals were seeded on next plates . The remaining animals were concentrated and placed in a −80°C freezer for future genomic DNA isolation . Genomic DNA was collected from every odd generation using a Zymo DNA isolation kit ( D4071 ) . To quantify the relative proportion of each strain , we used a digital PCR based approach using a custom TaqMan probe ( Applied Biosciences ) . Genomic DNA was digested with EcoRI for 30 min at 37°C . The digested products were purified using a Zymo DNA cleanup kit ( D4064 ) and diluted to ~1 ng/μL for the following Taqman assay . Four TaqMan probes were designed using ABI custom software that targeted the dpy-10 ( kah82 ) , dpy-10 ( kah84 ) , npr-1 ( g320 ) , or SNP WBVar00209467 in glb-5 . These probes were validated using defined concentrations of DNA from animals containing each allele . The Taqman digital PCR assays were performed using a Biorad QX200 digital PCR machine with standard probe absolute quantification protocol . The relative allele proportion was calculated for each DNA sample using count number of the droplet with fluorescence signal ( Equation 1 ) . To calculate the relative fitness of the two strains using three to four measurements of relative fitness , we used linear regression to fit this data to a one-locus generic selection model ( Equations 2 and 3 ) , assuming one generation per transfer . ( 1 ) P ( A ) t= No . Allele ANo . Allele A+ No . Allele a ( 2 ) P ( A ) t= P ( A ) 0WAAtP ( A ) 0WAAt+ ( 1- P ( A ) 0 ) Waat ( 3 ) log ( P ( A ) 0P ( A ) t − P ( A ) 01− P ( A ) 0 ) = ( log ( WaaWAA ) ) t To measure bordering rates , 2-week-old NGM plates were removed from a 4°C cold room , seeded with 200 μL of E . coli OP50 and incubated for 2 days at room temperature . 150 adult animals were picked onto these assay plates and placed in either a 20°C incubator or a BioSpherix chamber for 3 hr . Bordering behavior was quantified using a dissecting microscope by identifying animals whose whole body resided within 1 mm of the border of the bacteria lawn . N2 and CX12311 L4 hermaphrodites were picked to fresh agar plates . Their adult progeny were synchronized using alkaline-bleach to isolate eggs . These eggs were washed three times using M9 buffer and placed on a tube roller overnight to allow eggs to hatch . About 400 L1 animals were placed on NGM agar plates seeded with non-uniform lawns of E . coli OP50 and incubated in a BioSpherix chamber set at 10% O2 or 21% O2 levels for 48 hr . The ~L4 stage animals were washed off and used for standard Trizol RNA isolation . Replicates were performed on different days . The RNA libraries for next-generation sequencing were prepared using an Illumina TruSeq Stranded mRNA kit ( 20020595 ) following its standard protocol . These libraries were sequenced using an Illumina NextSeq 500 platform . Reads were aligned using HISAT2 using default parameters for pair-end sequencing . Transcript abundance was calculated using HTseq and then used as inputs for the SARTools ( Varet et al . , 2016 ) . Within this R package , edgeR is used for normalization and differential analysis . N2 cultured at 21% O2 is treated as wild type ( Chen et al . , 2014 ) . The genes showing significantly different expression ( log2 ( fold ) >1 or log2 ( fold ) < −1 , FDR adjusted p-value<0 . 01 ) were selected to perform Hierarchical Cluster analysis , and Principal Component analysis . Sequencing reads were uploaded to the SRA under PRJNA437304 . Animals were synchronized using alkaline-bleach . The eggs were washed by M9 buffer for three times and rotating on tube roller overnight to allow eggs to hatch . About 200 L1 animals were placed on NGM agar plates seeded with E . coli OP50 and cultivated at 20°C for 72 hr . In the pharyngeal pumping rates assays , the pharynges of 10 young adult animals ( 72 hr after place L1 on NGM agar plate ) were observed for 30 s each in three separate trails . To measure the pharyngeal size , young adult animals were placed onto agar pad and immobilized by 25 mM NaN3 . For each strain , pharyngeal sizes of 30 animals from three different plates were imaged under 40x objective lens using z-stack DIC microscope . The diameter of pharyngeal metacorpus , diameter of terminal bulb diameter , procorpus length , and isthmus length were measured using ImageJ software . To measure reproductive timing , animals were synchronized by picking 10 adult animals onto an NGM plate , allowing them to lay eggs for two hours , and then removing the adult animals from the plate . These offspring were then monitored using a 12x dissecting microscope at indicated time points to count the number of animals with oocytes and fertilized eggs in their uterus . A subset of these animals was washed off at indicated time points and fixed in 95% ethanol . The nuclei were stained with 1 . 5 μg/mL DAPI solution in Vectashield antifade mounting medium ( VECTOR H-1200 ) for 10 min in the dark before visualization . Each spermatheca was imaged by z-stack fluorescence microscopy using a 100x lens to determine whether spermatogenesis had started or to count the number of sperm produced by the hermaphrodite . Reproductive rate and body size measurements were measured as described previously ( Large et al . , 2016 ) . For crude pheromone assays , crude pheromone was prepared as described previously ( Zhuo et al . , 2017 ) . The crude pheromone was resuspended in ethanol and stored in −20°C . A dauer formation assay was performed to test the efficacy of crude pheromone . A 1/333 ( v/v ) crude pheromone level could induce a high >80% rate of dauer formation in N2 animals grown on 20 uL of heat killed E . coli OP50 bacteria ( 5 mg/mL ) . For the feeding assay and reproductive timing assays , the animals were grown on NGM plates for three generations on plates containing the indicated concentrations of crude pheromone ( or ethanol control ) . +indicates a 1/10 , 000 ( v/v ) crude pheromone and ++indicated a 1/2000 ( v/v ) ratio of crude pheromones . The plates were then dried in biosafety cabinet for 1 . 5 hr , seeded with 200 μL of overnight culture of the E . coli strain OP50 , incubated overnight , and used immediately for experiments . 35 mm Petri dishes evenly seeded with OP50 E . coli Bacteria for 24 hr before the start of assay . Individual L4 hermaphrodites were placed in the center of the plate and cultivated in BioSpherix chamber in 10% O2 or 21% O2 level at 21o C for 3 hr . The plates were placed on a grid that has 105 squares which cover the whole plate . The number of full or partial squares that contained animal’s tracks was quantified and the exploration fraction was calculated ( Equation 1 ) . ( 1 ) Exploration fraction= No . grids contained tracks105 All raw data are included in figure source data tables . All replicates were biological replicates using animals grown independently for multiple generations . The number of biological replicates were chosen using power analysis based upon the standard deviation from previous assays . To assess statistical significance , we performed one-way ANOVA tests followed by Tukey’s honest significant difference test to correct for multiple comparisons or the Wilcoxon-Mann-Whitney nonparametric test for pairwise comparisons . The Friedman test was used to compare the reproductive timing assays . The exact test used is listed in the legend for each panel . These files show a single generation ( 3 days ) of growth of the N2 or CX12311 grown strain in the presence of 21% or 10% environmental O2 .	Why do humans walk on two feet ? And what makes us smarter than our ape ancestors ? The answers to these questions , and countless others about the particular traits of any number of species , is often said to be natural selection – a process where genes that ensure the survival of a species are favored of others . But it is not always the answer . Other evolutionary forces , such as random changes to the frequency of certain gene variants , restrictions on the development of a certain trait and pleiotropy ( where one gene influences other , seemingly unrelated traits ) can also cause differences between species . Designing experiments to test whether a trait difference is due to natural selection or other factors is notoriously difficult . However , the humble nematode worm , Caenorhabditis elegans , has proven to be particularly useful in this respect . One subtype or strain of C . elegans with certain changes to its genes is used internationally as a ‘reference strain’ , to ensure results between labs are comparable . This strain , N2 , has been bred in the laboratory for hundreds of generations , isolated from its wild counterparts . N2 shows several differences in behavior from the wildtype , including its feeding habits . Wild C . elegans tend to feed together socially , whereas N2 prefers to feed alone . In 1998 and 2009 , researchers – including some involved in the current study – have identified the genetic modifications responsible for this change in behavior . Now , Zhao et al . set out to determine whether this was due to natural selection , and if so , was there a benefit to solitary feeding in laboratory conditions that was driving this genetic change ? Zhao et al . found that the genetic changes in the N2 strain gave the worms a considerable advantage in the artificial environment . However , experiments to modify the conditions the animals grew in revealed that the solitary feeding habits were not necessary for the fitness advantage . In other words , the changes in feeding habits were a symptom of the genetic changes that gave N2 a selective advantage , but they were not the cause . In other words , the changes in feeding behavior were not a result of natural selection , but rather of pleiotropy . The findings highlight that not every change in a trait is down to natural selection and must therefore be put to the test . With declining costs of DNA sequencing , researchers can now easily identify genes and regions of DNA that are likely to be under selection . However , they must be careful before leaping to the conclusion that behavioral differences linked to genetic changes are adaptive . In addition , the findings show that the laboratories relying on N2 as a model organism should be aware that the strain has evolved fundamental differences in its brain connections compared with the wildtype .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2018	Changes to social feeding behaviors are not sufficient for fitness gains of the Caenorhabditis elegans N2 reference strain
Renal medullary carcinoma ( RMC ) is a rare and deadly kidney cancer in patients of African descent with sickle cell trait . We have developed faithful patient-derived RMC models and using whole-genome sequencing , we identified loss-of-function intronic fusion events in one SMARCB1 allele with concurrent loss of the other allele . Biochemical and functional characterization of these models revealed that RMC requires the loss of SMARCB1 for survival . Through integration of RNAi and CRISPR-Cas9 loss-of-function genetic screens and a small-molecule screen , we found that the ubiquitin-proteasome system ( UPS ) was essential in RMC . Inhibition of the UPS caused a G2/M arrest due to constitutive accumulation of cyclin B1 . These observations extend across cancers that harbor SMARCB1 loss , which also require expression of the E2 ubiquitin-conjugating enzyme , UBE2C . Our studies identify a synthetic lethal relationship between SMARCB1-deficient cancers and reliance on the UPS which provides the foundation for a mechanism-informed clinical trial with proteasome inhibitors . Renal medullary carcinoma ( RMC ) was first identified in 1995 and is described as the seventh nephropathy of sickle cell disease ( Davis et al . , 1995 ) . RMC is a rare cancer that occurs primarily in patients of African descent that carry sickle cell trait and presents during adolescence with symptoms of abdominal pain , hematuria , weight loss and widely metastatic disease . Due to the aggressive behavior of this disease and the small numbers of patients , no standard of care exists . Patients are generally treated with multimodal therapies including nephrectomy , chemotherapy and radiation therapy . Despite this aggressive regimen , the mean overall survival rate is only 6–8 months ( Alvarez et al . , 2015; Beckermann et al . , 2017; Ezekian et al . , 2017; Iacovelli et al . , 2015 ) . Recent studies have implicated loss of SMARCB1 in RMC ( Calderaro et al . , 2016; Carlo et al . , 2017; Cheng et al . , 2008 ) . SMARCB1 is a tumor suppressor that when conditionally inactivated in mice leads to rapid onset of lymphomas or brain tumors ( Han et al . , 2016; Roberts et al . , 2002 ) . Furthermore , SMARCB1 is a core member of the SWI/SNF complex where alterations of one or more members have been identified in up to 20% of all cancers ( Helming et al . , 2014; Kadoch et al . , 2013 ) including malignant rhabdoid tumors ( MRTs ) and atypical teratoid rhabdoid tumors ( ATRTs ) . MRTs and ATRTs harbor few somatic genetic alterations other than biallelic loss of SMARCB1 and occur in young children ( Chun et al . , 2016; Gröbner et al . , 2018; Lee et al . , 2012; Ma et al . , 2018; Torchia et al . , 2016 ) . In contrast , RMC patients present as adolescents/young adults , are primarily of African descent and have been found to have fusion events in SMARCB1 and gene mutations in ERG , PDGFRB , MTOR , and ERBB2 ( Calderaro et al . , 2016; Carlo et al . , 2017 ) . Other pathways implicated in this disease included loss of TP53 and VEGF/HIF1A ( Swartz et al . , 2002 ) . An unresolved question is whether these cancers depend upon loss of SMARCB1 . Furthermore , there is an unmet need to identify therapeutic targets to provide better treatments for these patients . Here , we have developed and characterized faithful cell lines of this rare cancer . We demonstrate that RMC depends on loss of SMARCB1 for survival and , through integrated genetic and pharmacologic studies , we uncover the proteasome as a core druggable vulnerability in RMC and other SMARCB1-deficient cancers . From September 2013 until September 2018 , three patients who had a diagnosis of renal medullary carcinoma ( RMC ) were consented to IRB approved protocols ( Materials and methods ) . All patients were of African descent and adolescents . We first attempted to create a patient-derived xenograft from each patient by implanting tissue in the sub-renal capsule or subcutaneously in immunodeficient mice but these samples did not form tumors after 6 months of monitoring . We then attempted to develop cell lines from these patients and generated cell lines from two of the three patients ( CLF_PEDS0005 and CLF_PEDS9001 ) ( Materials and methods ) . For the first patient ( CLF_PEDS0005 ) , we obtained the primary tissue from our local institution at the time of the initial nephrectomy . We generated a short-term culture normal kidney cell line , CLF_PEDS0005_N , and a tumor cell line , CLF_PEDS0005_T1 ( Figure 1—figure supplement 1a ) . In addition , we obtained fluid from a thoracentesis performed when the patient relapsed 8 months into therapy . We isolated two cell lines that grew either as an adherent monolayer , CLF_PEDS0005_T2A , or in suspension , CLF_PEDS0005_T2B . Each of these tumor cell lines expressed the epithelial marker , CAM5 . 2 , and lacked expression of SMARCB1 similar to that observed in the primary tumor ( Figure 1—figure supplement 1b ) . For the second patient ( CLF_PEDS9001 ) , we partnered with the Rare Cancer Research Foundation and obtained samples through a direct-to-patient portal ( www . pattern . org ) . The primary tumor tissue from the second patient was obtained at the time of the initial nephrectomy . From this sample , we generated the tumor cell line , CLF_PEDS9001_T1 . Cell lines were generated from patients who received 4–8 weeks of neoadjuvant chemotherapy prior to their nephrectomy . Sequencing and cytogenetic efforts have identified deletion of one allele of SMARCB1 along with fusion events in the second allele of SMARCB1 in RMC patients ( Calderaro et al . , 2016; Carlo et al . , 2017 ) . We performed WES ( CLF_PEDS0005 ) or whole genome sequencing ( WGS; CLF_PEDS9001 ) on the primary kidney tumor tissues . In both patients , we confirmed the presence of sickle cell trait but also found tumor purity was <20% , which , like prior studies , prevented the identification of the fusion events ( Figure 1—figure supplement 1c ) . This low tumor purity is attributable to the stromal desmoplasia seen in RMC ( Swartz et al . , 2002 ) . We then performed WES on the normal cell line ( CLF_PEDS0005_N ) or whole blood ( CLF_PEDS9001 ) and compared it to the primary tumor cell lines ( CLF_PEDS0005_T1 and CLF_PEDS9001_T ) and metastatic cell lines ( CLF_PEDS0005_T2A and CLF_PEDS0005_T2B ) . We found a low mutation frequency ( 1–3 mutations/mb; Materials and methods; Figure 1a ) in the tumor cell lines similar to that of other pediatric cancers and cell lines such as MRT , ATRT and Ewing sarcoma ( Cibulskis et al . , 2013; Johann et al . , 2016; Wala et al . , 2018 ) . We found that only the metastatic cell lines harbored mutations in TP53 and TPR ( Materials and methods; Supplementary file 1 ) ( Cibulskis et al . , 2013 ) . Using copy number analysis , we confirmed the heterozygous loss of SMARCB1 . In agreement with prior studies , we failed to find an identifiable mutation or deletion to account for the loss of the second SMARCB1 allele with WES . We used dual-color break apart FISH and found that a fusion event led to loss of the second SMARCB1 allele as seen in prior studies ( Materials and methods; Supplementary file 2 ) ( Calderaro et al . , 2016; Carlo et al . , 2017 ) . We then performed WGS to assess for structural variations that would not be captured by WES to elucidate the breakpoint of the rearrangement in SMARCB1 ( Figure 1b; Materials and methods ) . For CLF_PEDS0005_T1 , we found a large deletion between BCR and MYH9 which is predicted to lead to loss of one allele of SMARCB1 ( Figure 1c ) along with a balanced translocation between intron 1 of SMARCB1 to the intron region following the C-terminal end of C1orf116 , yielding a non-functional allele ( Figure 1d and Supplementary file 3 ) . For CLF_PEDS9001_T1 , we found a large deletion between TTC28 and VPREB that would lead to loss of one allele of SMARCB1 ( Figure 1e ) along with a balanced translocation that leads to fusion of intron 10 of PLEKHA5 to intron 6 of SMARCB1 ( Figure 1f and Supplementary file 4 ) . Both translocations involved inactivation of the C-terminal end of SMARCB1 ( Figure 1—figure supplement 1d ) . We confirmed these findings by Sanger sequencing of the breakpoint , by qRT-PCR and by immunoblotting to demonstrate loss of SMARCB1 expression ( Supplementary file 5 , Figure 1—figure supplement 1e–f; Materials and methods ) . We then assessed previously identified breakpoint SMARCB1 sequences with the breakpoints identified in this study and failed to find alignments , fragile sites or other repetitive DNA elements that were shared amongst these sequences ( Figure 1—figure supplement 1g ) . Taken together , we have developed in vitro cell line models from two patients with RMC which faithfully recapitulate known genomics of this disease . We performed RNA-sequencing and transcriptomic profiling to compare the RMC models to other renal tumors or tumors that harbor loss of SMARCB1 . Specifically , we compared the Therapeutically Applicable Research to Generate Effective Treatments ( TARGET ) RNA-sequencing data from pediatric renal tumors ( e . g . Wilms Tumor , Clear Cell Sarcoma of the Kidney , and Malignant Rhabdoid Tumor ) or normal kidney tissues with the RMC models using t-distributed stochastic neighbor embedding ( tSNE ) ( Materials and methods ) . The normal cell line , CLF_PEDS0005_N , clustered with TARGET normal kidney tissues and RMC cell lines from both patients clustered with the TARGET Rhabdoid Tumor samples ( Figure 2a ) . These observations showed that these RMC cell lines share expression patterns with patients with MRTs . To determine if the RMC cell lines clustered separately among other SMARCB1-deficient cancers , we performed gene expression analysis of our RMC models and compared them to publicly available datasets of MRT cell lines , patients with RMC , MRT or ATRT or synovial sarcoma ( a cancer driven by the fusion oncoprotein SSX-S18 which displaces SMARCB1; Materials and methods ) ( Barretina et al . , 2012; Calderaro et al . , 2016; Han et al . , 2016; Johann et al . , 2016; Richer et al . , 2017 ) . Using tSNE , we found that the RMC cell lines closely mapped to a French cohort of RMC , MRT and ATRT patients ( Figure 2b ) . These observations demonstrated that RMC cell lines and SMARCB1 deficient patients express similar gene expression programs . We then assessed the consequences of re-expressing SMARCB1 . Specifically , we generated doxycycline-inducible open reading frame ( ORF ) vectors harboring SMARCB1 and stably infected our RMC models , G401 ( MRT cell line ) and HA1E ( SMARCB1 wild-type immortalized epithelial kidney cell line ) ( Hahn et al . , 1999 ) . We confirmed that the addition of doxycycline used in our studies did not affect the proliferation of the parental cell lines ( Materials and methods and Figure 2—figure supplement 1a–b ) . We used these inducible cell lines to assess the biochemical stability of the SWI/SNF complex by using 10–30% glycerol gradient sedimentation followed by SDS-PAGE ( Figure 2c–d and Figure 2—figure supplement 1c ) . In HA1E SMARCB1 wild-type cells , the SWI/SNF complex members SMARCB1 , ARID1A and SMARCA4 were robustly expressed at higher molecular weights ( e . g . fractions 13–16 ) . In G401 SMARCB1 deficient cells , the SWI/SNF complex is smaller and seen at lower molecular weights ( e . g . fractions 10–14 ) . Furthermore , expression of SWI/SNF complex members was modestly decreased in G401 , consistent with our prior studies ( Nakayama et al . , 2017; Wang et al . , 2017 ) . In our RMC cell lines , we found that the majority of ARID1A and SMARCA4 was observed in fractions 11–13 similar to what we found in the MRT cell line G401 . In addition , we found increased expression and a shift of ARID1A and SMARCA4 to larger fractions 13–15 upon re-expression of SMARCB1 in the RMC lines similar to what we observed in the SMARCB1 wild-type HA1E cell line . We concluded that the composition of the SWI/SNF complex is similar between RMC and other SMARCB1 deficient cancers . We then used the inducible cell lines to measure the consequence of SMARCB1 re-expression on the viability of the cells . We also generated cell lines with inducible expression of a LacZ control to compare with re-expression of SMARCB1 . In the HA1E SMARCB1 wild type cells , we found no significant difference in viability between induction of SMARCB1 versus induction of LacZ using direct counting of viable cells ( Figure 2e ) . In contrast , induction of SMARCB1 in the MRT G401 SMARCB1 deficient cells decreased the number of viable cells by 41% . Similar to G401 , we found that re-expression of SMARCB1 in each of the RMC models led to significant decreases in cell viability ( 37–62% ) ( Figure 2e; Figure 2—source data 1 ) . These observations suggest loss of SMARCB1 is required for the proliferation and viability of RMC cells . Since MRT cell lines arrest and senesce when SMARCB1 is re-expressed ( Betz et al . , 2002 ) , we looked for evidence of senescence by staining for senescence-associated acidic β-galactosidase in the RMC cells when SMARCB1 was re-expressed . Following 7 days of SMARCB1 or LacZ re-expression , we stained the cells for β-galactosidase ( Materials and methods ) . We failed to observe cells expressing β-galactosidase upon expression of LacZ in the RMC cells , but when SMARCB1 was expressed , we found 44 . 6% ( ±17% ) of the RMC cells stained for β-galactosidase ( Figure 2f and Figure 2—figure supplement 1d ) . These studies showed that re-expression of SMARCB1 in RMC cells may also lead to senescence . We then assessed what genes were differentially expressed upon SMARCB1 re-expression in the RMC and MRT cell lines as another way to assess the similarity between these two cancers . We performed RNA-sequencing on the doxycycline-induced SMARCB1 RMC cell lines and compared them to the uninduced cell lines or doxycycline-induced LacZ cell lines . We then re-analyzed our previously published studies of MRT cells with SMARCB1 re-expression and compared them to our RMC cells ( Wang et al . , 2017 ) . We found 1719 genes to be significantly different ( false discovery rate of <0 . 25 ) in the RMC cells and 2735 genes in MRT cells . We identified 527 genes that significantly overlapped between the RMC and MRT cell lines ( hypergeometric p-value less than 4 . 035e-63; Supplementary file 6 ) . We compared this group of genes to the genes differentially expressed between MRT tumor and normal tissues from TARGET ( n = 6 , 311 ) . We identified 257 genes that overlapped with the 527 significantly differentially expressed genes induced by re-expression of SMARCB1 ( Supplementary file 6 ) . Using this list of 257 genes , we performed gene ontology ( GO ) -based Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) and identified significantly enriched genes sets ( q-value <0 . 1 ) , including those related to the cell cycle and the ubiquitin-proteasome system ( UPS ) . We then analyzed the kinetics by which these gene expression changes occur after SMARCB1 was re-expressed . Specifically , we analyzed 5 genes of these 257 genes that are implicated in regulation of the G1/S ( RRM2 , TOP2A ) or G2/M ( PLK1 , CCNB1 , UBE2C ) phases of the cell cycle . For PLK1 and CCNB1 , we observed a gradual decrease in expression over the course of 120 hr whereas RRM2 , UBE2C and TOP2A exhibited a more profound decrease in expression after the first 24 hr and then a modest decrease over the following 96 hr ( Figure 2—figure supplement 1e–f ) . These findings confirm that changes in the transcriptome following SMARCB1 re-expression in RMC cell lines are similar to other SMARCB1 deficient cancer cell lines . In sum , these observations indicate that the RMC cell lines are functionally similar to those derived from other SMARCB1 deficient cancers . MRT , ATRT and RMC are aggressive and incurable cancers . We performed genetic ( RNAi and CRISPR-Cas9 ) and pharmacologic screens to identify druggable targets that would decrease proliferation or survival for these cancers . Specifically , we used the Druggable Cancer Targets ( DCT v1 . 0 ) libraries and focused on targets that were identified by suppression with RNAi , gene deletion with CRISPR-Cas9-based genome editing , and perturbation by small molecules ( Hong et al . , 2016; Seashore-Ludlow et al . , 2015 ) . We accounted for off-target effects in the RNAi screens by using seed controls for each shRNA . We performed these three orthogonal screens on both metastatic models of RMC , CLF_PEDS0005_T2A and CLF_PEDS0005_T2B ( Figure 3a ) . We introduced the shRNA DCT v1 . 0 lentiviral library into these two cell lines and evaluated the abundance of the shRNAs after 26 days using massively parallel sequencing ( Materials and methods ) . We confirmed depletion of known common essential genes such as RPS6 ( Figure 3—figure supplement 1a ) . We then analyzed the differential abundance between the experimental and seed control shRNAs to collapse individual shRNAs to consensus gene dependencies with RNAi Gene Enrichment Ranking ( RIGER ) ( Luo et al . , 2008 ) . Of 444 evaluable genes , 72 genes scored with a RIGER p-value<0 . 05 in CLF_PEDS0005_T2A and 74 genes scored in CLF_PEDS0005_T2B . In parallel , we introduced the CRISPR-Cas9 DCT v1 . 0 lentiviral library to determine the differential representation of the CRISPR-Cas9 sgRNAs between 6 and 23 days to identify genes depleted or enriched in this screen by massively parallel sequencing ( Materials and methods ) . We confirmed that the distribution of sgRNAs among biological replicates was highly correlated ( Figure 3—figure supplement 1b–e ) . When compared to the controls , there was significant depletion of essential genes such as RPS6 ( Figure 3—figure supplement 1f–g ) . We used RIGER to collapse the individual sgRNAs to consensus gene dependencies and found 124 genes ( of a total of 445 evaluable genes ) and 136 genes ( of a total of 445 evaluable genes ) with a RIGER p-value<0 . 05 in CLF_PEDS0005_T2A and CLF_PEDS0005_T2B cell lines , respectively . We performed a small-molecule screen using a library of 440 compounds that have known targets in the RMC cells ( CLF_PEDS0005_T2A and CLF_PEDS0005_T2B ) ( Hong et al . , 2016 ) . This library includes 72 FDA approved compounds , 100 compounds in clinical trials and 268 probes based on our prior studies . We calculated an area under the curve ( AUC ) based on an 8-point concentration range and considered AUCs < 0 . 5 as significant . Of the evaluable compounds , 75 ( 18% ) compounds significantly decreased cell viability in CLF_PEDS0005_T2A and 82 ( 20% ) compounds significantly decreased cell viability in CLF_PEDS0005_T2B . We then looked for genes or targets of the small molecules that scored in all three of the RNAi , CRISPR-Cas9 and small-molecule screens . We identified 21 genes in CLF_PEDS0005_T2A and 27 genes in CLF_PEDS0005_T2B ( Supplementary file 7 ) of which 19 genes scored in both screens ( Figure 3a ) . Among the 19 genes were components of the ubiquitin-proteasome system ( e . g . PSMB1 , PSMB2 , PSMB5 , PSMD1 , PSMD2 , and CUL1 ) , regulators of the cell cycle ( CDK1 , CDK6 , KIF11 and PLK1 ) and genes involved in nuclear export ( KPNB1 and XPO1 ) . To eliminate small molecules and targets that affect normal renal tissue , we screened the normal cell line ( CLF_PEDS0005_N ) with the small-molecule library . We calculated the robust Z-scores for these screens in relationship to the Cancer Cell Line Encyclopedia ( CCLE ) to normalize the responses to various compounds ( Barretina et al . , 2012; Rees et al . , 2016; Seashore-Ludlow et al . , 2015 ) . We then compared the results of this small-molecule screen with the RMC cancer cell lines ( CLF_PEDS0005_T2A and CLF_PEDS0005_T2B ) . We found that the tumor cells were differentially sensitive ( up to two standard deviations ) upon treatment with proteasome inhibitors , bortezomib and MLN2238 , when compared to the normal cell line ( Figure 3b; Figure 3—source data 1 ) . These findings suggest that the vulnerability to proteasome inhibition may be dependent on loss of SMARCB1 . To validate the dependency of SMARCB1 deficient tumors to the ubiquitin-proteasome system , we assessed the consequences of inhibiting proteasome function on survival of the primary tumor cell line , CLF_PEDS0005_T1 , by deleting components of the proteasome with CRISPR-Cas9 . We compared these findings with a model of undifferentiated sarcoma , CLF_PEDS015T , that does not harbor mutations in SMARCB1 ( Hong et al . , 2016 ) . We scaled the results based on the non-targeting sgRNA negative controls and positive controls targeting RPS6 , a common essential gene ( Hart et al . , 2015 ) . Compared to the control sgRNAs , there was an average decrease of 29% in viability in CLF_PEDS015T while there was an average decrease of 74% in CLF_PEDS0005_T1 ( Figure 3c; Materials and methods ) . Although deletion of the proteasome members affected proliferation in all of the models ( two tailed t-test p=1 . 1e-5 for CLF_PEDS015T and p=5 . 4e-20 for CLF_PEDS0005_T1 ) , we found that suppression of proteasome components affected the RMC model CLF_PEDS0005_T1 to a statistically greater degree ( two tailed t-test p=7 . 1e-8 ) . We subsequently validated that gene deletion by CRISPR-Cas9 of PSMB5 , one of the primary targets of proteasome inhibitors , in the CLF_PEDS0005_T2A and CLF_PEDS9001_T1 cell lines led to decreased viability ( Figure 3—figure supplement 1h–i ) . We then determined whether this vulnerability to proteasome inhibition was specific to the loss of SMARCB1 . We treated the normal cell line , CLF_PEDS0005_N , and an early passage of CLF_PEDS9001_T1 while it was a heterogenous population and retained SMARCB1 ( Figure 3—figure supplement 1j ) with bortezomib or the second-generation proteasome inhibitor , MLN2238 . We observed significantly decreased sensitivity to the proteasome inhibitors in the SMARCB1 retained isogenic cell lines as compared to the SMARCB1 deficient cell lines ( Figure 3d–e ) . We then treated our SMARCB1-inducible RMC and MRT cell lines with DMSO , bortezomib or MLN2238 . Re-expression of SMARCB1 led to a decrease in sensitivity to bortezomib or MLN2238 as compared to the isogenic SMARCB1 deficient lines ( Figure 3f–g ) . The observed differential resistance to SMARCB1 re-expression was between 2–3-fold with either bortezomib or MLN2238 ( Figure 4—figure supplement 1a–b ) . We concluded that re-expression of SMARCB1 partially rescued the sensitivity of MRT or RMC cell lines to proteasome inhibition . We then compared the results of small-molecule screens performed in SMARCB1-deficient cancer cell lines in CCLE to the rest of the CCLE cell lines ( n = 835 ) . We found that SMARCB1-deficient cell lines were significantly more sensitive ( two-tailed t-test p-value=0 . 011 ) to treatment with MLN2238 than non-multiple myeloma CCLE cell lines ( Figure 4a; Figure 4—source data 1 ) ( Rees et al . , 2016; Seashore-Ludlow et al . , 2015 ) . The degree of sensitivity was similar to that of multiple myeloma cell lines which are known to be sensitive to proteasome inhibition ( Dimopoulos et al . , 2016 ) . These findings confirm that SMARCB1 deficient cell lines are selectively vulnerable to proteasome inhibition . We then performed in vitro studies to confirm the findings from these high throughput small molecule screens . We treated an additional 6 SMARCB1 deficient cell lines ( 4 MRT and 2 ATRT ) with bortezomib . We compared these results to H2172 , a lung cancer cell line that was not sensitive to proteasome inhibition in CCLE small-molecule screens , and RPMI8226 , an established multiple myeloma cell line that is responsive to proteasome inhibition ( Hideshima et al . , 2001 ) . We found that our SMARCB1 deficient cell lines exhibited single digit nanomolar sensitivity to proteasome inhibition similar to that observed in the multiple myeloma cell line RPMI8226 ( Figure 4—figure supplement 1c ) . In contrast , we found that the IC50 in H2172 was at least 3-fold higher . Since there are no SMARCB1 wildtype pediatric kidney cancer cell lines in CCLE , we compared the sensitivity to bortezomib or MLN2238 in our RMC models with wildtype SMARCB1 patient-derived Wilms tumor cell line , CLF_PEDS1012_T ( Figure 4b–c and Figure 3—figure supplement 1j ) . We found that CLF_PEDS1012_T was more resistant to proteasome inhibition as compared to our RMC models and MRT cell line , G401 . These findings suggest that SMARCB1-deficient cells are more sensitive to proteasome inhibition . We then studied how SMARCB1 loss leads to a dependency on the ubiquitin-proteasome system . Since activation of c-MYC has been observed in SMARCB1-deficient cancers ( Cheng et al . , 1999; Genovese et al . , 2017 ) , we assessed how c-MYC levels are altered upon proteasome inhibition . Compared to RPMI8226 , a multiple myeloma cell line that relies on c-MYC for survival ( Tagde et al . , 2016 ) , we failed to observe suppression of c-MYC protein levels following bortezomib or MLN2238 treatment ( Figure 4—figure supplement 1d ) . We then assessed c-MYC expression levels in the G401 , CLF_PEDS9001T , and CLF_PEDS00005_T1 cell lines following treatment with MLN2238 and found that c-MYC levels were increased after MLN2238 treatment ( Figure 4—figure supplement 1e ) . In our models of RMC and MRT , these findings suggest proteasome inhibition does not lead to suppression of c-MYC . We then assessed if the SWI/SNF complex is altered upon treatment with a proteasome inhibitor . We treated uninduced and induced SMARCB1 cells with DMSO or a proteasome inhibitor . We failed to see a consistent significant change in total or nuclear protein when immunoblotting for SWI/SNF complex members ( SMARCE1 , SMARCD1 , SMARCD1 , SMARCC1 , SMARCC2 , SMARCA4 and ARID1A ) other than increases in SMARCB1 levels upon doxycycline treatment ( Figure 4—figure supplement 2a–b ) . These findings suggest that inhibition of the proteasome in SMARCB1 deficient cancers and its subsequent resistance upon SMARCB1 expression does not alter the total or nuclear levels of SWI/SNF complex members . ER stress has been implicated as a mechanism by which proteasome inhibitors act on multiple myeloma cells ( Obeng et al . , 2006 ) . We saw an increase in protein expression of markers of ER stress , GRP78 and IRE1α , following treatment with MLN2238 ( Figure 4—figure supplement 2c ) . Upon re-expression of SMARCB1 and subsequent treatment with a proteasome inhibitor , we did not see changes in either GRP78 or IRE1α protein levels ( Figure 4—figure supplement 2c ) . These observations suggest that although ER stress markers are elevated upon proteasome inhibition in SMARCB1-deficient cell lines , they are not rescued by SMARCB1 re-expression . We then performed GO-based GSEA ( Subramanian et al . , 2005 ) on the 527 significantly altered genes upon re-expression of SMARCB1 ( Supplementary file 6 ) to identify classes of gene function enriched in this group of genes . We identified numerous gene sets that involved the ubiquitin-proteasome system ( Supplementary file 8 ) . We then performed RNA-sequencing on G401 and the RMC cell lines treated with DMSO or MLN2238 . We identified 1758 genes which were significantly ( FDR < 0 . 1 ) up- or down-regulated upon treatment with MLN2238 ( Supplementary file 9 ) . We compared the 527 differentially expressed genes identified upon re-expression of SMARCB1 with the 1758 differentially expressed genes identified upon treatment with MLN2238 and identified 92 genes which overlapped . Of these genes , we identified 63 genes which were differentially expressed with re-expression of SMARCB1 and were inversely differentially expressed with treatment with MLN2238 ( Figure 4d; Figure 4—source data 2 ) . From this refined gene set , we performed GO-based GSEA ( Subramanian et al . , 2005 ) and found significant enrichment ( adjusted p-value ranging from 0 to 0 . 0061 ) in gene sets involving the cell cycle ( Supplementary file 10 ) . We subsequently assessed the cell cycle by DNA content with cells treated with DMSO or MLN2238 for 24 hr as proteasome inhibitors have been found to cause a G2/M cell cycle arrest in lymphomas , colorectal carcinomas , hepatocellular carcinomas , and glioblastoma multiforme ( Augello et al . , 2018; Bavi et al . , 2011; Gu et al . , 2017; Yin et al . , 2005 ) . We observed a significant shift in cells to G2/M ( two-tailed t-test p-value 0 . 0005; Figure 4e ) . Upon re-expression of SMARCB1 , we saw that this phenotype was rescued ( Figure 4f ) . By 48 hr , we saw a significant increase in markers of programed cell death such as Annexin V and PI positive cells ( Figure 4g ) . We found that treatment with a proteasome inhibitor led to increased cleaved caspase-3 levels in addition to changes to Annexin V and PI , suggesting that inhibition of the ubiquitin-proteasome system leads to programmed cell death ( Figure 4h ) . We then asked whether restoration of SMARCB1 expression inhibited cleaved caspase-3 activation . We found that induction of cleaved caspase-3 was less pronounced when SMARCB1 was re-expressed ( Figure 4h ) . These observations suggest that proteasome inhibitors initially lead to a SMARCB1-dependent G2/M cell cycle arrest and subsequent programmed cell death . To assess whether RMC cells exhibit an increased proclivity to undergo cell death after cell cycle arrest , we asked whether treatment of SMARCB1-deficient cancers with cell cycle inhibitors led only to cell cycle arrest or arrest followed by cell death . Specifically , we used nocodazole , an anti-mitotic agent that disrupts microtubule assembly in prometaphase , and RO-3306 , a CDK1 inhibitor which disrupts the CDK1-cyclin B1 interaction during metaphase ( Vassilev et al . , 2006; Wolf et al . , 2006 ) . We treated both G401 and CLF_PEDS9001_T with nocodazole or RO-3306 for 24 hr and observed accumulation of cells in G2/M as well as increased cyclin B1 and cleaved caspase-3 ( Figure 4—figure supplement 2d–e ) similar to what we observed after treatment with MLN2238 . By 72 hr , we found that treatment with either nocodazole or RO-3306 induced cell death in the majority of cells ( 65–90% ) ( Figure 4—figure supplement 2f ) , similar to what we observed when we treated cells with MLN2238 ( Figure 4c ) . These observations suggest that SMARCB1-deficient cell lines are susceptible to programmed cell death following treatment with a cell cycle inhibitor and that the cell cycle arrest observed after treatment with MLN2238 leads to programmed cell death . We subsequently searched for genes related to the ubiquitin-proteasome system and SMARCB1 function . We defined a set of 204 genes that were upregulated when comparing the log2 fold change between SMARCB1 deficient cells and SMARCB1 re-expressed cells in RMC and MRT cell lines ( Supplementary file 6 ) . We then took this set of 204 genes and examined the Project Achilles ( genome scale CRISPR-Cas9 loss of function screens ) DepMap Public 18Q3 dataset ( Meyers et al . , 2017 ) to determine whether any SMARCB1 deficient cell lines required expression of these genes for survival . This dataset included loss of function screens from 485 cancer cell lines and included three ATRT SMARCB1-deficient cancer cell lines: COGAR359 , CHLA06ATRT and CHLA266 . We found that SMARCB1 deficient cancer cell lines required UBE2C , an ubiquitin-conjugating enzyme , for survival . We noted that these cell lines were in the top 5% of cell lines ( n = 485 ) that required UBE2C for survival ( empirical Bayes moderated t-test p-value=0 . 00016; Figure 5a–b; Figure 5—source data 1 ) . These observations suggested that cancer cell lines that lack SMARCB1 were also dependent on UBE2C . Since the 3 cell lines profiled were ATRTs , we validated that the RMC cell lines were also dependent on UBE2C for survival . We generated sgRNAs specific for UBE2C and assessed viability by cell counting following gene deletion . We saw a significant decrease in cell viability in SMARCB1 deficient cell lines as compared to urothelial carcinoma cell line , JMSU1 ( SWI/SNF wild type ) , or non-small cell lung cancer cell line , A549 ( SMARCA4 mutant ) ( two-tailed t-test p-values 4 . 5e-5 and 4 . 6e-5; Figure 5c and Figure 5—figure supplement 1a ) . UBE2C serves as the E2 enzyme which adds the first ubiquitin ( Ub ) to cyclin B1 for degradation ( Dimova et al . , 2012; Grice et al . , 2015 ) . Cyclin B1 degradation is required in G2/M at the end of metaphase to enter anaphase ( Chang et al . , 2003 ) . Our integrated RNAi , CRISPR-Cas9 and small molecule screens identified that our RMC models required expression of PLK1 and CDK1 , genes involved in G2/M , for survival ( Figure 3a–b ) , and prior studies have identified that inhibition of PLK1 in ATRT or MRT cells leads to arrest in G2/M ( Alimova et al . , 2017; Morozov et al . , 2007 ) . Treatment of the RMC cell lines with MLN2238 led to accumulation of cyclin B1 as compared to cyclin D1 suggesting that MLN2238 inhibits degradation of cyclin B1 ( Figure 5d ) . When we re-expressed SMARCB1 , we found that cyclin B1 levels were unchanged upon MLN2238 treatment . Although APC/C serves as the E3 ligase for cyclin B1 , genetic deletion of APC/C in Project Achilles showed that APC/C was an essential gene across all cancer cell lines . These findings suggest SMARCB1 deficient cancer cells require UBE2C expression for survival , in part by regulating cyclin B1 stability . These studies identify a lethal interaction between suppressing the UPS and SMARCB1-deficient cancers in vitro . The doses used in this study were based on in vitro studies of multiple myeloma or lymphoma cell lines ( Chauhan et al . , 2011; Garcia et al . , 2016; Hideshima et al . , 2003; Hideshima et al . , 2001 ) . For patients with primary or refractory multiple myeloma , use of proteasome inhibitors has led to significant clinical responses ( Jagannath et al . , 2004; Moreau et al . , 2016; Richardson et al . , 2005; Richardson et al . , 2003 ) . We reasoned that if our SMARCB1 deficient cancers were susceptible to proteasome inhibitors at similar in vitro dosing , we would also see similar in vivo responses . We first determined whether these doses led to proteasome inhibition by assessing the ability of these cell lines to cleave Suc-LLVY-aminoluciferin . We found that treatment of the SMARCB1 deficient cell lines with either bortezomib or MLN2238 led to inhibition of the proteasome to a similar extent observed when the multiple myeloma cell line RPMI8226 was treated ( Figure 5—figure supplement 1b; Materials and methods ) . We also simulated the pharmacodynamics of proteasome inhibitors in vivo by treating cells in vitro with a pulse dose of proteasome inhibitors as has been performed in multiple myeloma and chronic myeloid leukemia cell lines ( Crawford et al . , 2014; Kuhn et al . , 2007; Shabaneh et al . , 2013 ) . We found that upon treatment with MLN2238 , SMARCB1 deficient cells arrested in G2/M , which led to cell death as measured by Annexin V/PI staining and led to accumulation of cyclin B1 similar to treatment with a continuous dose of MLN2238 ( Materials and methods; Figure 5—figure supplement 1c–f and Figure 5—figure supplement 2a–c ) . We subsequently performed in vivo studies to confirm the effect of proteasome inhibition in tumor xenografts . We used the rhabdoid tumor cell line , G401 , for our in vivo studies because we noted that the primary tumor cell lines ( CLF_PEDS0005_T1 and CLF_PEDS9001_T ) did not form subcutaneous xenograft tumors in immunodeficient mice ( Figure 5—figure supplement 2d–e; Materials and methods ) . We allowed tumors to achieve an average volume of 148 mm3 and then treated mice with either vehicle or MLN2238 at the maximum tolerated dose at 7 mg/kg twice a week in the Taconic NCr-nude mouse strain . Treatment with MLN2238 over 26 days induced significant tumor stabilization or regression as compared to vehicle treated tumors as assessed by absolute tumor volume ( two-way ANOVA test with p-value<0 . 0001; Figure 5e–f ) and did not induce significant changes in body weight as compared to vehicle-treated tumors ( two-tailed t-test p-value 0 . 154; Figure 5—figure supplement 2f ) . Furthermore , mice treated with MLN2238 survived significantly longer [p-value 0 . 0489 by log-rank ( Mantel-Cox ) test; Figure 5g] . We noted that several tumors in the treatment arm had a suboptimal response to MLN2238 ( Figure 5e–f ) . We harvested tumors from two pairs of mice that either showed regression or no response to MLN2238 and assessed cleaved caspase-3 and cyclin B1 levels . We found increased cyclin B1 and cleaved caspase-3 by immunoblotting in the tumor that responded to MLN2238 but did not observe increased cyclin B1 accumulation or activation of cleaved caspase-3 in the tumor without response ( Figure 5—figure supplement 2g ) suggesting that adequate inhibition of the proteasome was not achieved in mice with a suboptimal response . Combined , these results demonstrate that MLN2238 induces a cytostatic response in SMARCB1-deficient tumors in vivo . We have developed faithful patient-derived models of RMC which have been genomically validated using WGS , WES , RNA-sequencing and gene expression profiling . We have shown that these models are dependent upon the loss of SMARCB1 for survival . Re-expression of SMARCB1 in RMC leads to a significant decrease in cell counts and a senescence phenotype . Biochemically , re-expression of SMARCB1 in RMC leads to stabilization of the SWI/SNF complex in the same manner as re-expression of SMARCB1 in MRT . Diagnostically , patients with RMC are often misdiagnosed with renal cell carcinoma ( RCC ) due to the rarity of RMC , the lack of access to SMARCB1 histological stains and unknown sickle cell status ( Beckermann et al . , 2017 ) . Although SMARCB1 is currently included in targeted sequencing efforts nationwide ( AACR Project GENIE Consortium , 2017 ) , our studies along with prior studies ( Calderaro et al . , 2016; Carlo et al . , 2017 ) suggest that conventional target exome sequencing may fail to identify patients with RMC . Patients with RMC and other SMARCB1 deficient cancers have a poor prognosis despite aggressive multi-modal therapy . Using genetic and pharmacologic screens in these RMC models , we identified the ubiquitin-proteasome system as a specific vulnerability in RMC . When we looked more broadly at other SMARCB1 deficient cancers such as MRT and ATRT , we found that these models were similarly sensitive to inhibition of the ubiquitin-proteasome system . Re-expression of SMARCB1 partially rescued the sensitivity to proteasome inhibitors in RMC and MRT models . Prior studies have implicated MYC signaling and downstream activation of ER stress as a mechanism for sensitivity to proteasome inhibitors in Kras/Tp53 mutant pancreatic cancers with Smarcb1 deficiency ( Genovese et al . , 2017; Moreau et al . , 2016 ) . However , the background of mutant KRAS may be contributing to these findings as mutant HRAS or KRAS cancers are sensitive to enhanced proteotoxic stress and ER stress . Furthermore , KRAS mutant cancers depend on several proteasome components in genome scale RNAi screens ( Aguirre and Hahn , 2018 ) . Our studies have identified that the ubiquitin-proteasome system is a core vulnerability among a compendium of druggable targets as tested by orthogonal methods of RNA interference , CRISPR-Cas9 gene deletion or small molecule inhibition . We found that proteasome inhibition in SMARCB1-deficient cancer cell lines results in G2/M arrest due to inappropriate degradation of cyclin B1 . Although in multiple myeloma cells , tumor regression has been observed in xenografts following treatment with MLN2238 ( Chauhan et al . , 2011 ) , we found that treatment with MLN2238 of SMARCB1 deficient xenografts led to a cytostatic response . This finding is similar to what has been observed in xenograft models of non-small cell lung cancer ( 14 tumor models ) and colon cancer ( 6 tumor models ) ( Chattopadhyay et al . , 2015 ) . We note that these studies were performed in mice that tolerated MLN2238 treatments at 11–14 mg/kg ( Chauhan et al . , 2011 ) , a dose which we were unable to achieve in the Taconic Ncr-nude mice and may have led to the observed heterogeneous tumor response to MLN2238 treatment . Nevertheless , these studies still support the importance of testing this hypothesis in patients , particularly since there are no standard therapies for SMARCB1-deficient cancers . There have been case reports of one adult and two children with RMC who exhibited extraordinary responses for 2–7 years following diagnosis after empiric therapy with bortezomib either as monotherapy or in combination with chemotherapy ( Carden et al . , 2017; Ronnen et al . , 2006 ) . Our findings suggest that testing oral proteasome inhibitors such as MLN2238 for patients with RMC and potentially more broadly across SMARCB1-deficient cancers is warranted . Patients assented or families consented to IRB approved protocols . Patient PEDS0005 whole blood , adjacent normal kidney and tumor tissue were obtained within 6 hr as part of the nephrectomy following neoadjuvant chemotherapy . Upon relapse , pleural fluid was obtained from a palliative thoracentesis . The tumor and adjacent normal kidney tissue was minced into 2–3 mm3 cubes . CLF_PEDS0005 primary tissue was then dissociated as previously described and cultured in both RPMI media containing 10% FBS ( Sigma ) or DMEM/F-12 media containing ROCK inhibitor , Y-27632 , insulin , cholera toxin , 5% FBS , and penicillin/streptomycin ( Liu et al . , 2012 ) . Samples were gently passaged when cultures achieved 80–90% confluence . The normal kidney cell line was named CLF_PEDS0005_N . The tumor kidney cell line was named CLF_PEDS0005_T1 . For the pleural fluid sample , samples were grown initially in conditioned media as previously published ( Liu et al . , 2012 ) . Adherent and suspension cells were continuously passaged when cells reached confluence . By passage 5 , cells were noted to be growing as adherent and suspension cells , and these were sub-cultured to yield CLF_PEDS0005_T2A and CLF_PEDS0005_T2B , respectively . Samples were then transitioned to DMEM/F-12 media or RPMI as above at passage 13 . Patient PEDS9001 whole blood , adjacent normal kidney and tumor tissue were obtained similarly to PEDS0005 . Discarded tissue from the nephrectomy following neoadjuvant chemotherapy was sent to our institution within 24 hr of resection . For the normal kidney and tumor tissue , samples were minced to 2–3 mm3 and plated onto six well plates ( Corning , NY ) . Following mincing , tumor samples were cultured without further digestion . Tumor samples were grown in culture in the DMEM/F-12 media as described above to yield CLF_PEDS9001_T1 , while the normal samples yielded a cell culture that matched the tumor cell line . All immunohistochemical staining was done in the clinical histopathology laboratory at Boston Children’s Hospital with appropriate positive controls performed with each run . Antibodies used included anti-cytokeratin CAM5 . 2 ( BD Biosciences , 349205 ) and SMARCB1/BAF47 ( BD Biosciences , 612110 ) . We developed a custom dual-color breakapart FISH probe , using BAC probes surrounding SMARCB1 at 22q11 . 23: RP11-662F7 ( telomeric to SMARCB1 , labeled in green ) and RP11-1112A23 ( centromeric to SMARCB1 , labeled in red ) ( Empire Genomics , Buffalo , NY ) . The probe set was hybridized to a normal control to confirm chromosomal locations and to determine the frequency of expected fusion signals in normal cells . 50 nuclei were scored by two independent observers ( n = 100 per cell line ) in the CLF_PEDS0005 and CLF_PEDS9001 models . We performed whole exome sequencing ( WES ) from genomic DNA extracted from whole blood , normal/tumor tissues , and our patient-derived cell lines as noted in the text . One microgram of gDNA ( as measured on a Nanodrop 1000 ( Thermo Fisher Scientific ) ) was used to perform standard ( ~60 x mean target coverage for normal ) or deep ( ~150 x mean target coverage for tumor and cell lines ) WES . Illumina ( Dedham , MA ) chemistry used . We performed PCR-free whole genome sequencing ( WGS ) from gDNA extracted from whole blood , normal/tumor tissues , and our patient-derived cell lines as noted in the text . Two micrograms of gDNA were used to perform standard ( for normal ) or deep ( for tumor and cell lines ) coverage . Illumina ( Dedham , MA ) HiSeq X Ten v2 chemistry was used . We achieved an average depth of coverage of 38x for the germline DNA and 69x for the tumor cell line DNA . For Figure 2a , samples were processed using Illumina TruSeq strand specific sequencing . We performed poly-A selection of mRNA transcripts and obtained a sequencing depth of at least 50 million aligned reads per sample . For SMARCB1 re-expression RNAseq experiments and MLN2238 vs DMSO-treated experiments , samples were collected as biological replicates or triplicates . RNA was extracted using Qiagen RNeasy Plus Mini Kit ( Qiagen , Hilden , Germany ) . RNA was normalized using the Qubit RNA HS Assay ( Thermo Fisher Scientific ) . Five hundred ng of normalized RNA was subsequently used to create libraries with the Kapa Stranded mRNA-seq kit ( Kapa Biosystems , KK8420; Wilmington , MA ) . cDNA libraries were then quantitatively and qualitatively assessed on a BioAnalyzer 2100 ( Agilent , Santa Clara , CA ) and by qRT-PCR with Kapa Library Quantification Kit . Libraries were subsequently loaded on an Illumina HiSeq 2500 and achieved an average read depth of 10 million reads per replicate . WGS - Samples were aligned to Hg19 . Structural variation and indel Analysis By Assembly ( SvABA ) v0 . 2 . 1 was used to identify large deletions and structural variations . ( Wala et al . , 2018 ) . WES – Samples were aligned to Hg19 . Samples were analyzed using GATK v4 . 0 . 4 . 0 for copy number variation ( CNV ) , single nucleotide polymorphism ( SNP ) and indel identification across our RMC samples simultaneously using filtering parameters set by GATK ( Broad Institute , Cambridge , MA ) ( McKenna et al . , 2010 ) . MuTect 2 . 0 was used to identify candidate somatic mutations and these were filtered based on the Catalogue of Somatic Mutations in Cancer ( COSMIC ) ( Forbes et al . , 2015 ) . RNA – CLF_PEDS0005 and CLF_PEDS9001 samples and TARGET Wilms and Rhabdoid tumor samples ( dbGaP phs000218 . v19 . p7 ) were aligned or re-aligned with STAR and transcript quantification performed with RSEM . The TARGET initiative is managed by the NCI and information can be found at https://ocg . cancer . gov/programs/target . These normalized samples were then analyzed with t-SNE ( Maaten , 2014 ) . The following parameters were used in the t-SNE analyses: perplexity 10 , theta 0 , iterations 3000 . RNA sequencing samples in the SMARCB1 re-expression studies were subsequently aligned and analyzed with the Tuxedo suite ( e . g . aligned with TopHat 2 . 0 . 11 , abundance estimation with CuffLinks , differential analysis with CuffDiff and CummeRbund ) ( Trapnell et al . , 2010 ) . For the comparison to previously published work ( Wang et al . , 2017 ) , the published RNA sequencing samples along with our samples were re-aligned with TopHat 2 . 0 . 11 and analyzed with the Tuxedo suite . For samples treated with DMSO or MLN2238 , samples were aligned as above and analyzed with DESeq2 ( Love et al . , 2014 ) . gDNA was extracted using QIAamp DNA mini kit ( Qiagen ) . We performed a mixing study of our RMC cell lines with gDNA isolated from the G401 MRT cell line and then performed PCR amplification . We determined that the lower limits of detection of these fusions with our methods were ~1% of tumor cell line gDNA with a minimum 50 ng of gDNA . We subsequently performed the same PCR reactions with 100 ng of gDNA from the tumor tissue samples . Samples were gel purified and submitted for Sanger sequencing ( Eton Bio ) . We found that the sequences from the tumor tissue samples matched those of the cell lines , confirming that the genomic alterations that we found in the cell lines reflect those found in the original tumor . Primers utilized were CLF_PEDS0005 chr1 forward ( ATAAGACATAACTTGGCCGG ) , CLF_PEDS0005 SMARCB1 reverse ( TTTTCCAAAAGGTTTACAAGGC ) , CLF_PEDS9001 chr12 forward ( AAAAGCATATGTATCCCTTGCT ) , CLF_PEDS9001 SMARCB1 reverse ( CCTCCAGAGCCAGCAGA ) . RNA was extracted as above and normalized using the Nanodrop to one microgram . One microgram of RNA was then added to the High Capacity cDNA Reverse Transcription Kit ( Thermo Fisher Scientific ) and PCR reactions were performed as per manufacturer’s recommendations . cDNA was then diluted and added to primers ( Supplementary file 11 ) and Power SYBR Green PCR Master Mix ( Thermo Fisher Scientific ) . Samples were run on a BioRad CFX384 qPCR System in a minimum of technical quadruplicates . Results shown are representative of at least two biological replicates . We performed Affymetrix Human Genome U133 Plus 2 . 0 on our RMC cell lines . We then combined the following GEO datasets using a GenePattern module with robust multi-array ( RMA ) normalization GSE64019 , GSE70421 , GSE70678 , GSE36133 , GSE94321 ( Barretina et al . , 2012; Calderaro et al . , 2016; Johann et al . , 2016; Richer et al . , 2017; Wang et al . , 2017 ) . We utilized COMBAT and then tSNE to account for batch effects and to identify clusters of similarity ( Chen et al . , 2011; Johnson et al . , 2007; Maaten , 2014 ) . Nuclear extracts and gradients were performed as previously published ( Boulay et al . , 2017 ) . Briefly , 500 micrograms of nuclear extract from approximately 30 million cells were resuspended in 0% glycerol HEMG buffer containing 1 mM DTT , cOmplete protease inhibitors and PhosStop ( Roche ) . This was placed on a 10–30% glycerol gradient and ultracentrifuged at 40 k RPM for 16 hr at 4C . Following centrifugation , fractions of 550 µL were collected . Samples were then prepared with 1x LDS Sample Buffer ( Thermo ) . Samples were run on a 4–12% Bis-Tris gel and then transferred by immunoblotting in tris-glycine-SDS buffer with methanol . Immunoblots were subsequently blocked with Licor Blocking Buffer ( Lincoln , NE ) and then incubated with antibodies as noted in the methods section . Immunoblots shown are representative of at least two biological replicates . Primary cell lines were authenticated by Fluidigm or WES/WGS sequencing or by qRT-PCR . Cells were tested for mouse antibody production ( Charles Rivers Laboratories; Wilmington , MA ) and mycoplasma using the Lonza MycoAlertPLUS Mycoplasma Detection Kit ( Morristown , NJ ) . Established cell lines were authenticated by Fluidigm SNP testing . Cell lines were refreshed after approximately 20 passages from the frozen stock . Bortezomib , MLN2238 , nocodazole and RO-3306 were purchased from SelleckChem ( Houston , TX ) for the in vitro studies . Compounds were resuspended in DMSO and frozen down in 20 microliter aliquots to limit freeze-thaw cycles . Compounds were added as noted in the figure legends . In vitro studies used 15 nM for bortezomib and 100 nM for MLN2238 or are otherwise specified in the text . For the pulse experiments , we used 2 . 5 micromolar of MLN2238 . pDONR223 SMARCB1 was Sanger sequenced ( Eton Bio ) and aligned to variant 2 of SMARCB1 . SMARCB1 was subsequently cloned into the inducible vector pLXI401 or pLXI403 ( Genomics Perturbation Platform at the Broad Institute , Cambridge , MA ) by Gateway Cloning . LacZ was used as a control . Lentivirus was produced using tet-free serum ( Clontech , Mountain View , CA ) . Cell lines were infected with lentivirus to generate stable cell lines . Cell lines were then confirmed to re-express LacZ or SMARCB1 by titrating levels of doxycycline ( Clontech ) . Parental cell lines were treated with increasing doses of doxycycline to determine the toxicity to cells and measured by Cell-TiterGlo after 96 hr . Cells were then grown with or without doxycycline in a six well plate . Cells were counted by Trypan blue exclusion on a ViCELL XR ( Beckman Coulter , Brea , CA ) every 4–5 days . Results shown are the average of at least three biological replicates . Cells were plated in a six well dish and treated with or without doxycycline for up to 7 days . Senescence was assessed with the Senescence β-Galactosidase Staining Kit without modifications ( Cell Signaling Technologies , Danvers , MA ) . These were performed as previously published ( Hong et al . , 2016 ) . Briefly , we utilized the DCT v1 . 0 shRNA ( CP1050 ) and sgRNA ( CP0026 ) libraries from the Broad Institute Genetic Perturbation Platform ( GPP ) ( http://www . broadinstitute . org/rnai/public/ ) . Viruses from both pools were generated as outlined at the GPP portal . As CLF_PEDS0005_T2A and CLF_PEDS0005_T2B were expanded , we performed titrations with the libraries as outlined at the GPP portal . For the sgRNA pool , both cell lines were first transduced with Cas9 expression vector pXPR_BRD111 . We screened the DCT v1 . 0 shRNA library in biological replicates and the Cas9 expressing cell lines with the sgRNA pools at an early passage ( <20 ) and at a multiplicity of infection ( MOI ) <1 , at a mean representation rate above 500 cells per sgRNA or shRNA . gDNA was extracted and was submitted for sequencing of the barcodes . We achieved sequencing depths of at least 500 reads per shRNA or sgRNA . sgRNAs targeting the genes noted in the manuscript ( e . g . PSMB5 , UBE2C and controls; Supplementary file 12 ) were generated and introduced into the pXPR_BRD003 backbone . These were then sequence confirmed by Sanger sequencing ( Eton Biosciences ) . Lentivirus was produced and used for infection to generate stable cell lines expressing Cas9 . Cells were counted or harvested for protein as noted in the text . After indicated treatments , cell lysates were harvested using RIPA buffer ( Cell Signaling Technologies ) with protease inhibitors ( cOmplete , Roche ) and phosphatase inhibitors ( PhosSTOP , Roche ) . Antibodies used were as follows: ARID1A ( Santa Cruz; sc-373784 ) , α-tubulin ( Santa Cruz; sc-5286 ) , β-Actin ( C-4 ) ( Santa Cruz; sc-47778 ) , β-Actin ( Cell Signaling; 8457 ) , BAF57/SMARCE1 ( Bethyl Laboratories , A300-810A ) , BAF60a ( Santa Cruz; sc-135843 ) , BAF155 ( Cell Signaling; 11956 ) , BAF170 ( Santa Cruz; sc-166237 ) , SMARCA4 ( Santa Cruz; 17798 ) , Cleaved Caspase-3 ( Cell Signaling; 9664 ) , c-MYC ( Santa Cruz; sc-764 ) or c-MYC ( Cell Signaling; 9402 ) , cyclin B1 ( Cell Signaling; 4135 and 4138 ) , cyclin D1 ( Santa Cruz; sc-718 ) , GAPDH ( Cell Signaling; 2118S and 97166S ) , GRP78 ( Rockland Antibodies , Limerick , PA; 200–301 F36 ) , IRE1-alpha ( Cell Signaling; 3294 ) , lamin A/C ( Cell Signaling; 2032 ) , PSMB5 ( Abcam , Cambridge , MA; ab3330 ) , SMARCB1/SNF5 ( Bethyl A301-087A ) , UBE2C ( Proteintech , Rosemont , IL; 66087–1 ) . Results shown are representative of at least two biological replicates . Cells were treated using the conditions noted in the text . One million cells were spun down and resuspended in PBS . Cells were then subjected to FITC Annexin V and PI staining as described ( BD Pharmigen; 556547 ) . Another set of cells were subjected to PI/RNAse staining ( BD Pharmingen; 550825 or Invitrogen F10797 ) . Samples were analyzed within 1 hr with the SA3800 Spectral Analyzer ( Sony Biotechnology ) . Biological replicates were performed . Data were analyzed with FlowJo v10 ( FlowJo , Ashland , OR ) . We measured the cell’s ability to cleave Suc-LLVY-aminoluciferin utilizing Proteasome-Glo ( Promega ) following a one-hour treatment with the noted proteasome inhibitor and measured luminescence . Results shown are from at least two biological replicates . This research project has been reviewed and approved by the Dana-Farber Cancer Institute’s Animal Care and Use Committee ( IACUC ) , in compliance with the Animal Welfare Act and the Office of Laboratory Welfare ( OLAW ) of the National Institutes of Health ( NIH ) . Five million cells of G401 in 100 µL of a 50% PBS/50% Matrigel ( BD Biosciences ) mixture were injected subcutaneously into flanks unilaterally in Taconic NCr-Nude ( CrTac:NCr-Foxn1nu ) female mice at 7 weeks of age . When tumors reached approximately 150 mm3 , mice were randomized into various treatment groups: vehicle control ( 5% 2-hydroxypropyl-beta-cyclodextrin ( HPbCD ) ) or MLN2238 ( 7 mg/kg IV twice a week for 4 weeks ) . MLN2238 ( diluted in 5% 2-HPbCD ) was purchased from MedChem Express . Randomizations to the treatment arm occurred . Blinding was not performed . Statistical analysis was performed using the two-tailed t-test or Mantel-Cox as noted in the text .	Renal medullary carcinoma ( RMC for short ) is a rare type of kidney cancer that affects teenagers and young adults . These patients are usually of African descent and carry one of the two genetic changes that cause sickle cell anemia . RMC is an aggressive disease without effective treatments and patients survive , on average , for only six to eight months after their diagnosis . Recent genetic studies found that most RMC cells have mutations that prevent them from producing a protein called SMARCB1 . SMARCB1 normally acts as a so-called tumor suppressor , preventing cells from becoming cancerous . However , it was not clear whether RMCs always have to lose SMARCB1 if they are to survive and grow . Often , diseases are studied using laboratory-grown cells and tissues that have certain features of the disease . No such models had been created for RMC , which has slowed efforts to understand how the disease develops and find new treatments for it . Hong et al . therefore worked with patients to develop new lines of cells that can be used to study RMC in the laboratory . These RMC cells started dying when they were given copies of the SMARCB1 gene , which supports the theory that RMCs have to lose SMARCB1 in order to grow . Hong et al . then used a set of genetic reagents that can suppress or delete genes that are targeted by drugs , and followed this by testing a range of drugs on the RMC cells . Drugs and genetic reagents that reduced the activity of the proteasome – the structure inside cells that gets rid of old or unwanted proteins – caused the RMC cells to die . These proteasome inhibitor drugs also killed other kinds of cancer cells with SMARCB1 mutations . Proteasome inhibitors are already used to treat different types of cancer . Potentially , a clinical trial could be run to see if they will treat patients whose cancers lack SMARCB1 . Further work is also needed to determine the exact link between SMARCB1 and the proteasome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2019	Renal medullary carcinomas depend upon SMARCB1 loss and are sensitive to proteasome inhibition
Many organisms in nature have evolved mechanisms to tolerate severe hypoxia or ischemia , including the hibernation-capable Arctic ground squirrel ( AGS ) . Although hypoxic or ischemia tolerance in AGS involves physiological adaptations , little is known about the critical cellular mechanisms underlying intrinsic AGS cell resilience to metabolic stress . Through cell survival-based cDNA expression screens in neural progenitor cells , we identify a genetic variant of AGS Atp5g1 that confers cell resilience to metabolic stress . Atp5g1 encodes a subunit of the mitochondrial ATP synthase . Ectopic expression in mouse cells and CRISPR/Cas9 base editing of endogenous AGS loci revealed causal roles of one AGS-specific amino acid substitution in mediating cytoprotection by AGS ATP5G1 . AGS ATP5G1 promotes metabolic stress resilience by modulating mitochondrial morphological change and metabolic functions . Our results identify a naturally occurring variant of ATP5G1 from a mammalian hibernator that critically contributes to intrinsic cytoprotection against metabolic stress . Arctic ground squirrels ( AGS , Urocitellus parryii ) survive harsh winter environmental conditions through hibernation . By virtue of their profound ability to suppress metabolism and core temperature , with body temperatures dropping below 0°C , AGS are known as ‘extreme’ hibernators ( Barnes , 1989 ) . Hibernation in AGS can last 7 months and is characterized by drastic ( >90% ) reductions in basal metabolic rate , heart rate , and cerebral blood flow ( Buck and Barnes , 2000 ) . Curiously , hibernation is interrupted periodically by interbout arousal ( IBA ) episodes in which temperature and cerebral blood flow normalize rapidly ( Drew et al . , 2004; Karpovich et al . , 2009 ) . Nonetheless , AGS suffer no ischemic injury during hibernation or reperfusion injury during an IBA . Hibernating ground squirrels are resistant to ischemic and reperfusion injuries in numerous models , including brain and heart tissues after cardiac arrest in vivo and hippocampal slice models derived from animals during an IBA ( Dave et al . , 2009; Quinones et al . , 2016; Bhowmick et al . , 2017; Bogren et al . , 2014 ) . This resilience to reperfusion injury does not depend on temperature of the animal or season ( Bhowmick et al . , 2017 ) . In addition , AGS neural progenitor cells ( NPCs ) demonstrate resistance to oxygen and glucose deprivation ex vivo ( Drew et al . , 2016 ) . Together , these studies suggest that in addition to physiological adaptations , AGS possess cell autonomous genetic mechanisms that contribute to intrinsic tolerance to metabolic stress or injury . Proteomic and transcriptomic investigations have comprehensively catalogued the impact of season , torpor , and hibernation on cellular and metabolic pathways in several different tissues of hibernating ground squirrels , including the brain ( Quinones et al . , 2016; Ballinger et al . , 2016; Chang et al . , 2018; Gehrke et al . , 2019; Hampton et al . , 2013; Luan et al . , 2018; Andrews , 2019; Hindle et al . , 2014 ) . Although the mechanisms underlying hibernating ground squirrel ischemia and hypothermia tolerance in the brain are not fully elucidated , studies suggest that post-translational modifications , regulation of cytoskeletal proteins , and upregulation of antioxidants play a prominent role ( Bhowmick and Drew , 2017; Lee et al . , 2007; Tessier et al . , 2019 ) . Gene expression profiling and bioinformatic analyses also indicate the cytoprotective contributions of mitochondrial and lysosomal pathways in adapting to hypothermia and hypoxia in ground squirrel and marmot species ( Bai et al . , 2019; Ou et al . , 2018 ) . In neurons differentiated from 13-lined ground squirrel ( 13LGS ) induced pluripotent stem cells ( iPSCs ) , Ou and colleagues found that hibernating ground squirrel microtubules retained stability upon exposure to hypothermia . The authors identified mitochondrial suppression of cold-induced reactive oxygen species ( ROS ) and preservation of lysosomal structure are key features of ground squirrel cytoprotection , and that pharmacological inhibition of ROS production or lysosomal proteases recapitulates the hypothermia-tolerant phenotype in human cells ( Ou et al . , 2018 ) . Taken together , these studies provide important insights into pathways mediating AGS tolerance to metabolic stress . However , these studies have not focused on specific genes and proteins with cytoprotective effects uniquely evolved in hibernating ground squirrels . As such , we know very little about mechanistic details underlying genetic contribution to intrinsic stress resilience in ground squirrels . Using a cDNA library expression-based genetic screen combined with phenotypic analyses of cell survival and mitochondrial responses to stress as compared in mouse versus AGS NPCs , we identified AGS transcripts imparting ex vivo cytoprotection against various metabolic stressors . We further use CRISPR/Cas9 DNA base editing ( Koblan et al . , 2018 ) to determine functional importance of amino acid substitutions uniquely evolved in AGS , and identified AGS ATP5G1L32 as a causal contributor to stress resilience in AGS , suggesting potential for targeting this component of ATP synthase for neuroprotective treatments . When growing under identical cell culture conditions , AGS and mouse NPCs exhibit similar morphology , growth rates and expression of Nestin and Ki67 , markers for proliferating NPCs ( Figure 1A–B and Figure 1—figure supplement 1A-E ) . Although superficially indistinguishable , mouse and AGS NPCs demonstrate markedly different responses to metabolic stressors . When exposed to hypoxia ( 1% O2 ) , hypothermia ( 31°C ) , or rotenone ( 30 µM ) , AGS NPCs exhibit profound resistance to cell death compared with mouse NPCs ( Figure 1C ) , recapitulating resilient AGS phenotypes found in previous studies ( Dave et al . , 2009; Bhowmick et al . , 2017; Bogren et al . , 2014; Drew et al . , 2016 ) . Moreover , measurement of in vitro oxygen consumption of AGS NPCs after sequential exposure to mitochondrial toxins demonstrates strikingly higher ‘spare respiratory capacity’ in response to FCCP ( Figure 1D and Figure 1—figure supplement 1F , G ) , indicating a greater metabolic reserve for stressors ( Nicholls and Budd , 2000 ) . Mitochondrial citrate synthase and oxidative phosphorylation ( OXPHOS ) enzymatic activities were similar between the two species , with the exception of complex IV ( Figure 1—figure supplement 1H ) . Interestingly , functional improvements in mitochondrial function were also mirrored by changes in mitochondrial dynamic organization following exposure to FCCP at doses that lead to mitochondrial depolarization ( Figure 1E ) . At baseline , mouse and AGS cells had similar mitochondrial organization as evidenced by similar mean branch length and number of cells with fragmented mitochondria ( Figure 1F ) . Following FCCP treatment , mouse cells demonstrated marked increases in mitochondrial fission with concurrent decreases in mean branch length . By contrast , AGS cells appeared largely resistant to mitochondrial fission induced by FCCP ( Figure 1G ) . Together , these results demonstrate intrinsic differential cell survival and mitochondrial responses to metabolic stresses between mouse and AGS NPCs . To identify cytoprotective genes expressed in AGS , we constructed a normalized cDNA expression library from AGS NPCs and introduced the library to mouse NPCs by nucleofection ( Bertram et al . , 2012; Figure 2—figure supplement 1A-B ) . Screening of inserts revealed the average library insert size was 2 . 4 kB . To minimize false negatives due to incorrect splice isoforms , we performed screens in triplicate and maintained representation at 1000 cells/open reading frame . Two days after AGS cDNA library nucleofection , we exposed cells to hypothermia ( 31°C ) for 3 days , hypoxia ( 1% ) for 2 days , or complex I inhibition ( rotenone ) for 3 days , respectively ( Figure 2A ) . We then isolated plasmids from surviving cells , amplified cDNA insert sequences by PCR and used next-generation sequencing to identify a total of 378 putative cytoprotective genes , three of which ( Ags Atp5g1 , Ags Manf , and Ags Calm1 ) provided cytoprotection in all three examined metabolic stress conditions ( see Figure 2B and Supplementary file 1 ) . Since a portion of mouse NPCs survived metabolic stresses even without AGS cDNA library expression or as a result of protective secreted factors , we anticipated false positive hits without cell autonomous cytoprotective effects . Thus , in this study , we focused on characterizing the nuclear-encoded mitochondrial protein AGS ATP5G1 that conferred cytoprotective effects independently confirmed under all three metabolic stress conditions ( Figure 2B , E–G ) . ATP5G1 is one of three ATP5G isoforms making up the C-subunit of mitochondrial ATP synthases , and is regulated distinctly from ATP5G2 or ATP5G3 ( Gay and Walker , 1985; De Grassi et al . , 2006; Wigington et al . , 2016 ) . As most identified genes do not appear to be differentially expressed between mouse and AGS NPCs ( Ou et al . , 2018 ) , we hypothesized that resistance to metabolic stress may be related to uniquely evolved AGS proteins . Based on multiple sequence alignment of the ATP5G1 protein family in mammals , we observed three AGS-unique amino acid substitutions and two small insertions/deletions at the N-terminal region of AGS ATP5G1 , whereas the C-terminal membrane-spanning segment is largely invariant ( Figure 2C and Figure 2—figure supplement 1C ) . We expanded the analysis of AGS-unique amino acid substitutions to other cytoprotective protein variants identified from the screen of the AGS cDNA library . In particular , we analyzed uniquely evolved AGS proteins by comparing sequence alignments of the screened cytoprotective candidates for two species of ground squirrels ( AGS and the 13LGS , Ictidomys tridecemlineatus ) against nine other reference species across mammalian subclasses . We calculated the Jensen- Shannon Divergence ( JSD ) score , which captures sequence conservation and difference from the background amino acid distribution , and average ground squirrel-versus-other mammalian block substitution matrix ( BLOSUM ) −62 scores for each unique residue ( Capra and Singh , 2007 ) . High JSD and low BLOSUM62 scores indicate chemically significant amino acid substitutions , and as such potentially important functional AGS adaptations . We found that the leucine-32 residue of AGS ATP5G1 in place of the otherwise highly conserved proline is unique to hibernating ground squirrels , and on conservation analysis scored among the highest of all AGS-unique amino acid substitutions in identified cytoprotective protein candidates from our screen ( Andersson et al . , 1997; Figure 2D and Supplementary file 2 ) . The N-terminal region of ATP5G proteins can undergo cleavage , but also modulate mitochondrial function directly , by unknown mechanisms ( Vives-Bauza et al . , 2010 ) . Although the three C-subunit proteins are identical in sequence , they cannot substitute for one another and are all required to constitute a fully functional C-subunit ( Vives-Bauza et al . , 2010; Sangawa et al . , 1997 ) . To determine the relative levels of ATP5G1 , −2 , and −3 in mouse and AGS NPCs , we performed qRT-PCR analysis with species and transcript-specific primers . We found that in both mouse and AGS NPCs , expression of Atp5g3 or Atp5g2 is greater than that of Atp5g1 , consistent with prior reports in human and mouse tissues ( Gay and Walker , 1985; Vives-Bauza et al . , 2010 ) . We found that the relative abundance of the Atp5g1 isoform is elevated nearly twofold in AGS NPCs ( Figure 2—figure supplement 1D ) . However , the relative abundance of the mature ATP5G ( subunit C ) protein or oligomycin sensitivity of complex V activity , is not different in mouse and AGS cells ( Figure 2—figure supplement 1E-F ) . Overexpression of the AGS variant of ATP5G1 in mouse NPCs confers cytoprotection in cells exposed to hypoxia , hypothermia , or rotenone ( Figure 2E–G ) . We found that this protective response is not present in NPCs overexpressing ATP5G1L32P . Conversely , overexpression of the human ATP5G1P32L , which mimics the wild-type AGS ATP5G1 variant , leads to enhanced cytoprotection in these conditions of metabolic stress compared to that of human ATP5G1 . The ATP5G1 substitutions did not alter the mitochondrial localization of ATP5G1 when expressed in either mouse or AGS NPCs ( Figure 3—figure supplements 1 , 2 ) . In addition , overexpression of the AGS variant of ATP5G1 recapitulated key features of the AGS resilient mitochondrial phenotype , including increasing spare respiratory capacity and reducing mitochondrial fission with reduced fragmentation and increased branch length of mitochondria in response to FCCP ( Figure 3 , Figure 3—figure supplement 3A ) . Interestingly , NPCs expressing the AGS ATP5G1L32P variant demonstrated reduced spare respiratory capacity and increased mitochondrial fragmentation compared to the AGS ATP5G1 over-expressing NPCs . Overexpression of human ATP5G1P32L improved survival to metabolic stressors and reduced mitochondrial fragmentation , but compared to AGS ATP5G1L32P spare respiratory capacity was not significantly improved . This may indicate that improving spare respiratory capacity itself is not the sole mechanism conferring resilience to metabolic stressors . Of note , expression of AGS ATP5G1 with two other identified AGS-unique amino acid substitutions ( N34D , T39P ) did not affect survival of mouse NPCs exposed to hypoxia , hypothermia , or rotenone ( Figure 3—figure supplement 3B–D ) . Together , these results reveal cytoprotective effects of AGS Atp5g1 when ectopically expressed in metabolic stress-susceptible mouse NPCs , and identify functional importance of the leucine-32 residue of AGS ATP5G1 uniquely evolved in AGS . Species-specific substitutions of amino acid residues at sites deeply conserved in mammals indicate either relaxed selective constrains at the sites during evolution or potentially adaptive significance functionally specific for that species . As ectopic expression may not fully reflect endogenous functions , precise manipulation of endogenous genetic loci is required to determine definitive causal contribution of ATP5G1L32 to the metabolic resilience of AGS . Using the recently reported adenine DNA base editor ( ABEmax; 22 ) , we successfully generated AGS cell lines homozygous for ATP5G1L32P by introducing a cytosine-to-thymine substitution in the ( - ) strand of Ags Atp5g1 ( Figure 4A , B ) . We isolated three clonal AGS NPC lines harboring the desired knock-in mutation ( ABE KI ) and two clonal lines that underwent editing and remained homozygous for the wild-type allele ( ABE WT ) . Compared to ABEmax-treated AGS cells without successful knock-in ( Figure 4—figure supplement 1A ) , ABE KI cell lines did not demonstrate differences in Atp5g1 mRNA expression , protein abundance , or complex V activity ( Figure 4—figure supplement 1B-C , Figure 4I ) . However , knock-in of the L32P residue resulted in markedly reduced survival of AGS NPCs following exposure to hypoxia , hypothermia , or rotenone ( Figure 4C ) . In addition , we found the ABE KI AGS NPCs exhibited marked reduction in ‘spare respiratory capacity’ and altered mitochondrial dynamics in response to FCCP treatment ( Figure 4D–H and Figure 4—figure supplement 1C ) . Although overall ATP5G protein abundance is unchanged ( Figure 4—figure supplement 1D-E ) , we used clear-native gel electrophoresis ( Kovalčíková et al . , 2019; Wittig and Schägger , 2009 ) and identified a reduced presence of ATP synthase dimers relative to the total amount of ATP synthase in ABE KI cells ( Figure 4J–K ) . Further biochemical experiments are necessary to delineate the specific mechanisms of how the AGS leucine-32 substitution affects the assembly or stability of ATP synthase complex proteins . Nonetheless , genetic evidence in our study based on ectopic expression and specific CRISPR base editing of endogenous loci demonstrates causal roles of the AGS leucine-32 substitution in cytoprotection . Collectively , these results identify a naturally occurring cytoprotective AGS variant that contributes to cytoprotection against various metabolic stresses likely by modulating mitochondrial function . Previous studies have indicated that hibernating organisms evolved numerous physiological and cellular mechanisms enabling survival during the stressed metabolic conditions accompanying hibernation ( Bai et al . , 2019; Ou et al . , 2018; Ballinger et al . , 2017 ) . However , we still know little about the mechanistic details of how AGS protein-coding genetic variants contribute to intrinsic cytoprotective functions . We show that ex vivo cultured AGS NPCs can recapitulate remarkable intrinsic resilience to hypoxia , hypothermia , and other metabolic stressors . Additionally , using an unbiased cDNA expression screening and bioinformatic strategy , we identified numerous AGS transcripts and uniquely evolved AGS amino acid substitutions potentially contributing to cytoprotection . We focused on discerning the protective effect of AGS ATP5G1 , a nuclear-encoded mitochondrial protein , given that it was one of only three genes identified in all three metabolic stress paradigms and the prominent mitochondrial resilience phenotype of AGS NPCs . We hypothesize that analogous to amino acid substitutions in several human proteins providing adaptive benefits ( Simonson et al . , 2010; Song et al . , 2014; Xiang et al . , 2013; Yates and Sternberg , 2013 ) , substitutions in AGS ATP5G1 may underlie AGS adaptive mechanisms contributing to its robust cytoprotective phenotype . Using the dCas9 ABE technology , we validated a unique AGS ATP5G1L32 amino acid substitution in the N-terminal region of ATP5G1 that leads to improvements in mitochondrial physiologic parameters . Thus , our study used CRISPR base editing in non-model organism hibernator cells , for the first time to our knowledge , to identify a naturally occurring cytoprotective protein variant from AGS . CRISPR edited ATP5G1L32P did not fully abolish the metabolic resilience phenotype in AGS NPCs , indicating that other gene variants may also be involved . The robust ex vivo paradigm of AGS phenotypes established from our study makes it tractable to investigate additional gene and protein variants that contribute to the metabolic resilience phenotype in AGS . Further understanding the gene variants and mechanisms responsible for the AGS phenotype has important implications for novel neuroprotective treatments in ischemic diseases as well as promoting survival of neural stem cell grafts ( Bernstock et al . , 2017 ) . Mitochondrial metabolic dysfunction is central to ischemia and reperfusion injury . Physiologic , transcriptomic , and proteomic studies have highlighted the importance of ketone and fatty acid metabolism in hibernating states ( Brown and Staples , 2014; Xu et al . , 2013 ) as well as pointed to a role for specific post-translational protein modifications in the differential regulation of metabolic pathways in hibernation ( Ballinger et al . , 2016; Chung et al . , 2013; Herinckx et al . , 2017 ) . Specific variants of neuroprotective proteins have also been identified to be upregulated in ground squirrels during hibernation including s-humanin , however , the phenotypic or mechanistic consequences of these variants are not known ( Szereszewski and Storey , 2019 ) . We expanded this body of knowledge , by identifying altered mitochondrial dynamics and enhanced spare respiratory capacity in cells of AGS as potentially adaptive cellular mechanisms in hibernating animals . This mitochondrial phenotype is likely responsible for the broad resilience of AGS cells against a wide range of metabolic stressors . Spare respiratory capacity , as measured by FCCP-stimulated oxygen consumption , represents a marker for cellular metabolic reserves , correlates with metabolic resilience ( Nicholls and Budd , 2000 ) and is thought to be determined by the oxidative phosphorylation machinery ( Pfleger et al . , 2015; Yadava and Nicholls , 2007 ) . Notably , human and mouse NPCs and neural cells have been reported to have diminished spare respiratory capacity as they may respire maximally at baseline ( Khacho et al . , 2016; Lorenz et al . , 2017 ) . However , AGS demonstrate marked elevations in spare respiratory capacity compared to mouse cells , which likely explains marked AGS NPC survival even under complex I inhibition by rotenone ( Yadava and Nicholls , 2007 ) . While the elevated spare respiratory capacity is likely the result of AGS adaptations in numerous metabolically active proteins , the importance of the ATP5G1 variant is highlighted by our experimental evidence demonstrating improvement in spare respiratory capacity in mouse NPCs over-expressing AGS ATP5G1 variants and decreased spare respiratory capacity in AGS NPCs with ATP5G1 L32P knock-in . A critical role for ATP5G1 in cellular energetics is also supported by recent work uncovering ATP5G1 as one of the major effectors of the transcription factor , BCL6 , in regulating adipose tissue energetics as well as maintaining thermogenesis in response to hypothermia ( Kutyavin and Chawla , 2019; Senagolage et al . , 2018 ) . Mitochondrial fission and fusion are regulated by cellular metabolic state and a host of regulatory proteins , many of which have been implicated in cell survival response to stresses ( Labbé et al . , 2014 ) . While metabolic stresses often lead to mitochondrial fission followed by apoptosis , mitochondrial fusion and resistance to fission in response to stress are anti-apoptotic ( Abdelwahid et al . , 2007; Chen et al . , 2007 ) . Fusion is hypothesized to allow for complementation of damaged and dysfunctional mitochondria , and in states of metabolic stress , hyperfusion of mitochondria helps maintain mitochondrial membrane potential and cell viability ( Gomes et al . , 2011 ) . The increase in fusion and improvement in cell survival in mouse NPCs over-expressing AGS ATP5G1 and loss of resilient metabolic phenotypes in AGS cells carrying ATP5G1L32P underscore the importance of this pathway in altering mitochondrial morphologic response to metabolic stresses and increasing the metabolic oxidative capacity of cells . In mammals , many of the approximately 1000 nuclear-encoded mitochondrial proteins contain a unique mitochondrial targeting sequence ( MTS ) providing a high degree of specificity in regulating mitochondrial import and sorting . These mitochondrial targeting and processing functions are regulated by the highly conserved mitochondrial membrane translocating protein complexes ( TOM and TIM ) and MTS cleaving proteins , mitochondrial processing peptide ( MPP ) and mitochondrial intermediate peptide ( MIP ) . Processing of ATP5G1 and its incorporation of the mature peptide into oligomeric c-rings and Complex V-Fo appear to involve cleavage by MPP and stabilization by TMEM70 ( Kovalčíková et al . , 2019 ) . We did not find evidence that the ATP5G1 MTS sequence variations from AGS and human/mouse affected the mitochondrial localization or cleavage of the immature protein . This is likely due to evolutionarily conserved mitochondrial import sequence motifs and the putative ATP5G1 MPP/MIP cleavage site ( xRx↓ ( F/L/I ) xx ( S/T/G ) xxxx↓; see Figure 2—figure supplement 1C; Gakh et al . , 2002 ) . Interestingly , under native gel electrophoresis conditions in ABE KI NPC mitochondria , we observed a reduction in ATP synthase dimers relative to total ATP synthase . Additional supporting evidence is required to understand the mechanistic basis of this effect . We speculate that the AGS variant alters ATP5G1 processing which subsequently affects downstream dimerization of ATP synthases . Suprastructual alterations in ATP synthase organization are known to be critical to mitochondrial morphology and formation of the mitochondrial permeability transition pore ( MPTP ) ( Nesci and Pagliarani , 2019 ) . Though the exact nature of the relationship between ATP5G1 and the MPTP is controversial , many studies demonstrate improved bioenergetic responses and cell survival with ATP synthase dimerization ( Bonora et al . , 2017; Daum et al . , 2013; García-Aguilar and Cuezva , 2018 ) . Others have postulated that the cleaved ATP5G1 N-terminal mitochondrial targeting sequence modulates mitochondrial function downstream of Complex IV distinct from the functionally active C-terminal protein ( Vives-Bauza et al . , 2010 ) . Although increased abundance of the Atp5g1 transcript in AGS compared with mouse NPCs could contribute to the altered mitochondrial function as in prior investigations of regulation of ATP synthase in mouse brown adipose tissue ( Andersson et al . , 1997 ) , the ABE ATP5G1L32P KI cells did not demonstrate a difference in Atp5g1 mRNA transcript abundance , further supporting the notion that AGS ATP5G1L32P contributes to cytoprotection likely via post-transcriptional processing of ATP5G1 . Precise mechanisms of how AGS ATP5G1L32P affects mitochondrial function and metabolic stress resilience phenotypes await future investigations . Further unraveling of the mechanisms underlying AGS mitochondrial and cellular resilience to metabolic stress or injuries holds the hope of finding novel cytoprotective strategies that may lead to improved treatments for human diseases . Systematic investigation of additional cytoprotective genes and amino acid substitutions identified from AGS should provide important insights into the mechanism and pathways underlying intrinsic stress resilience to metabolic stresses . The use of CRISPR gene editing technologies coupled with phenotypic analysis in AGS NPCs is a new and powerful approach to evaluate causal roles of genetic variants in conferring phenotypic traits of AGS traditionally intractable to study . Identification and analysis of such causal variants for stress resilience in AGS may help develop pharmacological , gene therapy , or CRISPR/genome editing-based therapeutic strategies to treat human ischemic disorders , including stroke and heart attack . AGS NPCs ( Neuronascent , Gaithersburg , MD , USA ) and mouse NPCs ( gift of Song lab , Baltimore , MD ) have been previously described ( Drew et al . , 2016; Ma et al . , 2009 ) . They were grown under standard conditions at 37°C and 5% CO2 with NeuroCult basal media ( STEMCELL , Vancouver , BC , CA ) with EGF ( 50 ng/ml , PeproTech , Inc , Rocky Hill , NJ , USA ) , FGF ( 100 ng/ml , PeproTech , Inc ) , heparin ( 0 . 002% ) , and proliferation supplements ( STEMCELL ) . Early passage cultures ( P2 ) were expanded and frozen and thawed in batches for use in experiments . These cultures contain cells ubiquitously expressing the NPC marker , Nestin , and the proliferation marker , Ki-67 ( Figure 1—figure supplement 1 ) . For in vitro modeling of metabolic stress , cells were exposed to either: ( i ) 1% hypoxia in a specialized incubator ( Nuaire , Plymouth , MN , USA ) saturated with Nitrogen/5% CO2; ( ii ) hypothermia in standard incubators maintained at lower temperatures; and ( iii ) complex I inhibition with the addition of rotenone to cell media . For cell proliferation determination , wells were seeded in triplicate with 50 , 000 cells . On subsequent consecutive days , cells were detached with Accutase ( STEMCELL ) and counted by automated cytometry ( Nanoentek , Waltham , MA , USA ) . The pHAGE-ATP5G plasmids were generated by direct PCR and PCR fusions; and the point mutation plasmids generated using Q5 site-directed mutagenesis ( New England Biolabs , Beverly , MA , USA ) . For lentiviral transfection , the plasmids with packaging plasmids were co-transfected into HEK293FT ( with a ratio of 2:1 . 5:1 . 5 ) using Turbofect reagent ( Thermo Fisher Scientific Inc , Waltham , MA , USA ) according to the manufacturer’s instructions . Lentivirus-containing medium was filtered from the post-transfection supernatant and used for transduction of HEK293T cells or mouse NPCs . All lentivirus-infected cells were cultured in the medium containing Polybrene ( 4 μg/ml; Sigma Aldrich , St . Louis , MO , USA ) for 8 hr before changing media . Forty-eight hours after transduction , the cells were selected with 10 µg/ml Blastidicin S ( Thermo Fisher Scientific Inc ) . ATP5G1L32P NPCs were generated using the dCas9 base editor , ABEmax ( gift from David Liu , Addgene #112095 ) , as previously described ( Koblan et al . , 2018 ) . Briefly , a synthetic sgRNA ( TCCTCTAGTCTATTCAGGAA ) was selected by manual inspection of the AGS Atp5g1 sequence for a PAM ( NGG ) site near the desired edit on the ( - ) strand of the gene . AGS NPCs were nucleofected ( Amaxa 4D , program DS113 ) in P3 solution ( Lonza , Alpharetta , GA , USA ) containing pCMV ABEmax ( 500 ng/200 , 000 cells ) . Following a 48 hr recovery period , the same cells were nucleofected with the synthetic sgRNA sequence above ( 100 pmol , Synthego , Menlo Park , CA , USA ) . Cells were expanded and then clonally plated . Clones were screened by PCR as the desired base edit also introduced a new BfaI restriction enzyme cutting site . Sanger sequencing was used to confirm the two WT and three KI clone sequences utilized . Potential off-target effects of CRISPR/Cas9 cleavage were analyzed by Sanger sequencing of the top 5 predicted off-target genomic locations [https://mit . crispr . edu] , which demonstrated a lack of indels for all clones used in subsequent analysis . Mouse and AGS cells were plated in 24 or 96-well plates and grown to 70% confluence . Cells were exposed to metabolic stress paradigms as above , and detached and floating cells collected by centrifugation and washed with 1 ml PBS . The collected cells were resuspended with 200 μl PBS with addition of 0 . 2 μl Sytox blue ( 1 µM; Thermo Fisher Scientific ) or propidium iodide ( 2 μg/ml ) for an additional 5 min . The fluorescence intensity was measured for individual cells using automated cytometry ( Nanoentek ) or flow cytometry ( BD Biosciences , San Jose , CA , USA ) within 20 min of staining , and the percentage of cell death quantified using the FlowJo software . RNA was isolated from AGS NPC cells grown under standard conditions . A normalized cDNA library was generated by a commercial research partner ( Bio S and T , Montreal , QC , Canada ) from RNA extracted from AGS NPCs . Library quality and normalization is shown in Figure 2—figure supplement 1A and B . For library screening , plates containing 1 × 107 mouse NPCs cells were grown in triplicate and nucleofected with 200 , 000 clones each . Plates were exposed to one of three metabolic stress conditions ( hypoxia , hypothermia , or rotenone treatment ) for 48–96 hr . Following this treatment , plasmid DNA was purified from surviving cells and PCR-amplified AGS cDNA inserts subjected to next-generation sequencing . Resulting fastq files were trimmed ( Trim Galore ! ) and mapped to the Ictidomys Tridecemlineatus genome ( SpeTri2 . 0 ) using HISAT2 . Mapped reads were subjected a custom pipeline for analyzing amino acid substitutions ( https://github . com/evanmlee/MaLab_spec_subs; copy archived at https://github . com/elifesciences-publications/MaLab_spec_subs; Singhal , 2020 ) . Briefly , protein sequences of mapped genes were queried by gene symbol and downloaded from OrthoDBv10 for 10 species ( 13LGS , Mus musculus , Rattus norvegicus , Sorex araneus , Pongo abelii , Homo sapiens , Equus caballus , Bos taurus , Oryctolagus cuniculus , Sus scrofa ) . OrthoDB data was filtered by matching records against accepted GeneCards aliases for each gene ( Kriventseva et al . , 2019 ) . Multiple records per species were resolved using maximum percent identity against the accepted human , mouse , and 13LGS sequences , such that only one record per species was used for alignment . AGS protein sequences were downloaded from the Entrez Protein database . Multiple AGS isoforms were resolved by best identity match to the OrthoDB sequence data . The final protein sequence set was aligned with KAlign 2 . 04 ( Lassmann and Sonnhammer , 2005 ) . From aligned sequences , GS-specific residue substitutions were defined as amino acid variants present in 13LGS and AGS sequences and present in no other included species . For each GS-specific residue , sequence weights , JSD , and average GS-versus-outgroup BLOSUM62 scores were calculated as described previously ( Capra and Singh , 2007 ) . BLOSUM62 scores were used instead of point-accepted mutation scores in order to prioritize protein sequence changes with higher probability of potential chemical and functional difference . JSD was used to capture sequence conservation and difference from the background amino acid distribution . BLOSUM62 scores were calculated for GS residues against all other mammalian species sequences and averaged to give GS vs Outgroup BLOSUM62 . For the entire screened cytoprotective protein dataset , JSD and BLOSUM62 score were plotted for individual genes of interest against the remaining dataset . Analysis of mitochondrial respiratory potential was performed using a flux analyzer ( Seahorse XFe96 Extracellular Flux Analyzer; Seahorse Bioscience , North Billerica , MA , USA ) with a Seahorse XF Cell Mito Stress Test Kit according to the manufacturer’s instructions . Basal respiration and ATP production were calculated to evaluate mitochondrial respiratory function according to the manufacturer’s instructions . After the measurement , cells were harvested to count the cell number , and each plotted value was normalized relative to the number of cells used . Briefly , NPCs were seeded ( 25 , 000 cells/well ) into each well of XFe96 cell culture plates and were maintained in standard culture media . After 2–3 days in culture , cells were equilibrated in unbuffered XFeassay medium ( Seahorse Bioscience ) supplemented with glucose ( 4 . 5 g/L ) , sodium pyruvate ( 25 mg/L ) and transferred to a non-CO2 incubator for 1 hr before measurement . Oxygen consumption rate ( OCR ) was measured with sequential injections of oligomycin , FCCP , and rotenone/antimycin A . Analysis of mitochondrial respiratory chain complex I , II , and IV activity was measured in mitochondrial extracts using complex enzyme activity colorimetric or absorbance-based assays ( ab109721 , ab10908 , ab109911; Abcam , Cambridge , MA ) . Complex V activity was measured with Complex V Mitocheck kit ( Cayman Chemical , Ann Arbor , MI , USA ) and citrate synthase activity with a Citrate Synthase Enzyme Assay ( Detroit R and D , Detroit , MI ) . Mitochondrial extracts ( 50 μg ) were obtained as previously described ( Clayton and Shadel , 2014 ) and used to measure time-dependent absorbance alterations on a multi-well plate reader ( SprectraMax , Molecular Devices , San Jose , CA , USA ) . Mitochondrial membrane potential was evaluated by loading 1 × 105 cells in triplicate with the lipophilic positively charged dye tetramethylrhodamine ethyl ester ( TMRE , 50 nM ) . For depolarization control wells , 1 µM FCCP was added . Excitation and emission wavelengths ( 530 and 580 nm , respectively ) were measured on a multi-well plate reader . Mitochondrial localization of ATP5G1 constructs as well as morphology and fission/fusion is assessed in mouse and AGS NPCs nucleofected with mCherry or mEmerald-mito7 ( Gift from Michael Davidson , Addgene #55102 , 54160 ) as a mitochondrial marker and grown on glass coverslips in standard media ( Olenych et al . , 2007 ) . Cells are allowed to recover for 48 hr and then fixed with paraformaldehyde ( 4% ) one hour following treatment with FCCP ( 1 μM ) or DMSO . High magnification images of cells are captured by confocal microscopy ( DM6 , Leica , Wetzlar , Germany ) and mitochondrial morphological characteristics were assessed with the Mitochondrial Network Analysis ( MiNA ) toolset in J-image as previously described ( Valente et al . , 2017; Martín-Maestro et al . , 2017 ) . Briefly , the plugin converts confocal images to binary pixel features and analyzes the spatial relationship between pixels . The parameters analyzed are: ( i ) individual mitochondrial structures; ( ii ) networked mitochondrial; and ( iii ) the average of length of rods/branches . Twenty randomly chosen fields containing 30–50 cells were used to quantify the morphological pattern and network branch lengths of mitochondria . We classify the mitochondrial morphology as fragmented when the appearance is completely dotted with branch lengths < 1 . 8 μm . For SDS-PAGE , Laemmli loading buffer ( Bio-Rad Lab , Hercules , CA , USA ) plus 5% β-mercaptoethanol was added to protein extracts from cell pellets reconstituted in cell lysis buffer ( Cell Signaling Technology , Danvers , MA ) before heating at 95°C for 5 min . Around 30 μg of whole cell protein lysate samples were separated on 4–15% mini-PROTEIN GTX precast gels , and transferred to nitrocellulose membranes ( Bio-rad ) . For native electrophoresis , 20 μg of mitochondrial protein extracts were resuspended in buffer containing 50 mM NaCl , 2 mM 6‐aminohexanoic acid , 50 mM imidazole , 1 mM EDTA ( pH 7 ) , solubilized with digitonin ( 2 g/g protein ) for 20 min on ice , and centrifuged for 20 min at 30 , 000 g to remove cell debris . Supernatants were removed and 10% glycerol and 0 . 01% Ponceau S were added as previously described ( Kovalčíková et al . , 2019; Wittig and Schägger , 2009 ) . Samples along with a high molecular weight native marker ( GE Healthcare Life Sciences , Marlborough , MA ) were separated on 4–15% precast gels in 4°C with current limited to 15 mA and transferred to polyvinylidene difluoride membranes ( Wittig and Schägger , 2009 ) . Immunoblotting was performed after blocking in TBS ( Tris-buffered saline ) containing 5% non-fat milk and 0 . 1% Tween-20 . Membranes were incubated overnight with primary antibodies diluted in blocking solution at 4°C , followed by incubation with secondary antibodies at room temperature for 1 hr . Immunoreactivity was visualized by the ECL chemiluminescence system ( Bio-rad ) on standard film . The antibodies were ATP5A ( ab-14748 , 1:1000 , Abcam ) , ATP Synthase C-subunit ( ab-181243 , 1:1000 , Abcam ) , and citrate synthase ( #14309 , 1:1000 , Cell Signaling Technology ) . For immunocytochemistry of mammalian cells , AGS and mouse NSC/NPC cells were seeded on laminin-coated coverslips ( Neuvitro , Vancouver , WA , USA ) within 24-well plates . The cells were fixed with 4% paraformaldehyde in PBS , washed with PBS , and permeabilized with 0 . 02% Triton X-100 in PBS for 10 min . Blocking was done with 5% BSA in PBS for 1 hr , followed by incubation with antibodies against Nestin ( MAB2736 , 1:50 , R and D Systems , Cambridge , MA , USA ) or Ki-67 ( NB600-1252 , 1:500 , Novus Biologicals , Littleton , CO , USA ) in blocking buffer overnight at 4°C . The Nestin antibody was detected using goat anti-mouse AlexaFluor 488 or 647 ( 1:1000; Jackson ImmunoResearch Laboratories Inc , West Grove , PA , USA ) and the Ki-67 antibody was detected using AlexaFluor 488 goat anti-rabbit ( 1:1000; Jackson Immunoresearch ) or Cy3-conjuated donkey anti-rabbit ( 1:500; EMD Millipore , Burlington , MA , USA ) in blocking buffer . Cells were washed with PBS after primary and secondary antibody staining . Stained cells were overlaid with Fluoroshield mounting medium with DAPI ( Abcam ) to label nucleus . Fluorescence microscopy was performed with a Leica confocal microscope using the following fluorescence filters: DAPI ( 405 nm excitation ) ; Cy3 ( 551 nm excitation ) ; AlexaFluor 647 ( 651 nm excitation ) ; and GFP/AlexaFluor 488 ( 488 nm excitation ) . For comparison across conditions , identical light-exposure levels were used . RNA was extracted from approximately 200 , 000 mouse or AGS NPCs per condition according to manufacturer instructions ( Quick-RNA MiniPrep kit; Irvine , CA , USA ) . Total RNA was reverse transcribed into cDNA ( Bimake , Houston , TX , USA ) , and real-time PCR was performed ( LightCycler96 , Roche , Basel , CHE ) with SYBR Green ( Thermo Fisher Scientific ) as a dsDNA-specific binding dye . Quantitative RT-PCR conditions were 95°C for denaturation , followed by 45 cycles of: 10 s at 95°C , 10 s at 60°C , and 20 s at 72°C . Species-specific primers for each transcript were used ( for list see Table 1 ) . Melting curve analysis was performed after the final cycle to examine the specificity of primers in each reaction . Relative abundance of each Atp5g isoform as a fraction of total Atp5g was calculated by ∆∆CT method and normalized to Rpl27 . Data were analyzed using GraphPad Prism Software ( Graphpad , San Diego , CA ) and presented as means ± S . E . unless otherwise specified , with P-values calculated by two-tailed unpaired Student’s t-tests or two-way ANOVA ( comparisons across more than two groups ) adjusted with the Bonferroni’s correction . No randomization or blinding was used and no power calculations were done to detect a pre-specified effect size .	When animals hibernate , they lower their body temperature and metabolism to conserve the energy they need to withstand cold harsh winters . One such animal is the Arctic ground squirrel , an extreme hibernator that can drop its body temperatures to below 0°C . This hibernation ability means the cells of Arctic ground squirrels can survive severe shortages of blood and oxygen . But , it is unclear how their cells are able to endure this metabolic stress . To answer this question , Singhal , Bai et al . studied the cells of Arctic ground squirrels for unique features that might make them more durable to stress . Examining the genetic code of these resilient cells revealed that Arctic ground squirrels may have a variant form of a protein called ATP5G1 . This protein is found in a cellular compartment called the mitochondria , which is responsible for supplying energy to the rest of the cell and therefore plays an important role in metabolic processes . Singhal , Bai et al . found that when this variant form of ATP5G1 was introduced into the cells of mice , their mitochondria was better at coping with stress conditions , such as low oxygen , low temperature and poisoning . Using a gene editing tool to selectively substitute some of the building blocks , also known as amino acids , that make up the ATP5G1 protein revealed that improvements to the mitochondria were caused by switching specific amino acids . However , swapping these amino acids , which presumably affects the role of ATP5G1 , did not completely remove the cells’ resilience to stress . This suggests that variants of other genes and proteins may also be involved in providing protection . These findings provide the first evidence of a protein variant that is responsible for protecting cells during the metabolic stress conditions caused by hibernation . The approach taken by Singhal , Bai et al . could be used to identify and study other proteins that increase resilience to metabolic stress . These findings could help develop new treatments for diseases caused by a limited blood supply to human organs , such as a stroke or heart attack .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "genetics", "and", "genomics" ]	2020	Cytoprotection by a naturally occurring variant of ATP5G1 in Arctic ground squirrel neural progenitor cells
The Dna2 nuclease-helicase maintains genomic integrity by processing DNA double-strand breaks , Okazaki fragments and stalled replication forks . Dna2 requires ssDNA ends , and is dependent on the ssDNA-binding protein Rpa , which controls cleavage polarity . Here we present the 2 . 3 Å structure of intact mouse Dna2 bound to a 15-nucleotide ssDNA . The nuclease active site is embedded in a long , narrow tunnel through which the DNA has to thread . The helicase domain is required for DNA binding but not threading . We also present the structure of a flexibly-tethered Dna2-Rpa interaction that recruits Dna2 to Rpa-coated DNA . We establish that a second Dna2-Rpa interaction is mutually exclusive with Rpa-DNA interactions and mediates the displacement of Rpa from ssDNA . This interaction occurs at the nuclease tunnel entrance and the 5’ end of the Rpa-DNA complex . Hence , it only displaces Rpa from the 5’ but not 3’ end , explaining how Rpa regulates cleavage polarity . Dna2 has nuclease and helicase activities and plays key roles in maintaining genomic integrity . It is involved in the nucleolytic processing of 5’ flaps during Okazaki fragment maturation , of DNA double-strand breaks ( DSBs ) during homologous-recombination mediated repair , and of stalled replication forks in the intra-S-phase checkpoint ( Bae et al . , 2001; Cejka et al . , 2010; Hu et al . , 2012; Nimonkar et al . , 2011; Zhu et al . , 2008 ) . Dna2 was first identified as a replication mutant required for viability in yeast , and was subsequently shown to be involved in trimming long 5’ RNA-DNA flaps from Okazaki fragments during replication ( Budd et al . , 1995 ) . Most flaps are cleaved by Fen1 concomitant with their generation by strand displacement during Pol δ synthesis on the lagging strand . Flaps that escape early cleavage , or those extended by the Pif1 helicase , get long enough to be coated by the Replication Protein A ( Rpa ) , which renders them resistant to Fen1 ( Bae et al . , 2001; Pike et al . , 2009; Stith et al . , 2008 ) . Dna2 , which can displace Rpa from ssDNA ( Stewart et al . , 2008 ) , trims the flap to a length too short for stable Rpa binding , and restores Fen1 processing ( Ayyagari et al . , 2003; Bae et al . , 2001; Gloor et al . , 2012 ) . The lethality of Dna2 deletion in yeast is attributed to persistent Rpa-coated flaps , which recruit Ddc2 ( ATRIP in metazoa ) and activate the Mec1 ( ATR in metazoa ) DNA-damage checkpoint ( Chen et al . , 2013; Zhu et al . , 2008 ) . Long-flap processing by Dna2 is dependent on Rpa removing ssDNA secondary structure ( Stewart et al . , 2008 ) , an essential Rpa function in many other aspects of DNA metabolism ( Chen et al . , 2013; Fanning et al . , 2006; Symington and Gautier , 2011 ) . Also paralleling other Rpa-dependent processes , the ability of Dna2 to act on Rpa-coated ssDNA is dependent on direct Dna2-Rpa interactions ( Bae et al . , 2003 , 2001 ) . The yeast Dna2△405N mutation that reduces Rpa binding also reduces 5’ flap cleavage and DSB resection in vitro , and renders yeast temperature-sensitive for growth ( Bae et al . , 2003 , 2001; Niu et al . , 2010 ) . In DSB resection , Dna2 acts redundantly with Exo1 ( Gravel et al . , 2008; Mimitou and Symington , 2008; Zhu et al . , 2008 ) . Resection of the 5’ terminated DNA strand results in a long track of 3’ overhang ssDNA , which forms a nucleoprotein filament with the Rad51 strand-exchange protein and initiates homologous recombination ( Symington and Gautier , 2011 ) . In vitro , Dna2 , Rpa and the helicase Sgs1 ( BLM in mammals ) constitute the minimal complex that can carry out long-range resection . Resection is dependent on the nuclease activity of Dna2 and the helicase activity of Sgs1/BLM . Rpa is essential for supporting the helicase activity of Sgs1/BLM in part by sequestering the unwound strands , and also for regulating Dna2 by blocking its 3’ to 5’ exonuclease activity ( Cejka et al . , 2010; Nimonkar et al . , 2011; Niu et al . , 2010 ) . In cells , Rpa depletion eliminates long-range DSB resection , and also causes the loss or inappropriate annealing of short 3’ ends generated by Mre11 ( Chen et al . , 2013 ) . In addition to these functions , Dna2 is implicated in preventing the regression of stalled replication forks , which otherwise can generate aberrant structures resembling recombination intermediates and lead to genomic instability ( Hu et al . , 2012 ) . This is dependent on the Dna2 nuclease activity , consistent with the ability of Dna2 to cleave fork structures with regressed leading or lagging nascent strands in vitro ( Hu et al . , 2012 ) . Dna2 contains a PD- ( D/E ) XK superfamily nuclease motif ( Budd et al . , 2000 ) and a 5’ to 3’ helicase domain ( Bae and Seo , 2000 ) . It is a ssDNA endonuclease that requires a free end for cleavage , and does not cleave dsDNA , single-stranded gaps , D-loops or RNA ( Bae and Seo , 2000; Kao et al . , 2004 ) . In vitro , isolated Dna2 cleaves ssDNA starting at either end , with multiple rounds of incision degrading ssDNA in both the 5’ to 3’ and 3’ to 5’ directions ( Bae and Seo , 2000; Masuda-Sasa et al . , 2006 ) . With 5’ flap DNA , cleavage starts ~∼10 nucleotides ( nts ) from the ssDNA end and continues to within ∼5 nts of the duplex ( Bae et al . , 2001; Bae and Seo , 2000; Cejka et al . , 2010; Gloor et al . , 2012; Masuda-Sasa et al . , 2006 ) . It has been suggested that Dna2 loads at the free 5’ end of the flap and tracks in the 5’ to 3’ direction ( Kao et al . , 2004 ) . The helicase and ATPase activities of Dna2 are substantially weaker than those of other helicases ( Budd et al . , 2000; Masuda-Sasa et al . , 2006 ) . The helicase activity can be increased by a high ATP to Mg2+ ratio , but this also inhibits the nuclease activity through ATP sequestering Mg2+ ( Bae and Seo , 2000; Masuda-Sasa et al . , 2006 ) . The 5’ to 3’ polarity of the helicase translocation could , in principle , drive the tracking of Dna2 along the 5’ flap . However , ATPase mutations have a minimal effect on 5’ flap processing and DSB resection in vitro ( Cejka et al . , 2010; Niu et al . , 2010; Zhu et al . , 2008 ) . And , in vivo , ATPase-inactive yeast Dna2 mutants are viable , although they exhibit impaired growth ( Budd et al . , 2000 ) . To understand the mechanism of action and regulation of this multi-faceted enzyme , we first determined the structure of the intact Dna2-ssDNA complex . The structure revealed that the ssDNA has to thread through a tunnel to bind to Dna2 , with a polarity that precludes the cooperation of the helicase and nuclease activities . The requirement for DNA threading prompted us to investigate how Dna2 gains access to Rpa-coated DNA . We provide the structure of a complex between a Dna2 α helix and the Rpa70 OBN domain , both of which are flexibly-tethered and likely serve to recruit Dna2 to Rpa . We also establish a second Dna2-Rpa interaction that helps to displace Rpa from the 5’ DNA end , explaining how Rpa restricts the cleavage polarity of Dna2 . We determined the 2 . 3 Å crystal structure of full-length mouse Dna2 ( residues 1 to 1062 ) bound to a ssDNA substrate of 21 nucleotides ( nts ) , 15 of which are well ordered ( Figure 1A and Table 1 ) . The structure also contains an Fe4-S4 iron-sulfur cluster , ADP and two active-site Ca2+ ions , which do not support nuclease activity but can mimic magnesium coordination ( Yang et al . , 2006 ) . 10 . 7554/eLife . 09832 . 003Figure 1 . Structure of the Dna2-ssDNA complex . ( A ) Cartoon representation of the Dna2-ssDNA complex . The α1 helix , which packs with the hel2A domain of a crystallographic symmetry related protomer , is omitted from this view . The individual domains of Dna2 are colored separately as indicated in ( D ) , ssDNA is red , ADP is shown as sticks , calcium ions are shown as green spheres , the iron-sulfur cluster is shown in a space-filling representation ( nuc: nuclease domain; hel1A: helicase 1A domain , hel2A: helicase 2A domain , OB: oligonucleotide/oligosaccharide binding domain ) . ( B ) View looking up the vertical axis of ( A ) . ( C ) Schematic of the complex illustrating the relative arrangement of the Dna2 domains , and highlighting its cylinder-like shape . Colored as in ( A ) . ( D ) Linear representation of the Dna2 domains and their boundaries; colored as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 00310 . 7554/eLife . 09832 . 004Figure 1—figure supplement 1 . Dna2 secondary structure and sequence conservation . Helices are indicated as cylinders , b strands as arrows , segments lacking regular secondary structure as solid lines , and disordered regions as dashed lines . Secondary structure elements are colored as in Figure 1A . The N-terminal extensions , relative to mammalian Dna2 , of the other orthologs are not shown . In the alignment of yeast Dna2 , reliable homology starts with the β2 strand of the OB domain at residues 44 and 417 of the mouse and yeast orthologs , respectively . The ~360 residue , unique N-terminal segment of yeast Dna2 has an overall low hydrophobicity content indicative of lack of globular structure , and short regions of homology with closely related yeasts suggests that this region acts through linear epitopes , some of which may be acidic/amphipathic helices ( not shown ) . Residues that contact the DNA are marked by “d” , active site residues by “c” and iron-sulfur cluster ligands by “f” . The entire protein structure is within 0 . 9 Å Cα r . m . s . d . in DNA-bound and apo Dna2 crystals , and the overall DNA conformation is essentially identical in a different crystal form of Dna2 bound to a 21-nt ssDNA ( Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 00410 . 7554/eLife . 09832 . 005Figure 1—figure supplement 2 . Dna2 inter-domain interfaces . ( A ) The nuclease domain packs with the OB domain extensively , through both polar and hydrophobic residues . The four residues between the last b strand of the OB domain and the first b strand of the nuclease domain are buried in the interface between the two domains . The OB groove , where DNA or peptides bind in other OB fold proteins , is solvent exposed opposite from its surface that packs with the nuclease domain . Individual domains are colored as in Figure 1A . ( B ) The Dna2 nuclease core fold has a three-helix insertion ( residues 176 to 227 ) that packs with the helicase 1B domain . This three-helix insertion is also present in the AddB nuclease , where it packs with the N-terminal portion of the protein . ( C ) The nuclease domain also links to the helicase domain indirectly , through the stalk domain that is sandwiched between the two . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 00510 . 7554/eLife . 09832 . 006Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 006Data SetDna2-5’ overhang DNADna2-ssDNAapo Dna2apo Dna2 ( SeMet ) DNA2 α1-RPA 70NSpace groupP22121P212121P212121P212121C121a , b , c ( Å ) 87 . 2 , 118 . 5 , 149 . 3120 . 2 , 149 . 2 , 172 . 9120 . 9 , 148 . 6 , 170 . 5120 . 9 , 148 . 6 , 170 . 5134 . 3 , 50 . 9 , 76 . 5α , β , γ ( ° ) 90 . 0 , 90 . 0 , 90 . 090 . 0 , 90 . 0 , 90 . 090 . 0 , 90 . 0 , 90 . 090 . 0 , 90 . 0 , 90 . 090 . 0 , 103 . 9 , 90 . 0Resolution ( Å ) 50 . 0 - 2 . 35 ( 2 . 43 - 2 . 35 ) 30 . 0 – 3 . 11 ( 3 . 23 - 3 . 11 ) 50 . 0 – 3 . 0 ( 3 . 11 - 3 . 0 ) 30 . 0 – 3 . 4 ( 3 . 52 - 3 . 4 ) 60 . 0 – 1 . 5 ( 1 . 55 - 1 . 5 ) Rsym12 . 6 ( 65 . 6 ) 13 . 7 ( 87 . 5 ) 13 . 1 ( 78 . 7 ) 15 . 8 ( 59 . 3 ) 7 . 4 ( 80 . 1 ) Rpim6 . 3 ( 36 . 6 ) 6 . 6 ( 57 . 1 ) 6 . 8 ( 41 . 9 ) 4 . 0 ( 15 . 1 ) 5 . 1 ( 54 . 1 ) I/σ ( I ) 16 . 4 ( 1 . 9 ) 13 . 2 ( 1 . 4 ) 7 . 9 ( 1 . 3 ) 19 . 3 ( 4 . 7 ) 23 . 5 ( 3 . 2 ) Completeness ( % ) 99 . 0 ( 98 . 6 ) 99 . 0 ( 99 . 2 ) 98 . 3 ( 98 . 9 ) 100 . 0 ( 100 ) 84 . 3 ( 79 . 7 ) Redundancy5 . 0 ( 3 . 9 ) 6 . 1 ( 6 . 0 ) 4 . 4 ( 4 . 5 ) 15 . 1 ( 15 . 3 ) 2 . 7 ( 2 . 7 ) RefinementResolution ( Å ) 50 . 0-2 . 3630 . 0–3 . 1550 . 0–3 . 030 . 0-1 . 55No . of reflections57 , 87151 , 12254 , 10757 , 455Rwork/Rfree ( % ) 20 . 8/24 . 622 . 3/25 . 621 . 1/24 . 523 . 2/26 . 7Protein atoms8 , 29816 , 53616 , 5363 , 933DNA atoms29067400Cofactor atoms4170700Rmsd bond lengths ( Å ) 0 . 0090 . 0090 . 0090 . 007Rmsd bond angles ( ° ) B factors ( Å2 ) :proteinDNACa2+waterWilson1 . 4 65 . 5107 . 658 . 949 . 558 . 41 . 41 . 461 . 3Only the ssDNA is ordered . Values in parentheses are for the highest-resolution shell . The structure consists of a ∼310 residue domain that contains the PD- ( D/E ) XK nuclease motif , followed by a ∼450 residue , C-terminal helicase domain that has two RecA-like folds characteristic of the SF1 helicase subfamily ( domains 1A and 2A; Figure 1A , B ) . In addition , the structure reveals an OB ( oligonucleotide/oligosaccharide-binding ) fold domain N-terminal to the nuclease domain , and a β barrel domain that occurs between the nuclease and helicase domains and which is held in place by a stalk of two long alpha helices ( Figure 1 and Figure 1—figure supplement 1A ) . The overall structure has a cylindrical shape with a central tunnel through which the ssDNA threads ( Figure 1A , B ) . The base of the cylinder is formed by the nuclease domain , which adopts a doughnut-like structure with the active site embedded in the central tunnel ( Figure 1C ) . The β barrel and helicase 1A domains pack on top of the nuclease doughnut and extend the cylinder and central tunnel . The helicase 2A domain , which packs with the 1A domain as in the ADP states of other SF1 helicase structures , hangs over the tunnel opening at the top of the cylinder . Most of the DNA-binding sites of the nuclease and helicase 1A domains are inside the tunnel , whereas those of the helicase 2A domain are solvent exposed . The OB domain decorates the exterior of the nuclease domain , and is uninvolved in DNA binding . The nuclease domain is the hub that organizes the overall structure . It packs with the flanking OB and β barrel-stalk domains , as well as the helicase 1B domain ( Figure 1C and Figure 1—figure supplement 2 ) . The ssDNA is positioned with its 5’ end at the helicase domain , and its 3’ end at the nuclease domain ( Figure 1A ) . A 7-nt segment at the 5’ end contacts first the helicase 2A domain outside the tunnel and then the 1A domain inside the tunnel . The DNA crosses over from the helicase to the nuclease domains over the next two nucleotides , which are in the vicinity of the β barrel domain in the middle of the tunnel , but do not make any protein contacts ( Figure 1A ) . The subsequent 6-nt segment binds to the nuclease domain . Here , the first three nucleotides are fully enclosed by the tunnel , while the last three nucleotides contact the tunnel opening . The DNA bases stack continuously , except for a base step at the helicase , one at the nuclease and one at the transition between the two domains ( Figure 2C ) . Two of the three unstacked base steps are at pyrimidine-pyrimidine pairs , and this may contribute to the DNA binding at a well-defined register . An N-terminal acidic/amphipathic α helix ( α1; residues 1 to 13 ) packs with the helicase domain of a symmetry-related complex . The α1 helix is flexibly tethered to the rest of the protein , as the 6 residues that connect it to the OB domain have no electron density in the crystals . As shown later , this helix is one of the two Rpa-binding elements of Dna2 . 10 . 7554/eLife . 09832 . 007Figure 2 . Nuclease and helicase domain structures and DNA contacts . ( A ) Superposition of the Dna2 nuclease domain on the AddB nuclease domain . The N-terminal βαα extension ( residues 122 to 154 ) and the C-terminal αα extension ( residues 384 to 412 ) is colored in orange , with the corresponding elements of AddB in dark cyan . Green spheres are calcium ions . ( B ) DNA contacts and active site residues of the Dna2 nuclease domain . Hydrogen bonds are depicted as green dotted lines , calcium ions as blue spheres , water as a red sphere . ( C ) Diagram showing the contacts depicted in ( B ) and ( E ) . The residues are colored according to the domain they belong as in Figure 1D ( mc: main chain , sc: side chain ) . ( D ) Superposition of the Dna2 helicase domain on Ighmbp2 . Dna2 is colored as in Figure 1D . The Ighmbnp2 1A ( hel1A ) and 2A ( hel2A ) helicase domains are colored gold , its β barrel in light orange , and stalk dark orange . ( E ) DNA contacts of the helicase 1A ( pink ) and 2A ( cyan ) domains , showing residues that are involved in either hydrogen bond ( green dotted lines ) or van der Waals contacts . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 00710 . 7554/eLife . 09832 . 008Figure 2—figure supplement 1 . Electron density at the nuclease active site and structural similarity of the Dna2 helicase to Upf1-subfamily RNA/DNA helicases . ( A ) Top panel shows a stereo view of the mFo-DFc electron density at the active site , calculated with phases after omitting the two calcium ions ( blue spheres ) , the water molecule ( red sphere ) , a 3-nt segment centered on the scissile phosphate group and the side chain of Lys301 , and subjecting the rest of the entire structure to simulated annealing from 3000°K . The dark blue map is contoured at 5 σ , and the light green map at 2 . 5 σ . The orientation is similar to that of Figure 2B . For comparison , the corresponding residues of the λ nuclease-DNA complex ( Zhang et al . , 2011 ) are shown as black sticks ( the lysine is mutated to alanine in that structure ) , and the calcium ions and water molecule as black spheres . The Dna2 Ca-1 ion and water molecule have a coordination shell and position very similar to the first magnesium and associated water in λ nuclease , but the position of Ca-2 is different . The Ca coordination shell and residue labels are shown in bottom panel . For clarity , not shown are the electron density , λ nuclease , and the His164 side chain to which the green-dotted bond from Ca-1 would be connected . ( B ) Superposition of Dna2 helicase domains 1A ( pink ) and 2A ( cyan ) with Ighmbp2 ( gold ) . DNA is colored in red , RNA is colored in green . The only regions of the Dna2 helicase that do not have counterparts in the Upf1 subfamily are the ~40 residue 1B domain , which packs with the nuclease domain , and the ~20 residue 2B domain , which is distal from the rest of the protein and the DNA ( Figure 1D ) . ( C ) Superposition of individual Dna2 helicase domains 1A ( pink ) and 2A ( cyan ) with Upf1 ( gold ) . DNA is colored in red , RNA is colored in green . Motif III residues are shown as sticks . Supplemental discussion of helicase translocation . Based on the proposed SF1B translocation mechanism ( Saikrishnan et al . , 2009 ) , ATP binding and concomitant closure of the cleft between the 1A and 2A domains would be coupled to domain 2A releasing the DNA and rebinding it at a register shifted by 1 nt in the 5’ to 3’ direction , while domain 1A would retain its DNA . On ATP hydrolysis , domain 1A would be the one releasing its DNA and rebinding it at the +1 nt register , restoring the initial open-cleft conformation . In the Dna2-ssDNA structure , however , the rebinding of 1A to DNA at the shifted register will have to await the transient dissociation of the nuclease domain from DNA . The increased half life of the closed cleft would likely also reduce the rate of exchange of ADP for ATP , as the γ phosphate binding site of the helicase fold is far less solvent accessible in the closed cleft . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 008 As predicted , the Dna2 nuclease domain contains the core αβββαβ fold of the PD- ( D/E ) XK nuclease superfamily ( Aravind et al . , 2000; Pingoud et al . , 2005 ) . Based on the DALI server , its closest structural homologs are the bacteriophage λ exonuclease and the E . coli RecB and B . subtilis AddB nucleases , all of which are involved in the resection of DNA ends during homologous recombination ( Krajewski et al . , 2014; Singleton et al . , 2004; Zhang et al . , 2011 ) . These superimpose on the ∼310-residue Dna2 nuclease domain with a ∼2 Å root-mean-square deviation ( r . m . s . d . ) in the Cα positions of 128 , 112 and 106 residues , respectively . The most extensive similarity is exhibited by AddB , which shares with Dna2 the presence of the iron-sulfur cluster and most of the insertions and extensions that decorate the core fold ( Figure 2A ) ( Krajewski et al . , 2014; Yeeles et al . , 2009 ) . While most of these elements have diverged beyond a ∼2 Å r . m . s . d . , their secondary structures , arrangement and structural implications are closely related . In particular , both proteins have a β–α–α N-terminal extension and an α–α C-terminal extension that are stapled together by the iron-sulfur cluster ( Figure 2A ) . This iron-sulfur cluster domain supports a loop that crosses over the catalytic channel and converts it to a tunnel through which the DNA has to thread . λ exonuclease and RecB also have a crossover loop , but their N- and C-terminal extensions that anchor it are structurally divergent from Dna2 , and , more significantly , they lack the iron-sulfur cluster ( Singleton et al . , 2004; Zhang et al . , 2011 ) . The Dna2 nuclease domain binds to a total of four phosphodiester groups , two before and two after the scissile phosphate group ( Figure 2B , C ) . The scissile phosphate group contacts two calcium ions , one ( Ca-1 ) through a non-bridging oxygen , and another ( Ca-2 ) through the 3’ bridging oxygen of the preceding base . The Ca-1 ion has an octahedral coordination shell very similar to other PD- ( D/E ) XK nucleases ( Pingoud et al . , 2005; Zhang et al . , 2011 ) . It’s formed by the side chains of His164 , Asp278 and Glu299 , the Leu300 main chain carbonyl group , the non-bridging oxygen atom of the scissile phosphate group , and a water molecule , which also hydrogen bonds to Lys301 and is positioned for nucleophilic attack on the scissile phosphate ( Figure 2B and Figure 2—figure supplement 1A ) . Lys301 is buttressed by Gln314 , a motif IV residue characteristic of the RecB and λ exonuclease families ( Aravind et al . , 2000 ) . The relative position of the second , Ca-2 ion differs from other nucleases ( Figure 2—figure supplement 1A ) , and it is not clear whether it reflects the use of calcium instead of magnesium , or a divergent aspect of the nuclease mechanism of Dna2 . The base groups of the 6 nt segment stack in two sets of three , with Met157 and other crossover loop side chains wedging in between the Ade12 and Gua13 base groups ( Figure 2B ) . This is very similar to the λ exonuclease crossover loop , which marks the transition from double stranded to single stranded and is thought to play a key role in unwinding dsDNA , although λ exonuclease does so without a helicase domain and in the context of a homo-trimeric assembly that contacts the dsDNA of the substrate ( Zhang et al . , 2011 ) . Nevertheless , Met157 and the crossover loop may have an analogous function in the weak strand-separating activity of Dna2 . Dna2’s combination of the β barrel , stalk and helicase domains , and their relative arrangement are strikingly similar to RNA/DNA helicases of the Upf1 subfamily , which contains Upf1 , Ighmbp2 and Senataxin ( Fairman-Williams et al . , 2010 ) . The ∼600-residue assembly of the stalk , β barrel and helicase 1A and 2A domains can be superimposed on the structure of Ighmbp2 with a 1 . 5 Å Cα r . m . s . d . over 452 residues , with most of the non-superimposing residues accounted for by a ∼10° rotation of the β barrel , which otherwise is structurally conserved ( Figure 2D ) . The structural similarity between Dna2 and Ighmbp2 is actually more extensive than that between Ighmbp2 and Upf1 ( 1 . 8 Cα Å r . m . s . d for 436 residues ) ( Chakrabarti et al . , 2011; Lim et al . , 2012 ) . The Upf1-like subfamily is one branch of the SF1B family of helicases that translocate in the 5’ to 3’ direction ( Fairman-Williams et al . , 2010 ) . The other SF1B branch , represented by the DNA helicase RecD2 ( Saikrishnan et al . , 2009 ) , lacks the barrel and stalk domains . Furthermore , the individual RecD2 1A and 2A domains are not as similar to Dna2 , with alignment r . m . s . d . values of 1 . 8 Å for 134 residues and 1 . 9 Å for 97 residues , respectively , in contrast to the Dna2-Ighmbp2 alignment , where the corresponding values are 1 . 5 Å for 204 residues and 1 . 4 Å for 183 residues , respectively . Together , these structural observations suggest that Dna2 evolved by incorporating an ancestral Upf1-family helicase . DNA binds to Dna2 through both its phosphodiester and base groups ( Figure 2E ) . Dna2-base interactions include Van der Waals contacts from the motif III loop ( Leu795 and Val797 ) , which wedges in between the bases of the last two 2A-bound nucleotides . These contacts are consistent with the proposed role of motif III in preventing DNA sliding during translocation of SF1B helicases ( Saikrishnan et al . , 2009 ) . Dna2-phosphodiester interactions involve protein pockets that are rich in backbone amide and short-side chain hydroxyl groups ( Figure 2E ) . Three consecutive phosphodiester groups , near the 5’ end of the DNA , bind to the 2A domain ( Figure 2C , E ) . The fourth phosphodiester group is in the cleft between the 2A and 1A domains and does not contact the protein , while the fifth and sixth phosphodiester groups bind to pockets on the 1A domain ( Figure 2C ) . Contacts to the ribose groups are minimal , and they are consistent with the helicase accommodating the 2’ hydroxyl group of RNA . In fact , the DNA contacts as well as the phosphodiester backbone conformation are very similar to those of the Ighmbp2-RNA , Upf1-RNA complexes ( Figure 2—figure supplement 1B , C ) . This is in contrast to the RecD2-ssDNA structure , where extensive aromatic and van der Waals contacts with the sugar are thought to discriminate against RNA ( Saikrishnan et al . , 2009 ) . The helicase and ssDNA-dependent ATPase activities of Dna2 are considerably weaker than other helicases ( Budd et al . , 2000; Masuda-Sasa et al . , 2006 ) . One possible explanation , at least for the low ATPase/translocation rate , is the helicase domain being 5’ to the nuclease domain on the DNA . This would , in principle , make completion of the ATPase/translocation cycle dependent on the nuclease domain releasing its grip on the DNA ( see Figure 2—figure supplement 1 legend for discussion of translocation ) . The structure explains why the nuclease activity of Dna2 requires the ssDNA to have a free end ( Bae and Seo , 2000; Kao et al . , 2004 ) . The active site is embedded in a ∼10 Å wide portion of the tunnel , and the tunnel entrances leading to it are too narrow to accommodate dsDNA or a single-stranded loop of a gap substrate ( Figure 1A , B ) . Since isolated Dna2 can degrade ssDNA with either 5’ to 3’ or 3’ to 5’ polarity in vitro , the ssDNA must be able to enter and thread through either end of the tunnel . In 5’ flap processing , threading likely proceeds through initial , transient interactions at the nuclease domain entrance of the tunnel , with subsequent re-binding events occurring further along the tunnel interior . This threading is likely related to that of λ exonuclease , where a 5’ terminal phosphate binding site inside the enzyme is proposed to drive the forward movement of the DNA in an electrostatic ratchet mechanism ( Zhang et al . , 2011 ) . The 5’ to 3’ nuclease activity was shown to require ∼15 nts of ssDNA for optimal affinity and cleavage ( Bae et al . , 2001; Gloor et al . , 2012 ) . This suggests that the nuclease domain-DNA interactions do not provide sufficient binding energy or their half-life is too short relative to the catalytic step of cleavage . We thus presume that the 5’ end of the ssDNA threads through the nuclease portion of the tunnel without cleavage , until it reaches the helicase domain and engages the DNA-binding pockets there . To investigate this in more detail , we assayed the Dna2 affinity and cleavage of a series of 5’ overhang substrates with ssDNA lengths that extend successively from the nuclease to the helicase 1A and 2A domains . The ssDNA consisted of deoxythymidine nucleotides , which have minimal secondary structure and thus do not require Rpa for cleavage by Dna2 ( 5’ ( dT ) 6 to 5’ ( dT ) 24 ) ( Figure 3—figure supplement 1A ) . The short substrates that can span only the nuclease domain ( 5’ ( dT ) 6 and 5’ ( dT ) 8 ) have Dna2 affinities approximately two orders of magnitude lower than that of 5’ ( dT ) 24 , and they exhibit minimal cleavage ( Figure 3—figure supplement 1A , B ) . With the slightly longer 5’ ( dT ) 8 , low level cleavage occurs 7 and 8 nucleotides from the 5’ end , indicating that the ssDNA extends to the helicase 1A domain ( Figure 3—figure supplement 1A ) . The sites of cleavage indicate that Dna2 opens up ∼2 base pairs ( bps ) of the duplex , since the tunnel entrance 3’ to the scissile phosphate is too narrow to accommodate double-stranded DNA . As ATP is neither required , nor has a significant effect , this DNA unwinding is the result of threading , likely analogous to DNA unwinding by λ exonuclease ( Zhang et al . , 2011 ) . Extending the ssDNA by 2 nts , ( 5’ ( dT ) 10 ) , results in a major increase in Dna2 affinity and cleavage , with the cleavage sites indicating the engagement of the helicase 2A domain after the unwinding of ∼2 bps ( Figure 3—figure supplement 1A ) . With 5’ ( dT ) 17 where the ssDNA is long enough to reach both the 1A and 2A domains without duplex unwinding , there is a final increase in Dna2 affinity and cleavage that plateau at the levels of 5’ ( dT ) 24 . The 3’ to 5’ nuclease activity would require the 3’ end of the ssDNA to enter the tunnel at the helicase 2A domain . As the phosphodiester-binding sites of the 2A domain are fully accessible to bulk solvent , the structure suggests that 3’ end threading should be more efficient than 5’ end threading in the absence of ATP . Indeed , Dna2 cleaves a 3’ overhang ( dT ) 18 substrate at least 3-fold faster than the corresponding 5’ ( dT ) 18 ( Figure 3A ) . The structure further suggests that the reported inhibition of 3’ end cleavage by ATP is due to 3’ end threading being counteracted by the helicase domain moving the DNA in the opposite direction ( Figure 3A ) ( Bae and Seo , 2000; Masuda-Sasa et al . , 2006 ) . 10 . 7554/eLife . 09832 . 009Figure 3 . Dna2 nuclease activity . ( A ) Denaturing PAGE showing ATP inhibits the nuclease activity on 3’ overhang substrate , while slightly increasing it for the 5’ overhang substrate . Substrates are at 15 nM . For this and subsequent nuclease assays , cleavage was quantified by loss of substrate and plotted with ± s . d . error bars ( n = 3 ) . ( B ) Nuclease time course of 2 nM Dna2 with 10 nM 5’ RNA-DNA overhang or 5’ DNA overhang substrates . ATP or AMPPNP is at 1 . 3 mM . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 00910 . 7554/eLife . 09832 . 010Figure 3—figure supplement 1 . The ssDNA-length dependence of the DNA affinity correlates with cleavage rates . ( A ) Denaturing PAGE showing length dependency and cleavage sites of 5’ overhang DNA at a 30 nM concentration ( asterisk indicates FAM labeling of the 3’ end ) in the presence ( top panel ) or absence ( bottom ) of 1 . 3 mM ATP . Also shown are the dissociation constants ( Kd ) of Dna2 ( D278A ) for each substrate , according to the data shown in Figure 3—figure supplement 1B . ( B ) Electrophoretic mobility shift assays ( EMSA ) of Dna2 ( D278A ) binding to 5’ overhang DNA substrates . The graph shows fraction bound . Dotted lines are simulated binding curves based on the indicated Kd values . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 010 Taken together , these findings support the model that a major role for the helicase domain is augmenting the DNA affinity of Dna2 . As lack of ATP hydrolysis does not affect overall DNA binding by the helicase 1A and 2A domains , this model is consistent with the Dna2 ATPase-activity being dispensable for viability in yeast , and for DSB end resection in vitro , in contrast to the nuclease activity that is essential for both ( Bae and Seo , 2000; Budd et al . , 2000; Cejka et al . , 2010; Niu et al . , 2010; Zhu et al . , 2008 ) . RNA , which is not cleaved by Dna2 , can substitute for part of DNA in the length-dependency of cleavage , suggesting that it can interact with at least part of Dna2 ( Bae et al . , 2001; Bae and Seo , 2000 ) . In accord , we find that the affinity of Dna2 for a 5’ ( U ) 24 RNA overhang-DNA duplex is within an order of magnitude of its affinity for a comparable all-DNA substrate ( Figure 3—figure supplement 1B ) . This observance , coupled with the structural similarity of Dna2 to the Upf1 subfamily of RNA/DNA helicases raises the possibility that the helicase activity facilitates the bypassing of the 5’ RNA primer of Okazaki fragments . Indeed , ATP but not the non-hydrolyzable ATP analogue AMPPNP stimulates the cleavage of a 5’ ( U ) 12– ( dT ) 12 RNA-DNA overhang substrate by ∼50% ( Figure 3B , top panel ) . By contrast , ATP had a minimal effect on the cleavage of the corresponding all-DNA ( dT ) 24 substrate , while AMPPNP inhibited cleavage slightly ( Figure 3B , bottom panel ) . Rpa is a heterotrimeric protein that consists of the Rpa70 , Rpa32 and Rpa14 subunits . Genetic and biochemical studies in budding yeast indicated that Dna2 binds to the Rpa70 subunit , through an interaction between the N-terminal portions of the two proteins ( Bae et al . , 2003 ) . These studies further pointed to additional binding sites on Dna2 and Rpa70 , as deletion of the N-terminal interacting segments reduced but did not eliminate Dna2-Rpa association and the stimulation of the nuclease activity ( Bae et al . , 2003 ) . The N-terminal portion of yeast Dna2 ( residues 1 to 405 ) is poorly conserved in mammalian orthologs and also contains a ∼350-residue yeast-specific extension , although it appears to contain an acidic/amphipathic helix analogous to α1 of mouse Dna2 , and likely encompasses a portion of the OB domain ( Figure 1—figure supplement 1A ) . We thus tested whether a mouse Dna2 fragment consisting of the α1 helix and OB domain ( thereafter α1OB; residues 1 to 122 ) binds to mouse Rpa using a GST pull-down assay . As shown in Figure 4A , GST-α1OB but not isolated GST binds to heterotrimeric Rpa in a 30 μM stoichiometric solution ( lanes 1 and 2 ) . 10 . 7554/eLife . 09832 . 011Figure 4 . Dna2 physically interacts with Rpa . ( A ) GST pull-down assay showing that α1OB of Dna2 ( residues 1–122 ) interacts with Rpa70NAB ( 1–431 ) , but not the rest of Rpa heterotrimer ( Rpa ( -NAB ) ) . ( B ) Schematic drawing of the Rpa trimer , showing the OB domains as rectangles and the winged helix ( WH ) domain of Rpa32 as a sphere . ( C ) ITC curves for the human α1OB-Rpa70NAB , α1-Rpa70N and OB-Rpa70AB complexes . ( D ) Structure of the α1-Rpa70N complex . The α1 peptide is in yellow and Rpa70N in cyan . For clarity , the main chain amide group of L87 and carbonyl groups of M11 , K13 , F15 are not labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 01110 . 7554/eLife . 09832 . 012Figure 4—figure supplement 1 . Comparison of the Dna2 α1–Rpa70 OBN , p53–Rpa70 OBN and Dna2 OB structures . ( A ) Superposition of the α1–Rpa70 OBN structure on the fusion protein of Rpa70 OBN and p53 . In the fusion protein structure , the Rpa70 OBN ( gray ) groove has two p53 peptides , one ( cyan ) from its fused p53 and another ( green ) from a crystallographic-symmetry related fusion protein . While it was suggested that only the latter represented the solution-state complex , our structure has aspects of both p53-Rpa70 OBN interfaces . An acidic/amphipathic helix from the symmetry-related p53 superimposes with the Dna2 helix , and part of the second p53 peptide overlaps with the Dna2 β turn and associated contacts . α1 peptide is red . ( B ) The Dna2 OB groove ( cyan ) has an overall hydrophobic- and basic-residue content similar to that of the Rpa70 OBN ( pink ) . The L12 and L45 loops ( labeled ) , which in DNA-binding OB domains partially wrap around the DNA , but are truncated in the Rpa70 OBN domain are , respectively , truncated and absent in the Dna2 OB domain , resulting in a wide open groove . Residues mutated in Figure 5A are labeled . ( C ) ITC titration profile of human Dna2 α1 ( residues 1–20 ) binding to human Rpa70N ( residues 1–120 ) . ( D ) ITC titration profile of human Dna2 OB ( residues 21–122 ) binding to human Rpa70AB ( residues 181–422 ) . ( E ) ITC titration profile of human Dna2 α1OB ( residues 1–122 ) binding to human Rpa70NAB ( residues 1–422 ) . ( F ) The Dna2 OB domain does not exhibit any detectable DNA-binding at concentrations up to 80 μM in an EMSA assay with ( dT ) 20 ssDNA . A shifted , smeary band appearing at 160 μM OB has ~ 4% of the total DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 012 The Dna2-interacting N-terminal segment of yeast Rpa70 ( residues 1 to 180 ) consists of an OB fold domain ( named OBN ) that is a known protein-protein interaction site and a ~60 residue flexible linker ( Figure 4B ) ( Bochkareva et al . , 2005; Fan and Pavletich , 2012; Fanning et al . , 2006; Gomes and Wold , 1996 ) . The rest of Rpa70 consists of three OB folds that are the main DNA-binding domains ( named DBD-A , DBD-B and DBD-C; Figure 4B ) ( Bochkarev et al . , 1997; Fan and Pavletich , 2012 ) . Because DBD-A and DBD-B can also participate in protein-protein interactions ( Jiang et al . , 2006; Loo and Melendy , 2004; Yuzhakov et al . , 1999 ) , this raised the possibility that they account for the remainder of Rpa70’s Dna2 affinity . We tested this using a mouse Rpa70 fragment containing the OBN , DBD-A and DBD-B domains ( residues 1 to 431; thereafter Rpa70NAB ) . As shown in Figure 4A , GST-α1OB binds to Rpa70NAB ( lanes 3 and 4 ) but not to the Rpa heterotrimer lacking this fragment ( Rpa ( -NAB ) ; lanes 5 to 6 ) . The dissociation constant ( Kd ) of the α1OB–Rpa70NAB complex , determined by isothermal titration calorimetry ( ITC ) , is 12 ± 1 μM ( Figure 4C and Figure 4—figure supplement 1E ) . Using ITC , we found that the Dna2 α1 helix ( residues 1 to 20 ) binds to the OBN domain of Rpa70 ( residues 1 to 120; thereafter Rpa70N ) , while the Dna2 OB domain ( residues 21 to 122 ) binds to the Rpa70 fragment containing DBD-A and DBD-B ( residues 181 to 422; thereafter Rpa70AB ) , with Kd values of 34 ± 7 and 46 ± 10 μM , respectively ( Figure 4C and Figure 4—figure supplement 1 ) . The Kd values of these subcomplexes relative to that of α1OB–Rpa70NAB indicate only a low level of cooperativity , consistent with the two interacting elements on both Rpa and Dna2 being separated by flexible linkers . Further dividing Rpa70AB into the individual DBD-A and DBD-B polypeptides failed to show detectable binding to Dna2 OB under the same conditions , indicating both are required ( not shown ) . We co-crystallized a human Dna2 peptide ( residues 1 to 20 ) containing the α1 helix with the OBN domain of human Rpa70 ( residues 1 to 120 ) . In the 1 . 6 Å structure , residues 6 to 17 of Dna2 form a 2-turn amphipathic helix followed by a β turn , while the rest are disordered ( Figure 4D ) . The peptide binds to a shallow OBN groove that corresponds to the DNA-binding grooves of other OB domains ( Bochkarev et al . , 1997; Fan and Pavletich , 2012 ) . The only other OBN-peptide structure available is of OBN fused to the p53 transactivation domain peptide , where two p53 peptides occupy the OBN groove ( Bochkareva et al . , 2005 ) . The amphipathic helix of Dna2 overlaps with one of the p53 peptides , while the β turn coincides with part of the other p53 peptide ( Figure 4—figure supplement 1A ) . The mixed basic and hydrophobic character of the OBN groove complements the acidic-hydrophobic nature of the Dna2 peptide . Four arginine guanidinium groups and one backbone amide group of OBN contact two side chain carboxylate and three backbone carbonyl groups from α1 ( Figure 4D ) . One of these OBN arginine residues ( Arg43 ) splits the otherwise hydrophobic groove , demarcating two hydrophobic pockets . One pocket ( Ile33 , Met57 , Leu87 , Val93 and Ile95 ) binds to the hydrophobic face of the Dna2 helix ( Leu7 , Leu10 and Met11 ) , while the other pocket ( Leu45 and aliphatic portions of Arg31 , Arg43 , and Ser54 ) binds to Phe15 and Trp16 from the β turn of Dna2 . We do not know the structure of the Dna2 OB domain bound to the Rpa70 DBD-A and DBD-B domains , but the OB structure in intact Dna2 is consistent with a role in protein-protein interactions ( Figure 4—figure supplement 1B ) , and the isolated OB domain polypeptide does not exhibit any DNA-binding in EMSA at concentrations up to 80 μM ( Figure 4—figure supplement 1F ) . A common feature of proteins involved in Rpa-dependent processes is their ability to bind to Rpa , either directly or through accessory factors , and this is thought to form the basis for the displacement of Rpa from ssDNA ( Fanning et al . , 2006; Zou et al . , 2006 ) . Rpa displacement is best understood with the simian virus 40 ( SV40 ) T antigen and related viral replication proteins , where T antigen-Rpa interactions allosterically modulate Rpa’s ssDNA affinity ( Jiang et al . , 2006; Loo and Melendy , 2004; Yuzhakov et al . , 1999 ) . Because Dna2 has been shown to displace Rpa from 5’ flap DNA ( Stewart et al . , 2008 ) , we sought to address whether Dna2-Rpa interactions have an analogous , direct role in Rpa displacement , or whether they reflect a simple recruitment process that allows Dna2 to better compete with Rpa for ssDNA . We first confirmed that both the α1 helix and OB domain of Dna2 are required for the stimulation of the nuclease activity by Rpa . For this , we used a 5’ overhang substrate with a stem loop secondary structure that makes the Dna2 nuclease activity dependent on Rpa ( 5’SL24 ) ( Figure 5—figure supplement 1A ) . In keeping with the findings with the yeast Dna2△405N mutant , deletion of the Dna2 α1 helix reduced cleavage of the Rpa-coated 5’SL24 by a factor of ~2 compared to intact Dna2 ( Figure 5A ) . The analogous experiment to address the importance of the OB domain was not possible , as the OB-deleted Dna2 is insoluble ( not shown ) . We instead mutated three OB residues at positions corresponding to protein-protein contacts on the Rpa70 OBN domain ( Figure 4—figure supplement 1B ) . As shown in Figure 5A , two of these mutations ( I82A and R66E ) synergized with α1 deletion and reduced 5’SL24 cleavage further , whereas the third ( I110A ) had no discernible effect . To rule out that these mutations do not affect the structural integrity of Dna2 , we tested the Rpa-independent cleavage of the 5’ ( dT ) 24 substrate and found that it is not affected by neither the OB mutations nor the α1 deletion ( Figure 5B ) . 10 . 7554/eLife . 09832 . 013Figure 5 . Both Dna2-Rpa interactions are important for Dna2 stimulation , but only one is mutually exclusive with Rpa-ssDNA interactions . ( A ) Cleavage of a 5’ stem-loop overhang substrate ( 15 nM ) by wild type , α1-deleted , and OB mutant Dna2 in the presence of 15 nM Rpa . ( B ) Nuclease activity of the same set of enzymes as in ( A ) , but using a 5’ ( dT ) 24 overhang substrate in the absence of Rpa . ( C ) GST pull-down assay showing that α1OB and ( dT ) 24 ssDNA bind to Rpa in a partially mutually exclusive manner . Protein and DNA concentrations are in μM , and the bar graph shows the quantitation of Rpa70 binding relative to the lane in the absence of DNA , which is set to 1 . Error bars are standard deviations from three repetitions of each experiment . ( D ) The binding of α1 to the Rpa70N polypeptide is unaffected by ( dT ) 8 ssDNA . GST pull-down assay and quantitation as in ( C ) . ( E ) The binding of OB to the Rpa70AB polypeptide is abolished by ( dT ) 8 ssDNA . GST pull-down assay and quantitation as in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 01310 . 7554/eLife . 09832 . 014Figure 5—figure supplement 1 . Cleavage of a ssDNA containing secondary structure is Rpa dependent . ( A ) Cleavage of a 5’ overhang substrate that has a predicted 6 bp stem-loop secondary structure at the 5’ end of the ssDNA ( 5’ SL24 ) . Dna2 , at the indicated concentrations , failed to cleave 5’ SL24 ( 30 nM ) appreciably under conditions where ~95% of the corresponding 5’ ( dT ) 24 substrate was cleaved , irrespective of ATP . Addition of Rpa , at a stoichiometric amount to ssDNA , stimulated SL24 cleavage by over a factor of 20 , but had only a modest effect on ( dT ) 24 cleavage . In the presence of Rpa , the cleavage of the two substrates was within a factor of two , compared to being nearly two orders of magnitude apart in its absence . ( B ) Cleavage of a 178-nt long 5’ overhang substrate ( 80 nM ) by Dna2 ( 20 nM ) . As with the shorter 5’ overhang substrates , ATP ( 1 . 3 mM ) has only a minor stimulatory effect . Because mixed-sequence ssDNA of this length would invariably have some secondary structure ( the lowest energy secondary structure , predicted by the UNAfold server ( mfold . rna . albany . edu ) , is shown at bottom , right ) , cleavage is stimulated by Rpa ( 400 nM ) as with the 5’ SL24 substrate . For this specific experiment , samples are quantified by integration of final products in the box . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 014 We then addressed whether the α1 helix and OB domain play a direct role in Rpa displacement , which in principle can occur either through the allosteric destabilization of Rpa-ssDNA interactions as shown for T antigen , or through α1OB binding to an Rpa site that overlaps with or sterically hinders a ssDNA-binding site . We did not expect the isolated α1OB , in the absence of the DNA-affinity provided by the nuclease and helicase domains , to displace Rpa from ssDNA , as the affinities of the α1OB-Rpa and Rpa-ssDNA complexes differ by 5 orders of magnitude ( Kd values of ∼12 μM and ∼100 pM , respectively ) . Instead , we reasoned that if α1OB has a role destabilizing Rpa-ssDNA interactions , then this should be reflected in ssDNA interfering with α1OB-Rpa association . As shown in Figure 5C , addition of ( dT ) 24 substantially reduced Rpa binding to GST-α1OB ( lanes 1 to 3 ) , consistent with α1OB and ssDNA interacting with Rpa in a mutually exclusive manner . However , while one molar equivalent of ( dT ) 24 reduced the bound Rpa by a factor of ∼4 , the remaining Rpa was clearly above the background level of the GST-only reaction ( lanes 4 and 5 ) , suggesting that only one of the two Dna2-Rpa interactions is mutually exclusive with Rpa-ssDNA interactions . Consistent with this , addition of ( dT ) 8 had no discernible effect on the binding of GST-α1 to Rpa70N ( Figure 5D ) , whereas it eliminated the binding of GST-OB to Rpa70AB in a manner dependent on the stoichiometry of ( dT ) 8 to Rpa70AB ( Figure 5E ) . These findings indicate that the interaction between the Dna2 α1 helix and the Rpa OBN domain is a simple recruitment step , consistent with both of these elements being flexibly tethered to the remainder of their polypeptides and with their lack of ssDNA affinity . This simple recruitment would be important for Dna2 accessing ssDNA-bound Rpa , where the interaction between the Dna2 OB and Rpa DBD-A–DBD-B domains would not be initially available . The simple recruitment interaction would also increase the effective concentration of the Dna2 OB at the Rpa DBD-A–DBD-B , as they immediately follow the α1 helix and OBN domain , respectively ( Figure 6A ) . 10 . 7554/eLife . 09832 . 015Figure 6 . Dna2 displaces Rpa from 5’ but not 3’ overhang DNA . ( A ) Schematic of the proposed mechanism of Dna2 displacing Rpa from a 5’ overhang DNA . Dna2 is shown as a hollow cylinder , except for its α1 helix and OB fold domains that are shown as a rectangle and circle , respectively . The label “nuc” marks the nuclease tunnel entrance into which the ssDNA would thread , and “hel2A” marks the helicase tunnel exit where the 5’ end of the ssDNA would end up after threading . ( B ) Schematic illustrating that at a 3’ overhang DNA , Dna2-Rpa interactions do not result in a free ssDNA end that can thread into the Dna2 tunnel . ( C ) Nuclease-dead Dna2 ( D278A ) displaces Rpa from 5’- but not 3’ overhang DNA . The overhang consists of ( dT ) 26 and the DNA is conjugated to streptavidin ( shown by “S” ) through a biotin group at the end of the duplex . Column graph showing quantitation of Rpa32 plots the molar ratio relative to the reaction lacking Dna2 for each DNA substrate ( lanes 2 , 3 for 5’ overhang DNA , and lanes 5 and 6 for 3’ overhang DNA ) , or relative to lane 1 for the comparison of Rpa loading onto 5’- and 3’-overhang DNA . Quantitation of relative Dna2 loading is similarly shown in the last column graph . Error bars are standard deviations from three repetitions of each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 09832 . 015 This juxtaposition would then increase the probability of DBD-A–DBD-B transiently associating with OB and dissociating from ssDNA . This is plausible , because while the DNA affinity of DBD-A–DBD-B ( ~50 nM Kd ) is substantially higher than its Dna2 affinity , those of the individual DBD-A ( 2 μM Kd ) and DBD-B ( 20 μM Kd ) are not , and as with the intact Rpa heterotrimer , they are thought to associate with and dissociate from DNA sequentially ( Arunkumar et al . , 2005; Fan and Pavletich , 2012; Fanning et al . , 2006 ) . By itself , the transient displacement of DNA from DBD-A–DBD-B will not lead to the release Rpa from ssDNA . However , because the DBDA–DBDB is at the 5’ end of the Rpa-DNA complex ( Bochkarev et al . , 1997; Fan and Pavletich , 2012 ) , and the Dna2 OB domain is next to the nuclease tunnel entrance , the transiently free 5’ end of the DNA will be well-placed to start threading through the nuclease tunnel ( Figure 6A ) . The threading process then should be able to completely dissociate the already weakened Rpa-DNA complex ( Figure 6A ) . At a 3’ end of DNA , by contrast , the transient dissociation of DBD-A–DBD-B from DNA will expose an internal ssDNA lacking an end that can be trapped by the Dna2 tunnel , and DBD-A–DBD-B will revert to their DNA-bound state ( Figure 6B ) . This mechanism of Rpa displacement predicts that the inhibition of 3’ end cleavage by Rpa is due to the inability of Dna2 to displace Rpa there . To test this prediction , we conjugated 5’- or 3’- ( dT ) 26 overhang DNA that was biotinylated on the duplex end onto streptavidin beads , loaded it with Rpa , and then added nuclease-dead Dna2 ( D278A mutant ) . As shown in Figure 6C , Dna2 reduced the amount of Rpa bound to the 5’ overhang DNA by ∼30% ( lanes 2 and 3 ) , whereas it had a minimal effect on the Rpa bound to the 3’ overhang DNA ( ∼5% reduction; lanes 5 and 6 ) . The amount of Rpa displacement was proportional to Dna2 loading , which was substantial with 5’ overhang DNA ( ∼45% of the reaction lacking Rpa; lanes 1 and 3 ) , but negligible with 3’ overhang DNA ( ∼3%; lanes 4 and 6 ) . In the absence of Rpa , by contrast , the amount of Dna2 loading on the two DNA substrates was essentially identical ( lanes 1 and 4 ) , consistent with the inability of Dna2 to load onto the 3’ overhang being due to its failure to displace Rpa . The Dna2-ssDNA structure shows that the active site and most of the DNA-binding sites are enclosed in a narrow tunnel , necessitating the threading of the DNA through a tunnel end to access the DNA binding sites . The structure also indicates that the translocase activity of the helicase domain does not drive threading . A 5’ DNA end would have to thread halfway through the tunnel before it can access the helicase domain , while the threading of a 3’ end starting at the helicase 2A domain would be opposed by the 5’ to 3’ polarity of translocation , as we demonstrate . Instead of translocation , the helicase domain appears to be important for providing DNA affinity and for bypassing the 5’ RNA primer of Okazaki fragments . The structure also precludes the helicase domain tracking on DNA to any significant extent , because the nuclease domain is ahead , in the 5’ to 3’ direction of translocation . This is consistent with in vitro studies showing the ATPase activity to be dispensable for 5’ flap processing and DSB resection ( Cejka et al . , 2010; Niu et al . , 2010; Zhu et al . , 2008 ) , and the fact that this activity is rather low compared to bona-fide helicases ( Bae and Seo , 2000; Masuda-Sasa et al . , 2006 ) . In yeast , ATPase mutations do result in growth defects ( Budd et al . , 2000 ) , and it is possible this is due to the ATPase activity contributing to the bypassing of the RNA primer of Okazaki flaps , as suggested by our in vitro data . This may be reflected in the similarity of the Dna2 helicase domain to the Upf1 family of RNA/DNA helicases , which is extensive enough to indicate that Dna2 picked up an ancestral Upf1-like helicase during its evolution . The requirement for threading necessitates Dna2 having a mechanism to displace Rpa from ssDNA . As our proposed mechanism of Rpa displacement predicts , we find that Dna2 can displace Rpa from a 5’ but not a 3’ end , explaining how Rpa dictates the proper end polarity of the nuclease activity of Dna2 . Full-length mouse Dna2 was cloned into a pFastbac1 baculovirus vector engineered with a cleavable N-terminal GST tag and a non-cleavable C-terminal FLAG tag , and was expressed in Hi-5 insect cells ( Life technologies , Carlsbad , CA ) . The recombinant protein was purified first by GST-affinity chromatography and , after cleavage of the GST tag , by anion exchange and gel-filtration chromatography . Purified Dna2 was concentrated to ∼20 mg/mL in 20 mM Tris-HCl , 250 mM NaCl , 0 . 3 mM TCEP , pH 8 . 0 . All buffers were degassed before use . The various Dna2 mutants and seleno-methionine substituted protein were purified similarly . Seleno-methionine substituted Dna2 was expressed according to manufacturer’s protocol ( Expression systems , Davis , CA ) and was purified similarly . For the expression of the mouse Rpa heterotrimer , Rpa70 was cloned into a pFastbac1 vector and Rpa32/Rpa14 were cloned into a modified pFastBac-dual vector with Rpa32 fused to a cleavable N-terminal GST-tag . The Rpa heterotrimer was produced by co-infecting Hi-5 cells with both viruses , and was purified as described ( Fan and Pavletich , 2012 ) . The Rpa heterotrimer with truncated Rpa70 was expressed and purified similarly . GST-tagged mouse Dna2 α1 ( residues 1–20 ) , α1OB ( residues 1–122 ) and OB ( residues 21–122 ) and Rpa70NAB ( residues 1–431 ) , Rpa70N ( residues 1–122 ) , and Rpa70AB ( residues 191–431 ) fragments , as well as the corresponding human polypeptides used in ITC measurements , were cloned into a pGEX-4T vector and expressed in E . coli BL21DE3 cells . They were purified by GST affinity chromatography , ion exchange and/or heparin chromatography , and gel-filtration chromatography . The corresponding untagged polypeptides were expressed fused to an N-terminal 6-His-sumo tag in E . coli BL21DE3 cells . Following nickel affinity chromatography and cleavage of the 6-His-sumo tag by Ulp1 , they were further purified by ion exchange and/or heparin , and gel-filtration chromatography . Crystals of the Dna2-ADP complex were grown in 4°C using the hanging drop vapor diffusion method from a crystallization buffer of 80 mM MES , 250 mM Li2SO4 , 2 mM MgCl2 , 8–12% PEG MME 5000 , 0 . 5 mM TCEP , pH 6 . 5 , containing 12 mg/mL Dna2 and 1 mM ADP . Seleno-methionine substituted Dna2 was crystallized under similar conditions using seeding . Crystals of Dna2 bound to 21-nt ssDNA and ADP ( Dna2-ssDNA in Table 1 ) were grown from a crystallization buffer of 80 mM MES , 20 mM CaCl2 , 10 mM spermidine , 4–9% isopropanol , 0 . 5 mM TCEP , pH 6 . 5 , and 1 mM ADP . They contain two molecules in the asymmetric unit . Crystals of Dna2 bound to 5’ overhang DNA , which consists of a 17-nt 5’ overhanging ssDNA and a 6 base pair dsDNA ( Dna2-5’ overhang DNA in Table 1 ) , grew from a similar condition but in a different space group , and have one complex in the asymmetric unit and higher diffraction limits . As there is no electron density for the duplex , we presume it is disordered . All crystals were cryo-protected in crystallization buffer supplemented with 20–25% glycerol or ethylene glycol , and were flash-frozen in liquid nitrogen . The human DNA2 α1-RPA70 OBN complex was crystallized by mixing an 8 . 7 mg/ml solution of the RPA70N polypeptide ( residues 1–120 ) with a 3-fold molar excess of a synthetic DNA2 α1 peptide ( residues 1–20 ) from 50 mM Tris-HCl , 35% PEG 1500 , 2 mM TCEP , pH 8 . 0 . Crystals were cryo-protected in crystallization buffer supplemented with 20–25% glycerol and flash-frozen in liquid nitrogen . Diffraction data were collected at the 24IDC and 24IDE beamlines of the Advanced Photon Source ( Argonne National Laboratory ) and the X29 beamline of the National Synchrotron Light Source ( Brookhaven National Laboratory ) . Data sets were processed with the HKL2000 suite ( Otwinowski and Minor , 1997 ) . The structure of Dna2-ADP complex was determined using SAD with data collected at the selenium edge ( Bricogne et al . , 2003 ) . The phases were improved using solvent flattening and two-fold NCS averaging with multiple masks with the program DM ( Winn et al . , 2011 ) . The model was built using O ( Jones et al . , 1991 ) and Coot ( Emsley et al . , 2010 ) and refined with REFMAC5 ( Winn et al . , 2011 ) and PHENIX ( Adams et al . , 2010 ) using tight NCS restraints on atom positions . Initial phases for the two Dna2-ADP-ssDNA complexes were obtained by molecular replacement with PHASER ( McCoy et al . , 2007 ) using the apo-Dna2 structure as the search model , and the structures were refined using REFMAC5 ( Winn et al . , 2011 ) and PHENIX ( Adams et al . , 2010 ) , with TLS parameterization of temperature factors of the high resolution Dna2-5’ overhang DNA complex . The Ramachandran plot of the final model has 90 . 5% , 8 . 9% , 0 . 5% and 0% of the residues in the most favored , additional allowed , generously allowed and disallowed regions . The statistics from data collection and refinement are shown in Table 1 . Figures were generated using PyMOL ( http://www . pymol . org ) . Unless otherwise noted , experiments were performed in a 15 μL volume in 20 mM Tris-HCl , 125 mM NaCl , 6 mM MgCl2 , 1 . 3 mM ATP , 0 . 2 mg/mL BSA , 2% glycerol , pH 8 . 0 . Reactions were incubated for 30 min at 25°C and stopped by adding 0 . 5% SDS , 20 mM EDTA and 1 unit of proteinase K . Reactions were analyzed by 16% or 12% denaturing urea-PAGE . For Rpa-containing reactions , Rpa was incubated with DNA for 15 min at 4°C before the addition of Dna2 . For reactions using 6-FAM labeled DNA , wet gels were scanned using a fluorescent laser scanner ( Fujifilm FLA 5000 ) , and the bands were quantified with ImageGauge software ( Fujifilm ) . For GST pull-down experiments , 30 μM GST-tagged mouse α1-OB ( residues 1 to 122 ) or GST was incubated with an equal molar amount of full-length Rpa , Rpa70NAB or Rpa ( -NAB ) . Binding reactions ( 40 μL ) were carried out in 20 mM Tris-HCl , 80 mM NaCl , 0 . 3 mM TCEP , 2% glycerol , pH 8 . 0 at 4°C for 30 min before addition of glutathione beads . After 30 min , the beads were washed three times with binding buffer . Proteins were eluted with 20 mM glutathione and analyzed by SDS-PAGE . Other GST pull-down experiments were carried out similarly , with protein concentrations indicated in the main text or figure legends . For Rpa displacement experiments , 2 μM 5’- or 3’- ( dT ) 26 overhang DNA that was biotinylated at the duplex end was coupled to magnetic streptavidin beads . After washing , the beads were incubated with a 1 molar equivalent of Rpa for 20 min in 20 mM Tris-HCl , 125 mM NaCl , 2% glycerol , 0 . 3 mM TCEP , 0 . 01% Tween 20 , pH 8 . 0 , followed by the addition of a one molar equivalent of Rpa or Rpa-Dna2 mixture . The beads were incubated for 45 min with mixing , and after 3 washes the beads were boiled and analyzed by SDS-page . ITC experiments were carried out using a MicroCal ITC200 calorimeter ( Malvern Instruments Inc . , Westborough , MA ) at 20°C in a buffer of 20 mM HEPES , 80 mM NaCl , 0 . 2 mM TCEP , pH 7 . 5 . Binding reactions ( 10 μl ) containing 0 . 075 or 0 . 3 nM of 32P-labelled DNA substrates with increasing amounts of nuclease-dead Dna2 were carried out in 20 mM Tris-HCl , 80 mM NaCl , 0 . 3 mM TCEP , 0 . 2 mg/mL BSA ( New England Biolabs , Ipswich , MA ) , 2% glycerol , pH 8 . 0 . Reactions were incubated on ice for 30 min . , followed by electrophoresis at 4°C on 4% ( w/v ) native PAGE gels in 1x TB buffer . The dried gels were scanned using a phosphorimager ( GE Typhoon 7000 , GE Healthcare , Pittsburg , PA ) , bands were quantified with ImageGauge software ( Fujifilm ) , and the apparent dissociation constants ( Kd ) were calculated from the equilibrium expression of a one-site binding model . Curve fitting was done by minimizing the sum of the square of the differences between the observed fraction of bound DNA and the fraction predicted from the model .	DNA carries the genetic information that is essential for organisms to survive and reproduce . It is made of two strands that twist together to form a double helix . However , these strands can be damaged when the DNA is copied before a cell divides , or by exposure to radiation or hazardous chemicals . To prevent this damage from causing serious harm to an organism , cells activate processes that rapidly repair the damaged DNA . “Homologous recombination” is one way in which cells can repair damage that has caused both strands of the DNA to break in a particular place . In the first step , several enzymes trim one of the two DNA strands at each broken end to leave single stranded “tails” . Dna2 is one enzyme that is involved in making these tails , but it can only bind to single-stranded DNA so it only acts after another enzyme has made some initial cuts . The exposed single stranded DNA then searches for an intact copy of itself elsewhere in the genome , which promotes its repair . It is important that only one of the two DNA strands is trimmed at each end otherwise the repair will fail . A protein called Rpa is bound to the DNA and is required for Dna2 to correctly trim the DNA . However , it is not clear exactly how Rpa2 regulates Dna2 . Zhou et al . used a technique called X-ray crystallography to analyze the three-dimensional structures of Dna2 when it is bound to single stranded DNA and when it is bound to Rpa . The experiments show that Dna2 adopts a cylindrical shape with a tunnel through which the single-stranded DNA passes through . The region of Dna2 that is capable of trimming DNA – which is called the nuclease domain – is embedded within the tunnel . The entrance to the tunnel is too narrow to allow double-stranded DNA to enter , so this explains why Dna2 can only act on double-stranded DNA that already has a small single-stranded section at the end . Inside the tunnel , Dna2 displaces Rpa from one of the strands , which allows Dna2 to trim the DNA . However , other molecules of Rpa remain firmly bound to the other strand to protect it from Dna2 . These enzymes also act in a similar way to trim DNA before it is copied in preparation for cell division . Zhou et al . ’s findings provide an explanation for how Rpa determines which strand of DNA is trimmed by Dna2 . Further work is needed to understand how Dna2 and Rpa work with other enzymes to trim DNA .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Conclusions", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Dna2 nuclease-helicase structure, mechanism and regulation by Rpa
Here , we present a method for in-depth human plasma proteome analysis based on high-resolution isoelectric focusing HiRIEF LC-MS/MS , demonstrating high proteome coverage , reproducibility and the potential for liquid biopsy protein profiling . By integrating genomic sequence information to the MS-based plasma proteome analysis , we enable detection of single amino acid variants and for the first time demonstrate transfer of multiple protein variants between mother and fetus across the placenta . We further show that our method has the ability to detect both low abundance tissue-annotated proteins and phosphorylated proteins in plasma , as well as quantitate differences in plasma proteomes between the mother and the newborn as well as changes related to pregnancy . Several studies have presented draft maps of the human tissue proteome using mass spectrometry ( MS ) -based methods ( Kim et al . , 2014; Wilhelm et al . , 2014; Bekker-Jensen et al . , 2017 ) . Due to major analytical challenges in plasma , extensive MS-based plasma proteome studies have largely focused on meta-analysis of publicly available datasets . Examples include the Plasma Proteome Database ( PPD ) ( Nanjappa et al . , 2014 ) and the PeptideAtlas plasma proteome repository ( Schwenk et al . , 2017 ) . The recently curated plasma PeptideAtlas now compiles 178 MS datasets , to describe a total of 3509 proteins in plasma , including important in-depth plasma proteomics data-sets from recent years ( Keshishian et al . , 2015; Geyer et al . , 2016a; Geyer et al . , 2016b ) illustrating state of the art of proteome-wide MS-based plasma proteomics . One major challenge in plasma proteomics is the ability to analyze larger sample cohorts while retaining proteome coverage . Plasma datasets are often characterized by a high proportion of one-peptide identifications resulting in poor overlap between analyses . For example , in the Plasma Proteome Database 10 546 proteins have been reported from 509 publications; however , only 3784 of the identified proteins are present in two or more studies ( Nanjappa et al . , 2014 ) . The main analytical challenge when applying proteomics methods to plasma is the presence of a few extremely high abundant proteins that dominate the protein content . Roughly 55% of the total protein mass in plasma is made up by albumin alone and as few as seven proteins together make up 85% of the total protein mass . This can be compared with estimates from tissue and cellular data where 2300 housekeeping proteins are thought to make up 75% of the protein mass ( Kim et al . , 2014 ) . In general , high plasma proteome coverage is dependent on extensive fractionation . Consequently , aiming for high number of identified proteins comes at the cost of low sample throughput and vice versa . Hence , when larger cohorts have been analyzed the number of proteins monitored drastically decreases . Examples of high-throughput studies include 342 proteins monitored in 230 samples in a longitudinal study of twins using SWATH-MS ( Liu et al . , 2015 ) or relative quantification of 146 proteins in 500 subjects using iTRAQ labeling ( 5% peptide FDR ) ( Cole et al . , 2013 ) . Efforts to increase the throughput by shortening the analysis time of shotgun MS have also been made , for example by quantifying 285 proteins from 10 plasma samples in as little as 3 hr analysis time/sample ( Geyer et al . , 2016a ) . The same technology was later applied to a longitudinal study with 43 patients followed over 12 months , making in total 319 plasma samples measured in quadruplicates ( a total of 1294 samples ) , detecting on average 437 proteins per individual ( Geyer et al . , 2016b ) . In 2015 , a study by Keshishian et al . ( 2015 ) reported in total over 5000 proteins from 16 plasma samples using high pH reversed phase separation in combination with iTRAQ 4-plex labeling , making it the single most comprehensive plasma proteomics study to-date . However , despite these advances in the field , in-depth analysis of clinically-relevant-sized plasma cohorts remains a challenge . Here , we present a method to achieve unbiased , reproducible , in-depth plasma proteome analysis . High-resolution isoelectric focusing based pre-fractionation prior MS analysis ( HiRIEF LC-MS/MS ) has demonstrated capability of in-depth proteome analysis of cell and tissue samples ( Branca et al . , 2014 ) . We demonstrate that further development of HiRIEF facilitate in-depth plasma analysis to allow powerful liquid biopsy proteomics . Encouraged by the results , we performed a plasma proteogenomics analysis ( Nesvizhskii , 2014 ) to explore the possibility of detecting single amino acid variants ( SAAV ) in plasma . Presently , only a limited number of studies have applied proteogenomics to plasma ( Liu et al . , 2015; Johansson et al . , 2013; Chen et al . , 2012 ) and as of yet , no global studies have been performed to integrate genomic sequence information to the protein sequence variants in plasma to detect SAAV or other protein sequence alterations , as here presented as proof-of-principle . In the adaptation of the HiRIEF-LC-MS/MS method to plasma analysis , we analyzed female plasma , depleted of high-abundance plasma proteins , using a label-free MS approach . The optimization of the method included evaluation of the pH range of the peptide isoelectric focusing and amount of peptide sample load onto the strips; leading to in total 16 different HiRIEF conditions being assessed ( Table 1 , Supplementary file 1 ) . Additionally , the effect of MS analysis time alone on plasma proteome coverage was evaluated by an MS runtime control composed of 72 LC-MS/MS cycles of a depleted sample using identical settings and total MS time as for the HiRIEF fractions ( Figure 1a ) . This was done to make sure that the number of identifications post fractionation was not purely an inflation caused by MS analysis time alone or an effect of spurious and possibly false identifications caused in the search pipeline when searching a large number of MS-injections in parallel . When optimizing the HiRIEF methodology we identified on average 1505 proteins per condition ( min 904 , max 1888 ) , with strict 1% FDR cut-off at PSM , peptide and protein level ( Table 1 , Supplementary file 1 ) In comparison , the MS runtime control identified as few as 241 proteins . In the runtime control , less than 10 proteins composed 50% of the total protein intensity ( Figure 1—figure supplement 1a ) , hence indicating repetitive detection of high-abundance proteins in absence of further fractionation despite having depleted the 14 most abundant proteins . From the HiRIEF pH range evaluation of broad ( pH 3–10 ) , narrow ( pH 3 . 7–4 . 9 ) and ultra-narrow ranges ( pH 3 . 7–4 . 05 , 4 . 0–4 . 25 , 4 . 2–4 . 45 ) , we concluded that the pH range did not impact largely on protein identification numbers , which is in line with previously reported studies using HiRIEF on cells and tissue ( Table 1 ) ( Zhu et al . , 2018 ) . However , almost twice as many peptides were identified in the broad range compared to the ultra-narrow ranges , implying its usefulness for peptide-based quantitative analyses . As expected , the identification overlap across the different pH intervals was larger on protein level than on peptide sequence level ( Figure 1—figure supplement 1b ) . The overlapping proteins between different pH intervals were predominantly annotated as high-abundance plasma proteins ( Figure 1—figure supplement 1c–f ) . The overlap on peptide level corresponded to the predicted pI distribution , with peptides having an isoelectric point covered by several strips being detected repetitively in the corresponding strips ( Figure 1—figure supplement 1g–h ) . In the second step of the optimization , sample loads on the HiRIEF strip were evaluated in terms of analytical depth ( Figure 1b ) and protein identification numbers . This analysis clearly shows a correlation between peptide load on the strip and analytical depth , as well as total numbers of identified proteins when comparing 0 . 2 mg load to 1 mg , but reaching saturation at 4 mg ( Figure 1c ) , a feature which has previously not been shown in cellular/tissue analysis . To explore the potential of plasma phosphoprotein identification in the acidic HiRIEF range ( Panizza et al . , 2017 ) , we re-searched the data for phosphorylation modifications and found that on average 3 . 3% of proteins to contain phosphorylated peptides across all strips , including the clinically relevant EGFR . The highest number of phosphopeptides was found in strips covering the most acidic pI range ( Supplementary file 2 ) . Somewhat surprisingly , the distribution of protein phosphorylation sites differed from data from previous intracellular studies , showing a relatively high proportion of phospho-tyrosine and phospho-threonine modifications ( Lundby et al . , 2012; Olsen et al . , 2006; Sharma et al . , 2014 ) ( Supplementary file 3 ) . When summarizing the protein identifications from the different loads and pI intervals in total 3053 proteins were identified in the optimization experiments ( Supplementary file 4 ) . All raw data as well as result files ( protein , peptide and psm tables ) from the MS analysis are available in the public repository ProteomeXchange as described in the Materials and methods section . A comparison of our data to the 3509 proteins from the most recent version of the human plasma PeptideAtlas data set ( Schwenk et al . , 2017 ) shows that by using different pH intervals , plasma HiRIEF has the ability to cover at least 2236 of the 3509 proteins reported in the PeptideAtlas using gene centric comparison ( Figure 1—figure supplement 2 ) . Among the 611 plasma proteins detected exclusively by the HiRIEF approach , Golgi membrane proteins and MHC molecules were enriched ( Supplementary file 4 ) . This shows that the plasma HiRIEF method has the potential to both confirm both previously described plasma proteins and add novel components towards a more complete definition of the plasma proteome . To evaluate the sensitivity of the HiRIEF method , prostate-specific antigen protein ( PSA ) was spiked-in at a clinically relevant cut-off level of 4 ng/mL into the female plasma and analyzed on the ultranarrow , narrow and broad pH intervals described in the optimization . Interestingly , PSA was only detected in the 4 . 0–4 . 25 strip and not in the broader 3 . 7–4 . 9 or 3–10 strips , despite covering the same pI interval . This implies that specific , narrower pH intervals could be used to increase analytical depth for selected proteins . In line with this , analyzing the proportion of extracellular , membrane , cytoplasmic and nuclear proteins detected in the different pH intervals , a slightly higher proportion of intracellular proteins were detected in the ultranarrow ranges ( Figure 1—figure supplement 3 ) . Another advantage with the HiRIEF methodology is the predictability of the peptide pI ( Zhu et al . , 2018; Branca et al . , 2014 ) , which allows to design the fraction windows in concordance with the targets of interest . As a proof of concept , we tailored an MRM analysis of fractions 30–35 from the 4 . 0–4 . 25 strip that would theoretically contain the PSA peptide R . LSEPAELTDAVK . V and analyzed the selected fractions from the spike-in experiment using MRM analysis . In the experiment , we detected 670 × 10−18 moles of the PSA peptide , demonstrating that findings from the global analysis can be validated using targeted MS-methods by rational selection of HiRIEF fractions based on theoretical peptide pI ( Figure 1—figure supplement 4a–b ) . A key feature in plasma proteomics analysis is the ability to robustly detect and quantify proteins across samples in larger cohorts . For this purpose , we combined the HiRIEF methodology with TMT-10 plex labeling and analyzed two separate plasma cohorts collected at different hospitals , with four and five TMT-sets , respectively . To investigate the repeatability ( protein overlap between runs ) and quantitative reproducibility of the method we focused on the broad range strip ( pH 3–10 ) , as it would generate the highest number of identified peptides which , in turn , would give the best protein quantification accuracy . Due to the larger pI interval per fraction , the broad range strips give higher peptide focusing accuracy on strips ( Figure 1—figure supplement 5 ) , which is beneficial for the repeatability . On each strip , we loaded in total 1 mg of depleted TMT-labeled plasma ( 100 μg/TMT-label ) , which was in coherence with the yield from a single 40 μl crude plasma injection per sample on the depletion system ( Figure 1b ) . The first cohort contained samples from 30 healthy individuals , 15 men and 15 women . To connect the quantitative information between the sets and enable reproducibility calculations we created a pooled internal standard of depleted and digested plasma from all 30 individuals , which we included in the study . This allowed us to study the peptide level variability by including replicates of the pooled internal standards in each set - four TMT sets , 30 individual samples and 10 pooled internal standards ( Figure 1—figure supplement 6a ) . We identified in total 2587 unique proteins across the TMT-sets , of which 1313 ( 51% ) were detected in all four sets ( Figure 1—figure supplement 6b ) . Quantitative reproducibility was calculated both between and within TMT-sets based on the replicates of the pooled internal standard . The average peptide level technical CV ( coefficient of variation ) within one TMT set was 4 . 7% and the average CV between sets was of 7 . 3% ( Figure 1—figure supplement 6c , d ) . In the second cohort , we wanted to explore if we could reduce the MS analysis time , and still retain the proteome coverage . For this purpose , we tailored a condensed 3–10 HiRIEF LC MS/MS analysis approach where fractions in pI areas containing fewer peptides where pooled and analyzed together in the LC-MS/MS analysis . This reduced the number of fractions analyzed from 72 to 40 , and the MS analysis time by approximately 30 hr to 55 hr per HiRIEF plate . To test this condensed approach , we analyzed a second longitudinal cohort with plasma samples from 12 women making a total of five TMT-sets . In this analysis , we identified in total 2123 proteins , with 1135 being detected in in all five sets , showing that the reduced analysis time only had a minor impact on the number of identified proteins ( Figure 1—figure supplement 7a , b , c , d ) . Lastly , we compared the identified proteins from the nine TMT-sets from the two different cohorts to define a core set of proteins that could be robustly detected in plasma using the 3–10 strip . In this analysis , we used a gene-centric approach to avoid variation in protein grouping caused by protein inference . Using this approach , we identified in total 828 genes in all nine TMT sets containing samples from 42 individuals ( both men and women ) ( Figure 1—figure supplement 7e ) . To benchmark the performance of plasma HiRIEF with current state-of-the-art methodologies , we downloaded the available iTRAQ-4 plex , TMT-6 plex and TMT-10 plex datasets from Keshishian et al . ( Keshishian et al . , 2015; Keshishian et al . , 2017 ) and re-searched the raw data using the same search parameters as for the plasma HiRIEF data . Starting from 400 μl of plasma per sample and performing extensive IgY-based depletion in combination with high-pH reversed phase prefractionation into 30 fractions , Keshishian et al . report between 2066 and 4836 proteins identified per set , depending on the labelling approach and experiment , using approximately 90 hr of MS-time per set . Based on spiked-in peptides they present median peptide level CVs between 16–24% across different abundance ranges . Performing the same gene-centric core-set analysis , as described above for the HiRIEF data , on the re-searched data from the four iTRAQ 4-plex sets ( four patients ) and the iTRAQ 4-plex , TMT-6 plex , TMT-10 plex analysis ( one pooled plasma sample ) , a set of 1394 genes was robustly identified in all seven sets . In comparison to our data , we identified 1043 genes in seven out of the nine sets , with 42 individuals , starting with 40 μl plasma and 14-protein MARS depletion ( Figure 1—figure supplement 7e ) . While identifying slightly less proteins , we believe that the HiRIEF approach has an advantage in throughput , robustness and analytical cost , which makes it well suited for larger clinical cohorts . Next , we used plasma HiRIEF to explore the overall plasma protein inter-individual variability by analyzing the plasma from the 30 healthy donors ( 15 men , 15 women ) . To define which plasma proteins were tightly controlled or highly variable between individuals , we ranked all overlapping quantified proteins based on inter-individual CV ( % ) . When examining the classes of proteins associated with low or high variability by comparative GO enrichment analyses , we found that coagulation- and complement cascade proteins such as complement factor I ( CF1 ) and complement component C6 were tightly regulated , which was in line with previous findings ( Cominetti et al . , 2016 ) . Interestingly , transmembrane proteins and proteins coupled to receptor activity , such as the cancer-related EGFR and TGFBR3 , were also tightly regulated ( Supplementary file 5 ) . Large inter-individual variation was observed for lipoproteins and keratins ( the latter classified as probable contamination ) ( Supplementary file 6 ) . One example of proteins with large variability between individuals was lipoprotein A ( LPA ) , for which a genetic variation affecting its secretion into plasma is known ( Boerwinkle et al . , 1989; Utermann , 1989 ) ( Figure 2a ) . The HLA genes are similarly known to be highly polymorphic and showed very high level of variability , which could be due to the fact that the peptides were not properly assigned to the reference sequences in the database search . Two other highly variable proteins ( APOC2 and CETP ) showed strong gender correlation , with lower concentrations in women ( Figure 2b ) . Increased levels of CETP in women with Type two diabetes , but not in men , has been coupled to cardiovascular disease ( Alssema et al . , 2007 ) . Unsupervised clustering of the data showed distinct gender effects in the plasma proteome ( Figure 2c ) , that are concordant with previous studies ( Corzett et al . , 2010; Miike et al . , 2010 ) . Interestingly , the major gender-correlated cluster with relative up-regulation in male subjects contained proteins linked to lipid transport and binding . In fact , 15 of the 38 proteins in the cluster with significant gender-specific differential expression ( Student's t-test , p<0 . 01 , FDR corrected ) were proteins implicated as candidate cardiovascular disease genes , including seven potential drug targets and two FDA-approved drug targets ( classification according to the ProteinAtlas , www . proteinatlas . org ) ( Figure 2d , Figure 2—figure supplement 2 ) . To explore the potential of detecting biomarkers using plasma HiRIEF , we mined our data for the proteins included in the recently published CancerSEEK test ( Cohen et al . , 2018 ) . We identified five out of eight proteins in the optimization experiments ( covering several pI intervals ) and three of eight proteins in the cohort consisting of 30 healthy plasma donors analyzed only on the 3–10 strip ( Figure 3a , Supplementary file 7 ) . We then expanded the criteria and searched for FDA-approved drug targets , cancer-related proteins and possible tissue leakage proteins , all classified according to the Human Protein Atlas ( HPA ) ( Uhlén et al . , 2015 ) . The latter defined by using the most stringent human tissue proteome annotations provided by HPA - ‘tissue enriched' . We ranked the merged data from the optimization based on abundance and colored the proteins according to protein classes ( Figure 3b ) . As expected , classical plasma proteins were detected in the high-abundance range and potential biomarker classes ( FDA-approved drug targets , cancer-related proteins , the proteins from the CancerSEEK test and the spiked PSA ) were detected in the lower abundance ranges ( Figure 3b , Figure 3—figure supplement 1 , and Supplementary file 8 ) . Notably , using plasma HiRIEF , we could detect over 500 tissue-enriched proteins spanning over a wide range of concentrations , from highly abundant plasma proteins produced in the liver to less abundant proteins produced in the central nervous system or the pancreas ( Figure 3c , Figure 3—figure supplement 1 , and Supplementary file 8 ) . As a proof of concept to demonstrate the detection of potential tissue leakage proteins using the HiRIEF method , we searched for the presence of placenta-enriched proteins in the cohort with 30 blood donors ( non-pregnant , both men and women ) as we would not expect to detect any placenta proteins in these individuals . In this cohort , we detected low levels of eight proteins classified as placental enriched ( median #PSMs 2 , range 0–21 , the outlier protein IGF2; Figure 4a ) . We then obtained and analyzed plasma from a healthy female donor before pregnancy and during third trimester of pregnancy , again on the 3–10 strip as for the healthy donors . During pregnancy , we could detect 30 tissue enriched placental proteins in plasma , with median # of PSM´s 46 ( range 1–424 ) , which were not detected , or detected at low levels prior to pregnancy ( median #of PSMs 0 , range 0–31 , again the outlier protein being IGF2 ) ( Figure 4b ) . We continued the analysis to search for increased amounts of placental proteins in newborns , as the placenta is mainly of fetal origin . Two mother and newborn pairs were analyzed using plasma HiRIEF on HiRIEF 3–10 strips . In this cohort we found over 30 placenta-enriched proteins at high levels in the maternal plasma ( Figure 4c ) , but only a small increase of placental proteins among the newborns , with a median PSM ratio mother:newborn of 4:1 and 5:1 , respectively . Expanding the comparison between newborn and maternal plasma to the full plasma proteome , we compared protein abundances of mother- versus baby-proteomes and also the pregnant- versus not-pregnant proteomes of the female donor by plotting the protein abundances against each other ( Figure 4d ) . Among the proteins uniquely detected in both babies were Anti-Mullerian-Hormone ( AMH ) , which has reported serum level concentrations of 52 ng/mL in newborn boys ( Bergadá et al . , 2006 ) ( both newborns were boys ) compared to approximately 2 ng/mL in females during pregnancy and puerperium ( La Marca et al . , 2005 ) . Among the mother-unique proteins was corticotropin-releasing hormone ( CRH ) , a protein suggested as a trigger for parturition due to its rapid increase in circulating levels at the onset of parturition ( Zannas and Chrousos , 2015 ) . Among the pregnancy-unique proteins were numerous placenta-enriched proteins ( PAPPA2 , PSGs ) . Fetal hemoglobin was found at very high levels in the newborns ( 1501 and 317 PSMs , respectively ) , as expected , but a fraction of fetal hemoglobin was also detected in the mother’s blood ( 55 and 44 PSMs , respectively ) . The fetal-specific isoform of haemoglobin has previously been detected maternal plasma during pregnancy and transfer across the placenta has been proposed , however , it has also been shown that the level of fetal hemoglobin expressed in adults has a genetic component ( Thein et al . , 2007; Galarneau et al . , 2010 ) , making it difficult to determine if it is produced by the mother or transferred across the placenta . Fetal DNA can easily be detected in maternal plasma and is transferred across the placenta during pregnancy ( Lo et al . , 2007 ) . Several proteins are also suggested to pass the placenta ( Wölter et al . , 2016 ) , but providing evidence that such proteins have not been produced in the recipient has remained an analytical challenge . To pinpoint proteins that could have been transferred across the placenta , we adopted a proteogenomics approach ( Figure 4e ) . First , we wanted to know if we could detect single amino acid variants ( SAAVs ) in proteins in plasma that could be used as a traceable protein fingerprint , since this has not been previously shown in plasma . We therefore collected plasma and buffy coat from two mother-newborn pairs in connection to planned Caesarean section . The buffy coats containing white blood cells ( WBC ) were genotyped using SNP arrays and the corresponding plasma samples were analyzed by plasma HiRIEF ( 3–10 strip ) to detect SAAV using a customized database including all coding SNPs in dbSNP ( database generation as described in detail by Zhu et al . , 2018 ) . The MS data was subsequently curated using SpectrumAI within the iPAW pipeline to reduce false discoveries among the detected single amino acid substitutions ( Zhu et al . , 2018 ) . Indeed , in the plasma samples from the four individuals , we detected 229 peptides with unique SAAV ( mapping to in total 161 different genes ) , where the corresponding SNPs were also detected on DNA level in the SNP analysis ( Supplementary file 9 ) . In the cases where we detected a SAAV-containing peptide and the individual was heterozygous for the SNP , we also found the corresponding canonical peptide in 65% of the cases . These findings strengthened the hypothesis that SAAVs could be used as traceable protein fingerprints detected in plasma . We then explored if the curated SAAV peptides detected by MS that could not be explained by the individual’s own genotype could possibly be originating from a transfer across the placenta . Indeed , from the two mother-child pairs we found 32 cases with in total 26 unique peptides sequences from 24 proteins where the amino acid sequence could be explained by the genotype of other the individual , indicating a possible transfer of these proteins across the placenta . Mirror plots of spectra from the peptide in the donor ( with DNA support ) and the recipient ( without DNA support ) can be found in Supplementary file 13 . After manual inspection of the mirror plots , 5 out of the 32 cases where considered poor matches , leaving 21 proteins and 22 unique peptides . The mismatch rate is similar to the rate previously reported using synthetic peptides for sequence verification ( Zhu et al . , 2018 ) . For the majority of the proteins ( n = 11 ) , the transfer was detected in the direction of baby to the mother , but we also detected nine proteins potentially transferred from the mother to the baby . For three of these transferred proteins , we could see the same transfer directions in both of the mother-child pairs , strengthening our findings ( GAA from baby to mother , LRG from baby to mother and HSPG2 from mother to baby ) . For one additional transferred proteins , we saw opposing transfer directions in the two pairs , suggesting that the protein could transfer in both directions ( SERPINB11 ) ( Figure 4f ) . The transferred proteins were spread across the entire abundance range and did not stand out in terms of hydrophilicity/hydrophobicity ( as evaluated by Gravy score ) , size , pI or abundance ( Figure 4—figure supplement 1 and Supplementary file 10 ) . A detailed description of the transferred proteins , including peptides and PSMs can be found in Supplementary file 11 . To our knowledge , this is the first-time plasma proteogenomics has been used to study protein transfer across the placenta in-vivo providing a new possibility to study the molecular communication between mother and baby . In the last 5 years , the MS-based proteomics field has taken major steps forward in terms of proteome coverage and analytical depth . Identification and quantification of over 10 000 proteins can now routinely be performed in cell line and tissue samples , which is in line with the number of transcripts reported using RNA sequencing from same samples ( Thul et al . , 2017 ) . Global proteomics analysis of plasma on the other hand still remains a challenge . Compared to cellular and tissue analysis , plasma proteomics has in particular high demands on throughput and dynamic range . Inherent to plasma analysis is also a very high sample-to-sample variability and lack of genetic blueprint specific for plasma . Taken together , these challenges call for specific adaptions of MS based proteomics analysis for plasma . In their review ‘Revisiting biomarker discovery by plasma proteomics’ Geyer et al argues for a rectangular biomarker strategy , where both the discovery and validation cohort are analyzed using global MS-based proteomics ( Geyer et al . , 2017 ) . This is an appealing approach , as it enables seamless validation of protein species that would be difficult to validate using other technologies , such as proteins containing SAAV or larger panels of proteins . In the current paper , our aim has been to balance the demand on throughput and analytical depth , to develop method based on HiRIEF-MS that performs robustly in an in-depth plasma proteome analysis but also at same time has the capability to analyze larger clinical cohorts . We show that that the method has the capability to repeatedly detect tissue leakage proteins in plasma and that the robust high-precision fractionation enables quantification using several TMT-sets . We also show the method’s potential to detect phosphoproteins and single amino acid variant peptides adding a proteome dimension in to plasma analysis that is difficult to cover by antibodies due to lack of specific reagents . Lastly , we highlight the potential of HiRIEF-MS as a tool for plasma proteogenomics by showing its capability of deciphering the transfer of protein variants between mother and child during pregnancy . The approach provides new and unique possibilities for studies on signaling between the mother and the fetus as well as a great possibility for improving our understanding of pathological pregnancies . No statistical method was used to predetermine sample size . Descriptive statistics , t-tests and correlations were performed using GraphPad Prism version 6 ( GraphPad Software , La Jolla , CA ) , Microsoft Excel 2011 , and R statistical computing software ( https://www . r-project . org ) . Significance assessed by Student's t-test was FDR-adjusted ( Benjamini-Hochberg method ) to correct for multiple testing . Heatmap visualizations and hierarchical clustering ( Pearson correlation ) were performed using the online tool Morpheus ( software . broadinstitute . org/Morpheus ) . Annotations were extracted from BioMart ( www . ensembl . org/biomart ) and Ingenuity Pathway Analysis ( IPA ) software ( QIAGEN Redwood City , www . qiagen . com/ingenuity ) . Gene Ontology annotation of ranked lists was performed using the online tool GOrilla ( Gene Ontology enRIchment anaLysis and visuaLizAtion tool , http://cbl-gorilla . cs . technion . ac . il ) ( Eden et al . , 2009; Eden et al . , 2007 ) . Comparative gene ontology ( GO ) enrichment analyses of multiple lists were performed using the online tool ToppCluster , https://toppcluster . cchmc . org/publications . jsp ( Kaimal et al . , 2010 ) . We examined the presence of tissue leakage proteins , FDA-approved drug targets and cancer-related gene products in plasma by comparing our gene product data with the Human Protein Atlas ( www . proteinatlas . org ) tissue enriched ( enriched defined as at least five-fold higher mRNA levels in a particular tissue as compared to all other tissues , which is a stricter definition than ‘tissue enhanced’ ) proteome database . Figure 3c has been adapted from: https://commons . wikimedia . org/wiki/File:Female_shadow_anatomy_without_labels . png . In addition , we also mined our data for proteins used in a recently described multi-analyte blood test ( CancerSEEK test [Cohen et al . , 2018] ) . Data are available via ProteomeXchange ( http://www . proteomexchange . org/ ) with identifier PXD010899 .	Blood cells travel through the blood vessels in a soupy mixture of proteins called plasma . Most of these proteins are plasma-specific , yet small amounts of proteins can leak into the plasma from other body parts and may provide hints about what is going on elsewhere in the body . This could allow doctors to use plasma samples to assess health or detect disease . But so far developing methods to detect these leaked proteins has proved difficult . Plasma passing through the placenta can transfer proteins between a pregnant woman and her baby . Learning more about these protein exchanges may help scientists understand how the mother and baby adapt to each other and what triggers child birth . But , so far , they have been hard to study . Using DNA to help trace the origins of proteins found in mother or baby could make it easier . Now , Pernemalm et al . have used DNA sequencing in combination with protein analysis to identify proteins passed between two pregnant mothers and their babies . Comparing the genetic sequences of each mother and child made it possible to trace the origin of the proteins . For example , if a mother had a version of the protein that matched genes the child inherited from its father , they knew it passed from the baby to the mother . This approach found 24 proteins in plasma from two pregnant mothers that had likely passed through the placenta during pregnancy . Pernemalm et al . also analyzed the plasma of 30 healthy individuals and confirmed that it contained several proteins that had likely leaked from other organs , including the lungs and pancreas . Monitoring protein transfer between pregnant mother and baby may help scientists identify what triggers normal or premature deliveries . One advantage of the technique developed Pernemalm et al . is that it can analyze plasma proteins from large numbers of people , which could enable larger studies . More refinement of the technique may also allow scientists to identify leaked proteins in the plasma that provide an early warning of cancer or other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources" ]	2019	In-depth human plasma proteome analysis captures tissue proteins and transfer of protein variants across the placenta
Despite the crucial role of bacterial capsules in pathogenesis , it is still unknown if systemic cues such as the cell cycle can control capsule biogenesis . In this study , we show that the capsule of the synchronizable model bacterium Caulobacter crescentus is cell cycle regulated and we unearth a bacterial transglutaminase homolog , HvyA , as restriction factor that prevents capsulation in G1-phase cells . This capsule protects cells from infection by a generalized transducing Caulobacter phage ( φCr30 ) , and the loss of HvyA confers insensitivity towards φCr30 . Control of capsulation during the cell cycle could serve as a simple means to prevent steric hindrance of flagellar motility or to ensure that phage-mediated genetic exchange happens before the onset of DNA replication . Moreover , the multi-layered regulatory circuitry directing HvyA expression to G1-phase is conserved during evolution , and HvyA orthologues from related Sinorhizobia can prevent capsulation in Caulobacter , indicating that alpha-proteobacteria have retained HvyA activity . Genetic exchange is both fundamental to the adaptation of bacterial cells faced with ever-changing environmental conditions and the cause of the alarming dissemination of antibiotic resistance determinants among the bacterial pathogens . The underlying mechanisms include direct uptake of naked DNA ( transformation ) by bacterial cells as well as cell- or bacteriophage-based delivery systems ( respectively conjugation and generalized transduction ) ( Wiedenbeck and Cohan , 2011; Seitz and Blokesch , 2013 ) . Thus , uncovering mechanisms that curb genetic exchange could provide new entry points to help intervene with the spread of antibiotic resistances . While genetic exchange can be facilitated in response to changes in the number of cells in a population ( quorum sensing ) or other developmental states ( Seitz and Blokesch , 2013 ) , an important but yet unresolved question is whether genetic exchange can also be regulated by systemic cues , such as those directing cell cycle progression . Recent cytological experiments provide evidence that components of the pneumococcal natural transformation ( competence ) machinery can be linked to cell division , at least spatially ( Bergé et al . , 2013 ) , hinting that unknown mechanisms may indeed restrict genetic exchange in time or in space during the progression of the cell division cycle . A myriad of events are coordinated with progression through the eukaryotic cell cycle , but our understanding of such mechanisms and the factors that constrain them during the bacterial cell cycle are sparse . Microbial polysaccharidic capsules can also restrict bacteriophage-mediated genetic exchange . Typically , they mask bacteriophage receptor sites that are on or near the cell surface ( Hyman and Abedon , 2010 ) . Moreover , capsules are virulence factors in many Gram-negative and Gram-positive species , as they provide immune evasion by shielding or camouflaging the targets of host immune cells that are located on the surface of bacterial cells ( Schneider et al . , 2007; Kadioglu et al . , 2008 ) . While capsulation can be regulated by nutritional cues ( Kadioglu et al . , 2008; Yother , 2011 ) , cell envelope stresses ( Laubacher and Ades , 2008 ) or physical cues ( Sledjeski and Gottesman , 1996; Tschowri et al . , 2009; Loh et al . , 2013 ) , no systemic cues are currently known . As virulence regulators have recently been found to control bacterial cell cycle transcription ( Fumeaux et al . , 2014 ) , capsulation might also be regulated by the cell cycle . The synchronizable and capsulated alpha-proteobacterium Caulobacter crescentus ( henceforth Caulobacter ) is the pre-eminent model system for cell cycle studies ( Ravenscroft et al . , 1991; Skerker and Laub , 2004; Curtis and Brun , 2010 ) . It recently transpired that many of the emerging concepts of cell cycle control , and the underlying mechanisms such as those directing an asymmetric cell division ( Hallez et al . , 2004 ) , are also operational in other alpha-proteobacterial lineages ( Kobayashi et al . , 2009; Brilli et al . , 2010; Ardissone and Viollier , 2012; Pini et al . , 2013; Fumeaux et al . , 2014 ) . In Caulobacter this asymmetric cell division yields a motile and piliated swarmer ( SW ) cell that is in a G1-arrested state and a sessile stalked ( ST ) cell that resides in S-phase ( Figure 1A ) . The cellular buoyancy of the latter is higher than the former , a feature that has been exploited for synchronization of Caulobacter populations ( by density gradient centrifugation ) ( Ely , 1991 ) for cell cycle studies . A vast number of transcripts are cell cycle-regulated in Caulobacter ( Laub et al . , 2000 ) and in the alpha-proteobacterium Sinorhizobium meliloti ( De Nisco et al . , 2014 ) , a plant symbiont . Importantly , many transcripts of orthologous genes are cell cycle-regulated and this is in large part governed by the conserved and essential cell cycle transcriptional regulator A ( CtrA ) . CtrA activates transcription of many late S- and G1-phase genes that are repressed by the transcriptional regulators SciP or the MucR1/2 paralogs , respectively ( Fumeaux et al . , 2014; Gora et al . , 2010; Tan et al . , 2010 ) . In addition to acting as a transcriptional activator , CtrA functions as a negative regulator of gene expression and DNA replication initiation by binding to the conserved 5′-TTAA- ( N ) 7-TTAA-3′ motif ( CtrA box ) located in many Caulobacter and Sinorhizobium promoters ( Laub et al . , 2000 , 2002; Fumeaux et al . , 2014 ) and the Caulobacter origin of replication ( Cori ) ( Quon et al . , 1998 ) . 10 . 7554/eLife . 03587 . 003Figure 1 . Capsulation of Caulobacter cells is cell cycle regulated . ( A ) Schematic of the Caulobacter cell cycle and the regulatory interactions that determine the presence/absence of the capsule ( in blue ) . Phosphorylated CtrA ( CtrA∼P ) and MucR1/2 control expression of hvyA . The antagonistic kinase/phosphatase pair , DivJ ( yellow dot ) and PleC ( green dot ) , indirectly influences CtrA∼P and partitions with the stalked ( ST ) cell chamber or swarmer ( SW ) cell chamber , respectively . PleC promotes CtrA∼P accumulation in the SW cell . HvyA prevents encapsulation in SW cells . Pink denotes HvyA accumulation in the SW ( G1 ) cell compartment . Light blue indicates the presence of the capsule in ST ( S ) and pre-divisional ( PD ) cells . ( B ) Schematic of cell buoyancy upon centrifugation on density gradient for WT Caulobacter cells ( left ) . SW cells sediment in the lower band ( ‘heavy’ , in pink ) whereas ST and PD cells sediment in the upper band ( ‘light’ , in blue ) . ΔpleC and ΔhvyA cells are ‘light’ due to the constitutive presence of capsule ( middle ) . Upon transposon mutagenesis with himar1 Tn , we isolated ‘heavy’ non-capsulated mutants by cell density centrifugation ( right ) . ( C ) Caulobacter loci identified by the cell density screen . The upper panel represents the CCNA_00162-CCNA_00168 locus and the lower panel the CCNA_03921-CCNA_00472 locus on the mobile genetic element of Caulobacter NA1000 . The fragment deleted for each in-frame deletion is indicated ( Δ ) . Black triangles indicate Tn insertions obtained in the ΔpleC background and white triangles indicate Tn insertions obtained in the ΔhvyA background . CCNA_00166 ( hvyA ) is shown in pink , and the genes hit by our buoyancy screen for ‘heavy’ mutants are in blue . The asterisks show the ORFs identified as essential by Christen et al . ( 2011 ) . ( D ) Schematic of capsule polymerisation/export system based on the one for group 1 CPS in E . coli ( Collins and Derrick , 2007 ) . Putative functions were attributed to Caulobacter proteins based on the homology and conserved domains . CCNA_03998 , CCNA_00466 , and CCNA_00469 are putative glycosyltransferases; PssY , PssZ , and HfsE are polyisoprenylphosphate hexose-1-phosphotransferases; CCNA_00467 is a putative flippase; CCNA_00165 and CCNA_00470 are polymerases; CCNA_00163 and CCNA_00167 have homology to the tyrosine autokinase Wzc and its phosphatase Wzb , respectively , that regulate polymerisation and export; CCNA_00168 is the putative outer membrane lipoprotein required for translocation of the polysaccharide across the outer membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 003 Binding of CtrA to its target sites is stimulated 100-fold by phosphorylation of aspartate at position 51 ( D51 , CtrA∼P ) ( Siam and Marczynski , 2000 ) through a complex phosphorelay that controls both abundance and phosphorylation of CtrA as a function of the cell cycle ( Biondi et al . , 2006; Iniesta et al . , 2006 ) . CtrA∼P is present in G1-phase , proteolyzed during the G1→S transition to permit replication initiation and re-accumulates later in S-phase ( Domian et al . , 1997 ) . G1-phase transcription is also positively dependent on PleC ( Wang et al . , 1993 ) , a phosphatase that is sequestered to the SW cell pole at cell division ( Wheeler and Shapiro , 1999 ) . While PleC also regulates the cellular buoyancy properties , the molecular basis has never been determined . In this study , we use suppressor genetics to unearth the Caulobacter capsule as determinant of the buoyancy trait and we identify HvyA , a member of the poorly characterized bacterial transglutaminase-like cysteine protease ( BTLCP ) family , as a PleC-dependent negative regulator that is restricted to G1-phase to prevent capsulation at this time in the cell cycle . As the capsule protects Caulobacter cells from infection by the generalized transducing Caulophage φCr30 and no CRISPR/Cas ( clustered regularly interspaced short palindromic repeats–CRISPR associated ) -based adaptive immunity system to protect cells from invading genetic material is encoded in the Caulobacter genome ( Marks et al . , 2010 ) , HvyA is the first example of a factor restricting phage infection to a confined cell cycle phase . The switch in cellular buoyancy in NA1000 ( WT ) Caulobacter cells is cell cycle-regulated , but the underlying regulatory mechanism is elusive . We used a developmental mutant ( ΔpleC ) as an entry point to identify the genetic determinants conferring the change in buoyancy . Density gradient centrifugation of a WT culture yields ST and PD cells with the characteristic high buoyancy ( for simplicity , we refer to these cells as ‘light’ ) and SW cells with the characteristic low buoyancy ( ‘heavy’ cells , Figure 1B ) . As ΔpleC cells are exclusively ‘light’ , we simply sought himar1 transposon ( Tn ) insertions that render ΔpleC cells ‘heavy’ ( Figure 1B ) . After backcrossing such ‘heavy’ ΔpleC::Tn mutants , we mapped the Tn insertion sites to two loci ( Figure 1C ) . The first locus ( Figure 1C , upper panel ) encodes putative components of a group 1 ( Wzy ) -like capsular polysaccharide ( CPS ) export machinery , in which the saccharide precursors are first assembled on undecaprenol ( Und∼P , black zigzag in Figure 1D ) on the cytoplasmic membrane , flipped and assembled in the periplasm into a polymer that is then translocated across the outer membrane and anchored on the cell surface ( Whitfield , 2006 ) . Tn insertions were found in the genes encoding a putative capsular polysaccharide biosynthesis lipoprotein ( CCNA_00162 ) , a Wzc-like chain length regulator/tyrosine kinase ( CCNA_00163 ) , a putative O-antigen polymerase/ligase ( CCNA_00164 ) , a putative Wzb-like metallophosphatase ( CCNA_00167 ) , and a Wza-like outer membrane translocon ( CCNA_00168 ) , all commonly associated with capsular export systems . No Tn insertions were found in the other two genes within this cluster , CCNA_00166 and pssY . For the latter , this could be explained by a functional redundancy of pssY with the orthologs encoded by pssZ and hfsE , all encoding polyisoprenylphosphate hexose-1-phosphotransferases ( Toh et al . , 2008 ) . By contrast , CCNA_00166 is predicted to encode a putative bacterial transglutaminase-like cysteine protease ( BTLCP ) and we describe below that an in-frame deletion of CCNA_00166 ( ΔhvyA in Figure 1C ) in WT cells resulted in ‘light’ cells , akin to ΔpleC cells . Consistent with the results from the Tn analysis , an in-frame deletion ( Figure 1C ) in pssY did not render cells ‘heavy’ . By contrast , deleting CCNA_00162 , CCNA_00163 , CCNA_00164 , or ΔCCNA_00167 gave rise to ‘heavy’ WT ( and ΔpleC ) cells ( Figure 1C and Figure 2A , Supplementary file 1 ) . Moreover , complementation of ΔCCNA_00162 , ΔCCNA_00163 , ΔCCNA_00164 , ΔCCNA_00167 , or CCNA_00168::Tn mutant cells with a plasmid harbouring the corresponding gene under the control of the vanillate inducible promoter ( Pvan ) on a medium–copy number plasmid ( pMT335 ( Thanbichler et al . , 2007 ) ) restored the WT buoyancy phenotype , showing that the ‘heavy’ phenotype is attributable to the loss of CCNA_00162 , CCNA_00163 , CCNA_00164 , CCNA_00167 , or CCNA_00168 function ( Supplementary file 1 ) . 10 . 7554/eLife . 03587 . 004Figure 2 . Capsulation affects buoyancy , mucoidy , and bacteriophage sensitivity . ( A ) Sensitivity to bacteriophage φCr30 and buoyancy of Caulobacter WT ( NA1000 ) and different mutant strains . Mutation in CCNA_00163 or CCNA_00470 restores sensitivity to φCr30 in ΔpleC and ΔhvyA mutant backgrounds . ΔpleC , ΔhvyA , and the double mutant ΔpleC ΔhvyA are ‘light’ , whereas mutation in CCNA_00163 or CCNA_00470 renders cells ‘heavy’ ( also in a ΔpleC or ΔhvyA background ) . ( B ) Mucoidy of Caulobacter WT ( NA1000 ) and different mutant strains plated on PYE medium supplemented with 3% sucrose . WT , ΔpleC , and ΔhvyA are highly mucoid , whereas mutations in CCNA_00163 or CCNA_00470 confer a ‘rough’ non-mucoid phenotype in all three backgrounds ( WT , ΔpleC , and ΔhvyA ) . ( C ) Mucoidy of Caulobacter WT ( NA1000 ) , ΔpleC , or ΔhvyA cells over-expressing hvyA under control of Pvan on a medium copy number plasmid ( pMT335 ) or Pxyl on a low copy number plasmid ( pMT375 ) . Over-expression of hvyA confers the typical non-mucoid ‘rough’ colony phenotype on PYE agar plates supplemented with 3% sucrose , while the WT , ΔpleC , and ΔhvyA cells have a mucoid ‘smooth’ colony appearance . Sensitivity to bacteriophage φCr30 and buoyancy of Caulobacter WT ( NA1000 ) , ΔhvyA , ΔpleC , and ΔmucR1/2 strains carrying pMT335-hvyA or pMT375-hvyA are shown in Figure 2—figure supplement 1–2 . RsaA extracted from the same Caulobacter strains shown in Figure 2 is displayed in Figure 2—figure supplement 3 . The effect of proteinase K treatment on CCNA_00168 in Caulobacter WT , ΔhvyA , and ΔCCNA_00163 is shown in Figure 2—figure supplement 4 . The effect of the ΔhvyA mutation on swarming motility of Caulobacter is shown in Figure 2—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 00410 . 7554/eLife . 03587 . 005Figure 2—figure supplement 1 . Over-expression of hvyA renders ΔpleC cells ‘heavy’ and sensitive to φCr30 . Sensitivity to φCr30 and buoyancy of C . crescentus WT , ΔpleC or ΔhvyA cells harbouring hvyA on plasmid , under control of Pvan ( pMT335 , medium copy number ) or Pxyl ( pMT375 , low copy number ) . Over-expression of hvyA restores sensitivity to φCr30 and ‘heavy’ cell buoyancy to the ΔpleC or ΔhvyA cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 00510 . 7554/eLife . 03587 . 006Figure 2—figure supplement 2 . Over-expression of hvyA renders ΔmucR1/2 cells ‘heavy’ . Buoyancy of ΔmucR1/R2 cells harbouring hvyA on plasmid . Over-expression of hvyA from Pvan or Pxyl restores the ‘heavy’ phenotype to ΔmucR1/R2 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 00610 . 7554/eLife . 03587 . 007Figure 2—figure supplement 3 . S-layer is correctly assembled in Caulobacter mutants affected in capsule production . ( A ) RsaA , the main component of Caulobacter S-layer , is present in mutants affected in capsule synthesis/export . RsaA was extracted by EGTA treatment and run on 7 . 5% SDS-PAGE , followed by Coomassie staining . Mutations in pleC , hvyA , or CCNA_00163 do not affect the presence of RsaA . Molecular size standards are indicated in blue on the left , with the corresponding values in kDa . RsaA is indicated by the arrowhead on the right . ( B ) The amount of RsaA extracted by EGTA treatment is not affected by hvyA over-expression or in mutants affected in capsule synthesis/export . Proteins extracted by EGTA treatment were run on 7 . 5% SDS-PAGE , followed by Coomassie staining . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . RsaA is indicated by the arrowhead on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 00710 . 7554/eLife . 03587 . 008Figure 2—figure supplement 4 . Capsule protects cell surface proteins from proteinase K treatment . Immunoblot anti-CCNA_00168 ( outer membrane protein ) on whole cells ( WT , ΔhvyA , or ΔCCNA_00163 ) treated with proteinase K . Samples were taken at different time points ( 0 , 15 , 30 , 45 , and 60 min ) after adding proteinase K . The immunoblot shows that CCNA_00168 is degraded by proteinase K more rapidly in the non-capsulated ΔCCNA_00163 mutant than in WT or ΔhvyA cells . This indicates that the absence of the capsule in the ΔCCNA_00163 mutant strain makes the proteins in the outer membrane more accessible to proteinase K . Molecular size standards are indicated in blue on the left , with the corresponding values in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 00810 . 7554/eLife . 03587 . 009Figure 2—figure supplement 5 . Loss of hvyA affects swarming motility . ΔhvyA cells show a motility defect on soft agar plate ( PYE supplemented with 0 . 3% agar ) . We hypothesize that encapsulation of SW cells in the ΔhvyA mutant could interfere with flagellar rotation , as mutation of ΔCCNA_00167 restores WT motility to the ΔhvyA cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 009 The second cluster of genes resides on a 26-kbp mobile genetic element ( MGE ) that has previously been implicated in buoyancy ( Marks et al . , 2010 ) ( Figure 1C , lower panel ) . Specifically , we recovered Tn insertions in genes predicted to encode a homolog of the putative N-acetyl-L-fucosamine transferase WbuB from E . coli that is involved in O-antigen ( O26 ) synthesis ( D'Souza et al . , 2002 ) ( CCNA_03998 ) , a polysaccharide polymerase ( CCNA_00470 ) and a GDP-L-fucose synthase ( CCNA_00471 ) . The three genes are near other coding sequences for polysaccharides biosynthesis proteins , including two other putative glycosyltransferases ( CCNA_00466 and CCNA_00469 ) , a Wzx-like polysaccharide flippase/translocase ( CCNA_00467 ) and a sugar mutase homolog ( CCNA_00465 ) ( Marks et al . , 2010 ) . Consistent with a previous genome-wide Tn analysis showing that these genes cannot be disrupted ( Christen et al . , 2011 ) , we were unable to engineer in-frame deletions in CCNA_00466 or CCNA_00467 in the absence of a complementing plasmid . As the entire 26-kb MGE is dispensable for viability , it appears that inactivation of either one of these four genes results in synthetic toxicity due to a polysaccharide intermediate or the sequestration of Und∼P that is also required for peptidoglycan synthesis ( Yother , 2011 ) . To confirm that CCNA_03998 , CCNA_00470 , and CCNA_00471 are indeed buoyancy determinants , we engineered in-frame deletions in CCNA_03998 or CCNA_00470 ( CCNA_00471 was not tested ) in WT or ΔpleC mutant cells and found the resulting single or double mutants to be ‘heavy’ ( Figures 1C and 2A , Supplementary file 1 ) . The association of the ‘heavy’ phenotype with a capsule synthesis and/or export defect , the resemblance of CCNA_03998 to the predicted N-acetyl-L-fucosamine transferase WbuB ( D'Souza et al . , 2002 ) and the fact that D-fucose is a known component of the extracellular polysaccharide or capsule of C . crescentus ( Ravenscroft et al . , 1991 ) prompted us to test if these mutations also affected colony mucoidy , a phenotype typically associated with the presence of capsule or exopolysaccharides . Indeed , all ‘heavy’ mutants exhibited a non-mucoid ( ‘rough’ ) colony phenotype on PYE agar plates supplemented with 3% sucrose , while the WT or ‘light’ mutants ( ΔpleC ) had a mucoid ( ‘smooth’ ) colony appearance ( Figure 2A , B , Supplementary file 1 ) . In support of this result , we purified capsule from WT Caulobacter , ΔCCNA_00163 ( ‘heavy’ and ‘rough’ ) , ΔCCNA_00166 ( ΔhvyA , ‘light’ and ‘smooth’ ) , and ΔhvyA ΔCCNA_00163 ( ‘heavy’ and ‘rough’ , see below ) cells . As the Caulobacter capsule is primarily composed of neutral monosaccharides including fucose , mannose , galactose , and glucose ( Ravenscroft et al . , 1991 ) , we used glycosyl compositional analysis as proxy to quantify capsular material from ‘heavy’ and ‘light’ strains ( See ‘Materials and methods’ ) . As shown in Table 1 , the expected sugars ( determined as % of total carbohydrate weight in the preparations ) were abundant in preparations from the WT and the ‘light’ mutant ( ΔhvyA , Table 1 ) , whereas those from the ‘heavy’ mutants ( i . e . the ΔCCNA_00163 single mutant and the ΔhvyA ΔCCNA_00163 double mutant ) contained far less fucose , galactose , and mannose ( up to 37-fold , 23-fold , and 5 . 5-fold reductions , respectively , in the ΔhvyA ΔCCNA_00163 double mutant compared to the ΔhvyA single mutant ) . We also observed a significant reduction in galacturonic acid in the preparations from the ‘heavy’ mutants vs WT or ‘light’ cells , raising the possibility that this saccharide is also a constituent of the NA1000 capsule ( Table 1 ) . 10 . 7554/eLife . 03587 . 010Table 1 . Glycosyl composition of per-O-trimethylsilyl ( TMS ) derivatives of methyl glycosides performed on purified capsular polysaccharides from WT Caulobacter ( NA1000 ) , the single mutants ΔCCNA_00166 ( ΔhvyA ) and ΔCCNA_00163 and the ΔhvyA ΔCCNA_00163 double mutantDOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 010NA1000ΔhvyAΔCCNA_00163ΔhvyA ΔCCNA_00163Glycosyl residueMass ( μg ) Weight ( % ) Mass ( μg ) Weight ( % ) Mass ( μg ) Weight ( % ) Mass ( μg ) Weight ( % ) Ribose0 . 80 . 40 . 10 . 10 . 30 . 30 . 50 . 8Rhamnose2 . 31 . 21 . 10 . 66 . 46 . 22 . 74 . 3Fucose19 . 810 . 326 . 514 . 70 . 00 . 00 . 20 . 4Xylose0 . 00 . 00 . 10 . 10 . 00 . 00 . 00 . 0Glucuronic Acid0 . 00 . 00 . 00 . 00 . 80 . 80 . 30 . 5Galacturonic acid28 . 414 . 932 . 618 . 15 . 65 . 42 . 33 . 7Mannose23 . 212 . 126 . 614 . 83 . 93 . 81 . 72 . 7Galactose30 . 115 . 737 . 020 . 61 . 41 . 40 . 50 . 9Glucose64 . 333 . 652 . 929 . 449 . 948 . 523 . 738 . 3N-Acetyl galactosamine2 . 01 . 10 . 00 . 02 . 32 . 21 . 01 . 7N-Acetyl glucosamine16 . 08 . 43 . 01 . 629 . 528 . 727 . 945 . 2N-Acetyl mannosamine4 . 12 . 20 . 00 . 02 . 22 . 10 . 71 . 2Σ=191 . 3179 . 9102 . 861 . 8Mass is expressed in μg and weight % is relative to the total carbohydrate . Taken together , our results show that the loss of capsule synthesis and/or export renders cells ‘heavy’ and ‘rough’ and that the loss of capsulation is epistatic to the loss of PleC in buoyancy control . On these grounds we predicted that PleC , directly or indirectly , regulates one of the newly identified buoyancy determinants . Since the buoyancy switch is cell cycle-regulated and since all ΔpleC cells are ‘light’ ( Figure 1B ) , we reasoned that PleC is required to turn off capsule synthesis in G1-phase SW cells . As PleC is also required to activate motility and PilA ( the structural subunit of the pilus filament ) expression , when ΔpleC divides two daughter cells that are capsulated , non-piliated and non-motile are formed . By contrast division of WT yields one piliated , motile , and non-capsulated ( ‘heavy’ ) G1-phase SW progeny and one capsulated ( ‘light’ ) S-phase ST cell ( Figure 1B ) . If PleC indeed restricts capsulation temporally , it might control expression of a negative regulator of capsulation . Interestingly , PleC is required for the accumulation of the transcript of CCNA_00166 ( referred to as hvyA due to its requirement to render cells ‘heavy’ ) ( Chen et al . , 2006 ) . The hvyA transcript peaks during the G1-phase and encodes a 272-residue protein ( Figure 3A ) harbouring a classical N-terminal Sec-dependent signal sequence ( SS ) , but lacking discernible hydrophobic sequences or a lipidation signal for retention in the membrane , suggesting that it is periplasmic . The C-terminal part of HvyA features a BTLCP domain . This domain is thought to introduce intra- or inter-molecular crosslinks by transamidation , forming γ-glutamyl-ε-lysine isopeptide bonds between Gln and Lys residues , to hydrolyse amide bonds by the reverse protease reaction and/or to execute deamidation/esterification reactions of glutamine residues ( Lorand and Graham , 2003; Ginalski et al . , 2004 ) . Cysteine proteases typically feature a Cys-His-Asp catalytic triad ( C192 , H226 , and D241 for HvyA , based on sequence alignment , Figure 3—figure supplement 1 ) for the formation of a thioester bond intermediate by the reaction of the active site thiol ( from the Cys ) with Gln , followed by the transfer of the acyl group to an amine substrate ( from the Lys ) ( Lorand and Graham , 2003 ) . As an in-frame deletion in hvyA ( ΔhvyA ) phenocopied the buoyancy defect of ΔpleC cells ( yielding exclusively ‘light’ mucoid cells , Figures 1b , 2a , 2b ) , we conclude that HvyA is required for capsule-mediated buoyancy control in Caulobacter . While expression of WT HvyA from Pvan ( pMT335-hvyA ) reversed the buoyancy defect of ΔhvyA and ΔpleC cells , analogous plasmids encoding the predicted catalytic mutants ( C192S/A , H226Q/A , or D241A ) were unable to do so , although all the HvyA variants accumulated to comparable steady-state levels as the WT protein on immunoblots ( Figure 3C , Figure 3—figure supplement 2 ) . Thus expression of catalytically active HvyA is necessary and sufficient to mitigate the buoyancy defect of ΔpleC cells . As a genetic selection for ‘heavy’ ΔhvyA::Tn mutants was answered by Tn insertions in the same genes that render ΔpleC cells ‘heavy’ ( Figure 1C ) , we reasoned that mutations in capsule synthesis and export genes are epistatic to both the ΔhvyA and ΔpleC mutations . To confirm this notion , we engineered ΔhvyA ΔCCNA_00163 , ΔhvyA ΔCCNA_00167 , and ΔhvyA ΔCCNA_00470 double mutants as well as a ΔpleC ΔhvyA ΔCCNA_00167 triple mutant and found all resulting mutants to be ‘heavy’ and non-mucoid ( ‘rough’ ) on PYE sucrose plates ( Figure 2A , B and Supplementary File 1 ) . Importantly , constitutive expression of HvyA ( from a Pvan- or a Pxyl-plasmid ) in WT , ΔpleC , or ΔhvyA cells also renders cells ‘heavy’ and ‘rough’ ( Figure 2C , Figure 2—figure supplement 1 ) . On the basis of these findings we hypothesized that HvyA is a G1-specific negative regulator of capsulation whose expression is dependent on PleC ( Figure 1A ) . 10 . 7554/eLife . 03587 . 011Figure 3 . HvyA is a bacterial transglutaminase-like cysteine protease ( BTLCP ) homologue and its catalytic activity is required for function . ( A ) Schematic of HvyA domains: signal sequence ( SS ) and BTLCP domain ( in red ) are indicated . C192 , H226 , and D241 constitute the putative catalytic triad ( C192S/A , H226Q/A , and D241A alleles are non-functional; the D241N allele is functional , consistently with some BTLCP family members having a C/H/N catalytic triad ) . Residues identified in the buoyancy screen for non-functional variants are indicated below ( blue , non-functional; green , partially functional ) . ( B ) Schematic of the buoyancy screen for HvyA non-functional variants . The Pvan-hvyA::TAP fusion ( on plasmid ) was subjected to random mutagenesis , then introduced into Caulobacter cells that were subjected to multiple rounds of enrichment for ‘light’ phenotype by centrifugation on density gradient . ( C ) Immunoblot anti-HvyA-TAP on periplasmic proteins extracted by EGTA treatment . The HvyA alleles mutated in putative catalytic residues are expressed and exported to the periplasm like the WT protein . Immunoblot against Caulobacter β-lactamase ( CCNA_02223 , Bla on the lower panel ) is a control for periplasmic proteins . Molecular size standards are indicated in blue on the left , with the corresponding values in kDa . ( D ) ΔhvyA strains harbouring hvyA catalytic mutants under control of Pvan on plasmid were tested for sensitivity to φCr30 . Over-expression of the C192S or H226Q alleles does not restore sensitivity to φCr30 , indicating that these alleles are non-functional . ( E ) Immunoblot anti-HvyA-TAP on ΔhvyA cells harbouring mutagenized Pvan-hvyA-TAP and selected for ‘light’ buoyancy . The Pvan-HvyA-TAP mutant alleles selected are still over-expressed . Molecular size standards are indicated in blue on the left , with the corresponding values in kDa . ( F ) Immunoblot anti-HvyA-TAP on EGTA fractions of the same clones shown in panel ( E ) . The HvyA-TAP mutant alleles selected are exported to the periplasm like WT HvyA-TAP . Molecular size standards are indicated in blue on the left , with the corresponding values in kDa . ( G ) ΔhvyA strains harbouring hvyA variants under control of Pvan on pMT335 were tested for sensitivity to φCr30 . The L240R , H226Y , and W178S alleles do not restore sensitivity to φCr30 , whereas the P263R , R161P and D194G variants partially restore sensitivity to φCr30 . The alignment of BTLCPs protein sequences from Caulobacter ( HvyA ) , S . meliloti ( SMc00998 ) , S . fredii NGR234 ( NGR_c12490 ) , and P . fluorescens ( PFL_0130 ) is shown in Figure 3—figure supplement 1 . Immunoblots against HvyA-TAP , CtrA , and β-lactamase on whole lysates and EGTA fractions of cells expressing HvyA point mutants are shown in Figure 3—figure supplement 2–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01110 . 7554/eLife . 03587 . 012Figure 3—figure supplement 1 . Conservation of the BTLCP domain in HvyA orthologs . Alignment of protein sequences for members of the BTLCP family . CAUCN , C . crescentus HvyA; RHIME , S . meliloti SMc00998; RHISN , S . fredii NGR_c12490; PSEF5 , P . fluorescens PFL_0130 . The three putative catalytic residues are indicated in red . The residues identified by our buoyancy screen for HvyA loss-of-function are indicated in blue ( non-functional ) and green ( partially functional ) above the sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01210 . 7554/eLife . 03587 . 013Figure 3—figure supplement 2 . Catalytic HvyA-TAP mutants are still exported to the periplasm . Immunoblots on total proteins ( right ) and EGTA fractions ( left ) of ΔhvyA cells harbouring Pvan-hvyA-TAP derivatives on plasmid ( WT or catalytic mutants ) . All the variants tested were produced and exported to the periplasm . Immunoblot anti-CCNA_02223 ( β-lactamase , Bla ) was used as control for the periplasmic proteins and immunoblot anti-CtrA as control for cytoplasmic proteins . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01310 . 7554/eLife . 03587 . 014Figure 3—figure supplement 3 . Over-expression of HvyA-TAP alleles . ( A ) Immunoblot anti-HvyA-TAP on WT and ΔhvyA cells harbouring mutagenized Pvan-hvyA-TAP derivatives on plasmid . Mutagenized Pvan-hvyA-TAP was introduced in WT or the ΔhvyA strain and ‘light’ cells were selected by centrifugation on density gradient . ‘Light’ clones were tested by immunoblot in order to verify that the HvyA-TAP fusion was still expressed . The clones indicated by arrows were selected for sequencing the hvyA-TAP and identify the mutations . Cells harbouring WT Pvan-hvyA-TAP were used as control ( C+ ) for the immunoblot . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . ( B ) Immunoblot anti-CtrA on EGTA fractions of ΔhvyA cells harbouring Pvan-hvyA-TAP derivatives on plasmid ( WT or mutants ) . CtrA is a cytoplasmic protein and it is not released by EGTA treatment . The samples shown in this immunoblot are the same shown in Figure 3F . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 014 As all previous efforts to visualize the capsule directly by light or electron microscopy had been unsuccessful ( Ravenscroft et al . , 1991 ) , we conducted negative stain fluorescence microscopy ( FM ) with fluorescein isothiocyanate ( FITC ) -coupled dextran to measure the zone of exclusion of FITC-dextran in capsulated ( ΔhvyA , ‘light’ mucoid ) and non-capsulated ( ΔCCNA_00163 , ‘heavy’ non-mucoid ) cells ( Figure 4A , B ) . Akin to the difference between capsulated and non-capsulated Streptococcus pneumoniae ( Hathaway et al , 2012; Schaffner et al , 2014 ) , the zone of exclusion of FITC-dextran was significantly smaller in the case of ΔCCNA_00163 compared to ΔhvyA cells , although the actual size of the cells by differential interference contrast ( DIC ) microscopy was comparable ( Figure 4A ) . The increase in the exclusion radius of the dextran polymer can be explained by the presence of a capsule on ΔhvyA cells and by its absence from ΔCCNA_00163 cells . Atomic force microscopy ( AFM ) ( Dufrêne , 2014 ) provided additional support for this interpretation . In these experiments , bacteria were immobilized on porous membranes , a method allowing AFM imaging of the bacteria in liquid medium . However , recording reliable images of live cells turned out to be very difficult for two reasons: ( i ) most cells were detached or pushed aside during scanning; ( ii ) strong interactions between the tip and the soft bacterial surface led to fuzzy images that were difficult to interpret . Therefore , cells were imaged after fixation with 4% paraformaldehyde , a protocol that is known to preserve cellular structures ( Chao and Zhang , 2011 ) . Figure 5A–D shows representative low-resolution deflection images of fixed cells from the ΔCCNA_00163 and ΔhvyA strains ( from two independent preparations for each ) . As opposed to ΔCCNA_00163 cells , which were readily imaged without apparent cell surface damage ( Figure 5A , Figure 5C ) , the surface morphology of ΔhvyA cells ( Figure 5B , Figure 5D ) was strongly altered by the scanning tip , and showed streaks in the scanning direction , reflecting strong interactions between the tip and the soft cell surface . Similar trends were observed when recording high-resolution images on top of individual cells ( Figure 5E–H ) . While the surface of ΔCCNA_00163 cell was smooth ( surface roughness on 0 . 06 µm2 areas of ∼0 . 9 nm ) and featureless , ΔhvyA cells were rougher ( roughness of ∼2 . 2 nm ) and showed streaks in the scanning direction , suggesting that soft , loosely bound material was pushed away by the tip . In light of earlier AFM studies ( Dague et al . , 2008 ) , we note that such smooth and rough morphologies are consistent with the presence of crystalline-like arrays of proteins and of an amorphous layer of polysaccharides , respectively . 10 . 7554/eLife . 03587 . 015Figure 4 . Negative fluorescence imaging of capsule . ( A ) Differential Interference Contrast ( DIC , top ) and fluorescence images showing FITC-dextran exclusion ( middle ) or mCherry ( bottom ) of mixed ΔCCNA_00163 ( expressing mCherry ) and ΔhvyA cells . The area of FITC-dextran exclusion of the non-capsulated ΔCCNA_00163 cells is much smaller than that of the capsulated ΔhvyA cells . ( B ) Statistical analysis of the area from which FITC-dextran was excluded . The analysis was performed as described in the ‘Materials and methods’ . ( C ) DIC ( top ) and fluorescence images showing FITC-dextran exclusion ( middle ) or mCherry ( bottom ) on NA1000 spmX-mCherry cells . SW and ST/PD cells were separated on density gradient before incubation with FITC-dextran . Area of exclusion of ST and PD cells ( ‘light’ cells , on the left ) is much bigger than that of SW cells ( on the right ) . SpmX-mCherry is absent from SW cells , accumulates at the SW to ST cell transition and labels the ST pole . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01510 . 7554/eLife . 03587 . 016Figure 5 . Nanoscale AFM imaging reveals the ultrastructure of Caulobacter cell surface . Contact-mode deflection images of ΔCCNA_00163 cells ( A , C , E , G ) and ΔhvyA cells ( B , D , F , H ) at low ( A–D ) and high ( E–H ) magnifications . White arrows indicate streaks generated by the AFM tip scanning the soft , loosely bound layer at the surface of ΔhvyA cells . Images were taken on bacteria from two independent cultures for each strain . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 016 Taken together our results show that ΔCCNA_00163 cells are devoid of capsule , while a soft layer of capsular polysaccharides covers ΔhvyA cells . The results above raised the possibility that HvyA normally prevents capsulation in G1-phase cells , and FITC-dextran staining of an NA1000 culture expressing a ST cell marker ( SpmX-mCherry ) indeed revealed that G1-phase cells ( i . e . SW cells isolated on density gradient ) did not exclude the polymer and are thus non-capsulated , whereas ST/PD cells ( ‘light’ cells on density gradient ) show a much bigger area of FITC-dextran exclusion ( Figure 4C ) . The results above suggested that the G1-phase SW cells are able to synthesize capsule if HvyA is absent . To test if capsule export proteins are normally present in G1-phase cells , we raised antibodies to CCNA_00162 , CCNA_00163 , CCNA_00164 , CCNA_00167 , and CCNA_00168 . Immunoblotting revealed that these components are present throughout the cell cycle of WT cells ( Figure 6A ) . Owing to the poor specificity of the antibodies raised against HvyA , we engineered cells encoding a functional mCherry ( mCh ) -tagged HvyA derivative ( mCh-HvyA ) in which the mCherry moiety is fused in-frame after the SS . mCh-HvyA is expressed from the native hvyA promoter at the hvyA locus in lieu of untagged HvyA . Immunoblotting using polyclonal antibodies to mCherry revealed that mCh-HvyA is indeed restricted to G1-phase ( Figure 6A ) . Antibiotic chase experiments showed that mCh-HvyA is a very unstable protein and is rapidly degraded akin to the unstable protein CtrA ( Figure 6B ) , indicating that accumulation of HvyA is dictated by the timing of its synthesis ( G1-phase ) . 10 . 7554/eLife . 03587 . 017Figure 6 . HvyA is an unstable protein and its presence is restricted to G1-phase ( SW cells ) . ( A ) Immunoblots showing protein levels in synchronized Caulobacter cells: CCNA_00162 , CCNA_00163 , CCNA_00164 , CCNA_00167 , and CCNA_00168 levels do not change over the cell cycle . mCherry-HvyA accumulates only in SW cells , akin to CtrA . CcrM accumulates in PD cells , and PilA levels increase upon re-accumulation of CtrA . The β-lactamase ( Bla ) was used as control as it is constitutively present along the cell cycle . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . ( B ) Immunoblots showing degradation of mCherry-HvyA ( indicated by the arrowhead ) after addition of chloramphenicol ( 2 μg/ml ) to stop protein synthesis . mCherry-HvyA is degraded as rapidly as the unstable protein CtrA ( lower panel ) . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . ( C ) Immunoblot anti-mCherry-HvyA showing that levels of the mCherry-HvyA fusion protein ( indicated by the arrowhead ) are significantly decreased in the ΔmucR1/2 mutant strain . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . Levels of CCNA_00162 , CCNA_00163 , CCNA_00164 , CCNA_00167 , and CCNA_00168 in Caulobacter cells over-expressing HvyA ( from Pvan ) are shown in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01710 . 7554/eLife . 03587 . 018Figure 6—figure supplement 1 . Over-expression of hvyA does not alter the levels of capsule polymerization/export components . Immunoblots showing that the abundance and migration of capsule polymerization/export components is not affected by over-expression of hvyA . CCNA_00162 , CCNA_00168 , CCNA_00163 , CCNA_00167 and CCNA_00164 are equally present in WT cells harbouring the empty vector ( pMT335 ) or Pvan-hvyA ( pMT335-hvyA ) . Molecular size standards are indicated in blue on the right , with the corresponding values in kDa . CCNA_00167 and CCNA_00164 are indicated by the arrowhead on the left of the corresponding panels . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 018 Next , we explored if the G1-specific regulation of HvyA is due to transcriptional regulation by PleC . Using a lacZ-based promoter-probe reporter in which the hvyA promoter is transcriptionally fused to the lacZ gene ( PhvyA-lacZ ) , we found that PhvyA is indeed positively dependent on PleC ( 41 . 1 ± 6 . 0% of WT activity , Figure 7A ) . Inactivation of the gene encoding the antagonistic DivJ kinase in ΔpleC cells mitigated this response ( 94 . 0 ± 19 . 9% of WT activity , Figure 7A ) . Consistent with the fact that PleC and DivJ regulate CtrA∼P levels , our recent chromatin immunoprecipitation coupled to deep-sequencing ( ChIP-Seq ) analyses indicated that CtrA indeed binds the hvyA promoter ( PhvyA ) and that this binding is strongly diminished in ΔpleC cells ( Figure 7—figure supplement 1 ) . Moreover , PhvyA-lacZ is poorly active in ctrA401 mutant cells ( 43 . 3 ± 1 . 9% of WT activity , Figure 7A ) that express a temperature-sensitive version of CtrA ( T170I ) in lieu of WT CtrA ( Quon et al . , 1996 ) . By contrast , PhvyA-lacZ is strongly de-repressed ( 574 ± 30% of WT activity , Figure 7A ) in the absence of the paralogous repressors MucR1 and MucR2 ( ΔmucR1/2 ) that directly bind PhvyA ( Figure 7A , Figure 7—figure supplement 1 ) and that silence many promoters of G1-phase genes ( i . e . those that are activated by CtrA∼P after compartmentalization ( Fumeaux et al . , 2014 ) ) . Thus , CtrA∼P and MucR1/2 directly activate and repress PhvyA , respectively ( see Figure 8 ) . Consistent with the latter , MucR1/2 also represses PhvyA-lacZ activity in ΔpleC cells over-expressing WT or a phosphomimetic variant ( D51E ) of CtrA ( Figure 7—figure supplement 2 ) . Thus , hvyA ( and other genes whose transcripts peak in G1 , such as pilA and sciP ) is expressed from a promoter that is temporally confined during the cell cycle via activation and repression by PleC/CtrA∼P and MucR1/2 , respectively . 10 . 7554/eLife . 03587 . 019Figure 7 . Transcriptional and translational regulation of hvyA depends on PleC-CtrA , SciP , and MucR1/2 . ( A ) Beta-galactosidase activity of PhvyA-lacZ transcriptional fusion . Left panel: transcription of hvyA is strongly reduced in ctrA401 ( T170I , temperature sensitive ) and ΔpleC strains compared to WT Caulobacter . Mutation of the kinase DivJ partially restores hvyA transcription in the ΔpleC ΔdivJ strain . Transcription from PhvyA-lacZ is significantly increased in the ΔmucR1/2 mutant . Right panel: beta-galactosidase activity of PhvyA-lacZ in ΔmucR1/2 cells complemented with Caulobacter mucR1 ( CC_R1 , WT or mutant derivatives Y97C and R85C ) or the MucR paralogs from S . fredii NGR234 ( Sf_a00320 and Sf_c07580 ) on plasmid under control of Pvan . Values are expressed as percentages ( activity in WT NA1000 or WT carrying the empty vector set at 100% ) . ( B ) Beta-galactosidase activity of PhvyA-hvyA::lacZ translational fusion in the same strains shown in panel ( A ) . Translation of hvyA is strongly reduced in cells expressing the ctrA401 allele and in the ΔpleC strain compared to WT Caulobacter . Mutation of DivJ partially restores hvyA translation in the ΔpleC ΔdivJ strain . PhvyA-hvyA::lacZ activity is also significantly decreased in the ΔmucR1/2 and ΔpleC ΔmucR1/2 mutants , consistently with the ‘light’ buoyancy of these strains . PhvyA-hvyA::lacZ activity is restored in the ΔmucR1/2 double mutant carrying ctrA ( T170A ) , sciP ( T24I ) , or sciP ( T65A ) alleles ( ctrA* and sciP* ) . Values are expressed as percentages ( activity in WT NA1000 or WT carrying the empty vector set at 100% ) . ( C ) Translational control of hvyA by CtrA and SciP . Beta-galactosidase activity of the PhvyA-hvyA::lacZ fusion in WT Caulobacter cells over-expressing sciP ( T65A ) or sciP ( WT ) from Pvan on plasmid . Whereas sciP ( T65A ) does not affect the activity of the translational PhvyA-hvyA::lacZ fusion , over-expression of sciP ( WT ) significantly decreases the activity of the fusion , consistently with the model presented in Figure 8 . The activity is expressed as percentage of the activity in WT cells carrying the empty vector . Occupancy of CtrA , MucR1 , and MucR2 at the hvyA promoter region as determined by ChIP-seq analysis is shown in Figure 7—figure supplement 1 . The effect of the CtrA ( D51E ) allele on PhvyA-lacZ is shown in Figure 7—figure supplement 2 . The ability of heterologous MucR to restore PhvyA-lacZ repression or PhvyA-hvyA::lacZ activity in Caulobacter ΔmucR1/2 is reported in Figure 7—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 01910 . 7554/eLife . 03587 . 020Figure 7—figure supplement 1 . Occupancy at the hvyA promoter region as determined by ChIP-seq analysis . ( A ) Traces show occupancy of MucR1 ( in blue ) , MucR2 ( in purple ) , CtrA in WT cells ( in green ) , and CtrA in ΔpleC cells ( in red ) . In the absence of PleC , the occupancy of the hvyA promoter by CtrA is significantly reduced . ( B ) Same as panel A , but at a higher magnification to show the CtrA and MucR2 peaks that are otherwise masked by the MucR1 peak . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 02010 . 7554/eLife . 03587 . 021Figure 7—figure supplement 2 . Transcriptional control of hvyA by PleC-CtrA . Beta-galactosidase activity of the PhvyA-lacZ transcriptional fusion in WT and ΔpleC cells harbouring ctrA ( WT ) or the phosphomimetic ctrA ( D51E ) allele under control of Pvan on plasmid . The assay shows that the phosphomimetic ctrA ( D51E ) allele is unable to restore hvyA transcription in ΔpleC cells . The activity is expressed as percentage of the activity in WT cells carrying the empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 02110 . 7554/eLife . 03587 . 022Figure 7—figure supplement 3 . Transcriptional and translational control of hvyA by heterologous MucRs . Beta-galactosidase activity of the PhvyA-lacZ ( transcriptional ) and PhvyA-hvyA::lacZ ( translational ) fusions in ΔmucR1/2 cells harbouring heterologous mucR genes under control of Pvan on plasmid . Cc_R1 , Caulobacter mucR1 ( WT or mutant alleles ) ; Cc_R1 long , original annotation of Cc_R1 ( CCNA_00982 ) ; Sf_a00320 , S . fredii NGR_a00320; Sf_c07580 , S . fredii NGR_c07580; At_ROS , A . tumefaciens mucR paralog; Bs_mucR , B . suis mucR paralog; Bh_mucR , B . henseleae mucR paralog . The assay shows that in the presence of a heterologous MucR increased translational activity of PhvyA-hvyA::lacZ is always accompanied with a commensurate repression of the transcriptional PhvyA-lacZ . The activity is expressed as percentage of the activity in WT cells carrying the empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 02210 . 7554/eLife . 03587 . 023Figure 8 . Model for regulation of HvyA synthesis and control of cell buoyancy/capsulation in Caulobacter . In WT cells , hvyA transcription is activated by CtrA and repressed by MucR1/2 , whereas translation is promoted by a ( still unknown ) factor ( X , itself under control of CtrA and SciP like many late S phase genes ) . This tight regulation restricts HvyA synthesis to G1-phase , where HvyA prevents the encapsulation of the SW cell . In a ΔmucR1/2 mutant , SciP is over-produced , which lowers the levels of the translational regulator of hvyA and prevents HvyA protein accumulation , despite de-repression of hvyA transcription . The presence of the ctrA ( T170A ) , sciP ( T24I ) , or sciP ( T65A ) alleles ( ctrA* or sciP* ) restores the synthesis of the factor X in the ΔmucR1/2 mutant background , and therefore restores HvyA synthesis and WT buoyancy . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 023 Remarkably , regulation of PhvyA by MucR is conserved in S . meliloti . We found that PhvyA-lacZ was strongly de-repressed ( 1683 ± 130% of WT activity ) in a mucR::Tn mutant derivative ( Rm101 ) of the S . meliloti WT strain ( Rm2011 , Figure 9A ) . Conversely , promoter probe assays in WT and mucR mutant cells of Caulobacter and S . meliloti revealed that the promoter of SMc00998 , the S . meliloti hvyA ortholog that can partially substitute for C . crescentus hvyA ( see below ) , is regulated by MucR in both these alpha-proteobacteria . The promoter probe plasmid PSMc00998-lacZ indicated that the SMc00998 promoter is strongly de-repressed in S . meliloti mucR::Tn ( 314 ± 25% of WT activity ) and the Caulobacter ΔmucR1/2 mutant ( 568 ± 25% of WT activity , Figure 9B ) compared to WT . Recent microarray data showed that the SMc00998 transcript peaks in G1-phase and that its promoter harbours a CtrA-binding site ( De Nisco et al . , 2014 ) . Moreover , our ChIP-Seq analysis indicates that CtrA associates with the promoter of the hvyA-like gene NGR_c12490 of Sinorhizobium fredii NGR234 ( Fumeaux et al . , 2014 ) . Thus , positive and negative transcriptional regulation by CtrA and MucR is potentially wide-spread in alpha-proteobacteria . 10 . 7554/eLife . 03587 . 024Figure 9 . Conservation of hvyA transcriptional control and BTLCP function in alpha-proteobacteria . ( A ) Beta-galactosidase activity of the PhvyA-lacZ transcriptional fusion in WT and ΔmucR1/2 cells compared to S . meliloti WT ( Rm2011 ) and mucR mutant ( mucR::Tn , Rm101 ) . The assay shows that the transcriptional repression of PhvyA-lacZ by MucR is conserved in S . meliloti . The activity is expressed as the percentage of the activity in ( Caulobacter or S . meliloti ) WT cells . ( B ) Beta-galactosidase activity of the PSMc00998-lacZ transcriptional fusion in WT and ΔmucR1/2 cells compared to S . meliloti WT ( Rm2011 ) and mucR mutant ( mucR::Tn , Rm101 ) . The assay shows that the HvyA paralog SMc00998 is also under transcriptional repression by MucR in both S . meliloti and Caulobacter . The activity is expressed as the percentage of the activity in ( Caulobacter or S . meliloti ) WT cells . ( C ) Sensitivity to φCr30 and buoyancy of C . crescentus ΔhvyA cells harbouring HvyA paralogs under control of Pvan on plasmid . Over-expression of NGR_c12490 or SMc00998 restores sensitivity to φCr30 and ( partially ) cell buoyancy , whereas NGR_c19800 , NGR_c36180 , or the A . tumefaciens paralog ( Atu0252 ) are unable to complement the ΔhvyA mutation . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 024 Buoyancy tests with ΔmucR1/2 cells unearthed another mechanism of cell cycle-control of HvyA synthesis: translational regulation . We uncovered this mechanism upon our erroneous prediction that ΔmucR1/2 cells should be ‘heavy’ ( non-capsulated ) . This prediction was based on the strong de-repression of PhvyA in the absence of MucR1/2 observed above and our result that overexpression of HvyA renders Caulobacter cells ‘heavy’ . To our surprise , we found that ΔmucR1/2 cells are in fact ‘light’ ( Figure 2—figure supplement 2 ) and only express trace amounts of mCh-HvyA ( Figure 6C ) . In support of the notion that a deficiency in HvyA underlies the buoyancy defect of this strain , the activity of a PhvyA-hvyA::lacZ translational fusion is indeed reduced in ΔmucR1/2 cells ( 19 . 5 ± 5 . 4% of WT activity , Figure 7B ) , despite the massive de-repression of PhvyA . Moreover , constitutive expression of HvyA ( on plasmid , from Pvan or Pxyl ) induced the ‘heavy’ ( non-capsulated ) phenotype in ΔmucR1/2 cells ( Figure 2—figure supplement 2 ) . Expression of WT MucR1 from pMT335 alleviated both the buoyancy defect and PhvyA de-repression of ΔmucR1/2 cells , while analogous plasmids encoding the mutant MucR1 derivatives Y97C or R85C were unable to do so ( Figure 7A , Figure 7—figure supplement 3 ) , indicating that these defects are indeed due to aberrant expression of a MucR1/2-dependent target gene . Remarkably , MucRs from alpha-proteobacterial lineages can control transcription and translation of hvyA . In fact MucR orthologs from Agrobacterium tumefaciens , Brucella suis , Bartonella henselae , or S . fredii NGR234 ( NGR_a00320 or NGR_c07580 ) can partially complement the Caulobacter ΔmucR1/2 mutant . Interestingly , in the presence of a heterologous MucR , increased translational activity of PhvyA-hvyA::lacZ was always accompanied with a commensurate repression of the PhvyA-lacZ transcriptional reporter ( Figure 7A , B , Figure 7—figure supplement 3 ) . Specifically , while S . fredii NGR_c07580 and A . tumefaciens ROS were functional in transcriptional and translational regulation of hvyA , B . henselae MucR was inactive . However , B . suis MucR and S . fredii NGR_a00320 displayed an intermediate level of activity ( 47 . 0 ± 3 . 0% and 38 . 2 ± 2 . 7% of WT activity for PhvyA-hvyA::lacZ and 302 . 0 ± 11 . 1% and 283 . 5 ± 20 . 6% of WT activity for PhvyA-lacZ , respectively; Figure 7A , B , Figure 7—figure supplement 3 ) . How might MucR regulate the translation of hvyA ? The results above , along with our finding that MucR1/2 acts pleiotropically , prompted us to speculate that another target of MucR1/2 is responsible for translational regulation of hvyA . Our experiments showed that the translational regulator of hvyA ( X in Figure 8 ) is an indirect target of MucR1/2 and is deeply integrated into the cell cycle circuitry . Indeed , we observed that mutations in the master regulator gene ctrA [ctrA ( T170A ) ] or its antagonist gene sciP [sciP ( T24I ) or sciP ( T65A ) ] ( Fumeaux et al . , 2014 ) are epistatic to the ΔmucR1/2 mutation , as they conferred near normal PhvyA-hvyA::lacZ translation and a normal cellular buoyancy ( capsulation ) pattern with ‘heavy’ and ‘light’ cells , while PhvyA was still de-repressed ( Figure 7A , B ) . In support of this view , over-expression of WT SciP from pMT335 cripples PhvyA-hvyA::lacZ translation ( 33 . 9 ± 4 . 7% of WT activity ) , while over-expression of SciP ( T65A ) only has negligible effects ( 83 . 8 ± 9 . 9% of WT activity; Figure 7C ) . Importantly , we identified mechanisms for both transcriptional and translational regulation of hvyA in Caulobacter cells and showed that these two functions can be genetically uncoupled . This regulatory complexity highlights the requirement of proper buoyancy and capsulation control during the cell cycle . Since SciP is the negative regulator of S-phase genes ( activated by CtrA at the PD cell stage before compartmentalization ) , while MucR1/2 negatively regulate G1-specific genes ( activated by CtrA after compartmentalization ) , the coordinated transcriptional and translational control of hvyA by CtrA , MucR1/2 , and SciP suggests that cells prepare themselves for the impending capsule-less SW cell phase by setting the stage for rapid translation of HvyA once the transcript is synthesized at compartmentalization . As it has been suggested that the MGE and the buoyancy phenotype can influence the resistance profile to the S-layer specific Caulophage φCr30 ( Edwards and Smith , 1991; Awarm and Smit , 1998 ) , we evaluated this link and the possible involvement of hvyA quantitatively by genome-wide Tn mutagenesis followed by deep-sequencing ( Tn-Seq ) of Caulobacter cells challenged or not with bacteriophage φCr30 . This analysis revealed that hvyA is one of the two major φCr30-resistance determinants , along with the rsaA locus encoding the RsaA subunit of the S-layer and CCNA_01057 that is predicted to function in S-layer assembly ( Awram and Smit , 1998 ) ( Figure 10A ) . In fact , Tn insertions in hvyA and the rsaA gene are >380 and >160 times over-represented vs other genes in φCr30-challenged cells compared to those without phage , while insertions in CCNA_01057 are over-represented >400-fold upon phage treatment of cells , consistent with the S-layer serving as φCr30-receptor ( Edwards and Smit , 1991 ) . As expected we observed that insertions in the PleC-regulated hvyA gene confer a growth advantage in the presence of φCr30 , but we also found Tn insertions in pleC itself to be enriched ( >90-fold , Figure 10A ) . The reduction of Tn insertion bias in pleC compared to hvyA may be related to the residual expression of hvyA in pleC mutant cells ( and thus lower fitness towards φCr30 ) compared to the complete absence of HvyA due to Tn insertions in hvyA . 10 . 7554/eLife . 03587 . 025Figure 10 . Inactivation of HvyA and ectopic presence of capsule protect cells from phage infection . ( A ) Tn insertion bias in coding sequences ( CDS ) of NA1000 + φCr30 relative to WT cells as determined by Tn-seq . Peaks show CDSs with the highest number of Tn insertions . CCNA_01057 is part of the S-layer locus and is required for S-layer assembly . CDSs with an insertion ratio higher than 50 are indicated ( CCNA_00497 , putative rhamnosyl transferase; CCNA_01199 , putative glucose-1-P thymidylyltransferase; CCNA_03316 , putative UDP-N-acetylglucosamine-4 , 6-dehydratase; CCNA_03744 , putative dTDP-glucose-4 , 6-dehydratase ) . Non-coding sequences are not included . ( B ) Model showing how capsule can interfere with φCr30 infection . In WT NA1000 , φCr30 can infect SW cells , whereas ST cells are protected by the capsule ( in blue ) that masks the S-layer ( φCr30 receptor ) . In a ΔhvyA mutant , both cell types ( SW and ST ) are capsulated , which significantly reduces the ability of φCr30 to infect these cells . Conversely , in a capsule-less mutant ( for example ΔhvyA CCNA_00163::Tn , like the mutants isolated in our buoyancy screen for ‘heavy’ cells ) , both cells types ( SW and ST ) can be infected by φCr30 . Tn insertion bias in coding sequences ( CDS ) of NA1000 cells relative to NA1000 + φCr30 as determined by Tn-seq is shown in Figure 10—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 02510 . 7554/eLife . 03587 . 026Figure 10—source data 1 . ( xlsx ) contains the insertion ratios obtained for the Tn-Seq experiment ( column F was used to create Figure 10A and column G Figure 10—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 02610 . 7554/eLife . 03587 . 027Figure 10—figure supplement 1 . Inactivation of capsule synthesis/export genes decreases Caulobacter fitness towards φCr30 . Tn insertion bias in coding sequences ( CDS ) of WT cells relative to NA1000 + φCr30 as determined by Tn-seq . Peaks indicate CDSs that received fewest insertions in NA1000 + φCr30 relative to untreated WT cells . The two loci that we identified as required for capsule biosynthesis/export are indicated ( CCNA_00162-CCNA_00168 , and CCNA_00460-CCNA_00481 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 027 In contrast to the positive effects of the insertions in hvyA upon φCr30 challenge , Tn insertions in CCNA_00162-00168 and CCNA_00460-00481 were strongly under-represented ( >220- and >120-fold , respectively ) indicating that these genes have the opposite fitness effect , as would be predicted from our epistasis experiments on capsulation ( Figure 10—figure supplement 1 ) . To confirm these Tn-Seq results , we conducted phage spot tests with serial dilutions of φCr30 on lawns of either WT , ΔhvyA , or ΔpleC cells and observed that deletion in hvyA or pleC did not yield cleared zones ( cell lysis ) where the phage had been spotted ( Figure 2A ) . Complementation of the ΔhvyA strain with WT hvyA restored φCr30 sensitivity , whereas the catalytic point mutants ( encoding C192S and H226Q ) were unable to do so ( Figure 3D ) . In addition , ΔhvyA or ΔpleC strains harbouring ΔCCNA_00163 or ΔCCNA_00470 mutations exhibited clearing zones ( lysis ) akin to the WT ( Figure 2A ) . Thus , a ΔhvyA or ΔpleC mutation renders the entire population ‘light’ , capsulated , and resistant to φCr30 , while either a ΔCCNA_00163 or ΔCCNA_00470 ( akin to ΔCCNA_0162 , ΔCCNA_00164 , ΔCCNA_00167 , CCNA_00168::Tn or ΔCCNA_03998; Figure 2A , Supplementary file 1 ) mutation mitigates this effect , rendering the double mutants ‘heavy’ , non-capsulated , and sensitive to φCr30 . Moreover , the fact that mutations in pleC , hvyA , or CCNA_00163 did not noticeably affect the release of the RsaA S-layer subunit from the cell surface ( Figure 2—figure supplement 3A ) indicates that the S-layer is present and properly assembled in all the mutants . On the basis of these results , we hypothesize that the capsule protects the S-layer from φCr30 ( as depicted in the model in Figure 10B ) . Consistent with a protective role of the capsule for the cell surface and our results in spot tests with φCr30 , we also observed higher frequencies of generalized transduction by φCr30 for non-capsulated strains compared to WT , ΔpleC , or ΔhvyA strains ( Table 2 ) . At the same time , treatment of intact cells with proteinase K showed that outer membrane proteins such as CCNA_00168 are more sensitive to the protease in ΔCCNA_00163 mutant cells compared to WT or ΔhvyA strain ( Figure 2—figure supplement 4 ) , which also supports a protective role of the capsule for the cell surface and illustrates its role in preventing φCr30 infection in Caulobacter . 10 . 7554/eLife . 03587 . 028Table 2 . Transduction frequencies of φCr30 lysates in different Caulobacter strainsDOI: http://dx . doi . org/10 . 7554/eLife . 03587 . 028StrainhvyA::pSA480CCNA_01524::TnNA1000 ( WT ) 344∼750ΔpleC64165ΔhvyA48162ΔCCNA_00163522≥1000ΔpleC ΔCCNA_00163527≥1000ΔhvyA ΔCCNA_00163487≥1000For transduction , cells were normalised according to the OD600 and infected with the same amount of φCr30; two different markers were transduced , a pGS18T derivative integrated at the hvyA locus ( pSA480 ) and a himar1 insertion in CCNA_01524 ( flagellar modification gene , unrelated to cellular buoyancy or capsule production ) . The number of colonies counted after 3 days of incubation at 30°C is reported . The experiment was repeated twice . The connection among capsulation , buoyancy , and φCr30 resistance phenotypes prompted us to probe these phenotypes in structure–function analyses with HvyA . To this end , we exploited the HvyA overexpression phenotype to seek HvyA mutant derivatives with impaired function . Briefly , a plasmid ( pUG52 ) expressing a HvyA-derivative with a C-terminal TAP ( tandem affinity purification ) tag ( Puig and et al . , 2001 ) under the control of Pvan was subjected to random mutagenesis and subsequently introduced into the WT and the ΔhvyA mutant strain . Whereas this plasmid normally renders cells ‘heavy’ ( non-capsulated ) due to overexpression of HvyA-TAP , serial enrichment of the mutant pool by density gradient centrifugation led to recovery of ‘light’ mutants ( Figure 3B ) . Following immunoblotting of candidate strains for those still overexpressing HvyA-TAP ( Figure 3E , Figure 3—figure supplement 3A ) , we recovered a number of mutants with single amino acid substitutions within the BTLCP domain of HvyA . Among these variants we recovered mutants in or very close to the putative catalytic site ( H226Y , D194G , L240R ) , as well as mutations in conserved ( W178S ) , similar ( R161P ) or non-conserved ( P263R ) residues throughout the BTLCP domain ( Figure 3A , Figure 3—figure supplement 1 ) . While the isolation of the mutations in the catalytic centre validated the forward genetic screen , upon re-testing the mutants three of them ( R161P , D194G , and P263R ) exhibited complete loss-of-function with respect to buoyancy but only partial loss-of-function with respect to φCr30 sensitivity under conditions of maximal induction of the Pvan promoter on the plasmid . In fact , these mutants were able to support replication of φCr30 ( as shown by cells lysis in Figure 3G ) , indicating that they are indeed partially functional and that the φCr30 spot assay is much more sensitive as readout , possibly because localized breaches in the capsule are sufficient to promote φCr30 infection at a high multiplicity of infection ( m . o . i . ) . By contrast , the other missense mutants from the screen ( W178S , H226Y , and L240R ) neither restored φCr30-sensitivity nor buoyancy ( Figure 3G ) . To confirm that mutations in the BTLCP do not affect sorting to the periplasm , we released periplasmic components by treating cells with EGTA ( ethylene-glycol-tetraacetic acid ) , which is known to destabilize the outer membrane ( Raetz et al . , 2007 ) . Such treatment effectively liberated the beta-lactamase encoded by Caulobacter ( CCNA_02223 ) along with WT or mutant HvyA-TAP variants , but not cytoplasmic proteins such as CtrA ( Figure 3C , F; Figure 3—figure supplement 2 , Figure 3—figure supplement 3B ) . Prompted by the identification of the functional site in the Caulobacter BTLCP , we asked if heterologous BTLCPs support HvyA function in Caulobacter . While this was not the case for the BTLCP from A . tumefaciens ( Atu0252 ) or two HvyA paralogs from S . fredii NGR234 ( NGR_c19800 and NGR_c36180 ) , the BTLCP homologs SMc00998 and NGR_c12490 from S . meliloti and S . fredii NGR234 , respectively , restored φCr30-sensitivity and partially compensated for the buoyancy defect of ΔhvyA cells ( Figure 9C ) . Thus , despite adaptation to dissimilar ecological niches during evolution , alpha-proteobacterial lineages have retained and exhibit related BTLCP activity . The intricate transcriptional and translational regulatory circuit acting on the unstable BTLCP homolog HvyA confines its presence , and thus activity , to a specific phase ( G1 ) in the Caulobacter cell cycle . An inhibitor of encapsulation , HvyA ensures that G1 cells are capsule-less and that upon its elimination during the G1→S transition encapsulation can commence . In light of this first demonstration of cell cycle-regulated capsulation and the importance of capsule in masking , preventing , or impeding immune responses in bacterial pathogens , our results raise the intriguing possibility of targeting specifically the G1-phase regulatory hierarchy to combat bacterial infections by crippling encapsulation mechanisms . This can be achieved either by inhibiting capsular functions such as export or biosynthesis or , alternatively , by potentiating or sustaining the action of capsule inhibitors , such as Caulobacter HvyA , throughout the cell cycle . Our genetic dissection of the regulatory complexity underlying HvyA synthesis , involving cell cycle-controlled transcription and translation , provides a possible entry point for achieving such dysregulation to prevent capsulation . Noteworthy are our results showing that the mechanisms directing hvyA transcription in G1-phase are also operational in other alpha-proteobacteria . Interference with HvyA proteolysis at the G1→S transition presents another strategy to obtain capsule-less cells . However , possible strategies in doing so await the identification of the periplasmic protease responsible for the proteolytic degradation of HvyA . In addition to offering protection from immune cells , capsulation is a known resistance mechanism to bacteriophage adsorption ( Hyman and Abedon , 2010 ) . Our data show that in Caulobacter the T4-like phage φCr30 is obstructed by capsulation . As the G1-specific inhibition of capsulation by HvyA is released with the removal of HvyA during the G1→S transition , the nascent S-phase cells are protected from infection by φCr30 . As φCr30 is a generalized transducing bacteriophage , this temporal confinement of capsulation also has the consequence of limiting φCr30-mediated horizontal exchange of genetic material to SW cells . Temporal control of capsulation may also serve to prevent steric interference on the flagellar rotation by the capsule in the motile G1-phase . Steric hindrance of flagellar function by extracellular polysaccharides is well documented in other bacteria ( Blair et al . , 2008; Jarrell and McBride , 2008 ) and , consistent with this notion , we observed a slight reduction in the motility of ΔhvyA cells compared to WT or the ΔhvyA ΔCCNA_00167 double mutant ( Figure 2—figure supplement 5 ) . In S-phase cells the capsule may confer protection from certain bacteriophages or other predators while also endowing S-phase cells with the buoyancy that maintains them near high oxygen tensions at the surface of aqueous environments . A strict aerobe , Caulobacter , and S-phase cells in particular may have demand for energy production by respiration . BTLCPs are particularly widespread in alpha-proteobacterial genomes and are also found in certain gamma-proteobacterial clades such as Pseudomonads and Vibrios , but they appear to act on specific targets in these lineages . In Caulobacter , the HvyA coding sequence is embedded in the CPS export locus , but in other alpha-proteobacteria HvyA-like BTLCPs appear to be encoded in other genomic context ( s ) , so they probably affect also phenotypes unrelated to surface polysaccharides , mucoidy , and/or buoyancy . However , BTLCPs are predicted to have signal sequences for export into the periplasm and are often encoded in the vicinity of trans-envelope transport systems or enzymes predicted to act in the periplasm , suggesting that they act on periplasmic or extra-cytoplasmic targets . Consistent with an enzymatic activity , mutations in predicted catalytic residues of HvyA abrogate function and we were able to demonstrate limited promiscuity among HvyA , NGR_c12490 , and SMc00998 in vivo , at least under conditions of over-expression . Such protein crosstalk is not unexpected and has also been documented between non-cognate histidine kinases and response regulators for example . The crosstalk among the three transglutaminase/protease-like enzymes HvyA , NGR_c12490 , and SMc00998 indicates similar targets in Caulobacter and Sinorhizobia . Unlike HvyA , NGR_c12490 , and SMc00998 , coding sequences are not embedded within genes encoding known surface polysaccharides export proteins . However , over-expression of HvyA did not alter the abundance or migration of the CPS export proteins CCNA_00162 , CCNA_00163 , CCNA_00164 , CCNA_00167 , and CCNA_00168 by SDS-PAGE immunoblotting , or the amount of assembled RsaA S-layer subunit ( Figure 2—figure supplement 3B , Figure 6—figure supplement 1 ) , indicating that HvyA is not required to maintain the steady state levels of these proteins . The distant HvyA relative LapG from Pseudomonas fluorescens cleaves the cell surface adhesin LapA that is encoded nearby in P . fluorescens , but not found in Pseudomonas aeruginosa , Legionella pneumophila , or C . crescentus genomes ( Newell et al . , 2009; Navarro et al . , 2011; Newell et al . , 2011; Chatterjee et al . , 2012 ) . While the target of the LapG ortholog from L . pneumophila is unknown , this enzyme can cleave P . fluorescens LapA in vitro ( Chatterjee et al . , 2012 ) . LapG is regulated at the level of activity by sequestration to the cytoplasmic membrane by the cyclic-di-GMP responsive trans-membrane receptor LapD ( Navarro et al . , 2011 ) . Moreover , the Ca2+ ions not only promote outer membrane integrity , but also stimulate the cleavage of LapA by LapG in P . fluorescens ( Boyd et al . , 2012 ) . Residues predicted to bind Ca2+ ions are conserved in HvyA , and our unbiased structure–function analyses indeed unearthed mutations in conserved residues that are important for HvyA function . Thus , while control of HvyA activity by Ca2+ ions could contribute to control of HvyA action on its proper targets within the envelope , the tight regulation of HvyA ( and its orthologs ) abundance during the alpha-proteobacterial cell cycle represents the major regulatory mechanism to constrain BTLCP activity temporally . Caulobacter crescentus NA1000 ( Evinger and Agabian , 1977 ) and derivatives were grown at 30°C in PYE ( peptone-yeast extract ) or M2G ( minimal glucose ) . Sinorhizobium fredii NGR234 ( Stanley et al . , 1988 ) was grown at 30°C in TY ( tryptone-yeast extract ) . Sinorhizobium meliloti Rm2011 and derivatives ( Casse , 1979; Becker et al . , 1997 ) were grown at 30°C in Luria broth ( LB ) supplemented with CaCl2 2 . 5 mM and MgSO4 2 . 5 mM . Escherichia coli S17-1 λpir ( Simon et al . , 1983 ) , EC100D , EC100D pir-116 ( Epicentre Technologies , Madison , WI ) , and Rosetta ( DE3 ) pLysS ( Merck KGaA , Darmstadt , Germany ) were grown at 37°C in LB . The E . coli mutator strain XL-1 Red ( Agilent Technologies Inc . , Cedar Creek , TX ) was grown at 30°C in LB . Motility assays , swarmer cells isolation , electroporations , biparental matings , and bacteriophage φCr30-mediated generalized transductions were performed as described ( Ely , 1991; Viollier and Shapiro , 2003; Viollier et al . , 2004; Chen et al . , 2005 ) . Nalidixic acid , kanamycin , gentamicin , and tetracycline were used at 20 ( 8 for S . meliloti ) , 20 , 1 ( 10 for E . coli and S . meliloti ) , and 1 ( 10 for E . coli and S . meliloti ) μg/ml , respectively . Plasmids for β-galactosidase assays were introduced into S . meliloti by bi-parental mating and into C . crescentus by electroporation . Transposon mutagenesis of ΔpleC and ΔhvyA strains was done by intergeneric conjugation from E . coli S17-1 λpir harbouring the himar1-derivative pHPV414 as previously described ( Viollier et al . , 2004 ) . All the nalidixic acid and kanamycin resistant clones were pooled and grown in liquid medium before isolation of ‘heavy’ cells by centrifugation on density gradient . The isolated ‘heavy’ cells were subjected to three subsequent steps ( of growth in liquid medium and centrifugation on density gradient ) in order to obtain enrichment in ‘heavy’ mutant cells . After the third step of selection , ‘heavy’ cells were plated to obtain single colonies , and single Tn insertion sites were mapped by partial digestion of genomic DNA with HinPI , religation , transformation of rescued plasmids into E . coli EC100D-pir116 and identification of insertion sites by sequencing as previously described ( Viollier et al . , 2004 ) . The nucleotide positions of all mapped Tn insertions are reported in Supplementary file 2 . Transposon mutagenesis of C . crescentus NA1000 was done by intergeneric conjugation from E . coli S17-1 λpir harbouring the himar1-derivative pHPV414 ( Viollier et al . , 2004 ) . Pools of Tn mutants of >100 , 000 kanamycin and nalidixic acid resistant clones were obtained for NA1000 grown either in the presence or absence of bacteriophage φCr30 . To enrich the Tn pool for φCr30-resistant clones , after conjugation NA1000 cells were embedded into soft agar containing the appropriate antibiotics as well as φCr30 ( at m . o . i . ≥2 ) . After incubation of the plates at 30°C for 48 hr , all the clones were pooled for each Tn collection and chromosomal DNA was extracted . Sequencing ( Illumina HiSeq 2000 ) and analysis were done as described previously ( Murray et al . , 2013 ) . For the production of antibodies , His6-SUMO-CCNA_00162 ( 51-422 ) , CCNA_00163 ( 101-300 ) -His6 , His6-SUMO-CCNA_00164 ( 481-620 ) , His6-HvyA ( 26-272 ) , His6-CCNA_00167 ( 1-108 ) , His6-SUMO-CCNA_00168 ( 41-198 ) , and CCNA_02223 ( 22-289 ) -His6 were expressed in E . coli Rosetta ( DE3 ) pLysS cells and the recombinant proteins were purified using Ni-NTA agarose ( Qiagen , Hilden , Germany ) . His6-SUMO-CCNA_00168 ( 41-198 ) and CCNA_02223 ( 22-289 ) -His6 were purified in the soluble fraction and directly used to immunize rabbits ( Josman LLC , Napa , CA ) . Purified His6-SUMO-CCNA_00162 ( 51-422 ) , CCNA_00163 ( 101-300 ) -His6 , His6-SUMO-CCNA_00164 ( 481-620 ) , His6-HvyA ( 26-272 ) , and His6-CCNA_00167 ( 1-108 ) were excised from 12 . 5% SDS polyacrylamide gels and used to immunize rabbits . For immunoblots , protein samples were separated on SDS polyacrylamide gel , transferred to polyvinylidene difluoride ( PVDF ) Immobilon-P membranes ( Merck Millipore ) , and blocked in PBS ( phosphate saline buffer ) , 0 . 1% Tween20 , and 5% dry milk . The anti-sera were used at the following dilutions: anti-CtrA ( 1:10 , 000 ) ( Domian et al . , 1997 ) , anti-PilA ( 1:10 . 000 ) ( Viollier et al . , 2002 ) , anti-CcrM ( 1:10 , 000 ) ( Stephens et al . , 1996 ) , anti-mCherry ( 1:10 , 000 ) ( Chen et al . , 2005 ) , anti-CCNA_00162 ( 1:20 , 000 ) , anti-CCNA_00163 ( 1:100 , 000 ) , anti-CCNA_00164 ( 1:10 , 000 ) , anti-HvyA ( 1:10 , 000 ) , anti-CCNA_00167 ( 1:10 , 000 ) , anti-CCNA_00168 ( 1:20 , 000 ) , anti-β-lactamase ( anti-Bla , 1:20 , 000 ) . Protein-primary antibody complexes were visualized using horseradish peroxidase-labelled anti-rabbit antibodies and ECL detection reagents ( Merck Millipore ) . Cultures ( 8 ml ) were grown to exponential phase ( OD600nm ∼ 0 . 6 ) , centrifuged and washed twice with HEPES 10 mM pH 7 . 2 . Cells were re-suspended in HEPES 10 mM pH 7 . 5 containing 10 mM EGTA ( ethylene glycol tetraacetic acid ) and incubated at room temperature for 10 min . Cells were then pelleted by centrifugation and 20 μl of supernatant loaded on a 7 . 5% SDS polyacrylamide gel , followed by Coomassie Blue staining . C . crescentus cells expressing mCh-HvyA from the hvyA locus on the chromosome were grown in PYE to exponential phase ( OD600nm ∼ 0 . 6 ) before adding chloramphenicol ( 2 μg/ml ) . CtrA or mCh-HvyA levels were monitored by immunoblotting of samples taken at different time points after addition of chloramphenicol . C . crescentus cells ( WT , ΔhvyA , or ΔCCNA_00163 ) were grown in PYE to exponential phase ( OD600nm ∼ 0 . 6 ) , pelleted by centrifugation , and re-suspended in 20 mM Tris , pH 7 . 5 , 100 mM NaCl . The susceptibility of surface proteins to proteolysis was determined by treating whole cells with 0 . 5 mg/ml proteinase K; after incubation at 37°C ( 15 , 30 , 45 , or 60 min ) , 1× protease inhibitors ( Complete EDTA-free , Roche , Switzerland ) were added . The cells were washed four times with 20 mM Tris ( pH 7 . 5 ) , 100 mM NaCl , 1× protease inhibitors , re-suspended in SDS-PAGE loading buffer and boiled . Protein samples were analysed by immunoblotting using antibodies to CCNA_00168 . In order to create hvyA alleles mutated in the predicted catalytic residues , the hvyA ORF was sub-cloned into pOK12 ( Vieira and Messing , 1991 ) as NdeI/EcoRI fragment from pMT335-hvyA . pOK-hvyA was used as a template for oligonucleotide site-directed mutagenesis . Two complementary oligonucleotide primers containing the desired mutation were designed for each point mutation ( Supplementary file 5 ) . PCR reactions were composed of 30 cycles , carried out under the following conditions: denaturation , 94°C for 1 min; annealing , 60°C for 1 min; extension , 68°C for 8 min . The PCR products were treated with DpnI to digest the template DNA and used to transform E . coli EC100D competent cells . The constructions obtained were verified by sequencing and sub-cloned as NdeI/EcoRI fragments into pMT335 . In order to create TAP-tagged versions of the point mutants , the hvyA mutant alleles were amplified from pOK12 with hvyA_N and hvyA_CTIF primers and cloned into a pMT335 derivative harbouring the TAP epitope cloned as EcoRI/XbaI fragment ( Radhakrishnan et al . , 2010 ) . The HvyA loss-of-function alleles ( R161P , W178S , D194G , H226Y , L240R , and P263R ) were obtained following random mutagenesis of pUG52 , which was passed through the E . coli mutator strain XL1-Red . The mutant pUG52 library was electroporated into WT NA1000 or the ΔhvyA strain . The transformants were pooled and subjected to centrifugation on density gradient , in order to isolate ‘light’ cells . These ‘light’ cells were grown in liquid medium and then subjected to three subsequent rounds of isolation by centrifugation on density gradient to obtain enrichment in ‘light’ cells . Cells isolated in the last density gradient were plated to obtain single colonies , from which we recovered the plasmid . Nine plasmids were sequenced and we identified six different mutations: clones L006 and L011 encoded the W178S variant , L009 encoded D194G , L101 encoded P263R , L102 and L110 encoded L240R , L104 and L111 encoded R161P , and L105 encoded H226Y . β-galactosidase assays were performed at 30°C . Cells ( 50–200 μl ) at OD660nm = 0 . 1–0 . 5 were lysed with chloroform and mixed with Z buffer ( 60 mM Na2HPO4 , 40 mM NaH2PO4 , 10 mM KCl , and 1 mM MgSO4 , pH 7 ) to a final volume of 800 μl . 200 μl of ONPG ( o-nitrophenyl-β-D-galactopyranoside , stock solution 4 mg/ml in 0 . 1 M potassium phosphate , pH 7 ) were added and the reaction timed . When a medium-yellow colour developed , the reaction was stopped by adding 400 μl of 1M Na2CO3 . The OD420nm of the supernatant was determined and the Miller units ( U ) were calculated as follows: U = ( OD420nm * 1000 ) / ( OD660nm * time [in min] * volume of culture used [in ml] ) . Error was computed as standard deviation ( SD ) . C . crescentus cells grown in 2L PYE were pelleted by centrifugation , washed twice with phosphate saline buffer ( PBS , pH 7 . 5 ) , and lyophilized . Lyophilized cells were used for the purification of capsular polysaccharides , which was performed using a modification of the method described by Ravenscroft et al . ( 1991 ) followed by glycosyl compositional analysis conducted by combined gas chromatography/mass spectrometry ( GC/MS ) of the per-O-trimethylsilyl ( TMS ) derivatives of the monosaccharide methyl glycosides produced by acidic methanolysis . Briefly , after treatment of cellular lysates with 95% ethanol for polysaccharide enrichment , contaminants such as DNA , RNA , and proteins were removed by successive digestion with DNase I , RNase A , and proteinase K . Every enzymatic digestion step was followed by dialysis against distilled deionized water . Samples were then subjected to ultracentrifugation ( 100 , 000×g , 18 hr , 4°C ) to pellet lipopolysaccharide ( LPS ) . The supernatant containing capsular polysaccharides was freeze-dried and used for glycosyl composition analysis . Inositol ( 20 μg ) was added , as internal standard , to 500 μg of each sample . Polysaccharides were first hydrolysed with 2 M trifluoroacetic acid ( TFA ) at 120°C for 2 hr . Methyl glycosides were prepared from the dry samples by mild acid treatment ( methanolysis in 1 M HCl in methanol at 80°C for 16 hr ) followed by re-acetylation with pyridine and acetic anhydride in methanol ( for detection of amino-sugars ) . The samples were then per-O-trimethylsilylated by treatment with Tri-Sil reagent ( Thermo Scientific Pierce , Rockford , IL ) at 80°C for 30 min ( York et al . , 1985; Merkle and Poppe , 1994 ) . Gas chromatography/mass spectrometry ( GC/MS ) analysis of the TMS methyl glycosides was performed on an Agilent 7890A GC interfaced to a 5975C MSD , using an Agilent DB-1 fused silica capillary column ( 30 mm × 0 . 25 mm ID ) . Fluorescence and DIC imaging of Caulobacter cells were conducted as previously described ( Radhakrishnan et al . , 2008 ) . FITC-dextran with average mass of 2000 kDa ( Sigma-Aldrich , St . Louis , MO ) was used to assess capsule thickness as previously described ( Gates et al . , 2004 ) . Briefly , cells were grown in PYE supplemented with 1% sucrose to a final OD600nm = 1 . 2 . 500 μl of each culture were mixed , harvested by centrifugation at room temperature ( 3000×g , 5 min ) , washed once with PBS , and resuspended in 30 μl of PBS . 10 μl of bacterial suspension was mixed with 2 μl of FITC-dextran ( 10 mg/ml in water ) , applied onto a microscope slide , and firmly covered with a coverslip . Cells expressing SpmX-mCherry were grown in PYE supplemented with 1% sucrose to a final OD600nm = 0 . 6 . SW and ST/PD cells were separated by centrifugation on density gradient , then washed with PBS , and incubated with FITC-dextran as described above . The samples were imaged as described for fluorescence and DIC images ( Radhakrishnan et al . , 2008 ) . Images were analyzed with the MATLAB-based open-source software MicrobeTracker ( Sliusarenko et al . , 2011 ) . Statistics were calculated using Graphpad Prism 4 and statistical significance was determined using a two-tailed Mann–Whitney test . Caulobacter cells grown overnight in liquid PYE were rinsed in PBS buffer and resuspended in 4% paraformaldehyde ( Sigma-Aldrich ) solution for 1 hr at room temperature for fixation . Cells were then rinsed in PBS buffer and filtered through polycarbonate porous membrane ( Millipore , Billerica , MA , pore size: 3 µm ) . AFM imaging was performed using a Nanoscope VIII Multimode ( Bruker Corporation , Santa Barbara , CA ) and oxide-sharpened microfabricated Si3N4 cantilevers with a nominal spring constant of ∼0 . 01 N/m ( Microlevers , Veeco Metrology Group ) . After filtering the cell culture , the filter was gently rinsed with the buffer , carefully cut ( 1 cm × 1 cm ) , attached to a steel sample puck using a small piece of double face adhesive tape , and the mounted sample was transferred into the AFM liquid cell while avoiding dewetting . Images were taken in PBS buffer in contact mode under minimal applied force . Images were analysed using Nanoscope 8 . 10 software ( Bruker , Santa Barbara , CA ) . Rms ( root mean square ) roughness values were calculated on 250 × 250 nm2 areas of the high magnification height images subjected to second order filtering . In-frame deletions and replacement of hvyA by a mCherry-hvyA N-terminal fusion ( strain SA1737 ) were created using pNPTS138 derivatives constructed as follows:pNPTS_Δ00162: PCR was used to amplify two DNA fragments flanking the CCNA_00162 ORF , by using primers 162_ko1/162_ko2 and 162_ko3/162_ko4 . The PCR fragments were digested with HindIII/BamHI and BamHI/EcoRI , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00163: PCR was used to amplify two DNA fragments flanking the CCNA_00163 ORF , by using primers 163_ko1/163_ko2 and 163_ko3/163_ko4 . The PCR fragments were digested with HindIII/BamHI and BamHI/EcoRI , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00164: PCR was used to amplify two DNA fragments flanking the CCNA_00164 ORF , by using primers 164_ko1/164_ko2 and 164_ko3/164_ko4 . The PCR fragments were digested with HindIII/BamHI and BamHI/EcoRI , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_ΔhvyA: PCR was used to amplify two DNA fragments flanking the hvyA ORF , by using primers hvyA_ko1/hvyA_ko2 and hvyA_ko3/hvyA_ko4 . The PCR fragments were digested with EcoRI/BamHI and BamHI/HindIII , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_PhvyA-mCh::hvyA: primers hvyA_up_H and hvyA_up_B were used to amplify a 921-bp fragment encompassing the region upstream of hvyA and the first 78 bp of the hvyA ORF ( encoding the signal sequence ) . Primers mCh_B and mCh_X were used to amplify the mCherry coding sequence ( without ATG and stop codon ) . Primers hvyA_down_X and hvyA_down_E were used to amplify a 719-bp fragment of hvyA ORF . The three PCR fragments were digested with HindIII/BamHI , BamHI/XbaI , and XbaI/EcoRI , respectively , and ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00167: PCR was used to amplify two DNA fragments flanking the CCNA_00167 ORF , by using primers 167_ko1/167_ko2 and 167_ko3/167_ko4 . The PCR fragments were digested with EcoRI/BamHI and BamHI/HindIII , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00167 ( ΔhvyA ) : in order to obtain an in-frame deletion of CCNA_00167 in the ΔhvyA background , primers 167_ko5/167_ko6 were used to amplify by PCR a 1085-bp fragment upstream of CCNA_00167 ( using genomic DNA of the ΔhvyA strain as template ) . The PCR fragment was digested with EcoRI/BglII and ligated into pNPTS_Δ00167 , restricted with BamHI and EcoRI . pNPTS_Δ03998: PCR was used to amplify two DNA fragments flanking the CCNA_03998 ORF , by using primers 3998_ko1/3998_ko2 and 3998_ko3/3998_ko4 . The PCR fragments were digested with EcoRI/BamHI and BamHI/HindIII , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00466: PCR was used to amplify two DNA fragments flanking the CCNA_00466 ORF , by using primers 466_ko1/466_ko2 and 466_ko3/466_ko4 . The PCR fragments were digested with MunI/BamHI and BamHI/HindIII , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00467: PCR was used to amplify two DNA fragments flanking the CCNA_00467 ORF , by using primers 467_ko1/467_ko2 and 467_ko3/467_ko4 . The PCR fragments were digested with HindIII/BamHI and BamHI/EcoRI , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . pNPTS_Δ00470: PCR was used to amplify two DNA fragments flanking the CCNA_00470 ORF , by using primers 470_ko1/470_ko2 and 470_ko3/470_ko4 . The PCR fragments were digested with EcoRI/BamHI and BamHI/HindIII , respectively , then ligated into pNPTS138 , restricted with HindIII and EcoRI . Bi-parental matings were used to transfer the resulting pNPTS138 derivatives into C . crescentus strains . Double recombination was selected by plating bacteria onto PYE plates containing 3% sucrose . Putative mutants were confirmed by PCR using primers external to the DNA fragments used for the pNPTS138 constructs . To inactivate the rsaA gene in the ΔhvyA , ΔCCNA_00163 , or ΔhvyA ΔCCNA_00163 mutant strains , plasmid pNPTS138_ΔrsaA was introduced into the strains by bi-parental mating . Clones that had undergone a single recombination event were selected on PYE plates containing kanamycin and verified by PCR . To created strain SA1984 ( ΔmucR1ΔmucR2 with hvyA replaced by N-terminal mCherry-tagged hvyA ) , a 719-bp fragment was amplified by PCR using primers hvyA_in_B and hvyA_E . The PCR fragment was digested with BamHI/EcoRI and ligated into pGS18T , restricted with the same enzymes . The resulting plasmid ( pSA480 ) was integrated into the hvyA locus in SA1737 ( strain SA1951 ) , and the mCh-hvyA fusion was transduced into the ΔmucR1ΔmucR2 strain by φCr30-mediated transduction and selection on PYE kanamycin plates . To create the PhvyA-hvyA::lacZ translational fusion ( pSA184 ) , a 531-bp DNA fragment , encompassing 513-bp upstream of hvyA and the first six codon of hvyA ORF , was amplified by PCR with primers PhvyA_B/PhvyA_P , digested with BglII and PstI , and ligated into pJC327 ( Chen et al . , 2006 ) , restricted with the same enzymes . To create the PhvyA-lacZ transcriptional fusion ( pSA205 ) , the fragment corresponding to the hvyA promoter region was excised from pSA184 with BglII and PstI , and ligated into pRKlac290 ( Gober and Shapiro , 1992 ) , cut with BamHI and PstI . To create the PSMc00998-lacZ transcriptional fusion ( pSA146 ) , a 542-bp DNA fragment was amplified by PCR with primers Sm998_B/Sm998_P from S . meliloti genomic DNA , digested with BamHI and PstI , and ligated into pRKlac290 , cut with the same enzymes . Plasmids for β-galactosidase assays were introduced into S . meliloti Rm2011 and Rm101 by bi-parental mating . To complement the ΔhvyA mutation , the hvyA ORF was amplified by PCR with primers hvyA_N and hvyA_E . The resulting PCR product was digested with NdeI and EcoRI and ligated into pMT335 ( Pvan , medium copy plasmid; pMT335-hvyA ) or pMT375 ( Thanbichler et al . , 2007 ) ( Pxyl , low copy plasmid; pMT375-hvyA ) , restricted with the same enzymes . To create the Pvan-hvyA-TAP fusion ( plasmid pUG52 ) , the hvyA ORF was amplified by PCR with primers hvyA_N and hvyA_CTIF ( without stop codon ) . The PCR fragment was digested with NdeI and EcoRI and cloned into a pMT335 derivative harbouring the TAP epitope cloned as EcoRI/XbaI fragment ( Radhakrishnan et al . , 2010 ) . Plasmids to complement the in-frame deletion mutants and for over-expression ( from Pvan ) were constructed as follows:pSA362: CCNA_00162 ORF was amplified by PCR with primers 162_N ( with NdeI site overlapping the start codon ) and 162_M ( with MunI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA361: CCNA_00163 ORF was amplified by PCR with primers 163_N ( with NdeI site overlapping the start codon ) and 163_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA401: CCNA_00164 ORF was amplified by PCR with primers 164_N ( with NdeI site overlapping the start codon ) and 164_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA62: CCNA_00167 ORF was amplified by PCR with primers 167_N ( with NdeI site overlapping the start codon ) and 167_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA324: CCNA_00168 ORF was amplified by PCR with primers 168_N ( with NdeI site overlapping the start codon ) and 168_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pUG35: CCNA_03998 ORF was amplified by PCR with primers 3998_N ( with NdeI site overlapping the start codon ) and 3998_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pUG28: CCNA_00466 ORF was amplified by PCR with primers 466_N ( with NdeI site overlapping the start codon ) and 466_M ( with MunI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA102: CCNA_00470 ORF was amplified by PCR with primers 470_N ( with NdeI site overlapping the start codon ) and 470_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA264: SMc00998 ORF was amplified by PCR from S . meliloti genomic DNA with primers Sm998_N ( with NdeI site overlapping the start codon ) and Sm998_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA142: NGR_c12490 ORF was amplified by PCR from S . fredii NGR234 genomic DNA with primers Sf12490_N ( with NdeI site overlapping the start codon ) and Sf12490_M ( with MunI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA141: NGR_c19800 ORF was amplified by PCR from S . fredii NGR234 genomic DNA with primers Sf19800_N ( with NdeI site overlapping the start codon ) and Sf19800_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA147: NGR_c36180 ORF was amplified by PCR from S . fredii NGR234 genomic DNA with primers Sf36180_N ( with NdeI site overlapping the start codon ) and Sf36180_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pSA309: Atu0252 ORF was amplified by PCR from A . tumefaciens genomic DNA with primers At252_N ( with NdeI site overlapping the start codon ) and At252_E ( with EcoRI site flanking the stop codon ) and cloned into pMT335 , restricted with NdeI and EcoRI . pMT335-Bh_MucR: a synthetic fragment encoding the mucR homolog from Bartonella henselae ( PRJBM_00467 ) was ligated into pMT335 ( using NdeI/EcoRI ) . Synthetic DNA fragment ( Integrated DNA Technologies ) encoding PRJBM_00467 ( codon optimized for C . crescentus ) ( 5′–3′ ) : CATATGGAGCACCGACCGGTGCTGGAAACCGAGTCGAATCTGGTCATCACCCTCGTCGCCGACATCGTCGCCGCGTATGTGTCGAACAACTCCATCCGTCCCACCGAGGTCCCCAGCCTCATCGCGGACGTCCATGCGGCTTTCCGCAAGGCCGGCAACGCCGACTTGACGGAAGTTGAGGTGGAGAAGCAGCGCCCTGCGGTCAACCCGAAGCGCAGCATCTTCCCGGACTACCTTATCTGCCTGGAAGATGGCAAGAAGTTCAAGAGCCTGAAGCGCCACCTGATGACGCACTATGGCATGCTGCCGGAAGAGTATCGCGAGAAGTGGCAGCTGGACTCTTCGTACCCCATGGTGGCCCCGAACTACGCGAAGGCCCGGTCGGCCCTGGCCAAAGAGATGGGCCTGGGGCGGAAGTCCAAGCGGAAAAAGACCAAGTGAATTCPlasmid pSA354 is a derivative of pCWR547 ( Radhakrishnan et al . , 2010 ) expressing His6-SUMO-CCNA_00162 ( 51-422 ) under the control of the T7 promoter . To construct pSA354 , a fragment encoding residues 51-422 of CCNA_00162 was amplified by PCR with primers 162_in_N and 162_in_S , digested with NdeI and SacI , and cloned into pCWR547 , restricted with the same enzymes . Plasmid pCWR508 is a derivative of pET-47b ( Novagen ) expressing CCNA_00163 ( 101-300 ) -His6 under the control of the T7 promoter . To construct pCWR508 , a fragment encoding residues 101-300 of CCNA_00163 was amplified by PCR with primers 163_in_N and 163_His_E ( that also encodes six His residues followed by a stop codon and EcoRI site ) , digested with NdeI and EcoRI , and cloned into pET47b , restricted with the same enzymes . Plasmid pSA352 is a derivative of pCWR547 expressing His6-SUMO-CCNA_00164 ( 481-620 ) under control of the T7 promoter . To construct pSA352 , a fragment encoding residues 481-620 of CCNA_00164 was amplified by PCR with primers 164_in_N and 164_in_S , digested with NdeI and SacI , and cloned into pCWR547 , restricted with the same enzymes . Plasmid pET-hvyA is a derivative of pET-28a ( Novagen ) expressing His6-HvyA ( 26-272 ) under the control of the T7 promoter . To construct pET-hvyA , a fragment encoding residues 26-272 of HvyA was amplified by PCR with primers hvyA_short and hvyA_E , digested with NdeI and EcoRI and cloned into pET-28 , restricted with the same enzymes . Plasmid pET-00167 is a derivative of pET-28a expressing His6-CCNA_00167 ( 1-108 ) under the control of the T7 promoter . To construct pET-00167 , a fragment encoding residues 1-208 of CCNA_00167 was amplified by PCR with primers 167_N and 167_in_E , digested with NdeI and EcoRI , and cloned into pET-28 , restricted with the same enzymes . Plasmid pSA342 is a derivative of pCWR547 expressing His6-SUMO-CCNA_00168 ( 41-198 ) under the control of the T7 promoter . To construct pSA342 , a fragment encoding residues 41-198 of CCNA_00168 was amplified by PCR with primers 168_short and 168_E , digested with NdeI and EcoRI , and cloned into pET-28 . The CCNA_00168 fragment was then sub-cloned into pCWR547 using NdeI/SacI . Plasmid pCWR496 is a derivative of pET-47b expressing CCNA_02223 ( 22-289 ) -His6 under control of the T7 promoter . To construct pCWR496 , a fragment encoding residues 22-289 of CCNA_02223 ( β-lactamase ) was amplified by PCR with primers bla_N and bla_His_E ( that also encodes six His residues followed by a stop codon and EcoRI site ) , digested with NdeI and EcoRI , and cloned into pET47b , restricted with the same enzymes .	Many bacteria have a tough outer coating known as capsule that protects them from untoward environmental conditions . This capsule also prevents viruses called bacteriophages from invading the bacterial cells , and it shields those bacteria that can infect humans from attack by our immune system . External conditions—such as a lack of nutrients and physical stresses—are known to trigger capsule formation . However , almost nothing is known about the signals from within the bacteria that control the formation of a capsule . Now , Ardissone et al . have used the capsulated bacterium called Caulobacter crescentus to show that capsule formation is regulated by the bacterial cell cycle . This cycle is a series of events and checkpoints that happen every time a cell divides to form two new cells . Ardissone et al . revealed that capsule cannot form during the first phase of the cell cycle . The bacterium only forms its capsule as this phase ends and before it copies its DNA and later divides in two . Ardissone et al . discovered that an enzyme called HvyA , which is only produced during the first phase of the cell cycle , prevents the capsule from forming . Inactivating the HvyA enzyme was also shown to make the bacteria impervious to infection by a bacteriophage . Furthermore , Ardissone et al . dissected the complicated steps involved in regulating the production of the HvyA enzyme and showed that such regulatory steps are also used by other species of bacteria . Without their capsules , bacteria can take up new genetic material from a number of sources that might help them adapt to a changing environment . Ardissone et al . 's findings suggest that by only exchanging genetic material during the first phase of the cell cycle , bacteria ensure that any useful DNA is taken up and copied along with their own DNA later in the cell cycle . Antibiotic resistance spreads between bacteria via the exchange of genetic material , making it increasingly difficult to treat bacterial infections . Interfering with the formation of the capsule during an infection could help overcome this problem by making the bacteria more vulnerable to attack either by our own immune system or by bacteriophages that can be used to treat bacterial infections . By investigating how genetic exchange and capsule formation are linked and regulated , the findings of Ardissone et al . might now open up new strategies to help combat bacterial infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2014	Cell cycle constraints on capsulation and bacteriophage susceptibility
How cells sense their mechanical environment and transduce forces into biochemical signals is a crucial yet unresolved question in mechanobiology . Platelets use receptor glycoprotein Ib ( GPIb ) , specifically its α subunit ( GPIbα ) , to signal as they tether and translocate on von Willebrand factor ( VWF ) of injured arterial surfaces against blood flow . Force elicits catch bonds to slow VWF–GPIbα dissociation and unfolds the GPIbα leucine-rich repeat domain ( LRRD ) and juxtamembrane mechanosensitive domain ( MSD ) . How these mechanical processes trigger biochemical signals remains unknown . Here we analyze these extracellular events and the resulting intracellular Ca2+ on a single platelet in real time , revealing that LRRD unfolding intensifies Ca2+ signal whereas MSD unfolding affects the type of Ca2+ signal . Therefore , LRRD and MSD are analog and digital force transducers , respectively . The >30 nm macroglycopeptide separating the two domains transmits force on the VWF–GPIbα bond ( whose lifetime is prolonged by LRRD unfolding ) to the MSD to enhance its unfolding , resulting in unfolding cooperativity at an optimal force . These elements may provide design principles for a generic mechanosensory protein machine . Platelets can serve as a natural model system for studying cell mechanosensing as they rapidly respond to changes in hydrodynamic forces and substrate stiffness due to vascular pathology ( Jackson , 2011; Qiu et al . , 2015 ) . Previous studies have suggested the role of GPIbα as a mechanoreceptor , for force exerted on it via its ligand VWF induces platelet signaling ( Ruggeri , 2015 ) . Conceptually , this coupled mechanical-biochemical process ( mechanosensing ) can be broken down into four steps: 1 ) Mechanopresentation: the receptor binding domain A1 is exposed by structural changes in VWF induced by elongational flow and collagen immobilization ( Ju et al . , 2015a; Springer , 2014 ) ; 2 ) Mechanoreception: GPIbα LRRD receives the force signal via engaging VWF-A1 to tether the platelet against shear stress; 3 ) Mechanotransmission: force is propagated from the LRRD through the mucin-like macroglycopeptide ( MP ) stalk ( cf . Figure 2A ) ( Fox et al . , 1988 ) and the MSD across the membrane to adaptor and signaling molecules ( e . g . 14-3-3ζ ) inside the platelet ( cf . Figure 7G ) ; and 4 ) Mechanotransduction: force induces mechano-chemical changes to convert mechanical cues to biochemical signals . Some of these steps have been characterized separately . For example , GPIbα forms catch-slip bonds with wild-type ( WT ) A1 in >15 pN , such that the bond lifetime first increases with force , reaches a maximum at ~25 pN , and decreases thereafter; whereas it forms slip-only bonds with type 2B von Willebrand disease ( VWD ) mutant ( e . g . A1R1450E ) , such that the bond lifetime decreases monotonically with force ( Ju et al . , 2013; Yago et al . , 2008 ) . As another example , force induces unfolding of the LRRD , which prolongs A1–GPIbα bond lifetime ( Ju et al . , 2015b ) , and of the MSD , which is hypothesized to play a role in platelet signaling ( Zhang et al . , 2015 ) . However , how these inter-connected steps are orchestrated to enable the information encoded by force to be translated into biochemical signals is still poorly understood . We used a biomembrane force probe ( BFP ) to recapitulate the above process in a single-cell and single-molecular bond level to address the following questions: 1 ) What molecular events would be induced in GPIbα and how these events are regulated mechanically ? 2 ) Whether , and if so , how changes in presentation of force by VWF-A1 mutation would affect the force reception by GPIbα and its response to force ? 3 ) What features of the force ( waveforms ) could be sensed by the platelet via GPIbα to initiate intraplatelet calcium fluxes ? 4 ) What proximal events may be responsible for transducing force into a biochemical signal ? By manipulating the mechanopresentation and mechanoreception steps then analyzing the resulting mechanotransmission and mechanotransduction steps , we gained insights into the inner workings of this GPIbα-mediated mechanosensory machine . Using an optical trap , Zhang et al . observed force-induced MSD unfolding in purified recombinant full-length GPIb-IX and a GPIbα stalk region construct ( Zhang et al . , 2015 ) . Using a BFP , we observed LRRD unfolding in glycocalicin ( GC ) ( Ju et al . , 2015b ) , the extracellular segment of GPIbα lacking the MSD ( Liang et al . , 2013 ) ( Figure 2A–C ) . Here we pulled GPIbα on platelets via A1 and observed two unfolding signatures , one in the ramping and the other in the clamping phases of the force trace ( Figure 1E ) . Unfolding that occurred in the ramping phase is termed ramped unfolding , which is featured by a sudden force kink at 5–20 pN as observed in previous studies of GPIbα unfolding ( Ju et al . , 2015b; Zhang et al . , 2015 ) . Similar to findings of protein unfolding studies ( Kellermayer et al . , 1997; Rief et al . , 1997; Tskhovrebova et al . , 1997; Zhang et al . , 2009a , 2015 ) , both the force-extension curves before and after unfolding were well fitted by the worm-like chain ( WLC ) model ( Figure 1—figure supplement 2 ) . Unfolding that occurred in the clamping phase is termed clamped unfolding , which is featured by an abrupt force drop ( Figure 1F ) . Although not observed in the previous studies of GPIbα unfolding ( Ju et al . , 2015b; Zhang et al . , 2015 ) , this feature has been described in protein unfolding studies using force-clamp experiments ( Oberhauser et al . , 2001; Tskhovrebova et al . , 1997 ) . Unfolding lengths derived from both signatures were measured from the probe bead position vs . time data ( Figure 1E insert , 1F and Figure 1—figure supplement 2 ) . The lengths of individual ramped unfolding events distributed tri-modally with three subpopulations ( Figure 2D and Figure 2—figure supplement 1; Materials and methods ) . The first subpopulation coincides with the ramped unfolding length distribution from WM23 vs . platelet experiments ( Figure 2E , white bars ) . WM23 binds the MP region below the LRRD ( Figure 2A ) , hence could unfold MSD only . The average unfolding force vs . length data from the WM23 experiment was well fitted by the WLC model , yielding a contour length of 25 . 99 ± 0 . 85 nm ( Figure 2H ) that matches the previously reported MSD contour length ( Zhang et al . , 2015 ) . The average unfolding force vs . length data from the A1 experiment overlaid well on the same WLC model fit ( Figure 2H ) . These results identify the first subpopulation in Figure 2D as MSD unfolding . The second subpopulation in Figure 2D matches the histogram of ramped unfolding lengths of GC pulled via A1 ( Figure 2E , blue bars ) that ranges from 18–56 nm and peaks at 36 nm ( length of leucine-rich repeats 3–6 ) . The average unfolding force vs . length plots derived from the A1 vs . platelet and A1 vs . GC experiments overlaid well on the same WLC model fit ( Figure 2I ) . The best-fit contour length ( 70 . 29 ± 3 . 56 nm ) matches the length of LRRD , calculated using a 4-Å contour length per residue ( Ju et al . , 2015b ) . These results identify the second subpopulation in Figure 2D as LRRD unfolding . The third subpopulation in Figure 2D can be identified as concurrent unfolding of both MSD and LRRD that occurred within too short a time elapse to be distinguished by our BFP as two separate events , because its maximum unfolding length ( 85 nm ) matches the sum of the observed maximum MSD and LRRD unfolding lengths . Similar tri-modally distributed ramped unfolding lengths were obtained by using mAb AN51 ( epitope mapped to the N-terminal flanking region above LRRD , cf . Figure 2A ) instead of A1 to pull the platelet GPIbα ( Figure 2F ) , and the second subpopulation also matches the ramped unfolding length distribution obtained using AN51 to pull GC ( Figure 2G , blue bars ) . These results are expected because the unfolding lengths are determined by the respective primary structures of the LRRD and MSD , and as such should not depend on the 'grabbing handle' used to pull GPIbα . The consistence of the A1 and AN51 results imparts confidence in our identification of the three subpopulations as unfolding of MSD , LRRD , and both , respectively . Interestingly , the two force waveforms induced unfolding of different GPIbα domains . Clamped forces unfolded only MSD as the lengths of clamped unfolding distribute as a single peak at 20 nm ( Figure 2E , G , red bars ) , matching the first subpopulation in Figure 2D , F , respectively , regardless of whether platelet GPIbα was engaged by A1 or AN51 . Furthermore , unfolding of LRRD in GC was induced only by ramped forces but not clamped forces ( Figure 2J , K ) . By comparison , pulling platelet GPIbα via WM23 with both ramped and clamped forces induced MSD unfolding events with similar occurrence frequencies and unfolding lengths ( Figure 2J , K ) . These results indicate that MSD can be unfolded by increasing forces as well as constant forces . By comparison , LRRD unfolding requires increasing forces . Some force-clamp cycles ( Figure 1C; Video 2 ) generated two consecutive unfolding events , one in the ramping and the other in the clamping phase ( Figure 1E ) . The respective unfolding lengths of the ramped and clamped unfolding events were 34–55 nm and 13–25 nm that totaled 47–80 nm , agreeing with those of the LRRD , MSD , and MSD+LRRD subpopulations in Figure 2D , F . Together , these results provide criteria to determine whether and which GPIbα domain ( s ) is unfolded ( Figure 3—source data 1A ) . 10 . 7554/eLife . 15447 . 009Video 2 . Force-clamp experiment mode with a bond lifetime event . The video consists of two parts in series . Part I is an animation ( produced by Adobe Flash; 12 fps ) , and part II is a video recording of a representative fluorescence BFP experiment ( recorded by a customized LabView program; 25 fps ) . Both parts show BFP force-clamp measurement cycles . In part I , the synchronized BFP illustration ( upper panel ) , A1–GPIbα interaction ( middle panel ) and 'Force vs . Time' signal ( lower panel ) of the same force-clamp cycle with a lifetime event are displayed in parallel . Phases of the BFP cycle are indicated in the lower panel . Part II shows two BFP cycles , which sequentially render a no bond event and a bond lifetime event . The pseudo-color epifluorescence images ( acquired at 1 fps ) are interpolated and superimposed onto the brightfield images to reflect the real-time intraplatelet Ca2+ level ( in a progressive sequence: blue , green , yellow , orange and red ) . Following the long lifetime event , calcium first rapidly elevates and then quickly decays , manifesting an α-type Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 009 To characterize the mechanical response of GPIbα , we measured the frequency , force and length of LRRD and MSD unfolding induced by a range of clamped forces exerted on platelet GPIbα or GC by A1WT or a type 2B VWD mutant A1R1450E . The ramped unfolding frequencies of both domains were extremely low at ≤10 pN but increased with the higher levels of clamped forces ( Figure 3A , B ) . Interestingly , LRRD , but not MSD , unfolded more frequently when platelet GPIbα ( Figure 3A ) and GC ( Figure 3—figure supplement 1 ) were pulled by A1WT than A1R1450E . The ramped unfolding forces of both domains increased with the clamped force and were indifferent to whether force was applied via WT or R1450E mutant of A1 ( Figure 3C , D ) . In general , a higher force was required to unfold LRRD than MSD . Surprisingly , pulling platelet GPIbα via different ligands generated distinctive MSD clamped unfolding frequency vs . force plots: increasing initially and decreasing after reaching maximal at 25 pN when pulled by A1WT , but decreasing monotonically when pulled by A1R1450E ( Figure 3E ) . These data suggest that the mechanoreceptor GPIbα may be able to interpret mechanical cues and discriminate ligands by responding to different force waveforms applied via different ligands with distinct LRRD and MSD unfolding frequencies . In addition , the distinctive force-dependences of two subpopulations of events that we deemed as respective LRRD and MSD unfolding provide further support for our criteria for their identification and classification . 10 . 7554/eLife . 15447 . 010Figure 3 . Force- and ligand-dependent cooperative unfolding of GPIbα LRRD and MSD . ( A–D ) Frequency ( A , B ) and force ( C , D ) of LRRD ( A , C ) or MSD ( B , D ) unfolding events occurred in the ramping phase induced by pulling via A1WT ( blue ) or A1R1450E ( red ) with indicated preset clamped forces . ( E ) Occurrence frequencies of MSD clamped unfolding induced by holding at indicated clamped forces with A1WT or A1R1450E bonds . ( F ) The degree of cooperativity , quantified by ∆P/P = P ( MSD+LRRD ) /[P ( MSD ) ×P ( LRRD ) ] -1 , is plotted vs . clamped force . P ( LRRD ) , P ( MSD ) and P ( LRRD+MSD ) are the observed occurrence frequencies of unfolding events of LRRD alone , MSD alone and LRRD+MSD , respectively . ( G , H ) Significance of cooperativity assessed by ( negative log10 of ) p-value of the χ2 test of the null hypothesis H0: MSD unfolding and LRRD unfolding are independent . The χ2 test was not performed at 10 pN since under this force LRRD unfolding did not occur and hence no unfolding cooperativity . N . D . = not detected ( A , C ) or not done ( F–H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01010 . 7554/eLife . 15447 . 011Figure 3—source data 1 . Statistics and cooperativity evaluation of the GPIbα domains unfolding . ( A ) Decision rules for and statistical summary of GPIbα domain unfolding in force-clamp experiment mode . Criteria for deciding whether or not ( + or − ) and which ( LRRD , MSD , or both ) GPIbα domain ( s ) was ( were ) unfolded are based on BFP profile signatures and the unfolding lengths . YES = observed , NO = not observed . NA = not applicable . ( B ) Related to Figure 3F–H . Evaluation of LRRD and MSD unfolding cooperativity . All probabilities were calculated from occurrence data in ( A ) . Observed joint probabilities were compared to their predicted counterparts based on the assumption that LRRD and MSD unfolded independently . For example , in 'WT A1 vs . Platelet' under 25 pN: The probability of LRRD unfolding is P ( LRRD ) = 3 . 4% + 6 . 9% + 2 . 76% = 13 . 06% . The probability of MSD unfolding is P ( MSD ) = 7 . 6% + 13 . 8% + 6 . 9% + 2 . 76% = 31 . 06% . The probability of MSD ramped unfolding is P ( MSD , ramp ) = 7 . 6% + 2 . 76% = 10 . 36% . The probability of MSD clamped unfolding is P ( MSD , clamp ) = 13 . 8% + 6 . 9% = 20 . 7% . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01110 . 7554/eLife . 15447 . 012Figure 3—figure supplement 1 . GC LRRD unfolding occurrence frequencies . Frequencies of LRRD unfolding events occurred in the ramping phase induced by pulling GC via A1WT ( blue ) or A1R1450E ( red ) with 10 , 25 , 40 and 60 pN preset clamped forces . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 012 The spatial separation of LRRD and MSD by the >30 nm long MP stalk and the distinctive dependences of their unfolding on the force waveform would seem to favor these two GPIbα domains to unfold independently . This hypothesis predicts that the probability for LRRD and MSD to unfold concurrently should be equal to the product of the respective probabilities for LRRD and MSD to unfold separately . To test this hypothesis , we estimated these probabilities from the observed unfolding occurrence frequencies . At 25 pN , the 34 . 5% of BFP force-clamp cycles with unfolding events consist of 7 . 6 , 17 . 2 , 6 . 9 , and 2 . 8% of unfolding of LRRD alone , MSD alone , LRRD and MSD sequentially , and concurrently ( Figure 3—source data 1A ) . Significantly , the frequency of observing both LRRD and MSD unfolding in the same binding cycle , calculated by pooling together both cases of two domains unfolding sequentially and concurrently , P ( MSD+LRRD ) , is much higher than the product of their respective occurrence frequencies , P ( MSD ) ×P ( LRRD ) , which is the joint probability for both to unfold assuming that they were independent ( Figure 3—source data 1B ) . These data suggest that the two GPIbα domains may unfold cooperatively , i . e . , one domain unfolding may increase the likelihood for the other to unfold . To quantify the degree of such cooperativity , we defined a relative probability difference , ∆P/P = [P ( MSD+LRRD ) - P ( MSD ) ×P ( LRRD ) ]/[P ( MSD ) ×P ( LRRD ) ] . ∆P/P > 0 indicates positive cooperativity between LRRD and MSD unfolding . No cooperativity was observed at 10 pN because this force was insufficient to induce appreciable LRRD unfolding . Pulling with A1WT by a 25 pN clamped force generated high cooperativity , and further increase in force decreased cooperativity ( Figure 3F ) . Remarkably , unfolding cooperativity was completely abolished at all forces when applied via the VWD mutant A1R1450E ( Figure 3F ) . We used χ2 test to determine if the hypothesis that MSD and LRRD unfolded independently should be rejected ( Materials and methods ) . At 25 pN , LRRD unfolding significantly enhanced MSD unfolding ( p = 3 . 09 × 10–4 ) . The χ2 test results are depicted as negative log p-values vs . force plots in Figure 3G , H for A1WT and A1R1450E , respectively . Interestingly , significant ( p = 0 . 05 , dashed horizontal lines ) unfolding cooperativity was observed only for A1WT at 25 and 40 pN . These data show that the cooperativity between LRRD and MSD unfolding is force- and ligand-dependent . To elucidate the mechanism underlying the force- and ligand-dependent unfolding cooperativity , we note that when the MSD unfolding events were separately analyzed according to their occurrence in the ramping or clamping phase , MSD clamped , but not ramped , unfolding was significantly ( p= 8 . 79 × 10–3 vs . 0 . 076 at 25 pN ) enhanced by LRRD unfolding ( Figure 3G ) , which occurred in the ramping phase only . This dominance of cooperativity by sequential rather than concurrent unfolding suggests a model for LRRD unfolding to impact MSD unfolding , which includes three ideas . The first idea has to do with the MSD time-to-unfold , tu ( cf . Figure 1F ) . Our force-clamp measurements revealed similar tu values induced by A1WT or A1R1450E pulling ( Figure 4A ) . The only exception is at 10 pN where a shorter tu was induced by A1WT than A1R1450E . This can be explained by their differential bond lifetimes ( Figure 4B , C ) . Compared to A1R1450E , the much shorter lifetime of GPIbα bond with A1WT at 10 pN may underestimate tu because early dissociation of GPIbα would prevent observation of slow MSD unfolding events . This reasoning provides the second idea for our model: MSD clamped unfolding should occur before A1–GPIbα dissociation . The third idea comes from our previous observation ( Ju et al . , 2015b ) that LRRD unfolding significantly prolongs GPIbα bond lifetime with A1WT ( Figure 4B ) but not A1R1450E ( Figure 4C ) . Combining these three ideas , our model proposes that the A1–GPIbα bond lifetime , regulated by force and prolonged by LRRD unfolding in respective ligand-specific manners , determines the occurrence of MSD clamped unfolding , which , despite its ligand-independent unfolding kinetics , generates a cooperativity pattern that maximizes at the optimal force of 25 pN for A1WT but not for A1R1450E . 10 . 7554/eLife . 15447 . 013Figure 4 . LRRD unfolding prolongs A1–GPIbα bond lifetime and facilitates MSD clamped unfolding . ( A–C ) Mean ± s . e . m . of MSD time-to-unfold ( tu , A ) and GPIbα bond lifetimes ( tb , B , C ) with A1WT ( blue ) or A1R1450E ( red ) were measured in the clamping phase at different forces in the absence ( − ) or presence ( + ) of LRRD unfolding in the same BFP cycle . No LRRD unfolding occurred at 10 pN; hence no bond lifetime was measured under the LRRD+ at this force . ( D ) 3D plot of the surface of joint probability density ( z-axis ) of GPIbα to dissociate from A1WT at tb ( x-axis ) and MSD to unfold at tu ( y-axis ) ( Materials and methods ) . Three planes , tu = 1 , 3 , and 5 s , under the probability density surface ( gray ) are shown in green or red , depending on whether they are on the left or right side of the tu = tb plane ( yellow ) . ( E , F ) Measured ( solid bars ) and predicted ( open bars ) frequency of MSD unfolding events occurred in the clamping phase induced by the indicated force exerted via A1WT ( E ) or A1R1450E ( F ) in the presence ( + ) or absence ( − ) of LRRD unfolding in the same BFP cycle . N . D . = not detected . Error bar = s . e . m . estimated by the multinomial distribution of events . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01310 . 7554/eLife . 15447 . 014Figure 4—source data 1 . MSD unfolding rates ( ku ) and the fraction ( w1 ) and off-rates ( k1 , k2 ) of GPIbα dissociating from A1WT or A1R1450E under different forces . w1 represents the fraction of binding events that dissociate with the off-rate k1 . The fraction of events that dissociate with the off-rate of k2 is simply calculated as w2 = 1-w1 . NA = not applicable . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01410 . 7554/eLife . 15447 . 015Figure 4—figure supplement 1 . MSD time-to-unfold distribution for A1WT and 3D probability density surface plot for A1R1450E . ( A ) Time-to-unfold ( tu ) distributions of A1WT MSD clamped unfolding with ( red ) and without ( black ) a preceding LRRD unfolding at 25 pN clamped force . The unfolding rate ku was calculated from the slope of ln ( survival frequency ) vs . time-to-unfold overlaid plot and the error were estimated from the 95% confident interval . ( B ) 3D plot of the surface of joint probability density ( z-axis ) of GPIbα to dissociate from A1R1450E at tb ( x-axis ) and the MSD to unfold at tu ( y-axis ) constructed based on Figure 4A , C ( distributed ensemble data of A1R1450E at 25 pN and without LRRD unfolding ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 015 To formulate the model mathematically , we multiplied the respective probability densities of the exponentially distributed MSD time-to-unfold ( tu ) ( Figure 4—figure supplement 1A ) and the dual-exponentially distributed lifetime ( tb ) of GPIbα bonds with A1WT or A1R1450E ( Ju et al . , 2013 ) to construct a joint probability density surface over the tu-tb plane ( Figure 4D and Figure 4—figure supplement 1B ) . The predicted MSD clamped unfolding probability is the volume under this surface over the region 0<tu<tb<∞ ( Materials and methods ) . When the model was tested against experiment , not only did the calculated force-dependent MSD unfolding frequency match the biphasic pattern for A1WT ( Figure 4E ) and the monophasic pattern for A1R1450E ( Figure 4F ) , but it also compared well with the observed occurrence frequencies numerically at all forces . Remarkably , the model predicts both the quantitative enhancement of MSD unfolding by LRRD unfolding for A1WT and the lack of enhancement for A1R1450E without a single freely adjustable fitting parameter . The excellent agreement between theory and experiment has provided strong support for our model and explained the data in Figure 4E , F . Platelet translocation on VWF signals through GPIbα to induce Ca2+ fluxes ( Mazzucato et al . , 2002; Nesbitt et al . , 2002 ) . We optimized the fluorescence BFP ( fBFP ) method ( Chen et al . , 2015; Liu et al . , 2014 ) for single-platelet calcium imaging and studied how platelet signaling was triggered by GPIbα mechanoreception via a sequence of intermittent single bonds under a range of clamped forces . The Ca2+ signals over the 200-s observation window of repeated platelet contact cycles were classified into three types ( Figure 5A , B ) : i ) null-type , featured by a basal trace with a maximum Ca2+ intensity increase ( normalized by its initial value ) ΔImax<0 . 05; ii ) α-type , featured by an initial latent phase followed by a spike ( mostly ΔImax>0 . 5 ) with a quick decay ( Video 2 ) ; iii ) β-type , featured by fluctuating signals around the baseline or gradually increasing signals to an intermediate level ( mostly ΔImax<0 . 5 ) followed by a gradual decay to baseline ( Video 3 ) . The null type reflects the baseline with background noise , while the α- and β-types match the previous characterization of platelet internal Ca2+release triggered by VWF–GPIbα bonds measured in flow chamber experiments ( Mazzucato et al . , 2002 ) . For each platelet , the calcium trace was overlaid with the sequential binding events , bond lifetimes , and their accumulation over the repeated platelet binding cycles ( Figure 5B and Figure 5—figure supplement 1A , B ) . 10 . 7554/eLife . 15447 . 016Figure 5 . Concurrent analysis of single-platelet Ca2+ flux and GPIb-mediated single-bond binding at 25 pN clamped force . ( A ) Representative epi-fluorescence pseudo-colored images of intraplatelet Ca2+ of null ( top row ) , α- ( middle row ) , and β- ( bottom row ) types at indicated times . ( B ) Representative time courses of normalized Ca2+ intensity of the null ( blue ) , α ( red ) and β ( yellow ) types . The concurrent measurement of bond lifetime events ( symbol ) and the cumulative lifetime ( curve ) for the platelet exhibiting α-type Ca2+ is overlaid . The pre-Ca2+ longest lifetime ( tmax ) and the maximum intensity increase of the α-type Ca2+ ( ΔImax ) are indicated . The time when a concurrent LRRD and MSD unfolding event occurred is indicated by the arrow . ( C , D ) Individual ΔImax values and their mean ± s . e . m . ( points , left ordinate ) and mean ± s . e . m . of tmax ( gray bars , right ordinate ) ( C ) and fractions ( D ) of Ca2+ types triggered by different stimulations . Each point in ( C ) represents results from one platelet and the numbers of platelets in each column are indicated in the corresponding bar in ( D ) , with matched colors to indicate Ca2+ types . ( E , F ) Scattergraphs of ΔImax vs . tmax for A1WT ( E ) and A1R1450E ( F ) . The solid lines are linear fits to respective data with corresponding Pearson coefficients indicated . The null-type Ca2+ data was excluded in the analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01610 . 7554/eLife . 15447 . 017Figure 5—figure supplement 1 . Concurrent analysis of single-platelet Ca2+ flux and single-bond dissociation from GPIb at 25 pN clamped force . ( A , B ) Concurrent measurement of single-platelet Ca2+ flux and kinetics of single GPIb bonds with anti-GPIbα ( A ) and anti-GPIbβ ( B ) . Top: representative epi-fluorescence pseudo-colored images of intracellular Ca2+ in platelets at indicated times . Bottom: Representative time courses of normalized intracellular Ca2+ intensity of the α- ( A ) and β- ( B ) types . The same keys are used as in Figure 5 . ( C , D ) Normalized maximum Ca2+ intensity increase ( ΔImax ) vs . pre-Ca2+ cumulative bond lifetime ∑ti ( C ) or pre-Ca2+ pre-Ca2+ average lifetime <t> ( D ) of platelet GPIbα bonds with A1WT at 25 pN clamped force . Dashed lines are linear fits to data . The Pearson coefficients of the correlation ( r ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01710 . 7554/eLife . 15447 . 018Figure 5—figure supplement 2 . Specificity-sensitivity analysis of optimal threshold . ( A ) Binary classifications of threshold performance outcomes on tmax–ΔImax data . ( B ) The receiver operating characteristic ( ROC ) plot: the true positive rate ( i . e . , sensitivity ) against the false positive rate ( i . e . , 1 - specificity ) at various tmax . The point of the optimal threshold t0 ( tmax = 2 s ) is identified by the black 'x' , achieving the best sensitivity 0 . 7059 and specificity 0 . 8095 . ( C ) Left: Mean ± s . e . m . of pre-Ca2+ MSD time-to-unfold ( tu ) measured from events with or without LRRD unfolding . Right: the tmax of platelets fluxing α- or β-type Ca2+ . Dashed line indicates the tmax threshold that best distinguishes the two Ca2+ types . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 01810 . 7554/eLife . 15447 . 019Video 3 . Force-ramp experiment mode with a bond rupture event . Similar to Video 2 , this video consists of two parts in series . In part I , the synchronized BFP illustration ( upper panel ) , A1–GPIbα molecular interaction ( middle panel ) and 'Force vs . Time' signal ( lower panel ) of the same force-ramp cycle with a ~65 pN rupture force event are displayed in parallel . Part II shows two BFP cycles , which sequentially render a no bond event and a bond rupture event . After the bond rupture event , low level calcium mobilization occurs right away , namely the β-type Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 019 Pulled by a 25 pN clamp force , A1WT–GPIbα bonds triggered much higher ΔImax than controls ( Figure 5C ) , showing 28 , 42 , and 30% of null- , α- and β-types , respectively ( Figure 5D ) . Control experiments that merely held aspirated platelets or contacted them by beads without coating any ligand showed null-type Ca2+ only ( Figure 5C , D ) . The α-type Ca2+ could also be triggered by pulling GPIbα with AN51 but not with an anti-GPIbβ mAb ( Figure 5C , D and Figure 5—figure supplement 1A , B ) , despite that GPIbβ is tightly connected to GPIbα within one GPIb complex and has been postulated to play a role in signaling through GPIb ( Strassel et al . , 2006 ) . These data demonstrated the necessity of GPIbα engagement to trigger intraplatelet Ca2+ and agree with the previous report that α-type Ca2+ peaks occur when platelets are transiently arrested in the whole blood flow ( Mazzucato et al . , 2002;Nesbitt et al . , 2002 ) The concurrent measurements of A1–GPIbα binding kinetics and intraplatelet Ca2+ allowed us to determine the pre-Ca2+ bond lifetimes ( Figure 5B ) , enabling single platelet correlative analysis of binding and signaling . Using the normalized maximum calcium intensity ΔImax to represent the Ca2+ level , we compared its correlations with three statistics of A1WT–GPIbα bond lifetimes occurred prior to calcium onset . We found that ΔImax correlates best with the pre-Ca2+ longest bond lifetime tmax ( Figure 5E ) , similarly well with the pre-Ca2+ cumulative lifetime ∑ti ( Figure 5—figure supplement 1C ) , but poorly with the pre-Ca2+ average lifetime <t> ( Figure 5—figure supplement 1D ) . Careful examination of many overlaid calcium and bond lifetime traces , as exemplified in Figure 5B , revealed that the ∑ti values are generally dominated by the tmax values , which are usually much longer than the rest of the pre-Ca2+ bond lifetimes and are immediately followed by the calcium onset before observing additional shorter bond lifetimes . In other words , for each platelet usually ∑ti could be approximated by tmax but <t> is of a smaller and variable value . This observation explains why calcium correlates equally well with tmax and ∑ti but not with <t> . Importantly , these results also suggest that a single long-lived GPIbα bond is sufficient to trigger Ca2+ in a platelet . This assertion has been further supported by the parallel analysis of the data for A1R1450E . Although for R1450E the tmax value was significantly shorter and the α-type Ca2+ population was greatly reduced , the ΔImax still showed similar correlation with tmax ( Figure 5F ) . Thus the pre-Ca2+ longest bond lifetime correlates the Ca2+ strength and type . We next asked whether the mechanoreceptor GPIbα is capable of sensing differences in the force waveform and discriminating the ligand through which force is applied . We first performed force-ramp experiment to generate a wide range of rupture forces using three ramping rates: 100 , 1000 and 10 , 000 pN/s . However , only low levels of β-type Ca2+ were resulted ( Figure 6A , B ) , showing no correlation with the largest rupture force prior to calcium onset ( Figure 6C , D , right ordinate ) , regardless of whether platelets were tested by A1WT ( Figure 6A , C ) or A1R1450E ( Figure 6B , D ) . In sharp contrast , much higher levels of Ca2+ of α- and β-types were induced by clamped forces applied to GPIbα via either A1WT ( Figure 6E , G ) or A1R1450E ( Figure 6F , H ) despite their much lower levels than the rupture forces seen in the force-ramp experiments ( Figure 6C , D ) . Concurrently , the longest of A1–GPIbα bond lifetimes that occurred prior to Ca2+ onset was measured on each platelet and averaged over all platelets in each group . This pre-Ca2+ longest bond lifetime , tmax , exhibited catch-slip bond behavior for A1WT and slip-only bond behavior for A1R1450E ( Figure 6G , H , right ordinate ) , just as the corresponding average bond lifetimes previously measured regardless of the intraplatelet calcium ( Ju et al . , 2013; Yago et al . , 2008 ) . Remarkably , the force-dependent pattern of calcium signals matched that of the pre-Ca2+ longest bond lifetimes for both A1WT and A1R1450E . The ligand-independent positive correlation of Ca2+ signal with tmax is consistent with a previously observed inverse correlation between the cytosolic Ca2+ level and the translocation velocity of platelets on immobilized VWF ( Nesbitt et al . , 2002 ) . This is expected because the platelet translocation velocity is an inverse metric of VWF–GPIbα bond lifetime ( Ju et al . , 2013; Yago et al . , 2008 ) . 10 . 7554/eLife . 15447 . 020Figure 6 . GPIbα can sense different force waveforms and discriminate different ligands . ( A–D ) Force-ramp fBFP experiment mode . Individual ΔImax values and their mean ± s . e . m . ( A , B , points ) , Ca2+ types ( C , D , stacked bars , left ordinate ) , and mean ± s . e . m . of pre-Ca2+largest rupture force ( C , D , black square , right ordinate ) are plotted vs . force ramping rate for A1WT ( A , C ) or A1R1450E ( B , D ) . ( E–H ) Force-clamp fBFP experiment mode . Individual ΔImax values and their mean ± s . e . m . ( E , F , points ) , Ca2+ types ( G , H , stacked bars , left ordinate ) , and mean ± s . e . m . of pre-Ca2+ longest lifetime ( G , H , black square , right ordinate ) are plotted vs . clamped force for A1WT ( E , G ) or A1R1450E ( F , H ) . Each point in ( A , B , E , F ) represents results from one platelet and the numbers of platelets in each column are indicated in the corresponding bar in ( C , D , G , H ) , with matched colors to indicate Ca2+ types . Error bar in ( C , D , G , H ) represents s . e . m . estimated by the multinomial distribution of events . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 020 The findings that durable force is important to both MSD unfolding and Ca2+ triggering prompted us to investigate the relationship between GPIbα domain unfolding and platelet signal initiation . We segregated the Ca2+ data generated by a 25 pN clamped force on A1WT–GPIbα bonds according to whether or not and , if so , which domain ( s ) was ( were ) unfolded prior to calcium onset . Platelets whose tests contained no unfolding event showed short tmax and low calcium of β- and null-types ( Figure 7A ) . Platelets whose tests contained at least one pre-Ca2+ MSD unfolding event but no LRRD unfolding showed slightly longer tmax and higher Ca2+ of mostly α-type . By comparison , only β-type Ca2+ was observed in the rare ( 2 . 6% ) cases where LRRD unfolded but MSD did not . Since in these cases the tmax values were much longer , this data excludes tmax to be the direct determining parameter for the Ca2+ type . Remarkably , the group with pre-Ca2+ unfolding of both LRRD and MSD exhibited long tmax and high Ca2+ of mostly α-type . 10 . 7554/eLife . 15447 . 021Figure 7 . Correlation between GPIbα domain unfolding and Ca2+ triggering at 25 pN clamped force . ( A–D ) Individual ΔImax values and their mean ± s . e . m . ( points , left ordinate ) in platelets triggered by A1 ( A , C ) or WM23 ( B , D ) binding , which were segregated into groups with ( + ) or without ( − ) unfolding of LRRD and/or MSD . Each point represents measurement from a platelet . The frequency of each unfolding combination to occur was indicated . ( A , B ) Data obtained from 25 pN force-clamp experiments . Corresponding tmax ( gray bars , right ordinate ) were overlaid with ΔImax . ( C , D ) Data obtained from 1000 pN/s force-ramp experiments . Corresponding pre-Ca2+ largest rupture force ( gray bars , right ordinate ) were overlaid with ΔImax . ( E , F ) Percentage of total events of three Ca2+ types in platelets in the same experiments as in ( A ) and ( B ) ( left bars ) as well as additional experiments performed in the presence of MPαC ( middle bars ) or MαCsc ( right bars ) . Error bar = s . e . m . estimated from the multinomial distribution of events . ( G ) A postulated model of GPIbα-mediated mechanosensing . Force applied via VWF-A1 induces GPIbα LRRD and MSD unfolding . GPIbβ head domain binds to the unfolded MSD and causes the dissociation of its cytoplasmic tail from GPIbα-associated 14-3-3ζ , which transduces signals across the platelet membrane and further downstream , finally leading to α-type Ca2+ . Each step of the mechanosensing process is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 02110 . 7554/eLife . 15447 . 022Figure 7—figure supplement 1 . Comparison of GPIbα bond lifetime and MSD clamped unfolding by A1WT and WM23 . The number of lifetime events ( A ) , their average lifetimes ( B ) , MSD unfolding occurrence frequency ( C ) , expected number of MSD unfolding events ( D ) , and calculated probabilities of observing at least one MSD unfolding event ( E ) over a 200-s interrogation time during which a platelet was repeatedly contacted by a BFP bead coated with A1 or WM23 . Data in panels A–C are presented as mean ± s . e . m . from n ≥ 60 events . The error bars in D are determined using Gaussian error propagation law . * indicates p < 0 . 05 by Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 02210 . 7554/eLife . 15447 . 023Figure 7—figure supplement 2 . Model of GPIb-mediated platelet mechanosensing . For a circulating platelet distal to an injured site without physical contact , its GPIb signaling will not be triggered ( A ) . Upon interacting with VWF , platelets tether and translocate on the sub-endothelial surface via sequential intermittent VWF-A1–GPIbα bonds that provide adhesive forces . Having relatively short lifetimes , these forces unlikely induce MSD unfolding on GPIbα , hence only trigger β-type Ca2+ ( B ) . LRRD unfolding under an increasing force prolongs bond lifetime and results in temporary platelet arrestment , providing a higher chance for MSD unfolding , which triggers α-type Ca2+ ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15447 . 023 Consistent results were obtained using WM23 to pull GPIbα to bypass LRRD unfolding ( cf . Figure 2A ) , showing significantly longer tmax and higher calcium for platelets with than without a MSD unfolding event and a clear α- vs . β-type signal distinction between them ( Figure 7B ) . The higher α-type Ca2+ triggering efficiency of WM23 than A1 ( compare MSD+ columns in Figure 7A and B ) may be explained , at least in part , by the slower kinetics of GPIbα dissociation from WM23 than A1 , which generated 70% more bond lifetime events by contacting a platelet for 200 s with WM23 than A1 ( Figure 7—figure supplement 1A ) . The bonds were also 74% more durable ( Figure 7—figure supplement 1B ) , resulting in a slightly higher ( although not significant ) MSD clamped unfolding probability per lifetime event for WM23 ( Figure 7—figure supplement 1C ) . The expected number of MSD clamped unfolding over a 200-s experimental period , calculated as the product of the number of lifetime events and the unfolding probability per lifetime event , was significantly higher for WM23 than A1 ( Figure 7—figure supplement 1D ) . The probability of platelet to have at least one MSD unfolding , calculated as 1– ( 1– unfolding probability per lifetime event ) ^ ( # lifetime events ) , is 40% and 60% for A1 and WM23 , respectively ( Figure 7—figure supplement 1E ) , close to the experimental results ( 51% vs . 69% , Figure 7A , B ) . Interestingly , despite their high levels , ramped forces generated very few MSD unfolding events and triggered only null/β- but not α-type Ca2+ regardless of whether A1WT or WM23 was used to pull ( Figure 7C , D ) . Together , these data indicate that both force-induced MSD unfolding and bond lifetime are necessary for inducing α-type Ca2+ signal . Using a sensitivity-specificity analysis ( Figure 5—figure supplement 2A , B ) that slides a putative threshold through the tmax vs . Ca2+ type data , we found tmax>2 s to be the best predictor for A1WT to trigger α- rather than β-type Ca2+ ( Figure 5E , dashed lines ) , which agrees with the fact that the 2 s threshold is much shorter than the average tmax of α-type Ca2+ , but exceeds that of β-type Ca2+ and MSD time-to-unfold ( Figure 5—figure supplement 2C ) . Thus , a longer-lived pre-Ca2+ bond favors MSD unfolding , thereby triggering α-type Ca2+; otherwise , it only triggers β-type Ca2+ . Together , our data suggests separate roles of LRRD and MSD unfolding in GPIbα signaling , with the former intensifying the Ca2+ level and the latter determining the Ca2+ signal type . To understand GPIbα-mediated mechanosensing requires analysis of not only ligand binding and domain unfolding in the extracellular segment of GPIbα , but also events in its cytoplasmic region . 14-3-3ζ is a cytoplasmic protein that has direct association with both GPIbα and GPIbβ C-termini ( Calverley et al . , 1998 ) and regulates GPIb signal transduction ( Dai et al . , 2005 ) . To investigate the role of 14-3-3ζ in GPIbα-mediated Ca2+ signaling , we perturbed the system with a myristoylated peptide ( MPαC ) that mimics the 14-3-3ζ binding sequence of GPIbα , thereby blocking the association of 14-3-3ζ with GPIbα cytoplasmic tail . Consistent with the previously reported signaling inhibition effect ( Dai et al . , 2005; Yin et al . , 2013 ) , MPαC reduced the fraction of α-type Ca2+ from 34 to 3% without affecting β-type Ca2+ , whereas a scramble peptide MαCsc had no effect ( Figure 7E ) . Similar results were obtained by pulling GPIbα via WM23 on platelets ( Figure 7F ) . Thus , GPIb–14-3-3ζ association , a biochemical event , is crucial for the transduction of MSD unfolding , a mechanical event , into intracellular signals . These observations indicate that GPIb–14-3-3ζ serves , at least in part , as a mechanotransducer ( Figure 7G ) . The mechanoreception of GPIbα has been supported by direct observations of transient intracellular Ca2+ spike ( termed type α/β peak ) upon platelet translocation on VWF under flow ( Mazzucato et al . , 2002; Nesbitt et al . , 2002 ) . However , many questions remain . Using fBFP real-time single-bond , single-platelet analysis of force-regulated ligand binding kinetics , receptor unfolding dynamics and intraplatelet calcium mobilization , we have: 1 ) identified , characterized and mathematically modeled the force- and ligand-dependent cooperativity between LRRD and MSD unfolding ( Figure 7G ) ; 2 ) defined an optimal magnitude and threshold duration of clamped force for platelet signal initiation via a single GPIbα bond ( Figure 7—figure supplement 2 ) ; 3 ) uncovered a mechanopresentation defect in a type 2B VWD mutant A1R1450E; 4 ) delineated the interplay among ligand engagement , GPIbα domain unfolding and signal triggering; and 5 ) revealed inhibition of GPIbα mechanotransduction by perturbing its cytoplasmic association with 14-3-3ζ . It is an interesting yet challenging problem to define the minimum mechanical stimulation for inducing signal transduction . We demonstrated that a single A1–GPIbα bond can induce calcium in a platelet without clustering GPIb by multimeric ligands . Specifically , in 83% of the cases where α-type Ca2+ was triggered , only a single MSD unfolding event was observed before the Ca2+ onset . Of these , 43% had only one pre-Ca2+ bond lifetime event . Thus , pulling a single GPIbα by a 25 pN force for >2 s to unfold MSD once is necessary and sufficient to induce α-type Ca2+ signals . By comparison , to trigger calcium in a naïve CD8+ T lymphocyte requires a sequence of intermittent bonds with a total of >10 s lifetimes under 10 pN accumulated in the first 60 s of contacts between T cell receptors and agonist peptide-major histocompatibility complex molecules ( Liu et al . , 2014 ) . In both cases a threshold of force duration is required , as ramped forces without a clamp phase are unable to trigger appreciable levels of α-type Ca2+ regardless of its magnitude . The binding defects of VWF-A1 with type 2B VWD mutations have long been recognized ( Ruggeri , 2004 ) . A recent study has shown a type 2B mutation , A1V1316M , causes additional signaling defects ( Casari et al . , 2013 ) . We showed that another type 2B mutation , A1R1450E , also has signaling defect . Interestingly , the defect to induce calcium has the same root as the binding defect , namely , the conversion of the wild-type catch-slip bond to the mutant slip-only bond ( Ju et al . , 2013; Yago et al . , 2008 ) . Consequently , force exerted on GPIbα by A1R1450E is less able to unfold LRRD , lasts shorter at 25 pN to unfold MSD less frequently , does not generate unbinding cooperativity between the LRRD and MSD , and induces lower level and frequency of α-type Ca2+ at >10 pN . Thus , the mechanical requirements for signal induction manifest as force-dependencies of VWF–GPIbα bond lifetime , MSD unfolding frequency , unfolding cooperativity , and Ca2+ level/type that display similar patterns for the same A1 construct ( WT or R1450E ) but different patterns between A1WT and A1R1450E . These findings show that the GPIbα mechanoreceptor can discriminate ligands and shed light to the biophysical mechanisms of type 2B VWD . Our new data on the interplay among VWF binding , GPIbα unfolding , and Ca2+ signaling have provided new insights into the inner workings of the A1–GPIb–14-3-3ζ molecular assembly ( Figure 7; Video 1 ) . By residing in the juxtamembrane stalk region , the MSD has been shown to be mechanosensitive , and hypothesized to play a role in activating platelets ( Zhang et al . , 2015 ) . In the present work , we found that MSD unfolding is required to trigger α-type Ca2+ , showing that this extracellular mechanical event is necessary for transduction of the information embedded in the force waveform into intracellular biochemical signals via the 14-3-3ζ connection ( Figure 7G ) . By overlapping with the ligand-binding site , the LRRD can feel the structural variation in the A1 and respond with an altered unfolding frequency and changed bond lifetime . Importantly , LRRD unfolding prolongs A1–GPIbα bond lifetime to facilitate MSD unfolding , thereby increasing the frequency of α-type Ca2+ and its level ( Figure 7—figure supplement 2 ) . Thus , our study has elucidated part of a mechanosensor that includes three components: 1 ) a MSD in the juxtamembrane region whose conformational change results in a binary decision of Ca2+ type , 2 ) a LRRD in the ligand-binding region whose conformational change leads to continuous alterations in ligand-binding duration , signal level and fractions of different signal types , and 3 ) a MP stalk that transmits force over a distance and provides coupling between the two unfoldable domains . The differential unfolding behaviors of the LRRD and MSD in response to distinct force waveforms provide a simple mechanical mechanism for unfolding cooperativity , by setting the response order such that LRRD unfolds first during force ramp to give more time for MSD to unfold during force clamp . These principles may be helpful for design of a generic mechanosensor , e . g . , using synthetic biology approaches . In addition to an increased MSD unfolding frequency , the cooperativity between LRRD and MSD unfolding may manifest as an increased LRRD unfolding extent . This has been suggested by the observation that the values of the first two peaks ( 20 and 36 nm ) do not add up to that of the third ( 65-70 nm ) in the unfolding length histograms . We note that the observed maximum unfolding length from the GC experiments ( 56 nm ) is smaller than the calculated contour length of LRRD ( 70 nm ) . Since LRRD consists of 8 leucine-rich repeats , this suggests that only some but not all of the repeats were unfolded in any given observation . Thus , the LRRD dataset contains mixed populations of partial unfolding events . By comparison , the MSD dataset may be a more uniform population as the observed maximum unfolding length from the WM23 experiments ( 27 nm ) matches the calculated contour length of MSD ( 26 nm ) . These considerations suggest possible explanations for the observation that events in which both LRRD and MSD unfold generate more length than the sum of lengths generated from events in which either LRRD or MSD unfolds: LRRD unfolding events with a higher number of unfolded leucine-rich repeats may facilitate MSD unfolding more effectively than those with a lower number of unfolded leucine-rich repeats . Alternatively , MSD unfolding , once happens , may induce more leucine-rich repeats in LRRD to unfold . Note that these two mechanisms are not mutually exclusive . Studies in mechanosensitive ion channels and enzymes have provided knowledge on how biomolecules respond to force and transduce mechanical stimulations into biochemical signals . For example , ion channels open and close in response to stress within the lipid bilayer or force within a protein link that can do work on the channel and stabilize its state ( Sukharev and Sachs , 2012 ) . Mechanosensitive enzymes or substrates , such as vinculin ( del Rio et al . , 2009; Grashoff et al . , 2010 ) or A2 domain of VWF ( Wu et al . , 2010; Zhang et al . , 2009b ) , change conformations in response to forces to expose a cryptic site to enable enzymatic reaction . By comparison , for the system studied herein , force signals are received by the receptor via ligand interaction , hence mediated by their binding kinetics . The process of transducing the extracellular mechanical events ( i . e . , LRRD and/or MSD unfolding ) across the cell membrane is likely mechanical rather than chemical ( i . e . , ion influx ) . The study of GPIb mechanosensing may help understand how mechanical force regulates platelet thrombotic functions . For example , in response to shear gradients resulting from flow perturbations , discoid platelets aggregate rapidly in a manner independent of agonist activation pathways ( Jain et al . , 2016; Nesbitt et al . , 2009; Yong et al . , 2011 ) . This intriguing phenomenon has significant implication in atherothrombosis and medical device thrombotic fouling . Other than the requirement for VWF–GPIbα binding , the underlying mechanism of such purely force-induced platelet thrombosis remains elusive . Here LRRD unfolding may play a role because it requires an increasing force ( resembles shear gradient ) and strengthens VWF–GPIbα bonds ( Ju et al . , 2015b ) ( Figure 7—figure supplement 2C ) . In addition , our findings may have broad implications since LRRD is a common structure shared by many adhesion and signaling receptors , e . g . , toll-like receptors ( Bella et al . , 2008 ) . Recombinant monomeric VWF-A1 ( residues 1238–1471 ) WT and type 2B mutant R1450E ( Cruz et al . , 2000; Morales et al . , 2006 ) generated by E . coli were gifts of Miguel A . Cruz ( Baylor College of Medicine , Houston , TX ) . Glycocalicin was purified from outdated platelets ( Fox et al . , 1988 ) . Three anti-GPIbα mAbs were used: AK2 ( Abcam , Cambridge , MA ) , AN51 ( Millipore , Billerica , MA ) and WM23 ( a gift from Michael Berndt , Curtin University , WA , Australia and Renhao Li , Emory University , Atlanta , GA ) . Anti-GPIbβ mAb LS-B3174 was purchased ( LifeSpan BioSciences , Seattle , WA ) . Anti-VWF-A1 mAb 6G1 was a gift from Michael Berndt . Myristoylated peptides ( MPαC , C13H27CONH-SIRYSGHpSL ) and myristoylated scrambled control peptide ( MαCsc , C13H27CONH-LSISYGSHR ) were produced as previously described ( Dai et al . , 2005; Yin et al . , 2013 ) . Human RBCs and platelets were collected abiding a protocol approved by the Institute Review Broad of Georgia Institute of Technology . RBCs were prepared as previously described ( Ju et al . , 2013a ) . To obtain fresh discoid human platelets , whole blood was drawn slowly from a vein of a healthy volunteer to fill in a 3 ml syringe preloaded with 0 . 5 ml ACD buffer ( 6 . 25 g sodium citrate , 3 . 1 g citric acid anhidrous , 3 . 4 g D-glucose in 250 ml deionized H2O , pH 6 . 7 ) . The whole blood was centrifuged at 150 g for 15 min without brake . Platelet-rich plasma was extracted and centrifuged at 900 g for another 10 min . The platelet pellet was resuspended into Hepes-Tyrode buffer ( 134 mM NaCl , 12 mM NaHCO3 , 2 . 9 mM KCl , 0 . 34 mM sodium phosphate monobasic , 5 mM HEPES , and 5 mM glucose , 1% bovine serum albumin ( BSA ) , pH 7 . 4 ) . For Ca2+ imaging experiments , isolated platelets were incubated with Fura-2-AM ( Life Technologies , Grand Island , NY ) at 30 μM for 30 min . For treatment with MPαC or MαCsc , the peptide pre-dissolved in DMSO was resuspended into the platelet suspension to reach a final concentration of 25 μM and incubated at 37°C for 30 min . A1WT , A1R1450E and antibodies were pre-coupled covalently with maleimide-PEG3500-NHS ( MW ~3500; JenKem , TX ) . As previously described ( Ju et al . , 2013a , 2015a ) , the modified proteins were then mixed with streptavidin ( SA ) -maleimide ( Sigma-Aldrich , St . Louis , MO ) in carbonate/bicarbonate buffer ( pH 8 . 5 ) and together linked to silanized borosilicate beads ( Thermo Fisher Scientific , Waltham , MA ) in phosphate buffer ( pH 6 . 8 ) . Site densities of ligands on beads were measured using the previously described flow cytometry method ( Ju et al . , 2015a ) . Our fBFP was developed to simultaneously measure the binding kinetics of single receptor–ligand bonds ( Ju et al . , 2015a , 2013 , 2015c ) and the mechanics of single protein conformational changes ( Chen et al . , 2012; Ju et al . , 2015b ) , as did our original BFP , and receptor-initiated intracellular signaling with a concurrent fluorescent imaging module ( Chen et al . , 2015; Liu et al . , 2014 ) . Bond formation , force application , receptor conformational change , and bond dissociation were enabled and monitored in controlled BFP cycles of a few seconds each . Intraplatelet calcium fluxes were ratiometracally imaged as a signaling readout . In a BFP cycle , the platelet was driven to approach , impinge and hold the probe with a 20–30 pN compressive force for a contact time of 2 s to allow for bond formation , and then retract ( ramp ) for bond detection ( Figure 1C and Figure 1—figure supplement 1A i–iv ) . Displacement of the probe bead was tracked , which reflected the force exerted on it . During the ramping phase , a bond event was signified by a tensile force signal ( Figure 1—figure supplement 1C , D ) , while no tensile force was detected in a no-bond event ( Figure 1—figure supplement 1B ) . Bond and no-bond events were enumerated to calculate an adhesion frequency in 50 repeated cycles for each bead and platelet pair . At least 3 bead–platelet pairs were measured and their adhesion frequencies were used to calculate mean ± s . e . m . values . To define the minimum requirement for GPIb mechanoreception , adhesions were adjusted to be infrequent ( <20% ) by titrating the densities of randomly distributed A1 and mAb on the probe beads ( Figure 1D ) . This ensured that most ( >89% ) platelet–bead binding events were mediated by non-clustered single-bonds ( Chesla et al . , 1998 ) . To quantify intraplatelet Ca2+ mobilization , we used ratiometric imaging with a light source that alternates two excitation wavelengths ( 340 ± 10 nm to excite Ca2+-bound Fura-2 , and 380 ± 10 nm to excite Ca2+-free Fura-2 ) . The emission light from the excited Fura-2 ( with or without Ca2+ binding ) was captured by a fluorescence camera . To maintain the physiological temperature ( 37°C ) inside the cell chamber , a custom-designed temperature control system made in house was integrated into the fBFP . Details about the Ca2+ imaging analysis and temperature control have been previously described ( Chen et al . , 2015; Liu et al . , 2014 ) . In the force-clamp mode , the target pipette was driven to repeatedly contact the probe bead for 2 s and retract at a constant speed ( 3 . 3 μm/s ) . Multiplying the BFP spring constant ( 0 . 3 pN/nm ) , this would translate to a linearly increasing force at a constant ramping rate ( 1000 pN/s ) . Upon detection of bond event , a feedback loop controls the retraction so that it would be paused at a desired clamped force ( 10 , 25 , 40 and 60 pN ) to wait for bond dissociation ( Figure 1—figure supplement 1C ) . After that the target pipette would return to the original position to complete the cycle ( Figure 1—figure supplement 1Av–vii ) . Each platelet was interrogated for a continuous time of 200 s to generate a force spectroscopy trace exemplified in Figure 1C before changing to a new pair of BFP bead and platelet . Lifetimes were measured from the instant when the force reached the desired level to the instant of bond dissociation ( Figure 1C ) ( Ju et al . , 2015a , 2013 ) . In the force-ramp mode , the force was loaded at different ramping rates ( 100 , 1000 or 10 , 000 pN/s ) . The target was retracted continuously until bond rupture without holding at a constant force ( Figure 1—figure supplement 1D ) . Unfolding of GPIbα in the ramping phase was signified by a sudden force stagnation or drop ( kink ) as opposed to the linearly increasing force signals ( Figure 1E ) . To determine the unfolding length , we derived a force vs . extension curve ( Figure 1E inset ) from the differential displacement between the BFP tracking system ( probe position ) and the piezoelectric actuator feedback system ( target position ) as previously described ( Chen et al . , 2012 ) . The unfolding length was given by the sudden extension increase without force increase , the result of which was comparable to the differential contour length derived by fitting the prior- and post-unfolding force-extension curves with the worm-like chain ( WLC ) model ( Bustamante et al . , 1994 ) ( Figure 1—figure supplement 2 ) . To reveal distinct populations , we used the nonparametric kernel density estimation to detect peaks in the data distribution of ensemble ramped unfolding lengths ( Freedman and Diaconis , 1981 ) ( Figure 2—figure supplement 1A , B ) and used a reliable data-based bandwidth selection method ( Sheather and Jones , 1991 ) to determine the optimal bin width as 5 nm for the data in Figure 2D , F . Using a 9 . 53 nm bin width determined by the Freedman-Diaconis formula and Freedman and Diaconis’ heuristic rule ( Freedman and Diaconis , 1981 ) for histogram analysis also revealed three peaks , although the valley separating the first two peaks consists of a single low fraction bin only ( Figure 2—figure supplement 1C ) . The first two peaks in Figure 2D were suggested as MSD and LRRD unfolding , respectively , based on their favorable comparisons to the respective WM23 vs . platelet and A1WT vs . GC data in Figure 2E , which allowed only MSD or LRRD to unfoldrespectively . To test these hypotheses , we analyzed the molecular mechanics by sorting the unfolding forces and lengths into subgroups from the respective WM23 vs . platelet and A1WT vs . GC experiments , plotted the average unfolding force vs . length data , and fitted the data to the WLC model ( Zhang et al . , 2009a , 2015 ) . The best-fit curves were then served as standards to calibrate the average unfolding force vs . length data from the A1WT vs . platelet experiment ( Figure 2H , I ) . The agreement between the data and the calibrated WLC curves rigorously verified the hypothetical identities of the first two peaks in Figure 2D . Unfolding of GPIbα in the clamping phase was signified by a sudden force decrease ( Figure 1E ) . The unfolding length was calculated from force change divided by BFP spring constant ( Figure 1F ) , similar to the integrin extension length measurement from the previous distance-clamp analysis ( Chen et al . , 2012 , 2016 ) . The duration from the beginning of the clamping phase to the beginning of unfolding was the time-to-unfold , tu ( Figure 1F ) . The cooperativity between LRRD and MSD unfolding at a clamped force ( e . g . 25 pN ) was determined by testing the null hypothesis that the two domains unfolded independently . The frequencies of LRRD unfolding and MSD unfolding pulled by A1WT were calculated using data from Figure 3—source data 1A ( take 25 pN for example ) : MSD = '+'MSD = '−'Row totalLRRD = '+'14 ( 9 . 7% ) 11 ( 7 . 6% ) 25 ( 17 . 2% ) LRRD = '−'25 ( 17 . 2% ) 95 ( 65 . 5% ) 120 ( 82 . 8% ) Column total39 ( 26 . 9% ) 106 ( 73 . 1% ) 145 ( 100% ) The Pearson's χ2 test was used to test the null hypothesis ( H0 ) that LRRD unfolding and MSD unfolding are independent . The χ2 statistic was calculated as follow:χ2=∑i=12∑j=12 ( Oij−Eij ) 2Eij=13 . 01 where Oij is the observed count and Eij is the expected count under the null hypothesis . The subscripts i and j denote LRRD and MSD respectively , whose values 1 and 2 denote with ( + ) and without ( - ) unfolding respectively . The system has ( 2−1 ) × ( 2−1 ) = 1 degree of freedom . The small p-value ( 3 . 09 × 10–4 from the above χ2 ) requires that we reject the null hypothesis and accept the alternative hypothesis that LRRD and MSD unfolding are not independent . In other words , cooperativity exists between LRRD and MSD unfolding when GPIbα on platelets was pulled by A1WT . Similar statistical analyses were employed to assess cooperative unfolding between LRRD and MSD when GPIbα was pulled by A1WT , A1R1450E or AN51 at different clamped forces . The levels of significance were presented as –log10 ( p-values ) ( Figure 3G , H ) . The measured MSD time-to-unfold tu distributed as a single exponential decay: pu ( tu ) =kue−kutu , where pu is the probability density and ku is the unfolding rate of MSD under a clamping force . By fitting the semi-log plotted experimental distribution with a straight line ( Figure 4—figure supplement 1A ) , the unfolding rate at 25 pN was evaluated from the negative slope or the reciprocal average time-to-unfold , ku = 1/<tu> = 0 . 870s-1 . Modeling the force-dependent MSD unfolding rate ( Figure 4A ) by the Bell equation ( Bell , 1978 ) , we found the zero-force unfolding rate and the width of the energy barrier to be 0 . 26 s−1 and 0 . 242 nm for pulling GPIbα via A1 on a live platelet . The first value is much larger and the second value is much smaller than the respective values previously obtained using an optical tweezer to measure ramped unfolding of MSD in purified GPIbα constructs ( Zhang et al . , 2015 ) . We previously reported that the A1–GPIbα bond lifetime tb distributed as a dual exponential decay with a fast- and a slow-dissociating off-rate ( Ju et al . , 2013 ) . We also recently showed that unfolding of LRRD prolongs A1–GPIbα bond lifetime ( Ju et al . , 2015b ) . These results were also observed in this work , which comprise individual bond lifetime measurements that give rise to the averaged results in Figure 4B , C . Therefore , the probability densities for a A1–GPIbα bond , with and without prior LRRD unfolding in the ramping phase , to dissociate at time tb during the clamping phase are: ( 1 ) ForLRRD−:p1 ( tb ) =w11k11e−k11tb+w12k12e−k12tb ( 2 ) ForLRRD+:p2 ( tb ) =w21k21e−k21tb+w22k22e−k22tb where kij and wij ( wi1 + wi2 = 1 ) denote , respectively , off-rates and associated fractions of bonds under a clamped force , with the first subscript indicating without ( Equation 1 ) or with ( Equation 2 ) a prior LRRD unfolding event and the second subscript indicating the fast ( Equation 1 ) or slow ( Equation 2 ) dissociation pathway . By fitting the above model to the lifetime ensemble data , the parameters were calculated ( Figure 4—source data 1 ) . Assuming that MSD unfolding and A1–GPIbα unbinding are independent events , the joint probability density for MSD unfolding at time tu and A1–GPIbα unbinding at time tb is p ( tu , tb ) =pu ( tu ) ×pi ( tb ) where i = 1 , 2 depending on whether LRRD unfolding occurs . This joint probability is depicted as a surface in Figure 4D and Figure 4—figure supplement 1B , using respective A1WT and A1R1450E data measured at 25 pN clamped force . The condition for observing MSD clamped unfolding is that the A1–GPIbα bond lifetime tb lasts longer than the time-to-unfold tu . Thus , the probability of observing MSD unfolding in the clamping phase Pui is the volume under the probability density surface over the region 0<tu<tb<∞ , which is marked by the vertical red planes in Figure 4D and Figure 4—figure supplement 1B . For instance , in the absence of prior LRRD unfolding , the probability of observing MSD unfolding in the clamping phase of 25 pN is: ( 3 ) Pu1=∫0+∞[p1 ( tb ) ∗∫0tbpu ( tu ) dtu]dtb=w11kuku+k11+w12kuku+k12=21 . 5% Similarly , in the presence of prior LRRD unfolding , ( 4 ) Pu2=w21kuku+k21+w22kuku+k22=46 . 2% The model was applied to predict the MSD clamped unfolding probabilities under different clamped forces pulled by A1WT and A1R1450E ( Figure 4E , F ) . For A1R1450E , the ensemble MSD clamped unfolding events were no longer segregated into LRRD- and LRRD+ groups , because few MSD clamped unfolding events occurred following LRRD unfolding due to the reduced bond lifetime . We used the sensitivity-specificity analysis to solve the optimal threshold t0 for pre-Ca2+ longest lifetime tmax separating α and β Ca2+ types . There are 4 possible outcomes ( Figure 5—figure supplement 2A ) : a false positive ( FP ) happens when tmax>t0 and a β-type Ca2+ was observed; a false negative ( FN ) happens when tmax≤t0 and an α-type Ca2+ was observed; a true positive ( TP ) happens when and an α-type Ca2+ was observed , and a true negative ( TN ) happens when tmax≤t0 and a β-type Ca2+ was observed . The sensitivity or true positive rate defines the fraction of true positive among all positive results , TP/ ( TP+FN ) , whereas the specificity or true negative rate defines the fraction of true negative among all negative results , TN/ ( TN+FP ) . The optimal threshold is solved by minimizing the total counts of false positive and false negative . To do that , a receiver operating characteristic ( ROC ) curve was created by plotting the TP rate ( sensitivity ) against the FP rate ( 1- specificity ) at various tmax values from which the optimal threshold t0 that achieved the best sensitivity and specificity was identified ( Figure 5—figure supplement 2B ) . Two-tailed Students’ t-test was used to assess significance for group comparisons . Pearson correlation coefficient was used as a measure for linear dependency between two variables ( Ca2+ level and binding kinetics ) . To determine errors in classification of different Ca2+ types ( null , β , α ) and in identification of unfolding ( no unfolding , LRRD , MSD , MSD+LRRD ) , we assume that observed counts ( n1 , n2 , … , nK ) follow a multinomial distribution with total counts n=n1+n2+…+nK and event probabilities ( p1 , p2 , … , pK ) . We then use the fraction of the i-th category , ni/n , as an estimate for the i-th event probability pi and ( ni/n ) ( 1−ni/n ) /n as the associated standard error s . e . m .	Platelets – the blood clotting cells – have the ability to detect , interpret and respond to mechanical forces , such as those generated by the flow of blood . The magnitude and duration of the forces detected by the platelets influences whether they form a blood clot . Understanding how the platelets respond to mechanical forces is therefore crucial for our knowledge of conditions such as thrombosis , where blood clots form inside vessels and block them . Clots that form within arteries are associated with heart attack and stroke , which account for around one third of all deaths worldwide . Cells can sense external forces via individual proteins on their surface and transmit the mechanical information across the cell membrane . This triggers signals within the cell that influence how it responds . However , the molecular details of these “mechanosensory” processes remain poorly understood . To patch up damaged blood vessels , platelets use a protein on their surface named glycoprotein Ibα ( GPIbα ) to bind to a plasma protein called von Willebrand factor that adheres to the vessel wall . This binding tethers the platelet to the blood vessel and activates it during clot formation . Previous studies suggested that mechanical force affects how this binding triggers the signals that activate platelets . Ju , Chen et al . used a homebuilt nanotool to pull on platelet GPIbα while it was bound to von Willebrand factor . This revealed that two distinct domains of the GPIbα protein unfold to relay information about the force , such as its magnitude and duration , to the platelet to trigger biochemical signalling inside the cell . The unfolding of each GPIbα domain has a distinct role in determining the quantity and quality of the signals . The unfolding events work synergistically – they occur together to produce an effect that’s greater than the sum of their individual effects . However , pulling on GPIbα via a mutant form of von Willebrand factor eliminated the synergy between the two unfolding events , therefore hindering the effective conversion of mechanical forces into biochemical signals . Notably , the two GPIbα domains unfolded by force exist in many protein families , including those involved in mediating cell adhesion and detecting signals . The biophysical tools developed by Ju , Chen et al . could be extended to analyze how mechanical cues are presented , received , transmitted and converted into biochemical signals in other cell types and biological systems . Furthermore , the structural insights gained from the platelet GPIbα system may help to design a generic mechanosensory protein machine .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2016	Cooperative unfolding of distinctive mechanoreceptor domains transduces force into signals
Transmission reduction is a key component of global efforts to control and eliminate malaria; yet , it is unclear how the density of transmission stages ( gametocytes ) influences infection ( proportion of mosquitoes infected ) . Human to mosquito transmission was assessed using 171 direct mosquito feeding assays conducted in Burkina Faso and Kenya . Plasmodium falciparum infects Anopheles gambiae efficiently at low densities ( 4% mosquitoes at 1/µl blood ) , although substantially more ( >200/µl ) are required to increase infection further . In a site in Burkina Faso , children harbour more gametocytes than adults though the non-linear relationship between gametocyte density and mosquito infection means that ( per person ) they only contribute slightly more to transmission . This method can be used to determine the reservoir of infection in different endemic settings . Interventions reducing gametocyte density need to be highly effective in order to halt human–mosquito transmission , although their use can be optimised by targeting those contributing the most to transmission . The malaria parasite is transmitted among humans by anopheline mosquitoes . Male and female transmission stages ( gametocytes ) are ingested by the mosquito and reproduce sexually in its stomach before developing into oocysts . Once oocysts establish , it is assumed that mosquitoes will form infectious sporozoites , so the proportion of mosquitoes developing oocysts is used as a measure of mosquito infectivity . In the human malaria parasite Plasmodium falciparum , mosquito infection is thought to increase with the number of gametocytes ingested by the mosquito ( Jeffery and Eyles , 1955; Graves et al . , 1988; Bousema and Drakeley , 2011 ) . However , several studies have failed to find an association ( Boudin et al . , 1993; Haji et al . , 1996 ) , and the precise shape of the relationship has never been rigorously quantified , particularly at very low gametocyte densities . Difficulties arise because estimates of gametocyte density have relied on microscopy , which may miss up to 80% of parasites ( Dowling and Shute , 1966 ) . More sensitive molecular methods such as Pfs25mRNA quantitative nucleic acid sequence–based amplification ( QT-NASBA ) have been developed ( Schneider et al . , 2004 ) . Unlike conventional microscopy , this technique enables gametocyte densities to be quantified over the entire epidemiologically relevant range . Transmission reduction is now a key component of global efforts to control and eliminate malaria ( Alonso et al . , 2011 ) . A wide range of novel transmission-reducing drugs and vaccines are currently under development , which aim to reduce malaria incidence by restricting human to mosquito transmission . The transmission-reducing ability of front-line therapeutics ( such as artemisinin-based combination therapies [ACTs] , or potential combinations of ACTs with gametocytocidal drugs ) is also gaining increased attention ( WHO , 2012; White , 2013 ) . It is likely that these interventions will be partially effective , but it remains unclear how much they need to reduce transmission from humans to mosquitoes in a particular location , and to which age groups they should be delivered in order to halt malaria transmission . Mathematical models of malaria transmission tend not to include explicitly the relationship between gametocytaemia and mosquito infectivity , opting , for simplicity , to assume density-independent transmission probabilities ( Smith et al . , 2007; Griffin et al . , 2010 ) , or fit functions to the relationship between the number of asexual parasites and the probability of a blood-feeding mosquito becoming infected ( Ross et al . , 2006 ) . Greater complexity , however , will be required to capture fully malaria population dynamics following the introduction of drugs and vaccines that specifically target transmission stages . For example , the World Health Organization ( WHO ) has recently recommended that in pre-elimination or elimination malaria programmes , single-dose primaquine ( 0·25 mg base per kg ) with an ACT should be given to all patients with falciparum malaria except pregnant women and infants <1 year old ( WHO , 2012 ) . More detailed mathematical models will allow the full impact of this change in strategy to be assessed in a range of different endemic settings . The optimal use of transmission-reducing interventions will also require a better understanding of the human reservoir of infection , which is likely to vary between different endemic settings . Various studies have attempted to estimate the relative contribution of different age groups to mosquito infection using skin or membrane feeding assays ( Muirhead-Thomson , 1957; Graves et al . , 1988; Boudin et al . , 1991; Githeko et al . , 1992; Drakeley et al . , 2000; Bonnet et al . , 2003 ) . These studies are logistically complicated and expensive , so more efficient methods of determining the reservoir of infection are required in order to target interventions optimally . A thorough understanding of the factors determining mosquito infection would allow mathematical models to predict the relative contribution of different groups to transmission from cross-sectional surveys , both before and after the introduction of transmission-reducing interventions . This article uses data from mosquito feeding assays conducted in Burkina Faso ( Ouédraogo et al . , 2009 , Dryad: Ouédraogo et al . , 2013 ) and Kenya ( Schneider et al . , 2007 , Dryad: Schneider et al . , 2013 ) to estimate the shape of the relationship between gametocyte density and mosquito infection ( processed data available at Dryad , Churcher et al . , 2013 ) . Other covariates that may influence infection such as asexual parasite density ( measured by microscopy ) and host age were also included to improve predictions of human to mosquito transmission . These results are combined with data from a cross-sectional survey conducted in a high transmission setting in Burkina Faso to predict the relative contribution of different age groups to overall malaria transmission . The relationship between the number of gametocytes in the blood and mosquito infection was found to be highly non-linear ( Figure 1A ) . Plasmodium falciparum infects mosquitoes at the very low gametocyte densities that predominate in natural infections and may not be detected using standard microscopy ( Bousema and Drakeley , 2011 ) . Infection rises rapidly with increasing gametocyte density , and by 1 gametocyte per microlitre , ∼4% ( 95% Bayesian credible interval [CI] , 3–5% ) of all mosquitoes develop oocysts . The subsequent increase in infection with increasing gametocyte density is best described by the Gompertz model ( deviance information criterion [DIC] = 1034 ) , which gave a significantly better fit than the linear ( DIC = 1073 ) , power ( DIC = 1059 ) , or hyperbolic ( DIC = 1062 ) functions ( Figure 1A ) . The best-fit model predicts that increasing density from 1 to 200 gametocytes per microlitre does not appreciably increase infection . Beyond 200 gametocytes per microlitre , infection rises again to finally plateau at ∼18% infected mosquitoes . Gametocyte density on an arithmetic scale was a better predictor of mosquito infection than gametocyte density on a logarithmic scale . Including information on the host’s asexual parasite density significantly improved model fit . Children with asexual parasite densities between 1 and 1000 parasites per microlitre were on average 27% ( CI , 19–58% ) less infectious to mosquitoes than those with no detectable asexual parasites , whereas those with asexual densities >1000 µl−1 were 77% ( CI , 14–140% ) more infectious . Including age improves the fit of the model suggesting that age is an important confounder , although the Bayesian credible intervals include 0 . Children >6 years old were 15% ( CI , 24–64% ) more likely to infect mosquitoes than those of younger ages , whereas no difference in the relationship between gametocyte density and mosquito infection was detected between Burkina Faso and Kenya . 10 . 7554/eLife . 00626 . 003Figure 1 . The relationship between gametocyte density and mosquito infection and the impact this will have on the effectiveness of gametocytocidal interventions . ( A ) The relationship between Plasmodium falciparum gametocyte density and the percentage of Anopheles gambiae mosquitoes that develop oocysts . Point colour , shading , and shape denote characteristics of the blood donor , such as location ( blue = Burkina Faso; red = Kenya ) , asexual parasite density as measured by microscopy ( no fill colour = none detectable , light shading = 1–1000 parasites per microlitre , dark shading ≥1000 µl−1 ) , or host age ( <6 years old = square , ≥6 years old = circle ) . The size of the point is proportional to the number of mosquitoes dissected . Coloured horizontal and vertical lines indicate 95% Bayesian credible intervals ( CIs ) around point estimates . The solid black line indicates the best-fit model , whereas the grey shaded area indicates the uncertainty around this line . The inset shows the relationship at very low gametocyte densities ( on a logarithmic scale ) . The outputs show the shape of the relationship for a child with no detectable asexual parasites . A full description of data used to fit the model are given in Figure 1—source data 1 . Regression coefficients and a measure of goodness-of-fit of the different models are given in Figure 1—source data 2 . Panel ( B ) shows how efficacious transmission-reducing interventions would need to be ( on the vertical axis ) at decreasing gametocyte density in order to reduce human to mosquito transmission ( contour lines ) . The best-fit line from ( A ) is used to illustrate the percentage reduction in mosquito infection that would be achieved according to the pre-intervention host’s gametocyte density ( on the horizontal axis ) for an intervention , which reduces gametocyte density by a given percentage ( which is assumed to be constant over different gametocyte densities ) . The colours represent the percentage reduction in mosquito infection that would be achieved , ranging from red ( low , 0–10% reduction ) to darker hues of blue ( high , 70–80% and 80–90% reduction , see legend ) . The 90–100% reduction in transmission is hardly visible and would correspond to nearly 100% efficacious interventions at the top of the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 00310 . 7554/eLife . 00626 . 004Figure 1—source data 1 . Description of the direct feeding assay data used to estimate mosquito infection . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 00410 . 7554/eLife . 00626 . 005Figure 1—source data 2 . Best-fit model and parameters . Lower DIC values indicate a more parsimonious fit to the data . All models were fitted with gametocyte density on the arithmetic scale . There was no evidence of any difference between study sites ( DIC of best-fit model allowing mosquito infection to vary between study location = 1042 ) . The best-fit model on the logarithmic scale was the Gompertz model ( DIC = 1053 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 005 The complex shape of the relationship between gametocyte density and mosquito infection will influence the success of transmission-reducing interventions and may explain why their ability to reduce the proportion of infected mosquitoes has been shown to depend on the gametocyte density of the host ( Churcher et al . , 2012 ) . For example , reducing gametocyte density by 99% in a host with 200 gametocytes per microlitre may not have much effect on their immediate contribution to transmission ( Figure 1B ) , although it will probably reduce the duration of infectivity . The same intervention efficacy at reducing gametocyte density in a host with 300 gametocytes per microlitre would cause an appreciable reduction in human to mosquito transmission from that individual . Figure 1B can be used to estimate how a reduction in the number of gametocytes will equate to a reduction in the proportion of mosquitoes becoming infected , and hence mosquito to human transmission . How this relates to the incidence of malaria and subsequent disease will depend , in part , on the degree of immunity in the human population . In the cross-sectional survey in Burkina Faso , children harbour more gametocytes than adults , with 10 year olds having five times as many gametocytes compared to the 50 year olds ( Figure 2A ) . However , the relationship between gametocyte density and mosquito infection established here means that adults are only 37% less likely to infect mosquitoes ( infecting on average 3 . 5% of them , Figure 2B ) . Results indicate that , at the time of the survey conducted in Burkina Faso , the peak in the reservoir of infection occurs at an earlier age than the peak in gametocyte density , which will improve the ( cost ) effectiveness of school-based programmes . 10 . 7554/eLife . 00626 . 006Figure 2 . Age patterns of gametocyte density and estimates of the age profile of the human reservoir of infection . ( A ) Results from a cross-sectional survey conducted in a high transmission setting in Burkina Faso showing how the mean number of gametocytes per microlitre of blood ( including 0s ) changes with host age . The distribution of gametocytes among hosts is highly overdispersed . A full description of data used to fit the model are given in Figure 2—source data 1 with the best-fit parameter estimates shown in Figure 2—source data 2 . The relationship between gametocyte density and oocyst prevalence shown in Figure 1A is used to predict the percentage of mosquitoes that will become infected after biting a host of a certain age . This is shown in panel ( B ) , which can be interpreted as the contribution of each age group towards the human to mosquito transmission . In both figures , the black solid line shows the best-fit model and the grey shaded area indicates the uncertainty ( 95% Bayesian credible Interval , CI ) having fitted the model to the individual data ( a total of 412 individuals ) . For illustrative purposes , the data are grouped into seven bins , namely 0–5 , 5–10 , 10–15 , 15–20 , 20–30 , 30–40 , 40–50 , and ≥50 year olds , and the size of the point is proportional to the number of individuals in the group . In Figure 1A , the 10–15 and 15–20 year groups appear lower than the best-fit line due to sampling artefacts generated by the highly overdispersed data . Vertical lines indicate the 95% CI around grouped estimates . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 00610 . 7554/eLife . 00626 . 007Figure 2—source data 1 . Description of the cross-sectional survey on 412 hosts carried out in Burkina Faso . Asexual parasite density was estimated by microscopy while gametocyte density was estimated using QT-NASBA . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 00710 . 7554/eLife . 00626 . 008Figure 2—source data 2 . Parameter estimates for the age profile of gametocyte density and the force of infection . DOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 008 The complex shape of the relationship between gametocyte density and mosquito infection elucidated here will influence our understanding of the population dynamics of falciparum malaria and will determine the success of transmission-reducing interventions . The efficiency with which P . falciparum gametocytes can infect mosquitoes means that transmission-reducing interventions , which reduce gametocyte density , will need to be highly effective in order to reduce human–mosquito transmission . Gametocytes can infect mosquitoes at very low densities , despite the mosquito needing to ingest both male and female parasites in the same blood meal . Substantial transmission was seen from hosts with <1 gametocyte per microlitre of blood . After this initial increase in infection , considerably more gametocytes are required in order to increase further the proportion of mosquitoes developing oocysts . The causes of this are unclear , but experimental systems may be informative . Data from the rodent malaria model Plasmodium berghei ( Sinden et al . , 2007 ) suggest that the phenomenon could be associated with parasite mortality during penetration of the mosquito gut wall or due to the mosquito’s immune response . Beyond 200 gametocytes per microlitre mosquito infection rises again ( although QT-NASBA measurement error makes the exact gametocyte density for this transition relatively uncertain ) . If an intervention can keep gametocyte density beneath 200 gametocytes per microlitre in a low transmission area , it may be sufficient to push the basic reproduction number of malaria beneath one and eliminate the disease . However , the highly overdispersed distribution of gametocytes between hosts ( adequately described by the negative binomial distribution ) means that in the population from Burkina Faso , 30% of transmission comes from hosts with densities beneath the microscopy detection threshold of 16 gametocytes per microlitre . This means that evaluating interventions solely on their ability to reduce microscopically detectable gametocytes , as has been suggested ( Graves et al . , 2012 ) , may give misleading results . Including information on asexual parasite density ( as measured by microscopy ) significantly improved the fit of the model , with hosts harbouring intermediate densities having the highest infectivity . It is unclear whether asexual parasites directly influence parasite infectivity or are associated with other ( here unmeasured ) variables such as the multiplicity of infection ( Nsango et al . , 2012 ) or factors influencing the blood environment such as immunity ( Bousema and Drakeley , 2011 ) . Plasmodium falciparum has asynchronous waves of asexual and sexual parasites . Circulating gametocytes take 2–3 days to mature before they can infect mosquitoes ( Lensen et al . , 1999 ) , which may explain the complicated relationship between infectivity and asexual density ( i . e . , intermediate asexual parasite densities might occur at the start of gametocyte production and have a relatively low proportion of infectious mature gametocytes ) . On average , older children were more infectious , supporting previous indications that transmission-reducing immunity may predominate in young children ( Drakeley et al . , 2006 ) . Figure 1A indicates that individual QT-NASBA gametocyte density estimates are relatively uncertain . Despite this , they are considerably more accurate than conventional microscopy . Molecular techniques may be highly precise in a state-of-the-art laboratory , but the practicalities of collecting and processing samples in non-ideal field settings increases the risks of incurring measurement error . Even though uncertainty is generally accepted by those conducting the experiments , measures of precision such as confidence intervals or standard errors around such density estimates are rarely reported in the scientific literature . Methods such as those presented in this article can allow robust quantitative insight to be gained from uncertain molecular methods . Care should be taken when interpreting gametocyte density estimates as it has been suggested that the marker used in this analysis ( Pfs25 mRNA ) might be parasite female specific ( Schneider , 2006 ) . Even if the current QT-NASBA method does only detect female gametocytes , it will neither change the qualitative conclusions of this study nor their application if other studies use the same technique . However , direct comparison of our results with those of studies using different methods for quantifying gametocyte density should be aware that the shape of the relationship may differ . Gametocyte density measurement error is likely to cause the majority of the uncertainty seen in model outputs , although the complexity of the membrane feeding assay is also likely to contribute . The proportion of mosquitoes infected by Plasmodium-infected blood is known to vary substantially within and between studies . This is in part due to methodological issues with the membrane feeding assay ( Bousema et al . , 2013 ) but also related to biological differences in the parasite-vector combination and the blood environment ( Bousema and Drakeley , 2011 ) . Further standardising the membrane feeding assay , improving the accuracy of QT-NASBA technique and the inclusion of additional covariates ( such as estimates of gametocyte maturity , which have been investigated in Plasmodium vivax; Chansamut et al . , 2012 ) , would improve the accuracy of the relationship between gametocyte density and mosquito infection . It would also permit the patterns described here to be checked for consistency across time and space , and enable a wider range of functional forms to be tested . Care should also be taken when interpreting the results of feeding assays as the mosquitoes used , and their biting behaviour is likely to be different from that seen in wild mosquitoes in field situations ( Bousema et al . , 2013 ) . Transmission reduction will become increasingly important as areas approach local elimination . Identifying host age groups that contribute most to the reservoir of infection will allow the optimal targeting of malaria control . In the high transmission site in Burkina Faso , young children harbour the majority of gametocytes but are only slightly more infective to mosquitoes than adults . The per person contribution of adults estimated here is considerably greater than that predicted by other mathematical models ( Ross et al . , 2006 ) and may increase further once age-dependent biting rates are taken into consideration ( Carnevale et al . , 1978; Ross et al . , 2006 ) . The contribution of different age groups to overall transmission will depend on local demography , age-dependent protection by malaria interventions ( such as the use of bed nets or time spent in the house and therefore personally protected by indoor residual spraying ) , and human to mosquito contact patterns . The last two of these are difficult to measure and poorly understood , reducing our ability to accurately predict the sources of infection . The reservoir of infection is likely to vary between endemicity settings and over time so multiple cross-sectional surveys may be required . The relationship between gametocyte density and mosquito infection will allow estimates of the reservoir of infection from cross-sectional gametocytaemia surveys without the need for logistically complicated mosquito feeding assays . The results of this article should be included within mathematical models that capture changes in gametocyte density with time ( Lawpoolsri et al . , 2009 ) to evaluate the full impact that different transmission-reducing interventions will have on transmission and prioritise the most appropriate candidates according to transmission setting . This article provides insights into the relationship between gametocyte density and mosquito infection that are needed to predict the outcome of transmission-reducing interventions . It shows that , given the ability of very low P . falciparum gametocyte densities to establish infection in A . gambiae , transmission-reducing interventions will need to be highly efficacious at reducing gametocyte density in order to halt human to mosquito transmission . The complex and non-linear shape of the relationship between gametocyte density and mosquito infection may explain why the ability of an interventions to reduce the proportion of infected mosquitoes has been shown to depend on the parasite load of the host ( Churcher et al . , 2012 ) . Gametocyte density also changes with host age so the effectiveness of transmission-reducing interventions that target gametocytes will also vary with age . This should be considered in clinical trials of transmission-reducing candidate drugs and vaccines . The highly overdispersed distribution of gametocytes in the host population and the non-linear relationship between gametocytes/asexual parasite density and mosquito infection means that the impact of different transmission-reducing interventions on overall transmission at the population level will be far from intuitive . Quantitative nucleic acid sequence-based amplification ( QT-NASBA ) is routinely used to estimate pathogen density . Like all diagnostic methods , it is prone to measurement error . To understand fully the associated uncertainty , it is important to appreciate the causes of the variability and how the quantification process might magnify uncertainty of point density estimates . Here , the uncertainty is quantified by repeatedly testing samples with known gametocyte density and fitting a hierarchical mathematical model . Nucleic acids are extracted from 50 µl of blood and then amplified in the presence of a fluorescence probe . The assay measures time to positivity ( TTP ) , which is the time it takes for the number of target amplicons detected to exceed a defined threshold ( Schneider et al . , 2004 ) . The relationship between TTP and gametocyte density is estimated by fitting a linear regression to TTP estimates generated using a sample with known gametocyte density ( a 10-fold dilution series of in vitro cultured gametocytes ranging from 106 to 101 gametocytes per millilitre ) . Let the observed TTP be denoted by Y then , ( 1 ) Y=β0+β1⁡ln⁡x+ε , where β0 and β1 are regression coefficients estimates , x is the ( known ) parasite density from the dilution series and ε represents a normally distributed random error with mean equal to 0 and constant variance , that is ε ∼ N ( 0 , σ2 ) . A full list of the parameters is given in Table 1 . The so-called statistical calibration or inverse regression problem ( Osborne , 1991 ) concerns the issue of making statistical inference on the value of an unknown ( log-transformed ) gametocyte density ) , denoted as ln x′ , from a new TTP observation , Y′ The classical approach ( Eisenhart , 1939 ) involves rearranging the regression model ( equation 1 ) so that ( 2 ) ln⁡x′=Y′−β0β1 , and substituting the regression parameters with their estimates β^0 and β^1 to yield an estimate lnx^′ . Here , we adapt this approach for use in a Bayesian hierarchical model . A number of different methods have been used to do this ( see Hoadley , 1970 and Hunter and Lamboy , 1981 and for discussion see Osborne , 1991 ) , each of which has a different method for dealing with the problem of high β^1 values ( i . e . , a gentle gradient , which when used in equation 2 can generate infinitely large confidence interval estimates ) . Here , we use the most parsimonious approach that does not require assignment of previous distributions to ( the unknown ) gametocyte densities . Rather , uncertainty in the estimated regression coefficients of equation 1 is propagated numerically via equation 2 to yield uncertainty in the estimated gametocyte densities . Since all the calibration line data have a relatively steep gradient , the difference between the different methods will be relatively minor , and all will generate sensibly tight confidence interval estimates . 10 . 7554/eLife . 00626 . 009Table 1 . Notation of statistical and mathematical modelsDOI: http://dx . doi . org/10 . 7554/eLife . 00626 . 009NotationDescriptionEquationYTime to positivity ( TTP ) readout generated by QT-NASBAEquation 1xKnown density of gametocytes per millilitre ( generated using dilution series ) Equation 1lnx^′Estimate of ( unknown ) gametocyte density on the logarithmic scaleEquation 2x^′Estimate of gametocyte density on the arithmetic scaleEquation 2β0Intercept of the calibration line fitted to the dilution seriesEquation 1β1Gradient of the calibration line fitted to the dilution seriesEquation 1σ2Intra-assay variance measuring the accuracy with which the calibration line fits the TTP estimates from the dilution seriesEquation 1gProportion of mosquitoes developing oocystsEquation 3ϕ ( x^′ , κ ) Saturating function determining the shape of the initial relationship between gametocytes and the proportion of mosquitoes developing oocystsEquation 4fi ( x^′ ) Function determining the shape of the relationship between gametocytes and proportion of mosquitoes developing oocysts . Subscript i indicates the functional form used , be it fα ( x^′ ) =α0+α1x^′α2 ( 1+α3x^′α2 ) , where constraining different parameters can generate a range of different shapes ( constant α1 = 0 , linear α2 = 1; α3 = 0 , power α3 = 0 , hyperbolic α1 > 0 α2 = 1 α3 > 0 , or sigmoid α2 > 1 ) or fγ ( x′^ ) =γ0+γ1⁡exp[γ2⁡exp ( γ3x′^ ) ] , , which generates a Gompertz ( sigmoid-like ) functionEquation 3μVector of regression coefficientsEquation 3zVector of dummy variables denoting donor blood characteristics , z1 = asexual parasite density ( 0 = undetected , 1 = low , 2 = high ) , z2 = host age ( 0 = younger than 6 years old , 1 = 6 or older ) , z3 = study locale ( 0 = Burkina Faso , 1 = Kenya ) Equation 3h ( A ) Function describing how gametocyte density and the reservoir of infection change with host age ( A ) . Shape determined by parameters τ , ψ , and ωEquation 5 Calibration lines can vary between runs and batches of reagents . Here , we define an experiment as being a single plate run with the same batch of reagents each of which will have its own dilution series and samples with unknown density . The accuracy with which the individual TTP estimates fit the log-linear calibration line can be used to estimate assay measurement error ( the intra-assay variability , σ ) by fitting a hierarchical mixed-effects model to multiple dilution series ( allowing estimates of β0 and β1 to vary among experiments to determine whether this significantly improves the fit of the model ) . The accuracy of gametocyte density estimates can also be improved by running multiple assays on the same sample and then taking the mean of all the ln x′ estimates . The application of equation 2 yields an estimate of gametocyte density , x^′ , on the logarithmic scale . Estimates on the arithmetic scale are generated by taking the exponent , that is x′^=exp ( lnx^′ ) . These estimates are used in the functions below to determine the relationship between gametocyte density and the proportion of mosquitoes developing oocysts . Plasmodium falciparum–infected blood was collected from children and fed through a membrane to A . gambiae sensu stricto mosquitoes that were dissected 7–9 days later to determine infection ( oocyst carriage ) . Data from 171 mosquito feeds on patients’ blood conducted in Burkina Faso ( Ouédraogo et al . , 2009 ) and Kenya ( Schneider et al . , 2007 ) were combined and used to fit a quantitative relationship between gametocyte density , asexual parasite density , host age , and mosquito infection ( proportion of mosquitoes developing oocysts ) . A full description of these data is given in Figure 1—source data 1 . Gametocyte density was quantified using QT-NASBA , and the uncertainty in the density estimates ( the intra-assay variability ) was quantified by fitting a hierarchical model to the 16 independent dilution series . In the Burkina Faso dataset , multiple assays were carried out on the same blood sample ( an average of 2 . 63 assays per unknown blood sample ) . The mean gametocyte density from these multiple assays was taken to increase the accuracy of the estimates . The precise shape of the relationship was determined by fitting a range of different functional forms ( a modified constant , linear , power , hyperbolic , sigmoid , and Gompertz functions ) to data presented in Figure 1—source data 1 and statistically determining which gave the best fit . The proportion of mosquitoes developing oocysts , denoted by g , can be described by equation 3 ( 3 ) g=ϕ ( x^′ , κ ) fi ( x^′ ) ( 1+μz1+μz2+μz3 ) . Function fi ( x^′ ) describes how infection changes with increasing estimated gametocyte density with subscript i indicating the different functions tested ( whose equations are given in Table 1 ) ; μ is a vector of regression coefficients and z1 , z2 and z3 are dummy variables denoting asexual parasite density , age of the blood donor for the membrane feeing assays , and location of provenance , respectively . Asexual parasite density ( estimated by microscopy ) was categorized as being either undetectable ( z1 = 0 ) , low ( <1000 parasites per microlitre of blood , z1 = 1 ) or high ( ≥1000 parasites per microlitre of blood , z1 = 2 ) . Hosts were classified into two different age categories ( <6 years [children , z2 = 0] or ≥ 6 yr [older children , z2 = 1 ) . Mosquito infection was allowed to vary between Burkina Faso and Kenya ( z3 = 0 for Burkina Faso , z3 = 1 for Kenya ) . The function ϕ ( x^′ , η ) determines the shape of the relationship at very low gametocyte densities and is motivated by the observation that the malaria parasite can adjust its sex ratio to optimise transmission ( Reece et al . , 2008 ) . Evidence indicates that transmission is possible even at very low gametocyte densities ( Schneider et al . , 2007; Ouédraogo et al . , 2009 ) , so ϕ ( x^′ , η ) allows infection to rise very quickly with increasing gametocyte density at a rate determined by parameter η , ( 4 ) ϕ ( x^′ , η ) =1− ( 1+x^′2η ) − ( η+1 ) . Equation 4 was originally derived to describe the probability that a host would contain both male and female parasites according to the mean number of parasites and the aggregation ( overdispersion ) parameter of the negative binomial distribution ( May , 1977 ) . Reproduction in malaria is more complex than in the ( helminth ) system for which equation 4 was devised; hence , parameter η is not biologically interpretable . To investigate whether mosquito infection is best predicted by gametocyte density on the arithmetic or logarithmic scale , both hypotheses were tested in the model , substituting an estimate of gametocyte density on the logarithmic scale ( lnx′^ ) for x^′ in equations 3 and 4 and comparing model fits . The model quantifying the uncertainty in gametocyte density estimates was fitted at the same time as the regression determining the relationship between gametocyte density ( and other covariates ) and the proportion of mosquitoes infected using Bayesian Markov Chain Monte Carlo methods . Fitting the models simultaneously enables the uncertainty in the gametocyte density estimates to be reflected in the uncertainty of the shape of the relationship . Blood samples were randomly collected from 412 hosts from a single village in Burkina Faso ( Ouédraogo et al . , 2010 ) . QT-NASBA assays were run on the samples , and the methods described above were used to convert assay results into estimates of gametocyte density . A description of these data is given in Figure 2—source data 1 . To facilitate visual inspection of the age profile of gametocyte density , and to generate summary statistics , the function h ( A ) was fitted to these data to describe how the mean number of gametocytes per microlitre of blood changes with host age ( A ) ( 5 ) h ( A ) =τ+ ( ψA−τ ) exp ( −ωA ) . Parameters τ , ψ , and ω were estimated assuming a negative binomial error structure to account for the high degree of overdispersion ( aggregation ) observed in gametocyte density estimates ( using an overdispersion parameter that did not change with host age or gametocyte density ) . As above , a hierarchical model was used to estimate the gametocyte density and its associated uncertainty at the same time as the age profile allowing the full uncertainty of the shape of the function to be expressed . The probability that a mosquito biting a host of a particular age will become infected can be estimated using the above model together with estimates of the host gametocyte and asexual parasite density . This was done using the cross-sectional data from Burkina Faso to generate a proxy for the human reservoir of infection at this specific time in this location ( although it does not take into account how vector biting rate may vary with host age ) . It is important to use individual host estimates of gametocyte density instead of mean estimates as the high degree of gametocyte overdispersion among hosts may accentuate the non-linear relationship between gametocyte density and mosquito infection . For each of the 412 hosts , point estimates were obtained of the percentage of feeding mosquitoes ( taking a single full blood meal ) that would develop oocysts according to the age and the gametocyte/asexual parasite density of the host ( equation 3 ) . Equation 5 was used to fit a relationship between host age and their contribution towards human to mosquito transmission . This best-fit line was used to predict the percentage of overall transmission that originated from hosts who had gametocyte density estimates below the detection threshold of microscopy . Here , we define the detection threshold as the minimum ( positive ) density that can be estimated by sampling a certain volume of blood . It is assumed that gametocytes are counted against 500 white blood cells ( WBC ) and that there are on average 8000 WBC per µl of blood . This gives an average detection threshold of 16 gametocytes per microlitre of blood . This is a conservative estimate of the threshold density since densities above this value may still give false negative results both due reading errors or random sampling of gametocytes on the microscope slide . Care should be taken when interpreting the results as mosquito membrane feeding experiments were not performed with blood from adult hosts . Evidence indicates that antibodies associated with transmission-blocking activity may be lower in older age groups ( Drakeley et al . , 2006 ) , which could mean that adult hosts could have a greater potential to contribute to overall transmission . Bayesian Markov Chain Monte Carlo techniques were used to fit the models in OpenBUGS ( Lunn et al . , 2009 ) . This approach allows the uncertainties in the assessments of gametocyte density and mosquito infection rates ( generated by different numbers of mosquitoes being dissected ) to be taken into consideration , propagated into parameter posterior distributions . All parameters were assigned uninformative priors and were run until convergence was reached using standard methodology ( Gelman and Rubin , 1996 ) . The most parsimonious models were selected by comparing deviance information criteria ( DIC ) values ( the lowest value giving the most parsimonious yet adequate fit ) ( Spiegelhalter et al . , 2002 ) . Uncertainly around the best-fit line is indicated by showing the 95% range of 10 , 000 runs randomly sampled from the posterior distribution . The distribution of gametocytes between hosts was highly overdispersed , with an aggregation parameter of the negative binomial distribution of 0 . 185 ( 95% Bayesian credible interval [CI] , 0 . 16–0 . 21 ) .	Malaria is one of the world’s most deadly infectious diseases . The most severe form is caused by the parasite Plasmodium falciparum , which can reside within red blood cells and thus evade the human immune system . Plasmodium is transmitted between humans by mosquitoes . When a mosquito takes a blood meal from an individual infected with the parasite , the insect ingests Plasmodium gametocytes ( i . e . , eggs and sperm ) , and these go on to reproduce in the gut of the mosquito . These parasites then move to the mosquito’s salivary glands , to be injected into the next person whom the mosquito bites . Although malaria is both preventable and curable , the mortality rates in many African countries remain high , especially among children . Reducing the transmission of malaria to mosquitoes is one of the primary goals in the global effort to control and eliminate the disease . While a range of drugs and vaccines that specifically try to reduce transmission are in development , non-medical interventions such as mosquito nets and insecticide spraying can quickly and effectively reduce infection rates . Here , Churcher et al . examine the dynamics of human to mosquito transmission of P . falciparum , and report that the ease with which mosquitoes become infected is not directly proportional to the density of parasite gametocytes in human blood . They found that the transmission occurs readily at very low gametocyte densities . Moreover , the transmission rate remains relatively stable as the density increases , before increasing significantly when the density reaches around 200 cells per microlitre . Churcher et al . also challenge the assumption that children are mostly responsible for transmitting the malaria parasite by suggesting that , in certain locations , there is a more significant role for adults than previously assumed . By identifying the groups that contribute most to transmission , and targeting resources to reduce gametocyte density in those individuals , it could be possible to greatly reduce the number of infected mosquitoes and , therefore , the number of infected humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2013	Predicting mosquito infection from Plasmodium falciparum gametocyte density and estimating the reservoir of infection
Skilled performance is characterized by precise and flexible control of movement sequences in space and time . Recent theories suggest that integrated spatio-temporal trajectories are generated by intrinsic dynamics of motor and premotor networks . This contrasts with behavioural advantages that emerge when a trained spatial or temporal feature of sequences is transferred to a new spatio-temporal combination arguing for independent neural representations of these sequence features . We used a new fMRI pattern classification approach to identify brain regions with independent vs integrated representations . A distinct regional dissociation within motor areas was revealed: whereas only the contralateral primary motor cortex exhibited unique patterns for each spatio-temporal sequence combination , bilateral premotor areas represented spatial and temporal features independently of each other . These findings advocate a unique function of higher motor areas for flexible recombination and efficient encoding of complex motor behaviours . Skilled performance in music , speech , or sports often involves long sequences of movements , which demand a precise sequential activation of different muscles in time . The ordering of these muscle activations—and hence the ordering of the movements of different body parts in space—is here referred to as the ‘spatial feature’ of a sequence . Additionally , movement sequences are often characterized by a stereotypical temporal structure or rhythm—their ‘temporal feature’ . The latter can either emerge spontaneously as part of chunk formation ( Sakai et al . , 2003 ) , or be directly relevant to the goal of the sequence , as in musical performance , dance , or speech ( Shin and Ivry , 2002; Lewis and Miall , 2003; Repp , 2005; Kotz and Schwartze , 2010; Bläsing et al . , 2012; Grahn , 2012; Penhune and Steele , 2012 ) . One of the hallmarks of human motor performance is the ease with which experts can modify the temporal and spatial features of learned motor skills . For example , a pianist is able to play the same tune using different variations of the rhythm , and a fluent speaker can change separately the word order or the rhythmic profile of speech for effective communication . How is such flexibility in skilled actions achieved neurally ? There has been a long-standing debate on whether a dedicated representation of temporal structure of skilled movements exists , or whether it is tightly integrated with a representation of its spatial features ( Conditt and Mussa-Ivaldi , 1999; Shin and Ivry , 2002; Ullén and Bengtsson , 2003; Medina et al . , 2005; Spencer and Ivry , 2009; Ali et al . , 2013 ) . Recent work suggests that spatio-temporal trajectories of movements can be learned and produced by a dynamical network of neurons that encodes patterned muscle dynamics , instead of by representing different parameters of a movement sequence separately ( Laje and Buonomano , 2013; Shenoy et al . , 2013 ) . This neural implementation has been advocated for the primary motor and premotor cortices and implies that temporal features are stored inseparably from the specific movement trajectory trained . From this perspective , a spatial sequence performed with two different temporal profiles would constitute two distinct behaviours and demand the training of independent neural generators . Alternatively , the motor system may parse movement sequences into their constituent spatial and temporal features , which then are represented independently . Such an encoding scheme would explain the ability of both animals and humans to flexibly recombine learned temporal patterns with a new spatial sequence and vice versa ( Ullén and Bengtsson , 2003; Ali et al . , 2013; Kornysheva et al . , 2013 ) . Neurally , sequence representations are characterised by the occurrence of sequence-specific tuning . For example , neurons in the supplementary motor area ( SMA ) vary their firing rate for specific movement transitions and whole sequences of movements rather than for individual movements in a sequence ( Tanji and Shima , 1994 ) . Inactivating the SMA , the primary motor cortex , the putamen or the dentate nucleus in monkeys disrupts sequential behaviour whilst sparing individual actions ( Tanji and Shima , 1994; Shima and Tanji , 1998; Hikosaka et al . , 1999; Lu and Ashe , 2005 ) , with a similar effect demonstrated for the SMA/pre-SMA in humans with non-invasive stimulation techniques ( Gerloff et al . , 1997; Kennerley et al . , 2004 ) . Recent evidence in primates also argues for the existence of neurons tuned to specific temporal intervals between movements in the same area , with a subset also tuned to the position of an interval in a sequence ( Merchant et al . , 2013 ) . However , it remains unknown whether these neurons are simply part of a dynamical network that represents spatial and temporal features in an integrated manner , or whether independent populations of neurons encode spatial and temporal features in isolation . Here , we used fMRI to study the sequential tuning of individual voxels in the human brain . We hypothesized that following training , specific neuronal sub-populations will become differentially active for different sequences , as has been observed in neurophysiological studies for spatial sequence features ( Tanji and Shima , 1994 ) . If such sequence-specific tuning is sufficiently clustered , it should be detectable with the relatively low spatial resolution of fMRI ( Kamitani and Tong , 2005; Swisher et al . , 2010 ) . Using a classification approach to evaluate these subtle differences in the local patterns of brain activity during sequence production , a recent imaging study ( Wiestler and Diedrichsen , 2013 ) indeed showed that such sequential tuning can be detected in the human brain in a range of motor and premotor areas . This study , however , did not reveal whether and how these areas represented spatial or temporal features of sequences . To this end , we developed a visually paced motor learning paradigm . Participants were trained on nine sequences consisting of unique combinations of three spatial and three temporal features ( Figure 1 ) . Half the participants were trained on the right and half on the left hand to probe whether possible differences between hemispheres reflected hemispheric specialisation or the difference between contra vs ipsilateral encoding . First , by looking at behavioural generalisation , we show transfer of trained temporal and spatial features to new combinations . Second , by employing separate classification procedures of fMRI voxel activity patterns and testing for generalization of patterns across temporal or spatial contexts , we were able to dissociate independent spatial and temporal from integrated representation profiles across the human motor system . 10 . 7554/eLife . 03043 . 003Figure 1 . Subjects were trained on nine sequences , which were unique combinations of three spatial ( finger order ) and three temporal sequence features . Sequences were presented in mini-blocks of three trials in a row . Each sequence began with the presentation of a warning cue ( square ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 003 We used a visually cued motor learning task to induce and assess the acquisition of sequences involving finger movements ( Kornysheva et al . , 2013 ) . Subjects were trained to produce nine finger sequences that were unique combinations of three temporal and three spatial sequence features ( Figure 1 ) randomly generated for each subject . Half the participants trained and performed the sequences with the left , the other half with the right hand . Over the course of 3 days of training the force responses on the keyboard triggered by the visual stimulus became faster ( Figure 2A ) . The average reaction time ( RT ) decreased from 410 ms ( SD: 70 ) to 288 ms ( SD: 48 , Figure 2B ) . To assess the specificity of the improvement in motor performance to the trained sequences , we also tested participants on sequences composed of untrained spatial and temporal features . Subjects also reduced the RT for untrained sequences from 456 ms ( SD: 90 ) to 346 ms ( SD: 64 ) , which suggests a general effect of visuomotor learning . However , the reduction was significantly smaller than that for the trained sequences ( F ( 1 , 30 ) = 13 . 342 , p=0 . 001 ) , and there was no interaction with the group ( right-hand-trained vs left hand , F ( 1 , 30 ) = 1 . 235 , p=0 . 275 ) . Overall error rates across conditions paralleled the RT findings ( Figure 2—figure supplement 1 ) . For trained sequences the error rate reduced from 46 . 1% ( SD: 26 . 7 ) to 6 . 3% ( SD: 4 . 9 ) . For untrained from 52 . 1% ( SD: 31 . 6 ) to 29 . 5% ( SD: 21 ) , suggesting that the RT findings were not due to a change in the speed-accuracy trade-off . During the fMRI session we only tested the nine trained sequences . The RTs increased as compared to the end of training ( Figure 2B , t ( 31 ) = 6 . 57 , p<0 . 0001 ) . This occurred despite the subjects' familiarity with a supine position on a mock MRI bed during training and may be related to the fMRI noise during the performance of the task , the lack of any auditory movement feedback occurring , as well as a more restricted mobility in an MRI environment . Importantly , however , across subjects individual RTs during fMRI session strongly correlated with RTs for trained sequences in the last training block ( r = 0 . 717 , p<0 . 0001 ) , indicating that the responses reflected consistent measures of behaviour . 10 . 7554/eLife . 03043 . 004Figure 2 . Reaction time ( RT ) results . ( A ) Two trial examples of force traces show faster finger responses to visual stimuli after ( ‘post’ ) as opposed to the beginning of training ( ‘pre’ ) . ( B ) Subjects showed general and sequence-specific learning during the training of the combined temporal and spatial sequences . The RT remained relatively stable across the fMRI session runs , albeit overall higher than at the end of training outside the MRI environment . ( C ) Post-test results . Left panel: repeating sequences nine times in the test phase yielded an immediate RT decrease for trained spatial sequences ( blue ) relative to untrained sequences ( black ) , and only delayed RT differences for trained temporal sequences ( red ) , in line with previous results ( Kornysheva et al . , 2013 ) . Right panel: a boxplot displaying RT results across subjects and all sequence repetitions in the post-test revealed significant RT advantages for the trained sequence , as well as the trained spatial and trained temporal feature conditions when compared to untrained sequences , suggesting that both the finger order and their relative timing were represented independently . A double asterisk ( ** ) indicates a significant difference between conditions with p<0 . 01 . Errorbars in the lineplots and maximum and minimum RTs ( boxplot whiskers ) are corrected for differences in the mean RT across individuals , and therefore represent the interindividual variability of relative RT differences across conditions . In the boxplot , upper and lower edges signify the 75th ( third quartile ) and 25th percentile ( first quartile ) , respectively . The median is designated as a horizontal white or black line in the box . Outliers ( equal or above 3interquartile range above the third quartile or below the first quartile ) are depicted as filled circles , suspected outliers ( 1 . 5interquartile range above the third quartile and below the first quartile ) are depicted as unfilled circles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 00410 . 7554/eLife . 03043 . 005Figure 2—figure supplement 1 . Error rate paralleled the RT results during the training ( A ) and fMRI showing clear sequence-specific advantages for the trained sequences , as well as sequences which retained the spatial features . Note that in contrast to the RT results in Figure 2C , the post-test ( B ) did not show a decrease of error rate for sequences with trained temporal , but new spatial features consistent with our previous behavioural findings ( Kornysheva et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 005 On the day following fMRI , we conducted a post-test to assess whether participants would be able to utilize both the learned spatial and temporal features when these were paired with novel untrained features . Based on previous studies ( Shin and Ivry , 2002; O’Reilly et al . , 2008; Brown et al . , 2013; Kornysheva et al . , 2013 ) , we expected evidence only for spatial , but not for temporal feature transfer in the first three trials . Indeed , during the training phase , in which each sequence was repeated only three times in a row ( Figure 2B ) , and during the first trials in the post-test ( Figure 2C ) the temporal transfer condition was not performed faster than untrained control sequence . However , consistent with two previous experiments ( Kornysheva et al . , 2013 ) , a delayed RT advantage for the temporal transfer condition emerged after a few repetitions of the new sequence combination ( Figure 2C , left panel ) . Averaged over all nine repetitions in the post-test , sequences which combined a trained temporal ( F ( 1 , 28 ) = 12 . 963 , p=0 . 001 ) or spatial ( F ( 1 , 28 ) = 20 . 830 , p=0 . 0001 ) feature with an untrained feature were performed faster than a repetition of a completely novel sequence ( Figure 2C , right panel ) . This suggests that training of timed finger sequences automatically results in independent storage of its spatial and temporal features . ANOVAs also revealed that these effects did not differ between the left and right-hand-trained groups ( p>0 . 127 ) . These results replicate earlier findings and can be explained by a model in which the temporal representation acts as a multiplicative go signal on a concrete spatial plan ( Kornysheva et al . , 2013 ) ; while both temporal and spatial sequences are represented independently , the temporal representations act at the output stage as a modulator on a spatial expectation . Thus , without a spatial expectation , temporal knowledge does not have an effect . For this reason , temporal transfer cannot manifest itself immediately , but only after a plan for the next movement has been formed ( Hikosaka et al . , 1999; Penhune and Steele , 2012 ) . Compared to rest , the visually cued production of trained finger sequences yielded increased activity in motor and visual areas . To analyse the imaging data from the left-hand and right-hand groups together , we grouped the hemispheres according to whether they were ipsi- or contralateral to the moving hand . As would be expected from earlier studies of visually trained skilled sequence production ( Wiestler and Diedrichsen , 2013 ) , increased activation was observed in the contralateral primary motor cortex ( hand knob area of M1 ) extending into the dorsal premotor cortex ( PMd ) , the ipsilateral lobule V-VI of the cerebellum , bilateral ( pre ) supplementary motor area ( SMA/pre-SMA ) , bilateral ventral premotor cortex ( PMv ) , left dorsolateral prefrontal cortex ( dlPFC ) , and visual areas—left posterior precuneus and secondary visual cortex ( V2 ) , as well as right associative visual and occipito-temporal cortices . To determine the tuning characteristic of population of voxels in different cortical areas , we utilized a system of four different cross-validated classification analyses ( ‘Materials and methods’ ) . The overall classifier ( Figure 3A ) was trained and tested on all nine sequences ( albeit from different imaging runs ) , and distinguished between all sequences independently of their component features . As confirmed by simulations of voxel activity patterns ( Figure 3E , black bar ) , this classifier shows above-chance accuracy whenever there are any reliable differences between the nine activity patterns , independently of the underlying tuning functions , and therefore can detect any sequence representation . 10 . 7554/eLife . 03043 . 006Figure 3 . Four classification procedures were employed to classify the voxel pattern of each searchlight ( 160 voxels , here reduced to 16 units for illustration purposes ) . ( A ) To test whether a voxel searchlight contained any sequential information , the overall classifier distinguished between the nine sequences independently . Classification was always cross-validated across imaging runs ( ‘Materials and methods’ ) . ( B ) To determine encoding of the spatial feature , the classifier was trained on data involving only two of the three temporal sequences , and tested on trials from a left-out imaging run in which the spatial sequences were paired with the remaining temporal sequence . ( C ) The temporal classifier followed the same training-test principle , but in an orthogonal direction . ( D ) The integrated classifier detected nonlinear encoding of the unique combinations of temporal and spatial features that could not be accounted for by linear superposition of independent encoding . The spatial and temporal mean patterns for each run were subtracted from each combination , respectively , to yield a residual pattern , which was then submitted to a nine-class classification . ( E ) Classification accuracy of the four classifiers on simulated patterns ( z-transformed , chance level = 0 ) . Results indicate that the underlying representation can be sensitively detected by contrasting the overall , temporal , spatial , and integrated classifiers . Importantly these classification procedures can differentiate between a non-linear integrated encoding of the two parameters as opposed to the overlap of independent temporal and spatial encoding . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 006 In contrast , a region that contains an independent representation of the order of finger presses in space should show consistent activity patterns for spatial features of sequences , independent of their temporal features . To detect such patterns , a spatial classifier was trained on trials where the three spatial features were combined with two different temporal features , and then tested on data in which these spatial features were combined with a new temporal feature ( Figure 3B ) . Therefore , this classifier can only yield above-chance accuracy , when the voxels show consistent tuning for the spatial features , independently of the temporal structure of finger sequences . The temporal classifier ( Figure 3C ) was defined in the same way , but classified temporal features combined over two spatial features , and tested on the left out spatial feature . Finally , the integrated classifier ( Figure 3D ) tested for a non-linear interaction between the tuning for temporal and spatial features by subtracting out information that could be explained by each component ( spatial or temporal ) separately . The classifier therefore only detects regions that show unique , idiosyncratic patterns for each of the nine sequences . Our pattern simulations demonstrated that this set of four classifiers could sensitively reveal the type of the underlying representation . Importantly , we could distinguish between a region containing a unique , integrated , representation of the two sequential features ( in which case only the integrated classifier should be above chance , Figure 3E , ‘Integrated encoding’ ) and a region containing a superposition of independent temporal and spatial sequence representations ( in which case both independent classifiers within one region are above chance , Figure 3E , ‘Independent encoding’ ) . To determine which cortical regions showed any differences between the activity patterns for the nine unique combinations of temporal and spatial features , we utilized the overall classifier . We found significant above-chance classification accuracy in the hand knob area of the contralateral M1 extending into primary sensory area ( S1 ) , PMd and PMv , the contralateral anterior insula , the ipsilateral Lobules V–VI of the cerebellum ( Figure 4—figure supplement 1 ) , the bilateral SMA , superior parietal and extrastriate cortices , the ipsilateral medial M1 , and inferior parietal cortex ( Figure 4A; Table 1 ) . Overall , the above threshold clusters were more widespread on the contralateral than on the ipsilateral side . These results were in line with the regions found in an earlier study ( Wiestler and Diedrichsen , 2013 ) , which used faster finger sequences that were produced from memory . Thus , our results indicate that similar encoding can be found for visually paced sequences involving longer temporal intervals between finger presses . From the overall classifier , however , we cannot yet determine how different features of the sequences were encoded . 10 . 7554/eLife . 03043 . 007Figure 4 . Searchlight classification results shown on an inflated representation of the cortical surface . ( A ) Group t-values indicate regions in which the overall classifier performed significantly above chance . ( B ) Significant group-level above chance classification of spatial ( blue ) , temporal ( red ) , and integrated ( green ) classifiers . Results are presented at an uncorrected threshold of t ( 31 ) > 3 . 37 . p<0 . 001 . CinS , cingulate sulcus; CS , central sulcus; IPS , intraparietal sulcus; PoSC , postcentral sulcus; SFS , superior frontal sulcus . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 00710 . 7554/eLife . 03043 . 008Figure 4—figure supplement 1 . Searchlight classification results in the cerebellum . Yellow scale indicates consistent above chance classification across the group by the overall classifier in ipsilateral Lobules V/VI ( Peak t ( 31 ) = 5 . 72 , pcluster<0 . 001 , MNI coordinates: 34 , −54 , −23 ) . Temporal , spatial , and integrated classifiers did not show significant encoding at an uncorrected threshold of t ( 31 ) > 3 . 37 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 00810 . 7554/eLife . 03043 . 009Figure 4—figure supplement 2 . Mean searchlight classification accuracy results displayed as in Figure 4 , split by group trained on the right and left hand . ( A ) Classification accuracy results ( z-values ) . Yellow scale indicates above chance classification by the overall classifier . Results are presented a threshold of z = 0 . 8 ( chance level: z = 0 ) . ( B ) Blue , red , and green colours stand for above chance classification of spatial , temporal , and integrated patterns , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 00910 . 7554/eLife . 03043 . 010Figure 4—figure supplement 3 . Classification accuracy of the main response function and temporal derivative . ( A ) An example of the main response function and temporal derivative for three sequence repetitions ( one mini-block ) . Note that in contrast to the temporal derivative that captures the temporal evolution of each sequence by returning to baseline between each sequence , the main response function used in the subsequent classification analysis remains elevated across the three sequence repetitions . ( B ) Classification accuracy of the main response vs the temporal derivative estimates in contra- and ipsilateral M1 and PMd . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 01010 . 7554/eLife . 03043 . 011Figure 4—figure supplement 4 . Maximum force for finger 1 ( thumb ) to 5 ( pinkie ) during fMRI . Each line depicts each participant's mean force across fingers for each of the nine trained sequences . Sequence specific idiosyncratic force profiles could be used to classify the sequences above chance ( p<0 . 001; acc: z-transformed accuracy , chance level = 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 01110 . 7554/eLife . 03043 . 012Table 1 . Areas showing above-chance classification accuracy for the decoding of sequences and their spatial and temporal featuresDOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 012MNIClassifierArea ( Brodmann area ) Area ( cm2 ) PclusterPeak t ( 31 ) XYZOverallContralateral M1/PMd/PMv ( BA4/BA6 ) 43 . 25<0 . 0019 . 40−36−2253 Superior parietal ( BA40/BA7 ) 15 . 80<0 . 0015 . 82−32−5456 Extrastriate vis cortex ( BA18 ) 8 . 53<0 . 0015 . 75−27−90−1 Extrastriate vis cortex ( BA19 ) 2 . 350 . 0025 . 57−36−8327 SMA ( BA6 ) 3 . 86<0 . 0015 . 27−8−12572 . 220 . 0024 . 73−42−82−13 Anterior insula ( BA48 ) 1 . 350 . 0364 . 35−35−10−21 . 810 . 0084 . 32−42212 Occipitotemporal area ( BA37 ) 1 . 750 . 014 . 10−40−62−11Ipsilateral Extrastriate vis cortex ( BA19 ) 20 . 84<0 . 0015 . 7734−89−6 PMd ( BA6 ) 12 . 72<0 . 0015 . 2421−1260 Superior parietal ( BA5 ) 3 . 76<0 . 0015 . 1919−5561 Superior parietal ( BA7 ) 3 . 27<0 . 0014 . 9230−5946 Medial M1 ( BA4 ) 4 . 84<0 . 0014 . 8614−4059 Occipitotemporal area ( BA37 ) 1 . 830 . 0144 . 5545−706 Extrastriate vis cortex ( BA19 ) 2 . 550 . 0024 . 2232−75−11 SMA/Pre-SMA ( BA6/BA32 ) 1 . 390 . 053 . 9381849IntegratedContralateral M1 ( handknob , BA4 ) 5 . 89<0 . 0015 . 39−33−2359SpatialContralateral Superior parietal ( BA7 ) 10 . 00<0 . 0016 . 93−31−5660 PMd ( BA6 ) 9 . 66<0 . 0016 . 20−31−1353 Inferior parietal ( BA40 ) 6 . 00<0 . 0015 . 78−39−3637 SMA ( BA6 ) 2 . 690 . 0025 . 70−9154Ipsilateral PMd ( BA6 ) 5 . 23<0 . 0014 . 6829−247 Inferior parietal/occipital3 . 98<0 . 0014 . 3133−6634 ( BA39/BA19 ) TemporalContralateral SMA ( BA6 ) 3 . 47<0 . 0015 . 74−8948 PMd ( rostral BA6 ) 6 . 38<0 . 0015 . 53−24−1558 Extrastriate vis cortex ( BA18 ) 11 . 00<0 . 0014 . 58−29−92−5 Extrastriate vis cortex ( BA19 ) 2 . 350 . 0064 . 34−35−8210Ipsilateral PMd ( rostral , BA6 ) 9 . 78<0 . 0015 . 9823−949 PMv ( BA6 ) 5 . 19<0 . 0015 . 2851−624 Posterior cingulate ( BA23 ) 2 . 440 . 0064 . 839−3031 Pre-SMA/anterior cingulate2 . 730 . 0044 . 7993442 ( BA32 ) PMd ( caudal BA6 ) 1 . 750 . 0344 . 6620−2657 Extrastriate vis cortex ( BA19 ) 1 . 750 . 0344 . 1342−851Results of surface-based random effects analysis ( N = 32 ) with an uncorrected threshold of t ( 31 ) > 3 . 37 , p<0 . 001 . p ( cluster . ) is the cluster-wise p-value for the cluster of that size . The p-value is corrected over the cortical surface using the area of the cluster ( Worsley et al . , 1996 ) . The cluster coordinates reflect the location of the cluster peak in MNI space . We then determined which regions encoded the nine sequences with a unique activity pattern without any consistent patterns for temporal or spatial features/components alone , using the integrated classifier . The only region that carried such integrated , non-additive encoding was the cortical output area—the M1 ( handknob area ) contralateral to the hand involved in producing the sequences ( Figure 4B , green , Table 1 ) . This region was clearly visible in the contralateral motor cortex of both the right and the left-hand groups ( Figure 4—figure supplement 2 ) . To test the encoding in motor-related areas in more detail , we analysed the data in four symmetrically defined regions of interest ( ROIs ) : primary motor cortex ( M1 ) , dorsal ( PMd ) , ventral premotor cortex ( PMv ) , and supplementary motor area ( SMA ) . Mean integrated encoding reached only significant above chance level in the contralateral M1 ( Figure 5 , p=0 . 0005 , Bonferroni correction at p=0 . 05/8=0 . 0063 ) , but not in any of the premotor regions ( p>0 . 155 ) . 10 . 7554/eLife . 03043 . 013Figure 5 . Classification accuracy ( z-values ) in anatomically and symmetrically defined motor regions of interest ( ROI ) . Integrated classification accuracy was significant above chance level in contralateral M1 only , whereas temporal and spatial classifiers showed higher accuracy in premotor areas , in a partly overlapping manner . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 01310 . 7554/eLife . 03043 . 014Figure 5—figure supplement 1 . Classification accuracy as in Figure 5 split by group trained on the right and left hand . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 014 We then searched for regions in which voxels would show consistent tuning for temporal or spatial features of the sequence , independently of the respective other component . For the spatial and temporal classifiers we found highest encoding outside M1 , particularly in premotor , as well as in parietal areas . The spatial classifier detected consistent patterns related to the order of the finger presses that remained unchanged when executed with different temporal features . Spatial classification accuracy was significantly above chance in the contralateral SMA , bilateral PMd , as well as superior and inferior parietal lobes ( Table 1 ) . The temporal classifier detected representations of temporal sequences , which did not change across different spatial sequences ( orthogonal to the spatial classification analysis ) . Clusters in bilateral PMd , contralateral SMA , ipsilateral PMv , anterior and posterior cingulate , and bilateral extrastriate visual areas were significant after correction for multiple tests ( Table 1 ) . These results suggest that while contralateral M1 exhibits mostly integrated encoding of temporal and spatial sequence features , the two sequence components are represented independently in the bilateral premotor cortex . An ROI ( M1 vs premotor ) × hemisphere ( contralateral vs ipsilateral ) × classifier ( integrated vs independent spatial and temporal ) × group ( right vs left hand ) mixed ANOVA indeed revealed a significant ROI × hemisphere × classifier interaction ( F ( 1 , 30 ) = 12 . 808 , p=0 . 001 ) . This effect did not interact with group ( p=0 . 75 ) , suggesting that although the effect was less pronounced for left-hand-trained participants , the distribution of integrated and independent sequence encoding was similar for left and right hand sequence production ( Figure 5—figure supplement 1 ) . The difference in representation can be better appreciated in Figure 6A , which shows the level of temporal , spatial , and integrated sequence feature encoding on a cross-section running from rostral PMd to the caudal end of the occipito-parietal junction ( cf . Figure 6—figure supplement 1 for profiles split by right and left hand groups ) . In contrast to temporal and spatial encoding , integrated encoding peaked at the level of the central sulcus in the contralateral hemisphere . To test for differences in the distribution of integrated and independent encoding , we used a Center of Gravity ( CoG ) analysis . We determined the CoG for integrated and independent ( averaged over spatial and temporal ) classification accuracies on the precentral part of the cross-section . Indeed , we found a more caudal CoG for integrated , and a more rostral CoG for independent encoding in the contralateral hemispheres of both left- and right-hand groups ( Figure 6C ) . A hemisphere ( contralateral vs ipsilateral ) × classifier ( integrated vs independent ) × group ( right vs left hand ) mixed ANOVA revealed a hemisphere × classifier interaction ( F ( 1 , 30 ) = 6 . 417 , p=0 . 017 ) , but no interaction with group ( p=0 . 409 ) . For the postcentral part of the cross-section , we found the reverse pattern—a more caudal CoG for independent as compared to integrated encoding ( Figure 6D ) . Again the hemisphere ( contralateral vs ipsilateral ) × classifier ( integrated vs independent ) interaction was highly significant ( F ( 1 , 30 ) = 13 . 394 , p=0 . 001 ) , and did not interact with group ( p=0 . 088 ) . These results therefore clearly suggest a difference in how primary motor and premotor , as well as parietal areas represent spatio-temporal finger sequences . 10 . 7554/eLife . 03043 . 015Figure 6 . Distribution of encoding in cortical cross-sections . Shown are profiles of integrated ( green ) , temporal ( red ) , and spatial ( blue ) classification accuracy ( z-values ) , ( A ) on a cross-section running from rostral premotor cortex , through the hand area , to the occipito-parietal junction and ( B ) on a profile running from the ventral , through the dorsal premotor cortex , to the SMA ( BA 6 ) . ( C ) Center of gravity ( CoG ) analysis across the precentral part ( between rostral PMd and central sulcus ) of the profile in A shows that independent temporal and spatial classification accuracy ( mean in purple ) is represented more rostrally to integrated classification accuracy in the contralateral hemisphere across right and left hand groups . ( D ) CoG analysis across the postcentral part of the profile shows the opposite pattern to C with independent classification accuracy represented more caudally , further away from the CS towards the parietal cortex as compared to integrated classification accuracy . This gradient was found in the contralateral hemisphere across right and left hand groups . ( E ) CoG analysis across the lateral premotor cortex shows a slight ventral bias for temporal compared to spatial classification accuracy across hemispheres and groups . BA , Brodmann area; CoG , center of gravity; IPS , inferior parietal sulcus; m , medial wall; M1 , primary motor cortex; OPJ , occipito-parietal junction; PMd , dorsal premotor cortex; PMv , ventral premotor cortex; PoCS , postcentral sulcus; PreCS , precentral sulcus ventral premotor cortex; S1 , primary sensory cortex; SFS , superior frontal sulcus; SMA , supplementary motor area . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 01510 . 7554/eLife . 03043 . 016Figure 6—figure supplement 1 . Distribution of encoding as in Figure 6 split by group ( A ) on a cross-section running from rostral premotor cortex , through the hand area , to the occipito-parietal junction and ( B ) on a cross-section running from the ventral , through the dorsal premotor cortex , to the SMA ( BA 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 016 Although temporal and spatial features were represented independently in premotor and parietal cortex ( i . e . , these regions showed specific activity patterns for one feature , independent of the respective other feature ) , spatially these representations overlapped to a certain degree , especially in caudal PMd ( Figure 4B ) . It is only through the use of multivariate analysis techniques that we were able to distinguish such overlap of independent representations from the integrated representation in the primary motor cortex . Inspection of the temporal and spatial representation maps , however , also suggests a difference in where temporal and spatial features are localised . Especially in the ipsilateral premotor cortex , it appears that temporal encoding was more pronounced in the ventral , whereas spatial representations are more evident in the dorsal premotor cortex . This gradient can be seen more clearly on a profile plot of encoding in the premotor cortex ( Figure 6B ) . A CoG across lateral premotor cortex indeed revealed that temporal encoding was centred more ventrally and spatial encoding more dorsally ( Figure 6E ) . A hemisphere ( contralateral vs ipsilateral ) × classifier ( temporal vs spatial ) × group ( right hand trained vs left hand ) mixed ANOVA showed a main difference between classifier ( F ( 1 , 30 ) = 5 . 836 , p=0 . 022 ) , but no interaction with group ( p=0 . 687 ) or hemisphere ( p=0 . 678 ) . In summary , we found that outside of primary motor cortex , both temporal and spatial features of a learned movement sequence are represented independently , albeit in partly overlapping areas . Furthermore , we found a functional gradient with temporal representation being stronger in ventral and spatial representation stronger in dorsal premotor areas . One potential concern regarding multivariate analysis for fMRI data is that the voxel patterns used for classification may simply reflect differences in the temporal profile of activation rather than differences in the spatial activation patterns related to sequence-specific encoding . Since voxels in M1 are known to show differential tuning for isolated finger movements ( Diedrichsen et al . , 2013b ) , some voxels may show higher activity early in each sequence , while others would peak late in the sequence . However , there is good evidence to suggest that this effect cannot account for the results reported here . First , if the classification reflected tuning to individual finger movements , all classifiers should show the highest accuracy in the contralateral hand area of S1 and M1 , where finger representations are the strongest ( Wiestler et al . , 2011 ) . This , however , could not be observed , as evidenced by a differential distribution of temporal , spatial , and integrated encoding across the brain . Second , each sequence trial was produced three times in a row ( mini-block ) , so that the main response function was elevated across the execution of three sequences and did not return to baseline before the end of the last trial ( Figure 4—figure supplement 3A ) . It therefore should be insensitive to differences in the temporal activation profile within each execution . In contrast , the temporal derivative of the main response function included in our first level general linear model ( GLM ) , captured variations of the temporal profile within each trial . For example , the derivative would indicate whether a voxel was activated more in the early or late phase of each sequence . For the main classification analysis , we discounted the derivative , as we wanted to isolate differences in spatial activity patterns , rather than different temporal profiles . However , the response derivate also allowed us to test for the information contained in different temporal activation profiles ( Figure 4—figure supplement 3B ) . As expected , based on the derivative , we observed increased classification accuracy of independent spatial and to some degree also temporal features in M1 , while in premotor areas such as the PMd , the independent classifiers performed much worse , in line with the evidence suggesting weaker finger representations in premotor areas ( Diedrichsen et al . , 2013b ) . The factors response type ( main response vs derivative ) , ROI ( M1 vs PMd ) , and hemisphere ( contralateral vs ipsilateral ) showed a significant interaction between the three factors for both spatial ( F ( 1 , 31 ) = 9 . 165 , p=0 . 005 ) and temporal encoding ( F ( 1 , 31 ) = 5 . 395 , p=0 . 027 ) . This suggests that the voxel patterns based on the main response estimates are unlikely to reflect differences in the temporal profile of the observed response . Finally , the above chance classification could reflect simple differences of movement parameters during the sequence execution rather than sequence encoding ( Todd et al . , 2013 ) . Despite employing the same fingers and the same temporal intervals across all nine sequences , as well as by controlling the number of runs and jumps between finger digits and intervals ( ascending or descending interval transitions ) , in some subjects minor , but systematic finger force differences between the trained sequences occurred , such as more force on the thumb in one sequence and on the index finger in a different sequence ( Figure 4—figure supplement 4 ) . Accordingly , force on the five fingers could be used to reliably classify the nine-trained sequences ( mean zacc = 2 . 25 , t ( 31 ) = 10 . 914 , p<0 . 001 ) . Importantly , however , the strength of force differences did not correlate with classification accuracy in contralateral M1 ( r = −0 . 210 , p=0 . 257 ) , such that simple differences in finger forces could not account for the finding of integrated feature encoding here . Instead , we hypothesized that the reported multivariate encoding of sequences in contralateral M1 would covary with the degree with which that participant showed sequence-specific learning , defined as the RT advantages for trained as opposed to untrained sequences at post-test . Indeed , the classification accuracy correlated with the amount of sequence-specific learning , ( r = 0 . 468 , p=0 . 008 ) . Thus , participants with higher behavioural learning effects also showed higher classification accuracy ( Figure 7A ) . No positive relationship could be revealed for ipsilateral M1 and either force differences or sequence learning ( r <−0 . 222 , p>0 . 186 , Figure 7B for correlation with sequence learning ) . This further supports that encoding in contralateral M1 is likely to be related to the sequential skill level . 10 . 7554/eLife . 03043 . 017Figure 7 . Correlation between sequence-specific learning ( RT advantages for trained relative to untrained sequences in the post-test ) and overall encoding in M1 . Learning significantly covaried with the overall encoding in the contralateral M1 r = 0 . 47 , p=0 . 008 ( A ) , but not in the ipsilateral M1 r = −0 . 25 , p=0 . 169 ( B ) . The correlation of sequence learning and contralateral encoding in M1 remained significant when taking all , and not only the task-activated voxels in contralateral M1 into account , ( r = 0 . 369 , p=0 . 041 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03043 . 017 Our study employed fMRI multivoxel pattern analysis that reflects the differential tuning of individual voxels ( Kamitani and Tong , 2005; Kriegeskorte et al . , 2006 ) to identify neural representations of spatial and temporal finger sequence features . We were able to dissociate independent feature representations in which voxel patterns related to spatial and temporal sequence features combined linearly , from integrated feature representations in which each spatio-temporal combination was associated with a unique activity pattern . We demonstrate that only the output stage of the cortical motor hierarchy , the primary motor cortex ( M1 ) contralateral to the moving hand , encoded spatio-temporal features of finger sequences in an integrated fashion . In contrast , bilateral medial and lateral premotor cortices showed partly overlapping , but mutually independent representations of the spatial and temporal features . The independent encoding of sequence features in higher order motor areas paralleled our behavioural findings—the nervous system's ability to flexibly transfer both spatial and temporal features from trained to new sequence contexts . The integrated sequence encoding found in the contralateral M1 is in line with electrophysiological data showing that 40% of neurons in the primary motor area in monkeys can exhibit tuning to sequences of muscle commands ( Matsuzaka et al . , 2007 ) , evidence that inactivation of M1 via muscimol can selectively disrupt sequential behaviour ( Lu and Ashe , 2005 ) , as well as previous sequence learning studies in humans ( Karni et al . , 1995; Penhune and Doyon , 2005; Steele and Penhune , 2010 ) . We found that the overall sequence encoding in the contralateral M1 covaried with the amount of behavioural advantages for the trained sequences , suggesting that our analysis uncovered skill-dependent representations . The fact that each spatio-temporal sequence combination had its unique activity pattern in M1 is consistent with a dynamical systems view which proposes that each movement is controlled by a subpopulation of neurons that form a dynamical network ( Laje and Buonomano , 2013; Shenoy et al . , 2013 ) . Instead of representing movement features separately , these networks are assumed to produce complex movement patterns based on a neural state-space trajectory , which is determined by the internal connectivity and external input to the circuitry ( Shenoy et al . , 2013 ) . Accordingly , for each unique spatio-temporal sequence a slightly different distribution of neurons is activated in M1 which in turn cause distinct voxel activity patterns for each of the studied sequence combinations ( Kamitani and Tong , 2005; Kriegeskorte et al . , 2006 ) . This integrated encoding in M1 is in line with our model , which suggests that the temporal and spatial sequence features are integrated non-linearly in the nervous system ( Kornysheva et al . , 2013 ) . While adequate for learning and producing specific spatio-temporal sequences , integrated encoding such as found in M1 alone would not allow the system to use learned spatial or temporal features independently . However , subjects showed behavioural advantages for untrained sequence in which only one of the trained features ( spatial or temporal ) was retained . With spatial and temporal features of movement trajectories emerging from the same local circuits , an integrated representation as proposed in dynamical systems models ( Shenoy et al . , 2013 ) cannot explain the flexible transfer or independent adaptation of spatial and temporal features reported here and in previous studies ( Ullén and Bengtsson , 2003; Ali et al . , 2013; Kornysheva et al . , 2013 ) . In contrast , our results indicate that higher order motor areas ( lateral and medial premotor cortices ) parsed the sequences into the two constituent features , in line with the modularity and flexibility we observed in behaviour . The independent albeit partly overlapping spatial and temporal encoding suggests a modular feature-separating storage for movement production . Although our experiment provides both behavioural and imaging evidence for the independent representation of temporal and spatial features of movements , it is possible that not all classes of movements lend themselves to such a separation . For example , the temporal profile of a force perturbation during a reaching movement is learned inseparably from the whole spatio-temporal trajectory ( Conditt and Mussa-Ivaldi , 1999 ) . One critical factor that distinguishes these classes may be whether movement kinematics are continuous as in reaching movements or fall into discrete phases , as induced by the current task ( Ivry et al . , 2002 ) . In contrast , for discrete sequential movements , as studied here , the evidence for an independent representation of the temporal structure is compelling . Importantly , our behavioural finding is unlikely to be an effect artificially induced by the orthogonal design , in which all three spatial sequence features were crossed with three temporal features , since flexibility in independently adapting spatial or temporal features of sequences has also been observed in a previous study which involved the training of only one particular spatio-temporal combination ( Kornysheva et al . , 2013 ) . Previous data demonstrated that the premotor cortex modulates its activity once movement sequences are made more complex in either their spatial or temporal structure ( Bengtsson et al . , 2004 ) , or in the presence of both features ( Sakai et al . , 2002; Brown et al . , 2013 ) . However , these studies did not establish how temporal and spatial features were encoded . Here , we address this issue , starting from a recent study that showed that voxel pattern analysis can uncover sequence representations in motor and premotor areas , which are specifically enhanced for trained sequences ( Wiestler and Diedrichsen , 2013 ) . This classification analysis was now extended to identify regions with voxel patterns related to spatial and temporal sequence features that were represented independently from the respective other feature . Our representational analysis allowed us for the first time to distinguish between integrated vs independent representations , even if the independent spatial and temporal representations overlapped spatially . While we validated this approach by voxel pattern simulations , the analysis hinges critically on the assumption that activity in independent neuronal populations within a single voxel combine additively in the observed BOLD signal . Is this assumption justified ? Functional imaging signals show clear non-linearities when studied over a large dynamic range ( Logothetis et al . , 2001 ) . However , the activity of single voxels only varied very slightly between different sequences in our paradigm . For this restricted range , therefore , we can be relatively confident that a linear approximation is reasonable ( Diedrichsen et al . , 2013a ) . Furthermore , because any non-linearity between neural signal and hemodynamic response should be relatively homogenous across adjacent cortical motor areas , the distinct representational dissociation found between M1 and PMd is likely to reflect neural rather than hemodynamic differences . How can the independent encoding of spatial and temporal features be implemented on a single cell level ? Neurons in primate premotor cortices have been shown to be tuned to specific spatial transitions between movement elements and whole sequences of movements increasing their firing rate prior to a movement based on its sequential context ( Mushiake et al . , 1991; Tanji and Shima , 1994; Shima and Tanji , 2000 ) . Such units have been suggested to be an important component in the organisation of skilled sequential behaviours , however whether timing between movements modulates the same neurons has not been systematically explored . Our findings of independent representations of the spatial and temporal features predict that the representation of movement order in space first reported by Shima and Tanji in the SMA ( Tanji and Shima , 1994 ) is likely to be independent of the exact timing of individual movements in the sequence . In other words , changing the temporal intervals between movements in a sequence should not interfere with the tuning of the neuron to a specific spatial movement transition or sequence . At the same time our findings predict independent encoding of the temporal features of movement sequences within the same areas as the spatial features . In the medial premotor cortex of monkeys , Merchant et al . ( 2013 ) found evidence for the encoding of sub-second intervals sensitive to the sequential context within a synchronization-continuation task . In addition to neurons tuned to interval duration , the authors found cells that are tuned to both a specific interval between movements and a specific position in a sequence , for example , a neuron increased its firing rate only to the 850 ms interval between the fifth and sixth lever push . This type of encoding would be detectable with our method since in the trained sequences the individual digits and temporal intervals occurred at unique positions . However , the intervals tested in each sequence were isochronous , and it is currently unknown whether this type of encoding would generalize to a sequential temporal pattern similar to the one employed in the current task . Furthermore , the timing of a motor responses can be related to the firing of neural units in the cerebellum , as for instance in the case of cerebellar Purkinje cells which decrease simple spike activity in a timed fashion depending on the trained interval between the conditioning stimulus and the conditioned response following training ( Jirenhed and Hesslow , 2011a , 2011b ) . Yet , whether this type of encoding can also be modulated by the sequential context is not known . In contrast to our findings in the cortex , the cerebellum did not yield significant results with regard to temporal encoding , although there was a clear significant overall encoding of the sequences in the ipsilateral lobule VI which forms reciprocal connections with contralateral premotor areas ( Buckner et al . , 2011; Bostan et al . , 2013 ) . This null result could be related to higher noise levels for subtentorial structures ( Wiestler et al . , 2011 ) or the more complex folding structure of the cerebellar cortex , which may push the informative signal variations below the threshold of effective spatial resolution . Despite partly overlapping temporal and spatial representations , we found a gradient with spatial features represented more in dorsal and temporal features more in ventral aspects of the lateral premotor cortex similar to studies involving spatial and temporal production ( Bengtsson et al . , 2004 ) and prediction ( Schubotz and von Cramon , 2001 ) . This regional distribution raises the possibility that the temporal representation uncovered here may not be an abstract representation of the temporal structure of sequences , but rather a representation of an additional effector system . Indeed , most subjects' introspective reports suggests subvocal rehearsal at the beginning of their training , albeit less so towards the end of training and during fMRI . However , a simple mapping of the temporal feature representations to the control of vocal sequencing and timing is unlikely . There was no direct involvement of more posterior and perisylvian primary vocalization centres . Moreover , cross-sections through the cortical Brodmann area 6 suggest that temporal encoding largely overlapped with spatial encoding along the premotor cortex , instead of being restricted to ventral-most premotor areas recruited for rhythmically structured vocal rehearsal ( Riecker et al . , 2002 ) . Instead , the nervous system may build on specialised processes in premotor areas for temporal control , which have originally evolved for the sequencing of the oro-facial and laryngeal musculature in speech production ( Schubotz , 2007 ) . Such a temporal representation in the premotor cortex could modulate the finger motor system at the level of M1 by interacting with premotor input from the spatial representation in a non-linear manner , as suggested in our multiplicative model ( Kornysheva et al . , 2013 ) , thereby modulating the integrated dynamical systems representation of the sequences in M1 . The presence of multiple temporal and spatial representations across the cortex suggests parallel computations related to the same sequential features . These may have complimentary functions . The medial vs lateral premotor cortex encoding of sequence features may be related to the concurrent encoding of both internally and externally ( visually ) driven sequential movements ( Goldberg , 1985; Mushiake et al . , 1991 ) . Within the lateral premotor cortex , temporal processing has been hypothesized to be associated with the degree of motor involvement; Chen et al . ( 2009 ) have shown that the ventral premotor cortex enables direct action-related encoding of temporal structure while the dorsal premotor cortex facilitates higher order temporal organisation of sequences , in line with the direct vs indirect transformation hypothesis by Hoshi and Tanji ( 2007 ) . Finally , multiple spatial representations may reflect different spatial sequence reference frames—such as movement sequences in extrinsic spatial coordinates in the rostral PMd and the posterior parietal cortex ( Brown et al . , 2013; Wiestler et al . , 2014 ) and in intrinsic spatial coordinates in the caudal PMd ( Wiestler et al . , 2014 ) . Taken together , such a diversity of spatial and temporal sequence representations across the network of premotor and parietal areas may enable flexible control of skilled behaviour that can adapt to the situational task requirements . Overall , the independent encoding of the spatial and temporal features of movement sequences in premotor areas endows the nervous system with the ability for adjustments of individual movement parameters—which would not be possible with a fixed integrated representation such as in M1 alone . For example , this separate encoding may explain why a pianist who learned a particular passage can effortlessly produce the same sequence of finger movements with a novel rhythmic structure—or combine the same rhythm with new variations of the sequence of notes . Finally , the decomposition into features in the premotor cortex provides a computational solution for representing longer and more complex sequences of actions . If the system utilized integrated encoding alone , such as in the primary motor cortex , it would have to represent all relevant permutations of spatial , temporal and other relevant parameters ( e . g . , amplitude and direction ) , which very quickly would lead to a combinatorial explosion . Instead , the premotor cortex may dedicate its resources to representing these features separately , creating a more compact and flexible representation . This special feature may explain why evolution endowed us with premotor areas , rather than simply with a larger primary motor cortex . This architecture has parallels in the ventral visual stream , in which lower visual areas encode specific combinations of simple features at specific spatial locations , whereas higher visual areas represent more complex visual arrangements—such as body parts and scenes—independently of their orientation or location of the stimulus ( Freeman and Simoncelli , 2011 ) . Thus , the decomposition of movements into features may be the cardinal function of premotor areas , and endow the system both with behavioural flexibility and the capacity to store long , complex sequences of movements . 32 neurologically healthy volunteers took part in this study ( 16 female ) , aged between 19 and 36 ( mean: 24 . 8 , SD: 5 . 6 ) . Half of the subjects were trained and scanned on the right and half on the left hand ( males and females balanced across groups ) . All subjects were right-handed according to the Edinburgh Inventory of Manual Preference ( Oldfield , 1971 ) with a mean score of 89 ( range: 70–100; SD: 11 . 9; both groups had the same mean score of 89 , right group SD: 10 . 4 , left group: SD 13 . 5 ) . None of them were professional musicians or athletes . All subjects were naive concerning the hypothesis of this study . Experimental procedures were approved by the research ethics committee of University College London . Written informed consent was obtained from each participant for data analysis and publication of the study results . Participants placed all five fingers of either the left or right hand on a keyboard , which was secured with a foam pillow on the participant's lap . The keyboard had five elongated keys , 20 mm wide , with a groove for each fingertip . A force transducer was mounted below each key and measured the force exerted by the fingers . The force transducers ( Honeywell FS series ) had a dynamic range up to 16 Newton ( N ) , with a repeatability of constant force measurements of <0 . 02 N . Signals from the force transducers were transmitted from the scanner room via a shielded cable . Filters in the scanner room wall prevented leakage of radiofrequency noise . In 21 out of 32 subjects , the force of the fingers of the untrained hand was recorded on a keyboard for the other hand to monitor potential mirror movements ( Diedrichsen et al . , 2013b ) . In the remaining subjects , the passive hand was placed on the pillow on the left or right leg , respectively . Force traces revealed no mirror movements on the contralateral hand . Participants viewed a projection screen mounted behind the scanner bore via a mirror . The screen showed a central cross , on which participants were instructed to fixate during the entire experiment . Participants executed isometric right or left finger presses against the non-movable keys . We implemented a visually cued motor learning task ( Figure 1 ) , ( Kornysheva et al . , 2013 ) to force the subjects into a specific spatial and temporal structure of movement . Subjects were presented with a sequence of white digits ( 1–5 ) in the middle of a black screen that was repeated three times in the training and fMRI sessions . The digits 1 , 2 , 3 , 4 , and 5 instructed to press the thumb , index , middle , ring , and little finger of the right or left hand , respectively . We expected that while the first execution of a sequence would be rather reactive and driven by the cue , the second and third execution would rely on a learned representation of the sequence in question . Indeed the second and third execution during fMRI was significantly faster than the first one ( t ( 31 ) = 9 . 608 , p<0 . 001 ) . Participants were instructed to perform the task as fast and accurately as possible . Each trial started with a warning cue ( ‘ ! ’; duration: 400 ms ) , followed by a sequence of five digits that was timed according to a sequence of five possible inter-stimulus-interval values ( ISI; 600 , 800 , 1000 , 1400 , 1700 ms ) . The first ISI commenced when the warning cue disappeared . Each digit remained on the screen until the onset of the next digit according to the respective ISI value or for 600ms after the onset of the last digit . The trial ended with an inter-trial interval of 1 . 6 s , making each trial 8 . 1 s long . Subjects received feedback on their performance throughout the experiment as follows: if the subjects pressed the correct button within the limits of 50 ms before the onset of the current and 50 ms before the onset of the next digit , that digit turned green; if the response was too early the next digit appeared in yellow; if the response was too late the digit turned turquoise; if the finger press was incorrect , the digit turned red . Subjects received a point only when all digits in a sequence turned green , that is , when they pressed the correct finger in the correct time window . After each block of 27 ( training and post-test ) and 54 ( fMRI test ) trials , respectively , subjects received feedback on their cumulative point score and the median RT in the last block . They were informed that the participant with the highest cumulative score ( weighted by their reaction time ) would receive an additional financial reward . Movements were instructed by sequences with a particular combination of digit order and timing ( ISI sequence ) . Ascending or descending digit run triplets ( e . g . , 2-3-4 ) were excluded from the pool of possible sequences . Identical triplets across sequences were prohibited . The number of ascending to descending and descending to ascending direction changes was fixed at 2 across all sequences ( e . g . , in the sequence and 5-2-4-3-1 the direction change is at 2-4 and 4-3 ) . The position of the five elements ( finger digits and temporal intervals ) had to be different for the three-trained spatial sequences and the three-trained temporal sequences , respectively . The sequences were randomly generated for each participant according to these criteria and matched across the right and the left hand training groups . The experiment was conducted over 5 consecutive days , with a training ( days 1–3 ) , fMRI ( day 4 ) , and post-test phase ( day 5 ) . Note that one of 32 subjects could not take part in a post-test . 3 of the 32 subjects had a delay of 1–4 days between fMRI and post-test . Finally , one subject had the last training session scheduled on the same day as the fMRI . The training phase took place on 3 days and took approximately 1 . 5 hr per day , involving 21 blocks of 27 trials each . 18 of 21 blocks contained the nine different combined sequences ( 3 spatial × 3 temporal structures ) presented three times in a row ( mini-blocks ) . Three additional probe blocks per day were introduced to measure RT advantages related to trained sequences , as well as independent temporal and spatial transfer to new sequences throughout the training phase . Each probe block contained three probe conditions . In the trained temporal condition the cues appeared with each of the trained temporal features , but indicating an untrained order of finger presses . The untrained order was different in each of the three timing probe blocks , but was repeated across the three trials of each mini-block . In the trained spatial condition , the visual stimulus cued sequences involving trained spatial sequence features of finger presses , but in combination with a new temporal feature , that is an untrained sequence of inter-stimulus-intervals . Equally , the new temporal sequence was different in each of the three order probe blocks , but did not change across the three trials of each mini-block . Finally , the untrained condition cued a sequence of finger presses and inter-stimulus intervals that were different from any other trained condition . Again , the novel combination of spatial and temporal features was different in each of the three untrained sequence blocks , but repeated three times in each mini-block . This made the probe blocks as similar as possible to the training blocks . Finally , the three training sessions were identical in terms of sequences and trial randomisation , ensuring that behavioural change could not be explained by differences in sequences or trial delivery . The probe blocks appeared in the beginning ( 1st block ) , the middle ( 11th block ) and the end ( 21st block ) of each training session . Whether subjects started the first training day with a probe block or a training block was counterbalanced across subjects . Note that Figure 2B displays the first probe block as 1st block and the first training block on 2nd block for all participants . Reaction times ( RTs ) for each response were defined as the time at which the force of a finger reached maximum velocity around the onset of the visual cue . Only correct responses were considered . Also , responses that occurred more than 100 ms before stimulus onset or more than 600 ms after stimulus onset were considered as errors and excluded from further analysis . Within each correct trial , we averaged the RT for all responses . We then used the median RT across trials for each individual and condition in the group analysis . Since one of the subjects in the right hand training group did not participate in the post-test , in the post-test analysis only , we excluded a subject trained on corresponding sequences on the left hand to ensure that sequences employed were matched across groups . However , the results of the post-test analysis did not change qualitatively when these subjects were included . Error rates were determined for each block and condition ( cf . Figure 2—figure supplement 1 ) . Data were acquired on a 3 T Siemens Trio system with a 32-channel head coil . Functional data comprised 6 runs of 190 vol each , using a 2D echo-planar imaging sequence ( repetition time [TR] = 2 . 72 s ) . The first 3 vol were discarded to allow magnetization to reach equilibrium . We acquired 32 slices in an interleaved sequence at a thickness of 2 . 7 mm ( 0 . 3 mm gap ) and an in-plane resolution of 2 . 3 × 2 . 3 mm2 . The matrix size was 96 × 96 . Trials were triggered every 2 . 97 TR ( every 95 slices ) . The slices were positioned to cover the cortical motor and premotor areas , as well as the cerebellum . The ventral prefrontal cortex , anterior temporal lobe , and the superior-most part of the parietal lobe were not covered in each subject . Field maps were obtained after the first functional run to correct for inhomogeneities in the main magnetic field ( Hutton et al . , 2002 ) . We also acquired a single T1-weighted anatomical scan ( 3D magnetization-prepared rapid gradient echo sequence , 1 mm isotropic , 240 × 256 × 176 mm field of view ) . The functional data were analysed using SPM8 ( http://www . fil . ion . ucl . ac . uk/spm/ ) , and custom written MATLAB code ( The MathWorks , Inc . , Natick , MA ) . First , we corrected for slice acquisition timing by shifting the acquisition to align with the middle slice of each volume . We then corrected for head movements using a 6-parameter motion correction algorithm . This step also included correction of possible image distortions using the acquired fieldmap data ( Andersson et al . , 2001; Hutton et al . , 2002 ) . The realigned functional data were then coregistered to the individual anatomical scan , using the automatic algorithm in SPM . The coregistration was visually checked , and the affine parameters were adjusted by hand to improve the alignment , if necessary . The preprocessed data were analysed using a general linear model . To remove the influence of movement-related artifacts , we used a weighted least-squares approach ( Diedrichsen and Shadmehr , 2005 ) . For each of the nine trial types , we defined one regressor per imaging run that captured the activation for each voxel across the three sequence executions . This main regressor consisted of boxcar functions for each execution of the sequence of that type , each starting at the moment of warning cue and lasting for 6 . 5 s ( 2 . 4 TRs ) , which was then convolved with the standard hemodynamic response function ( Figure 4—figure supplement 3 ) . Additionally , we included the temporal derivate of the response in the model , linearly independent of the main response . Only the main response was used for classification analysis , to determine differences in spatial activation patterns averaged over the sequence executions , rather than differences in the temporal profiles of the sequence . In addition , each error ( incorrect finger response ) was modeled as a regressor of no interest , one regressor per imaging run starting at the onset of the digit presentation associated with an erroneous response and lasting for 2 . 72 s ( 1 TR ) to account for errors at different positions in the sequence . For each imaging run and voxel , the analysis therefore estimated in 20 regression coefficients ( 9 sequence regressors , 1 error regressor plus 10 respective temporal derivative ) , from which the nine main sequence response estimates were used in the classification analysis . The searchlight analysis was performed taking into account each subject's individual anatomy . This approach was used to consider anatomical variations and achieve maximum accuracy in the localization of representation . The cortical searchlight analysis was implemented on each subjects' individual surfaces . The cerebellar searchlight was volume-based within the cerebellar grey matter as defined by SUIT . From the anatomical images , we obtained a surface reconstruction using the software Freesurfer ( Dale et al . , 1999 ) , which estimates the outer boundary of the gray matter ( pial surface ) and the white–gray matter boundary ( white surface ) . The surfaces were aligned via spherical registration to the Freesurfer average atlas ( Fischl et al . , 1999 ) . Individual data were then projected onto the group map via the individual surface . Correction for multiple tests was performed on the surface using Gaussian field theory ( Worsley et al . , 1996 ) . To detect sequence-specific representations in the neocortex , we used a surface-based searchlight approach ( Oosterhof et al . , 2011 ) . The corresponding toolbox is available on http://surfing . sourceforge . net . A circular region was defined on the cortical surface and the radius increased until exactly 160 voxel lay between the selected surface patches on the pial and white surfaces . It has been shown that a surface-based searchlight minimizes the spillover of multivoxel information from one region to the next across a sulcus and therefore allows for more regionally specific inferences ( Oosterhof et al . , 2011 ) than volume-based searchlights ( Kriegeskorte et al . , 2006 ) . The classification accuracy for each searchlight ( cf . classification procedures below ) was assigned to the center of each searchlight . A classification accuracy map was generated by moving the searchlight across the cortical surface . For the identification of sequence-specific representations in the cerebellar cortex , a volume-based searchlight approach was utilized ( Kriegeskorte et al . , 2006; Wiestler et al . , 2011 ) . Each searchlight consisted of 160 adjacent voxels and the calculations were restricted to voxels lying in the cerebellum using a masking algorithm in the SUIT toolbox ( Diedrichsen , 2006 ) . We implemented different classification procedures to extract information regarding the overall , temporal , spatial , and integrated encoding of sequences . We first employed an overall multi-class classifier , which tested for any differences in the activity patterns associated with nine unique classes ( combinations of three spatial and three temporal structures ) of sequences ( Figure 3A ) . The mean pattern for each class , and the common voxel-by-voxel co-variance matrix was determined from the training data set , consisting of five of the six imaging runs . A Gaussian-linear multi-class classifier ( for details see Wiestler et al . , 2011 and Source code 1 ) was then used to independently classify the nine patterns of the test data set ( the remaining imaging run ) . By rotating which runs served as training and test set , we obtained a cross-validated classification accuracy . Values above the chance level of 11% ( 1/9 classes ) indicated that there were some systematic differences between the activity patterns . A region that encodes the order of finger presses ( spatial sequence ) independently of the temporal sequence should show similar activation patterns for each spatial sequence across the three temporal sequences ( Figure 3B ) . To test for such a representation , we used a linear classifier that distinguished between three spatial sequences paired with one particular temporal sequence , while being trained on data from trials in which spatial sequences were paired with the two remaining temporal sequences . To guarantee independence of the beta estimates , training and test sets were drawn from separate imaging runs . The procedure therefore involved training to distinguish between three spatial sequences timed according to two temporal sequences from five runs and testing the difference between the spatial sequence patterns paired with the remaining different temporal sequence from the remaining run , resulting in a 18-fold cross-validation procedure . Significant above-chance classification accuracy performance ( 33 . 3% , 1/3 classes ) indicated the presence of systematically different local activation patterns for different spatial sequences . The temporal classifier ( Figure 3C ) was designed following the same principle , simply exchanging the role of spatial and temporal sequence . Finally , the integrated classifier isolated representations that code for the unique combination of temporal and spatial sequences . Like the overall classifier , it treated each of the nine unique combinations as a separate class . Critically , however , the mean temporal and the mean spatial class patterns of each run were subtracted from the overall pattern of the respective run . Since this was done for every run separately , subtracting the respective activity patterns did not induce dependence between training and test sets ( see also Figure 3E for simulation results ) . The residual patterns therefore reflected the interaction component between timing and order that cannot be attributed to a linear combination of the two factors . For better comparability across classifiers , as well as for group analysis , the classification accuracies were transformed to z-scores , assuming a binomial distribution of the number of correct guesses . We then tested these z-scores against zero ( chance level ) across participants . Whole brain results were corrected using a surface-based random effects analysis ( N = 32 ) with an uncorrected threshold of t ( 31 ) > 3 . 37 , p<0 . 001 and a cluster-wise p-value for the cluster of that size . The p-value was corrected over the cortical surface using the area of the cluster ( Worsley et al . , 1996 ) and further Bonferroni corrected for tests over two hemispheres ( pcorrected = p*2 ) . Significance in the cerebellum was assessed using a small volume correction ( SUIT ) . We validated this classification approach using simulations of activity patterns of 160 voxels . The activity pattern ( y ) for the ith spatial and jth temporal sequence for the kth imaging run was generated as: yi , j , k = si + tj + ii , j + rk + ei , j , k Each pattern component ( Diedrichsen et al . , 2011 ) was generated as a normal random vector over 160 voxels . By varying the variance of the spatial ( s ) , temporal ( t ) , and integrated ( i ) representations relative to the noise ( e ) , we obtained simulated data sets that were then submitted to the same classification approach used for the actual imaging data ( Figure 3E ) . To examine the distribution of different types of finger sequence representations ( spatial , temporal , integrated ) across motor areas and a possible interaction with the hand being trained , anatomical regions of interests were defined symmetrically in both hemispheres based on probabilistic cytoarchitectonic maps aligned to the average Freesurfer surface ( Fischl et al . , 2008 ) as areas with at least 35% probability of the respective field . Four motor areas with significant overall sequence encoding were considered for ROI analysis: bilateral primary motor ( M1 ) , supplementary motor area ( SMA ) , dorsal ( PMd ) , and ventral premotor cortex ( PMv ) . The hand region of primary motor cortex ( M1 ) was defined as Brodmann area ( BA ) 4 , 2 . 5 cm above and below the hand knob ( Yousry , 1997 ) . Dorsal premotor cortex ( PMd ) was defined as the lateral aspect of BA 6 , superior and PMv inferior to the middle frontal gyrus . The supplementary motor areas ( SMA/pre-SMA ) comprised the medial aspect of BA6 . The M1 , SMA , and PMd were identical to the ROIs used in previous work ( Wiestler and Diedrichsen , 2013 ) and the PMv was added as the lateral part of the premotor cortex ( BA6 ) ventral to PMd . We averaged data across all voxels in the anatomically defined ROIs . This approach enabled us to uncover the respective sequence representations in regions independent of their metabolic demand during the task . A z-score for the classification accuracy for overall , spatial , temporal , and integrated encoding across each in the respective ROI was determined for each subject . ROI analysis was performed on non-smoothed data of each individual . One-sample t tests were employed to probe encoding above chance level ( zero ) . The p-value was Bonferroni-corrected for eight comparisons in each ROI ( critical p-value was 0 . 006=0 . 05/8 ) . In addition to the ROI approach , we defined two symmetrical cross-sections to investigate the continuous profile of sequence feature encoding across cortical regions of both the contra- and ipsilateral hemispheres . The first cross-section through the surface map ran from the rostral end of PMd to the posterior superior parietal cortex . The second cross-section went through BA6 , from the ventral tip through dorsal premotor cortex into the SMA . To determine biases in sequence feature encoding within these cross-sections , we determined the center of gravity ( CoG ) for each subject , classification type , and hemisphere . The minimum classification accuracy value of the respective classifier across the profile of interest was first set to zero ( baseline ) before determining the CoG of the accuracy shapes . The CoG was calculated by computing the spatial average of the coordinates of all nodes in the cross-section , after weighting each with the normalised classification accuracy z at this point . The CoG analysis was performed for three subsections: ( 1 ) rostral PMd to central sulcus , ( 2 ) central sulcus to the lateral part of the superior parietal cortex hitting the medial wall , and ( 3 ) The lateral aspect of BA6 , starting at the level of the inferior frontal gyrus up to the crown of the cortex , excluding the portion of BA6 in the medial wall . We set out to probe the relationship between the amount of sequence learning across subjects and the encoding of the sequences in the contralateral primary motor cortex ( M1 ) that our classification approach revealed to be involved in integrating the spatial and temporal features of sequences , as well as the ipsilateral M1 as a control region . We considered the 20% most activated voxels in contralateral and ipsilateral M1 , thus taking only nodes that were most recruited during the task . Sequence-specific learning was defined as the individual RT advantage for the trained sequences compared to untrained sequences in the post-test phase ( cf . RT results ) . However , we also considered the possibility that the classification results may have been influenced by systematic differences in the execution of the sequences , specifically by differences in force produced by each finger . A region that is sensitive to these lower-level behavioural differences may look like it encodes sequential aspects of the task . To quantify the differences between sequences along these behavioural variables , we computed the accuracy of multivariate classification of the nine trained sequences ( overall classifier ) , using maximum forces at each of the five fingers as different data features instead of voxel activity . Higher classification accuracy would indicate more pronounced force differences between the sequences . Subsequently , two-sided Pearson's correlations between the overall encoding in the contralateral and ipsilateral M1 and the RT difference between untrained and trained sequences in the post-test ( RT advantage ) , as well as force classification accuracy were calculated .	Once a pianist has learned to play a song , he or she can nearly effortlessly reproduce the sequence of finger movements needed to play the song with a particular rhythm . A skilled pianist can also improvise , pairing the same keystrokes with a different rhythm or playing the same rhythm with a slightly different sequence of keys . This ability to flexibly modify and recombine sequences of physical movements in space and time enables humans to exhibit great creativity in music , language , and many other tasks that require motor skills . However , the underlying brain mechanisms that allow this flexibility are only beginning to be explored . Some scientists have theorized that networks of brain cells in the parts of the brain that control movement store a sequence in time and space as one inseparable unit . However , this theory doesn't explain why pianists and other skilled individuals can separate and recombine the physical movements and timing of a sequence in new ways . An alternate idea is that the brain captures the information necessary to execute a series of physical movements separately from the timing at which the movements are to be carried out . This would allow these features to be put together in new ways . Kornysheva and Diedrichsen taught a group of volunteers a series of finger movements paired with particular rhythms . Half the volunteers performed the task using their left hand and the other half with their right hand . After training the volunteers performed better when producing sequences they had been trained on , even in trials where either the rhythm or the finger sequence was slightly changed . The volunteers were also asked to perform the trained movements while their brain activity was monitored in a functional magnetic resonance imaging ( fMRI ) machine . Kornysheva and Diedrichsen looked for areas that showed similar patterns of increases and decreases in activity whenever a particular sequence was performed . This identified areas that showed unique patterns for each trained sequence combination of finger movements and rhythm , which could be distinguished from areas where the activity patterns for sequences remained similar across rhythms or across finger movements . Kornysheva and Diedrichsen found that a region of the brain that controls movement encodes sequences on the opposite side of the brain from the moving hand . In this part of the brain , the movement and timing were encoded together as one unit . However , in premotor areas—which are known to help individuals to plan movements—the timing and the finger movements appeared to be encoded separately in overlapping patches on both sides of the brain . This automatic separation appears to be a fundamental function of the premotor cortex , enabling behavioural flexibility and the storage of complex sequences of movements in space and time .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Human premotor areas parse sequences into their spatial and temporal features
The outer membrane ( OM ) of Gram-negative bacteria serves as a selective permeability barrier that allows entry of essential nutrients while excluding toxic compounds , including antibiotics . The OM is asymmetric and contains an outer leaflet of lipopolysaccharides ( LPS ) or lipooligosaccharides ( LOS ) and an inner leaflet of glycerophospholipids ( GPL ) . We screened Acinetobacter baumannii transposon mutants and identified a number of mutants with OM defects , including an ABC transporter system homologous to the Mla system in E . coli . We further show that this opportunistic , antibiotic-resistant pathogen uses this multicomponent protein complex and ATP hydrolysis at the inner membrane to promote GPL export to the OM . The broad conservation of the Mla system in Gram-negative bacteria suggests the system may play a conserved role in OM biogenesis . The importance of the Mla system to Acinetobacter baumannii OM integrity and antibiotic sensitivity suggests that its components may serve as new antimicrobial therapeutic targets . Gram-negative bacteria are enveloped by two lipid bilayers , separated by an aqueous periplasmic space containing a peptidoglycan cell wall . The inner membrane ( IM ) is a symmetric bilayer of glycerophospholipids ( GPL ) , of which zwitterionic phosphatidylethanalomine ( PE ) , acidic phosphatidylglycerol ( PG ) , and cardiolipin ( CL ) are among the most widely distributed in bacteria ( Zhang and Rock , 2008 ) . In contrast , the outer membrane ( OM ) is largely asymmetric and composed of an inner leaflet of GPL and an outer leaflet of lipopolysaccharide ( LPS ) or lipooligosaccharide ( LOS ) ( Pelletier et al . , 2013 ) . The OM forms the first line of defense against antimicrobials by functioning as a molecular permeability barrier . The asymmetric nature of its lipid bilayer and the structure of LPS/LOS molecules facilitates barrier function , as the core region of LPS impedes the entry of hydrophobic molecules into the cell while the acyl chains within the bilayer also help to prevent the entry of many hydrophilic compounds ( Bishop , 2014 ) . Although progress has been made in understanding many aspects of OM assembly – including the discovery of an LPS transport system and the machinery for proper folding and insertion of outer membrane proteins ( Okuda and Tokuda , 2011; Okuda et al . , 2016 ) – little is known about the molecular mechanisms for transport of the GPLs necessary for OM formation and barrier function . Acinetobacter baumannii is an important cause of antibiotic-resistant opportunistic infections and has significant innate resistance to disinfectants and antibiotics . Similar to other Gram-negative opportunistic pathogens such as Pseudomonas aeruginosa and Klebsiella spp . , individuals with breached skin or damaged respiratory tract mucosa are most vulnerable ( Chmelnitsky et al . , 2013; Abbo et al . , 2005 ) . We performed a genetic screen to identify genes important for the OM barrier of A . baumannii . This led to the identification of an ABC ( ATP-binding cassette ) transporter complex that promotes GPL export to the OM . Transporter disruption attenuates bacterial OM barrier function , resulting in increased susceptibility of A . baumannii to a wide variety of antibiotics . The homologous system for E . coli has previously been termed Mla for its suggested role in the maintenance of outer membrane lipid asymmetry via the removal of GPL from the outer leaflet of the OM to the IM . While this is a reasonable hypothesis , there is not direct biochemical evidence that the Mla system functions to return GPL from the OM to the IM . In this work , we present evidence that the A . baumannii Mla system functions to promote GPL movement from the IM to the OM . This conclusion is based on the observation that newly synthesized GPLs accumulate at the IM of mla mutants , akin to how LPS molecules accumulate at the inner membrane in bacteria with mutations in the lpt genes encoding the LPS ABC transport system ( Okuda et al . , 2016 ) . Given the broad conservation of Mla in prokaryotic diderm organisms , the anterograde trafficking function of Mla might be exploited by a variety of species . We identified strains with mutations in genes required for maintenance of the Acinetobacter baumannii OM barrier by screening transposon mutants for the development of a blue colony phenotype on agar plates containing the chromogenic substrate BCIP-Toluidine ( XP ) . Although A . baumannii carries an endogenous periplasmic phosphatase enzyme , colonies remain white on agar plates containing XP . We reasoned that lesions in genes necessary for the OM barrier function should result in a blue colony phenotype , as the XP substrate becomes accessible to the periplasmic enzyme ( Strauch and Beckwith , 1988; Lopes et al . , 1972 ) . Screening roughly 80 , 000 transposon-containing colonies for the blue colony phenotype yielded 364 blue colonies whose insertions were mapped to 58 unique genes ( Supplementary file 1 ) . We confirmed the results of the screen by assaying for OM-barrier defects using ethidium bromide ( EtBr ) and N-Phenyl-1-naphthylamine ( NPN ) uptake assays ( Helander and Mattila-Sandholm , 2000; Murata et al . , 2007 ) . We also tested for resistance to antimicrobials , including trimethoprim , rifampicin , imipenem , carbenicillin , amikacin , gentamicin , tetracycline , polymyxin B , and erythromycin . Greater than 85% of the strains identified in the screen demonstrated decreased OM barrier function compared to wild type . Out of the 58 strains with transposon insertions , 23 demonstrated OM permeability defects by NPN and EtBr uptake assays , and 49 out of 58 resulted in increased sensitivity to two or more antibiotics compared to the parent strain , indicating that the screen identified lesions causing OM barrier defects leading to increased permeability to small charged and hydrophobic molecules , including commonly used antibiotics . Four mutants with a blue colony phenotype contained unique transposon insertions in the genetic loci A1S_3103 and A1S_3102 , predicted to encode core components ( mlaF and mlaE ) of a multicomponent ABC transport system . These genes are within a five-gene operon that encodes for a conserved proteobacterial ABC transport system homologous to the E . coli mla system previously implicated in OM integrity ( Malinverni and Silhavy , 2009 ) . The A . baumannii operon includes: mlaF and mlaE , respectively predicted to encode the nucleotide-binding and transmembrane domains of an ABC transporter; mlaD , encoding a protein containing an IM-spanning domain and a predicted periplasmic soluble domain; mlaC , encoding a soluble periplasmic protein; and mlaB , an additional gene predicted to encode a cytoplasmic sulfate transporter and anti-sigma factor antiagonist ( STAS ) -domain protein ( Figure 1A ) . An additional putative OM lipoprotein is encoded on mlaA , or vacJ , which is clustered with the rest of the mla operon in some Gram-negative bacteria , although it is at a different chromosomal location in A . baumannii . MlaA has been functionally associated with the rest of the Mla components in E . coli , as mutations in mlaA yield comparable phenotypes to mutations in other components of the system ( Strauch and Beckwith , 1988 ) . Bioinformatic analysis predicts that the mlaC and mlaF genes respectively encode the soluble periplasmic component and cytoplasmic ATPase component of the ABC transport system , and we chose to focus on mutants of these genes for further experiments to elucidate the function of the mlaFEDCB operon . Chromosomal deletions were created by allelic exchange , and these mutations resulted in OM permeability defects as measured by EtBr uptake assays . We complemented the OM defect for the ∆mlaC and ∆mlaF deletion mutants by repairing the original deletion event in the chromosome and confirmed complementation of the observed permeability defect ( Figure 1B ) . Deletions in mlaF and mlaC also rendered A . baumannii increasingly sensitive to a variety of antibiotics as determined by MIC measurements ( Figure 1D ) . Increased sensitivity to antibiotics whose uptake is not mediated by OM porins is consistent with a direct effect on the membrane component of the OM permeability barrier ( Nikaido , 2003; Vaara , 1992 ) . In addition to OM defects , the mla mutants display phenotypes that may correlate with OM stress , including increased production of extracellular carbohydrates as evidenced by crystal violet staining of pellicles following growth in broth culture ( Figure 1—figure supplement 1A ) . These data indicate a role for Mla in the maintenance of the outer membrane barrier of A . baumannii . To exclude the possibility that the membrane defect was the result of the disruptive effect of a partially formed Mla protein complex , we engineered an enzymatically inactive ATPase component and expressed the defective enzyme from a plasmid . We reasoned that by expressing this allele in the wild type bacteria we could create a dominant-negative effect on Mla function . The cytoplasmic ATPase component of the Mla system , MlaF , contains the consensus sequence GxxxxGKT at residues 49–56 , characteristic of a Walker A motif . Downstream residues 173–178 contain the sequence LIMYDE , typical of a Walker B motif . The Walker motifs form highly conserved structures critical for nucleotide binding and hydrolysis ( Walker et al . , 1982 ) . The lysine residue of the Walker A motif is particularly essential for the hydrolysis of ATP . Mutations in this lysine residue are inhibited for nucleotide binding , and the mutated protein is rendered inactive ( Hanson and Whiteheart , 2005 ) . Additionally , ATPase mutants in the key lysine residue have been shown to have a dominant-negative effect on ATP hydrolysis when co-expressed with their wild-type ATPase counterparts , as typical ABC transporters have a structural requirement for two functional nucleotide-binding proteins which dimerize upon substrate transport ( Davidson and Sharma , 1997; Wilkens , 2015 ) . Therefore , we created a version of the MlaF coding sequence with a leucine substitution of the Walker A lysine residue ( MlaFK55L ) , and then cloned the mutated mlaF into the low-copy pMMBkan vector under control of the mlaF native promoter . We observed that expression of MlaFK55L in wild type A . baumannii had a dominant-negative effect on membrane permeability as measured by EtBr uptake ( Figure 1C ) , and expression of MlaFK55L also resulted in increased exopolysaccharide production as demonstrated by increased staining by crystal violet ( Figure 1—figure supplement 1B ) . Correspondingly , expression of MlaFK55L rendered A . baumannii more sensitive to a variety of antibiotics , resulting in reduced MICs when compared to A . baumannii expressing the empty pMMBkan vector ( Figure 1D ) . Therefore , expression of a defective ATPase results in a dominant-negative mutant with a comparable phenotype to deletion of components of the mla operon . These results demonstrate a requirement for ATP hydrolysis by MlaF for the maintenance of OM barrier function in A . baumannii , and indicate that the phenotypes of deletion mutants were likely a result of a lack of transport function , rather than formation of a toxic incomplete membrane protein complex . The genetic arrangement and conservation of the components of this ATPase-containing transport complex indicated it was likely that the individual components formed a higher order protein structure . To define whether the Mla components form a stable protein complex , we expressed the entire operon ( mlaFEDCB ) from A . baumannii ATCC 17978 in E . coli with a carboxy-terminal hexahistidine tag on the MlaB component . Affinity purification of MlaB revealed three additional bands , with sizes corresponding to MlaF , MlaD , and MlaE ( Figure 2—figure supplement 1 ) and confirmed by MALDI-TOF mass spectrometry analysis , indicating that these four proteins form a stable complex . We did not detect MlaC , suggesting it might interact only transiently with the other components , consistent with results recently reported by Thong et al . ( 2016 ) . We next used cryo-electron microscopy to characterize the architecture of the A . baumannii MlaBDEF complex ( abMlaBDEF ) . This complex is uniformly dispersed in vitreous ice ( Figure 2—figure supplement 2A ) , and 2D classification demonstrated the presence of a range of views suitable for structure determination ( Figure 2—figure supplement 2B ) . Following 2D- and 3D-classification , we obtained a final dataset of ~14 , 000 particles with which we obtained a structure to a resolution of 8 . 7 Å ( Figure 2—figure supplement 2D ) . The structure possesses significant visible features in agreement with the nominal resolution ( Figure 2—figure supplement 2C ) . Based on the bioinformatically-predicted localization of individual proteins and work recently performed on the similar E . coli Mla complex ( ecMlaBDEF ) ( Thong et al . , 2016 ) , we propose that MlaD localizes to the periplasmic side of the IM , MlaE forms the central transmembrane region , and MlaF and MlaB form the bottom layer on the cytoplasmic face of the IM with two visible hetero-dimers ( Figure 2—figure supplement 2E ) . We note that the structure of ecMlaBDEF , at lower resolution , was reported recently ( Ekiert et al . , 2017 ) . The overall features of both structures , solved independently , are identical , suggesting that they correspond to the correct structure for the complex . However , the limited resolution of the ecMlaBDEF complex structure did not allow modeling of its individual subunits , in contrast to the abMlaBDEF structure reported here . We note that a clear six-fold symmetry is present for the region of the map attributed to MlaD ( Figure 2B ) , despite the fact that we only imposed a 2-fold symmetry averaging . This agrees with the proposed hexameric state of its E . coli homologue ( ecMlaD ) ( Thong et al . , 2016 ) . We next modeled abMlaD , using an evolution restraints-derived structural model of ecMlaD ( Ovchinnikov et al . , 2017 ) as a template , and used our previously-published EM-guided symmetry modeling procedure ( Bergeron et al . , 2013 ) to model its hexameric state . The obtained abMlaD hexameric model is at a low-energy minimum ( Figure 2—figure supplement 3B ) and fits the EM map density well ( Figure 2B and Figure 2—figure supplement 4B ) . A crystal structure of the periplasmic domain of ecMlaD published recently ( Ekiert et al . , 2017 ) formed a crystallographic hexamer , suggesting that this corresponds to the native hexomeric arrangement for this domain . Our abMlaD hexameric model is very similar to the crystallographic ecMlaD structure ( Figure 2—figure supplement 3C ) , supporting the proposed domain arrangement in the MlaBDEF complex . We note , however , that one region of density in the EM map is not accounted for by our MlaD hexamer model ( Figure 2B ) . The localization of this extra density suggests that it corresponds to a ~ 45 amino-acid insert present between strands 4 and 5 of the abMlaD β-sheet ( Figure 2—figure supplement 4A ) . The role of this insert , uniquely found in the A . baumannii orthologue , is not known . We next modeled the structures of MlaB and MlaF and fitted their respective coordinates in the corresponding region of the EM map ( Figure 2C and Figure 2—figure supplement 3A ) . For both proteins , most helices are well resolved , which allowed us to place the models unambiguously . We then compared the conformation of the ATPase MlaF to that of the maltose transporter ATPase MalK , which has been trapped in several conformations of the transporter; that is the inward-facing state , the pre-translocation state , and the outward-facing state ( Khare et al . , 2009; Oldham et al . , 2013 ) . Interestingly , the arrangement of MlaF clearly resembles the pre-translocation state of MalK ( Figure 2D ) . This suggests that we have trapped a similar conformation of the abMlaBDEF complex . It is possible that MlaD and/or MlaF , for which there are no equivalent in other ABC transporters , stabilizes this conformation . Alternatively , it is possible that the presence of detergents , which were present to solubilize the complex , mimics the natural ligand in the transporter’s active site . Finally , the transmembrane ( TM ) region of the map is well resolved , and density for the transmembrane ( TM ) helices can be clearly identified . We therefore modeled abMlaE , using an evolution restraints-derived structural model of ecMlaE ( Ovchinnikov et al . , 2017 ) as a template , and fitted the obtained coordinates in the corresponding region of the map , with the orientation corresponding to the predicted topology . The resulting MlaE dimer model ( Figure 2D ) fits well to the EM map density ( Figure 2—figure supplement 4C ) , and clearly corresponds to a closed transporter , with no solvent channel between the subunits . Interestingly , we also noted clear density for three TM helices that likely correspond to the MlaD N-terminal helices ( Figure 3A ) . However , they lacked continuity , and we observed that only two form a direct interaction with MlaE . It is possible that this is due to heterogeneity in the orientation of MlaD relative to the rest of the complex . To verify this , we performed further 2D classification of the particles used for reconstruction ( Figure 3B ) , which revealed a range of positions for the MlaD region relative to the rest of the complex . We therefore performed further 3D classification leading to a smaller dataset of ~8000 particles . This produced a structure of lower resolution ( ~11 . 5 Å ) but with the six MlaD N-terminal TM helices clearly visible ( Figure 3B ) . While the periplasmic domain possesses 6-fold symmetry , the TM domains of MlaD do not appear symmetrical , with two forming close contacts with the density attributed to MlaE while the other four do not appear to contact any other proteins . This observation likely explains the asymmetry of contacts between the dimeric MlaE and the hexameric MlaD . A higher-resolution structure will be required to determine if additional contacts are formed between the outward-facing loops of MlaE and the periplasmic domain of MlaD . The crystal structure of MlaC has been solved from Ralstonia solanacearum . The structure contains a single phosphatidylethanolamine molecule oriented such that the hydrophobic acyl chains are located inside the protein while the hydrophilic head group is exposed ( Huang et al . , 2016 ) . More recently , the crystal structure for MlaC has been solved from E . coli and shown to bind lipid tails ( Ekiert et al . , 2017 ) . As noted in previous work performed on the E . coli Mla system , this is strong evidence that the substrates of the Mla system are GPL ( Malinverni and Silhavy , 2009 ) . In order to confirm that the periplasmic components of the Mla pathway in A . baumannii interact with GPL , we purified the soluble domains of both MlaC and MlaD by expressing histidine-tagged proteins followed by Ni-affinity FPLC purification . After overnight dialysis of the proteins , we performed Bligh-dyer chloroform extraction on the purified proteins to isolate any bound GPL and analyzed the results by LC-MS/MS . GPL analysis revealed both phosphatidylglycerol and phosphatidylethanolamine of varying acyl chain lengths . This suggests the possibility that the periplasmic substrate binding components of the system may bind diacylated GPL molecules with limited polar head group specificity ( Figure 4—figure supplement 1 ) . Given the OM defect of mla mutants , as well as the system’s apparent direct association with GPL , we chose to further characterize the overall membrane GPL composition of the mla mutants . Previous work on the Mla system in E . coli has demonstrated an increase in hepta-acylated lipid A in mla mutants , indicating activation of PagP that acylates GPL and lipid A in the outer leaflet of the OM in enterobacteria ( Malinverni and Silhavy , 2009; Dalebroux et al . , 2014 ) . From this data it has been suggested that the system may serve to maintain lipid asymmetry within the OM , although it is well known that GPL displacement to the OM outer leaflet is a general reflection of chemical damage to the OM ( Jia et al . , 2004; Bishop et al . , 2000; Dekker , 2000 ) . However , biochemical analysis of the membrane GPL composition for mla mutants has not been published for any organism to our knowledge , so we sought to apply our lab’s methods of GPL quantification to test the hypothesis of retrograde transport function . To determine whether A . baumannii mla mutations cause changes in the membrane GPL concentration , GPL were extracted from inner and outer membrane fractions separated by density centrifugation . As can be seen on Figure 5—figure supplement 3 , density centrifugation results in nice separation of the outer and inner membranes of Acinetobacter baumannii , with the OM contain the vast majority of OmpA and the inner membrane containing all the NAPPH oxidase . Thin-layer chromatography ( TLC ) and electrospray-ionization time-of-flight mass spectrometry ( ESI-MS ) were used to qualitatively assess GPL composition from these well separated membrane fractions . TLC and ESI-MS indicated ΔmlaC A . baumannii had a dramatically decreased abundance of all major phospholipid species in the OM compared to wild type . ( Figure 4A and Figure 4—figure supplement 2 ) . To better analyze the differences in membrane GPL , we quantified GPL by normal phase liquid-chromatography collision-induced-dissociation mass spectrometry ( LC-MS/MS ) . We quantified the ratio of individual GPL within each membrane by normalizing to an internal standard of known quantity . We then normalized the quantified GPL to the protein content of isolated IM and OM . Quantitative LC-MS/MS confirmed the overall reduction in outer membrane GPLs observed by ESI-MS and TLC , with the reduced levels observable across multiple GPL species for ΔmlaC mutants relative to wild type ( Figure 4B ) . Therefore , mutations in the components of the Mla system result in a decrease in OM GPL , whereas the retrograde transport hypothesis would predict an increase in OM GPL . Therefore , these results instead suggest a possible role for Mla in outward GPL trafficking . The overall decrease in outer membrane glycerophospholipids of A . baumannii mla mutants suggests that either the Mla system is functioning to deliver GPLs from the inner membrane to the outer membrane , or alternatively , mutations in the Mla system may disrupt the outer membrane in a manner that leads to the activation of outer membrane phospholipases , which then degrade GPL . Work performed on the Mla system in E . coli has demonstrated that disruption of genes in the Mla pathway results in activation of both the OM acyl-transferase PagP , which cleaves a palmitate moiety from GPL and transfers it to LPS and PG , creating a hepta-acylated LPS molecule and palmitoyl-PG and the OM phospholipase PldA ( Malinverni and Silhavy , 2009; Bishop et al . , 2000 ) . A . baumannii has no known PagP enzyme but similar activity of the multiple predicted OM phospholipases could account for the reduction in OM GPL as observed by TLC and quantitative mass spectrometry . Therefore , we designed a mass spectrometry-based assay to study intermembrane GPL transport using 13C stable isotope labeling ( Figure 5—figure supplement 1A ) , to better analyze the directionality of GPL transport by the Mla system between the bacterial membranes . When grown in culture with sodium acetate as the sole carbon source , many bacteria directly synthesize acetyl-CoA using the conserved enzyme acetyl-CoA synthase ( Kumari et al . , 2000 ) . Acetyl CoA , the precursor metabolite for fatty acid biosynthesis , is first converted to malonyl-CoA and enters the FasII ( fatty acid biosynthesis ) pathway that supplies endogenously synthesized fatty acids to macromolecules such as lipopolysaccharides , phospholipids , lipoproteins , and lipid-containing metabolites . By growing cultures in unlabeled acetate then ‘pulsing’ with 2-13C acetate and analyzing separated membrane fractions from set time points , we can observe the flow of newly synthesized GPLs between the IM and OM of A . baumannii ( Figure 5—figure supplement 1B ) ( Dalebroux et al . , 2014 ) . Upon introducing the 2-13C acetate as the sole carbon source , 13C-labeled GPL were immediately synthesized in the bacterial cytoplasm . We reasoned that continued growth in 13C acetate should result in a mixed pool of unlabeled and labeled IM GPL molecules . As the GPL are then fluxed from the IM to the OM , the likelihood that an individual GPL molecule is transported is directly proportional to the ratio of labeled to unlabeled GPL in the IM pool . As the bacteria continue to grow in 13C acetate , the ratio of labeled to unlabeled GPL in the IM will gradually increase as new GPL are synthesized and inserted in the IM . As such , the likelihood of transporting labeled GPL to the OM will also increase . A comparison of the ratios of labeled to unlabeled GPL in the IM and OM will thus reflect the efficiency of transport between the membranes , and analysis of transport in wild type A . baumannii will establish reference for transport efficiency with which to compare our mutants . Additionally , OM phospholipases , some of which may be activated upon membrane damage ( Istivan and Coloe , 2006 ) , will not distinguish between labeled and unlabeled GPL and therefore will not affect the ratio of labeled to unlabeled GPL obtained from this assay . Membrane separation and analysis of wild type A . baumannii revealed near-identical rates-of-change between the two membranes in ratios of 13C-labeled to unlabeled GPLs , indicating that newly synthesized GPLs are transported and inserted into the OM at a rate equivalent to their rate of synthesis and assembly within the IM . Furthermore , the ratios of labeled to unlabeled GPLs were nearly equal in the IM compared to the OM at the time points evaluated , indicating that GPL transport likely occurs rapidly , consistent with earlier pulse-chase experiments performed in E . coli that estimate the half-life of translocation of various GPLs at between 0 . 5 and 2 . 8 min ( Donohue-Rolfe and Schaechter , 1980 ) . By contrast , mutants in the Mla system accumulate newly synthesized GPLs in their IM at a greater rate than occurs in the OM as evidenced by the increasing disparity in the ratio of labeled to unlabeled GPLs between the IM and OM over time ( Figure 5A ) . The discrepancy in ratios of labeled to unlabeled GPLs between the IM and OM of ∆mlaF is apparent for PG and PE of varying acyl chain lengths corresponding to the most naturally abundant species C16:0/C16:0 , C18:1/C18:1 , or C16:0/C18:1 ( Supplementary file 2 ) . Further , the effects of MlaFK55L expression on GPL trafficking were similar to what was observed in the ∆mlaF strain ( Figure 5B ) . Therefore , ATP hydrolysis by MlaF appears to be a requirement for extraction of these GPLs from the IM of A . baumannii for subsequent transport to the OM . To better characterize the role of the periplasmic substrate binding component MlaC , we performed similar stable isotope pulse experiments to observe the flow of newly synthesized GPLs in the ∆mlaC strains . Stable isotope experiments on ∆mlaC mutants reveal IM accumulation of newly synthesized GPLs similar to the result in ∆mlaF mutants ( Figure 5—figure supplement 2A ) , indicating that in the absence of the periplasmic component GPLs are not efficiently removed from the IM by the remainder of the Mla system . We also sought to characterize the potential role of the putative OM-lipoprotein MlaA , which has been implicated as a component of the Mla system in E . coli . A chromosomal deletion strain of mlaA was created by allelic exchange , and complemented by expression of MlaA from a pMMB67EH-Kan plasmid . The results of the stable isotope pulse experiments in the ∆mlaA strain revealed results consistent with those obtained from ∆mlaC and ∆mlaF , in which the ratio of labeled to unlabeled GPL is consistently higher in the inner membrane than the outer membrane after one hour of exposure to 13C-acetate ( Figure 5—figure supplement 2B and C ) . These results are consistent with a model in which the IM-localized ABC transporter complex MlaBDEF first transfers GPLs to the periplasmic binding protein MlaC , which then transports GPL to the OM , whereupon MlaA facilitates GPL insertion into the OM . We performed a screen to identify A . baumannii proteins that are essential for its OM barrier that led to the identification of an ABC transport system whose ATPase activity maintains OM barrier function . IM and periplasmic components of this system can be purified , bind GPLs , and assemble into a defined protein complex with significant symmetry , indicating that this system could function to transport GPLs from the IM to the OM . Consistent with the possibility that Mla functions as an anterograde transporter , the OM of mutants show an overall reduction of GPL along with an excess accumulation of newly synthesized GPL on the IM . Therefore , these results lead us to propose that the function of the A . baumannii Mla system is the trafficking of GPL from the IM , across the periplasm , for delivery to the outer membrane ( Figure 6 ) . According to this model , ATP hydrolysis by MlaF provides the initial energy to extract GPL from the IM , while the substrate binding components MlaD and MlaC take up lipids for delivery to the OM . It has been observed by van Meer and colleagues that complete extraction of GPLs from the membrane bilayer requires a Gibbs free energy of ~100 kJ/mol ( Abreu et al . , 2004; van Meer et al . , 2006 ) , whereas ATP contains just 30 kJ/mol of energy . To account for the energy difference , a hydrophobic acceptor molecule is proposed to allow the lipids to fully dissociate from the rest of the ABC transporter and facilitate complete removal from the bilayer . The GPL-binding component , MlaD , contains an IM spanning domain and is shown here , and in orthologous systems , to be in complex with the MlaE and MlaF proteins within the IM ( Ekiert et al . , 2017; Roston et al . , 2012 ) . The close association of MlaD with the outer leaflet of the IM may allow it to extract lipids from the IM by hydrophobic interaction with the acyl chains after ATP hydrolysis by MlaF . Subsequent trafficking across the periplasm then involves the periplasmic GPL binding protein MlaC , which likely accepts GPL from MlaD and then carries them to the OM . We note however the observed effect of mlaC deletion on GPL accumulation in the IM , while statistically significant for most of the analyzed diacyl-glycerophospholipids , appears to be less than that of deletion of the ATPase component ( Figure 5C ) , suggesting that while MlaC may participate in transfer to the OM , there may be redundant mechanisms by which the IM complex can transport or remove IM GPL in the absence of MlaC . While the precise mechanism of GPL insertion into the OM is not yet known , work performed on the E . coli Mla system has shown that MlaC interacts with both the IM MlaFEDB complex , as well as with the putative OM lipoprotein MlaA , and our results support a role for MlaA in the function of the overall Mla system and delivery of GPL to the OM . In this work , we designed a method to monitor lipid transport between Gram-negative bacterial membranes using stable 13C isotope labeling . We considered the possibility of loss of GPL from the outer membrane due to outer membrane vesicle formation or , more likely , by the possible increased activation of outer membrane phospholipases . Both would have the effect of removing GPL from the outer membrane and would result in lower GPL levels in the outer membrane of mla mutants . Neither of these mechanisms would operate specifically on either labeled or unlabeled lipids . We understood the decrease in outer membrane GPL to be insufficient evidence of an anterograde transport function for mla , and for this reason we developed the stable isotope assay to control for these possibilities . The stable isotope assay gives insight into whether these results are due simply to mislocalization or degradation of outer membrane GPL , or if they can in fact be attributed to deficient anterograde GPL transport . This is because the peak intensity of each newly synthesized , C13-labeled GPL is normalized to the corresponding unlabeled version of that GPL species with each sample injected into the LC-MS/MS . The result is a ratio of labeled to unlabeled GPL for every membrane sample . With C13-acetate as the sole carbon source , we observe a gradual increase in the ratio of labeled GPL relative to unlabeled GPL over time . In wild type bacteria , these ratios for the inner and outer membranes track closely over time , which indicates that under these conditions GPL transport from the inner to the outer membrane occurs quite rapidly . In mla mutants the ratio is both higher and typically increases at a greater rate in the inner membrane . Phospholipases and budding outer membrane vesicles will not distinguish between labeled and unlabeled GPL species , and so will not impact the ratios obtained with this assay . Our results using this assay are consistent with the Mla system functioning as an anterograde GPL transporter , however they do not exclude the possibility of a dual role for Mla components in the maintenance of OM lipid asymmetry . Previous work performed on the orthologous Mla system in E . coli has been interpreted to suggest that the function of the system is to remove GPL from the outer leaflet of the OM for retrograde transport back into the cytoplasm based on the observation that E . coli mla mutants likely contain GPLs on the outer leaflet of the OM . ( Malinverni and Silhavy , 2009; Benning , 2008 ) . This proposed function was inferred from the observation that gene deletions resulted in an increased activation of the OM-phospholipase enzymes PagP and OMPLA , suggesting an increased amount of GPL in the outer leaflet of the OM ( Malinverni and Silhavy , 2009 ) . The interpretation of retrograde transport function was also based on the existence of an orthologous system in plant chloroplasts that transports lipids from the endoplasmic reticulum ( ER ) into the organelle . Many plants require this retrograde transport function because certain lipids in the chloroplast thylakoid membrane derive from GPL originating in the ER ( Hurlock et al . , 2014 ) . However , since Gram-negative bacteria synthesize GPL within the IM , retrograde transport of GPL would only be necessary for the recycling of GPL mislocalized to the OM outer leaflet . Although this is a reasonable inference based on data available at the time , we would point out that the directionality of transport by the E . coli Mla system had not been thoroughly probed experimentally using membrane analysis or with a functional assay of the type performed here . It is conceivable that the import function of the orthologous chloroplast TGD system is a result of adaptation to the intracellular environment , the system in this case having evolved to aid in the transfer of GPL from the nearby ER to the chloroplast . Furthermore , while it is possible that the Mla system in E . coli serves a different primary function than in A . baumannii , we demonstrate here that both complexes possess a similar architecture , pointing to a conserved function . The outer membrane defect phenotypes observed in E . coli mla mutants might also be explained by a disruption of OM structure stemming from decreased concentrations of OM GPL , leading to activation of the PagP enzyme . It is well established that for E . coli , GPL displacement to the OM outer leaflet and subsequent activation of these enzymes reflects OM instability and can be achieved by chemical disruption of the bilayer ( Jia et al . , 2004; Bishop et al . , 2000; Dekker , 2000 ) . It may be the case that the OM of E . coli mla mutants contain GPL in the outer leaflet , but the possibility remains that OM GPL can flip into the outer leaflet under conditions of OM damage resulting from an imbalance of LPS-to-GPL ratios , along with perhaps the corresponding disruption of OM proteins . However , final determination of the directionality of GPL transport by the Mla system in E . coli and other organisms will require intermembrane transport studies similar to what has been done here for A . baumannii , along with studies similar to those performed for the Lpt LPS transport system for which molecular transfer of LPS from molecule to molecule of the Lpt system is functionally defined . Following the introduction of the retrograde transport model for Mla function into the existing literature , a number of studies have examined the phenotypic effects of Mla disruption in various organisms . In a recent study in PNAS , Powers and Trent first obtained A . baumannii deficient in lipooligosaccharide ( LOS ) by selection in the presence of polymyxin B ( Powers and Trent , 2018 ) . They then performed an evolution experiment , passaging the strains in cultures containing polymyxin B over 120 generations , at which point they observed significantly improved growth in the populations . These evolved populations were also observed to have increased resistance to antibiotics including vancomycin , bacitracin , and daptomycin , and to appear more morphologically consistent relative to the unevolved strains when observed microscopically . Whole genome sequencing of the evolved strains revealed mutations in mla genes in seven of the 10 evolved populations . They also observed frequent disruptions in pldA , as well as in other envelope genes . To further study these effects , they then introduced clean deletions of mlaE and pldA to ATCC 19606 , and selected for LOS-deficient bacteria by plating on polymyxin B . These double mutants demonstrated improved growth and resistance to antibiotics but continued to display altered cellular morphology . The authors present their data as evidence in support of Mla as a retrograde transport system , and assume that a lack of removal of GPL from the outer leaflet is promoting the OM barrier . We would point out that lacking in their data is examination of the membrane glycerophospholipid ( GPL ) profile in their LOS-deficient mutants . It is assumed by the authors that mla and pldA mutations have the effect of stabilizing a symmetric outer membrane produced in the absence of LOS by allowing GPL to fill in the outer leaflet , resulting in improved growth and antibiotic resistance . Given that the data suggests that Mla and PldA are selected against when LOS is absent , examination of the outer membrane GPL content might have supported the authors’ conclusions if it revealed an increase in GPL in mla and pldA mutants . Absent such data , it is not obvious to us that the authors have sufficiently ruled out alternative explanations for their observed phenomena . For example , we would question the mechanisms regulating the homeostasis of both the inner and outer membranes and the entire periplasmic space in the absence of LOS . The authors acknowledge earlier work that observed an increase in expression of mla genes upon initial loss of LOS in 19606 ( Henry et al . , 2012; Boll et al . , 2016 ) . Genes in the mla pathway were shown to have an up to 7 . 5-fold increase in gene expression upon loss of LOS . Powers and Trent assert that the function of Mla is deleterious in the absence of LOS , but perhaps what is deleterious is the profound upregulation of mla expression in the absence of LOS , combined with an active PldA . If Mla is an anterograde transporter , we can imagine this might create a situation in which GPL are rapidly removed from the inner membrane and then degraded in the outer membrane in excess of what the cell can support and limiting both of these processes together simply allows the cell to achieve a new homeostasis . Understanding of the myriad processes regulating bacterial outer membrane assembly and integrity remains limited even when LOS is present , and so interpreting results such as these as providing direct evidence of function may exceed the limits of the data . The gene for MlaA , the proposed OM component , is at a different chromosomal location from the remainder of the mla operon . Recent structural studies on MlaA have revealed that MlaA forms a ring-shaped structure localized the inner leaflet of the OM , and have shown it to form stable complexes with the outer membrane proteins OmpF and OmpC ( Abellón-Ruiz et al . , 2017 ) . The proposed structure of MlaA in the OM supports the argument that MlaA is involved in removal of GPL from the outer leaflet , and it is suggested that GPL from the outer leaflet travel through a pore in MlaA where they are received by MlaC , yet our data reveals that A . baumannii ∆mlaA mutants are defective in delivery of GPL from the IM to the OM . These data can be reconciled by a model in which MlaA functions both to remove mislocalized GPL from the outer leaflet of the OM , and additionally serves to facilitate delivery of GPL to the OM by MlaC , perhaps by enabling MlaC localization to the surface of the inner leaflet . By this model , mutations in MlaA will be phenotypically similar to mutations in other components of the Mla system , and we would expect to observe a decreased rate of anterograde GPL transport . We would here point out that while previous work has implicated the Mla system in the maintenance of OM lipid asymmetry through observation of increased activity of PagP , the role of the MlaFEDB complex and MlaC in retrograde GPL transport has previously only been inferred from homology to the chloroplast TGD system . It is established that cellular mechanisms exist in Gram-negative bacteria to resist stressful conditions that lead to OM disruption . For example , OM phospholipase enzymes , such as PldA , are activated under conditions of membrane stress to digest GPL in the outer leaflet of the OM , as high levels of GPL in the outer leaflet destabilize the OM barrier function . The model of retrograde GPL transport by the Mla system proposes that growing cells expend cellular energy in the form of ATP in order to transport undigested GPL from the OM , across the periplasm , and back into the IM , at which point some of those same molecules will be transported back to the OM by an unknown mechanism . However , the available data points most clearly to a model of anterograde GPL transport by MlaFEDB and MlaC , facilitated in some way by MlaA . The first three genes of the mla operon – comprising an ATPase , permease , and substrate-binding components of the ABC transporter complex – are conserved in Mycobacteria spp , Actinobacteria , and chloroplasts , while the entire five-gene operon appears to be conserved in Gram-negative bacteria ( Casali and Riley , 2007 ) . Given the conservation of the system across Gram-negative species , our results may shed light on a generalized mechanism contributing to OM biogenesis . Additionally , we have here demonstrated that the function of this ABC transport system is crucial for maintaining the integrity of the A . baumannii OM . The fact that mla mutations are tolerated , and that levels of OM GPL are reduced but not abolished , suggests the intriguing possibility of additional undiscovered mechanisms of GPL delivery to the OM . Also of interest is the potential role of the increased exopolysaccharide observed upon disruption of the Mla system . It is possible this exopolysaccharide plays a partially compensatory role in A . baumannii resulting from decreased OM GPL , given that recent work has shown that A . baumannii exopolysaccharides can contribute to antibiotic resistance , likely through improved barrier function ( Geisinger and Isberg , 2015 ) . The progression towards a more complete understanding of intermembrane GPL transport and OM barrier function should ultimately have relevance in the development of novel drug targets to undermine emerging antibiotic resistance in Gram-negative pathogens . The emergence of antibiotic resistant Gram-negative bacteria for which few or no antibiotics are available therapeutically is an important medical concern . This issue is typified by current isolates of A . baumannii that can only be treated with relatively toxic colistin antibiotics . This has led many individuals and agencies to propose the development of single agent antimicrobials which could be used for organisms such as A . baumannii and P . aeruginosa that have significant antibiotic resistance . Therefore , work furthering the understanding of the OM barrier could lead to the development of drugs which target the barrier and allow the therapeutic use of many current antibiotics . Transposon mutagenesis and subsequent chromosomal deletions of mla genes were performed in Acinetobacter baumannii ATCC 17978 . To perform transposon mutagenesis a Mariner-based transposon vector was designed for use in Acinetobacter baumannii ATCC 17978 . The new transposon vector , derived from pBT20 , termed pMarKT , contains an outward facing pTac promotor as well as a selectable kanamycin resistance marker followed by an omega terminator within the Mariner arm sites ( Kulasekara et al . , 2005 ) . The plasmid backbone contains the Mariner transposase gene C9 Himar , a tetRA resistance marker from Tn10 , a p15A origin from pACYC184 , and an oriT site for mobilization . The plasmid was constructed by PCR of select fragments followed by restriction digest and ligation of the cleaved ends . The new transposon vector was confirmed by restriction digest and partial sequencing . Initial mutagenesis revealed that many hits occurred in the high affinity phosphate uptake transcriptional repressor phoU ( A1S_0256 ) . Subsequent rounds of mutagenesis were conducted on an ATCC 17978 phoU chromosomal deletion strain , and plated on high phosphate media to reduce the background level of cleavage of the chromogenic substrate . Chromosomal deletions were performed by allelic exchange using a pEX2tetRA vector , which was created by insertion of the tetRA tetracycline resistance marker from Tn10 into the pEXG2 plasmid ( Rietsch et al . , 2005 ) . Roughly 1000 bp regions upstream and downstream of the genes of interest were amplified for homologous recombination with the ATCC 17978 chromosome . Sucrose was used to counter-select against cells retaining the pEX2tetRA backbone , and deletions were confirmed by PCR . Complementation of deletions was accomplished by repairing the original deletion in the chromosome , again using the pEX system and allelic exchange . Donor E . coli containing the pMarKT transposon vector were suspended in LB broth to an OD600 of 40 and mixed with an equal volume of the recipient A . baumannii suspended to OD600 of 20 . 50 µL aliquots of this mixture were then plated in spots on a dried LB agar plate and incubated for 2 hr at 37°C ( Kulasekara et al . , 2005 ) . Each 50 µL spot resulted in about 80 , 000 colonies of A . baumannii containing Mariner transposon insertions . The mutants were plated on LB agar containing 1X M63 salts , 50 µg/mL kanamycin , 30 µg/mL chloramphenicol , and 40 µg/mL XP substrate . Plates were incubated for at least 36 hr at 30°C to allow for the appearance of the blue color from cleavage of the XP substrate . Sequencing of the transposon insertions was adapted from the method described in Chun et al . ( 1997 ) , including semi-arbitrary two-step PCR amplification of transposon regions followed by sequencing . Bacteria were grown in 5 mL LB cultures to mid-log OD600 ( 0 . 3–0 . 6 ) , then spun down and normalized in PBS to OD600 0 . 2 . Prior to measurement , CCCP was added at 200 µM to inhibit the activity of efflux pumps . Ethidium bromide was added immediately prior to measurement to final concentration of 1 . 2 µM in 200 µL total reaction volume . Permeability was assessed using a PerkinElmer EnVision 2104 Multilabel Reader using a 531 nm excitation filter , 590 nm emission filter , and a 560 nm dichroic mirror . Readings were taken every 15 s for 30 min with samples assessed in triplicate in a Greiner bio-one 96-well flat bottom black plate . MICs were determined in 96-well microtiter plates using a standard two-fold broth dilution method of antibiotics in LB broth . The wells were inoculated with 104 bacteria per well , to a final well volume of 100 μL , and plates were incubated at 37°C with shaking unless stated otherwise . Experiments were performed thrice using two technical replicates per experiment . MICs were interpreted as the lowest antibiotic concentration for which the average OD600 across replicates was less than 50% of the average OD600 measurement without antibiotic . Strains were inoculated to OD600 0 . 05 and grown overnight at 37°C in 2 mL LB broth with shaking in glass tubes . The next day , liquid was carefully decanted and the tubes left to dry for 2 hr at 37°C . Pellicles were stained with the addition of 0 . 1% crystal violet , then gently washed three times in dH2O . Crystal violet was solubilized in a 80:20 solution of ethanol:acetone and read at 590 nm . P values were determined from a Student’s t-test over three biological replicates per sample . The mlaFEDCB operon from the genome of A . baumannii ATCC 17978 was subcloned into the pET-28a vector ( Novagen , US ) with a hexahistidine ( −6HIS ) tag fused at the C-terminus of the MlaB protein . The nucleotide sequence of the operon was confirmed using DNA sequencing . The plasmid was transformed into E . coli RosettaBlue strain . Cells were grown at 37°C in LB medium until the cell density reached an OD600 of 1 . 0 . The temperature was then reduced to 16°C before induction with 1 mM isopropyl β-D-thiogalactoside ( IPTG ) . After growth at 16°C for 18 hr , cells were harvested by using centrifugation at 4 , 200 g . Cells were resuspended in ice-cold buffer A ( 20 mM Tris-HCl ( pH 8 . 0 ) , 150 mM NaCl , 5% ( v/v ) glycerol ) and subjected to three runs of homogenization at 10 , 000–15 , 000 psi using Avestin EmulsiFlex-C3 high pressure homogenizer ( Avestin , Ottawa , Ontario , Canada ) . The homogenate was centrifuged at 17 , 000 g for 10 min at 4°C , and then the supernatant was ultra-centrifuged at 100 , 000 g for 60 min . The membrane fraction was resuspended in buffer A supplemented with 1% ( w/v ) dodecyl-β-d-maltopyranoside ( DDM ) and was slowly stirred for 1 hr at 4°C . After another ultra-centrifugation at 100 , 000 g for 30 min , the supernatant was collected and loaded on 2 ml of Ni2+-nitrilotriacetate affinity resin ( Ni-NTA from Qiagen , Germany ) pre-equilibrated with buffer A supplemented with 5 mM imidazole and 0 . 025% ( w/v ) DDM . After incubating for 1 hr , the resin was washed with 50 ml buffer A supplemented with 20 mM imidazole and 0 . 025% ( w/v ) DDM . The protein sample was eluted with 10 ml elution buffer containing buffer A , 300 mM imidazole , and 0 . 025% ( w/v ) DDM , and was concentrated to 0 . 5 ml . The concentrated protein sample was then loaded onto a Superdex-200 column ( 10/30 , GE Healthcare , US ) pre-equilibrated with 20 mM Hepes ( pH 7 . 0 ) 150 mM Nacl 0 . 025% DDM . Peak fractions were collected and the pooled protein sample was concentrated to 1 mg/ml . Purified Mla complex at ~1 mg/ml was applied to glow-discharged holey grids , blotted for 6 s , and plunged in liquid ethane using a Vitrobot ( FEI ) . Images were acquired on a FEI Tecnai G2 F20 200 kV Cryo-TEM equipped with a Gatan K-2 Summit Direct Electron Detector camera with a pixel size of 1 . 26 Å/pixel . 500 micrographs were collected using Leginon ( Suloway et al . , 2005 ) spanning a defocus range of −1 to −2 µm . Movie frames were aligned with MotionCorr2 ( Zheng et al . , 2017 ) and the defocus parameters were estimated with CTFFIND4 ( 49 ) . 333 high-quality micrographs were selected by manual inspection , from which ~ 55 , 000 particles were picked with DOG in Appion ( Lander et al . , 2009 ) . Particle stacks were generated in Appion using a box size of 200 pixels . Several successive rounds of 2D and 3D classification were performed in Relion 2 ( Scheres , 2012; Kimanius et al . , 2016 ) using an initial model generated by Common Lines in EMAN2 ( 53 ) leading to a final stack of ~14 , 000 particles for 3D structure refinement in Relion . The structures of MlaB and MlaF were modeled using the threading server Phyre ( Kelley et al . , 2015 ) based on the structures of the anti-sigma factor antagonist tm1081 ( PDB ID 3F43 , 18% sequence identity to MlaB ) and the ABC ATPase ABC2 ( PDB ID 1OXT , 36% sequence identity to MlaF ) respectively . Two copies of each structural model were positioned in their putative location within the EM map using Chimera ( Pettersen et al . , 2004 ) and their position was optimised using the Fit to EM map option . The abMlaD and abMlaE structures were modelled on ecMlaD and ecMlaE structural models deposited in the Gremelin database ( Ovchinnikov et al . , 2017 ) , using Modeller . For abMlaD , the N-terminal TM helix and the insert region were modelled ab initio using the Rosetta suite ( DiMaio et al . , 2011 ) and positioned in their putative localisation in Chimera . The MlaD hexamer , as well as the MlaE dimer , were modelled with Rosetta using a EM-guided symmetry modelling approach described previously ( Bergeron et al . , 2013 ) . The final model was refined with Rosetta . Cells were resuspended in 20 mL of 0 . 5 M sucrose , 10 mM Tris pH 7 . 8 , 75 µg freshly prepared lysozyme ( Roche 10837059001 ) , and 20 mL of 0 . 5 mM EDTA , and kept on ice with gentle stirring for 20 min . Samples were homogenized ( Avestin EmulsiFlex-C3 ) and spun down at 17 , 000 g for 10 min to removed un-lysed cells prior to ultracentrifugation . Membranes were spun down using a Ti45 Beckman rotor at 100 , 000 g for 1 hr and then added to the top of a sucrose gradient . IM and OM were separated by 18 hr ultracentrifugation using a SW-41 rotor in a Beckman Coulter Optima L90X ultracentrifuge . Spheroplast formation and sucrose gradient separation of IM and OM was adapted from a method by Osborn et al . ( 1972 ) by use of a defined 73%–53–20% sucrose gradient as described in Dalebroux et al . ( 2015 ) . Our sucrose gradients contain three distinct concentrations of sucrose , and inner and outer membranes separate into distinct bands that are collected individually ( Figure 5—figure supplement 3F ) . To limit any potential mixing of the membranes , the inner membrane is collected from the top of the tube while the outer membrane is collected by puncturing the bottom of the tube and allowing the bottom band to be collected . The purity of membrane separation by this method was confirmed by NADH assay and by Western blotting for the A . baumannii OM-localized OmpA protein , with 10 µg of total protein loaded into each lane as measured by Bradford protein assay ( Figure 5—figure supplement 3 ) . GPLs from isolated membranes were extracted using a 0 . 8:1:2 ratio of water: chloroform: methanol as per the method of Bligh and Dyer ( Bligh and Dyer , 1959 ) . Two-dimensional TLC was performed using silica gel 60 plates and immersion in Solvent System A ( 60:25:4 CHCl3:CH3OH:H2O ) , followed by Solvent System B ( 80:12:15:4 CHCl3:CH3OH:CH3COOH:H2O ) in the orthogonal direction . Primers were designed to amplify the mlaC gene of ATCC 17978 , excluding the signal sequence for export from the cytoplasm , and the periplasmic domain of mlaD of ATCC 17978 , excluding the membrane-spanning domain . These fragments were cloned into pET29b and expressed with a carboxy-terminal hexahistidine ( −6HIS ) tag in BL21 E . coli with 2 hr induction . Cells were pelleted and resuspended in Tris-buffered saline containing 10% glycerol ( TBSG ) and protease inhibitor cocktail ( Roche , Complete EDTA-free ) . Cells were lysed by homogenization ( Avestin ) and ultracentrifuged at 100 , 000 g for 1 hr to spin down membranes . The supernatants were then applied to a 5 mL-HiTrap ( TM ) Chelating HP Ni-affinity column pre-loaded with 0 . 1 M NiSO4 and equilibrated with TBSG . The proteins were eluted from the column using FPLC ( Akta ) by applying a stepwise gradient of 25 mM , 50 mM , and finally 300 mM imidazole for protein elution . Elution was monitored by UV-absorption at 280 nm . The MlaC- and MlaD-containing fractions were then further purified by injecting into a HiLoad 120 ml-6/600 Superdex ( TM ) 200 preparative grade size-exclusion column equilibrated in TBSG using a flow rate of 1 mL/min . The purity of the collected protein fractions was confirmed by SDS polyacrylamide gel electrophoresis . Proteins were diluted to 2 mg/mL and dialyzed overnight in 1 L TBSG at 4°C with stirring . GPLs were extracted from 1 mg each of purified proteins MlaC and MlaD by the method of Bligh and Dyer and analyzed by LC-MS/MS as previously described . Retention of PG , CL , PE , and Lyso-CL was achieved at a flow rate of 0 . 3 mL/min using mobile phase A [CHCl3/CH3OH/NH4OH ( 800:195:5 v/v/v ) ] and mobile phase B [CHCl3/CH3OH/NH4OH ( 600:340:5 v/v/v ) ] . The chromatography method used is a three-step gradient as described in the SI Materials and methods of Dalebroux et al . ( 2014 ) . The samples were run on an Agilent Zorbax Rx-SIL silica column ( 2 . 1 × 100 mm , 1 . 8-Micron ) using an Agilent HPLC autosampler . Mass spectrometry was performed using an AB Sciex API4000 Qtrap with multiple reaction monitoring ( MRM ) . The identities of the major GPLs present in the A . baumannii membrane were predicted by parent ion scans . The Q1/Q3 transitions of glycerolphospholipids from cells grown in 2-13C acetate were determined using a Thermo Orbitrap LTQ . The integrated peak areas of both 13C-labeled and unlabeled GPLs from the AB Sciex API4000 Qtrap were used to calculate the ion-current ratios for each GPL species . The ratio of labeled GPL for each unique species can be calculated based on the following equation: Rlab = Ri Rb ( MacCoss et al . , 2001 ) Where Ri is the ion-current ratio of labeled GPL to unlabeled GPL within the sample and Rb is the ion-current ratio of samples before the administration of the tracer , 13C-acetate , and represents the natural background abundance of the stable isotope species within the bacterial membrane . Rlab approximates the molar ratio of labeled species to unlabeled species ( nlab/nun ) according to the equation ( nlab/nun ) = [Ri-Rb]/k , where k is the molar response factor of the instrument and is ideally equal to unity ( MacCoss et al . , 2001 ) . To demonstrate that OM phospholipases will not distinguish between labeled and unlabeled GPL and therefore will not affect the ratio of labeled to unlabeled GPL obtained from this assay , we compared ratios of labeled and unlabeled GPL from wild type A . baumannii and deletion mutants in pldA . Bacteria were grown carrying either the empty pMMB::kan vector , or expressing the Walker box mutant MlaFK55L . Accumulation of newly synthesized GPL was observed in those strains expressing MlaFK55L when compared to the vector control , across various species of GPL . Of strains expressing the vector control , on average 51 . 84 ± 1 . 07% and 52 . 07 ± 1 . 23% of newly synthesized PG C16:0/18:1 appeared on the inner membrane of wild type and ∆pldA , respectively , after one hour incubation with 13C acetate , while 66 . 33 ± 1 . 23% and 62 . 60 ± 1 . 70% of newly synthesized PG C16:0/18:1 accumulated at the inner membranes of wild type and ∆pldA expressing MlaFK55L . In vector controls strains , 48 . 53 ± 1 . 37% and 51 . 01 ± 0 . 55% of newly synthesized PG C16:0/16:0 appeared on the inner membrane of wild type and ∆pldA , respectively , after one hour incubation with 13C acetate , while 62 . 98 ± 1 . 01% and 60 . 41 ± 1 . 25% of newly synthesized PG C16:0/16:0 accumulated at the inner membranes of wild type and ∆pldA expressing MlaFK55L . In vector controls strains , 50 . 17 ± 1 . 31% and 50 . 49 ± 1 . 15% of newly synthesized PE C16:0/18:1 appeared on the inner membrane of wild type and ∆pldA , respectively , while 60 . 14 ± 0 . 93% and 62 . 06 ± 1 . 07% of newly synthesized PE C16:0/18:1 accumulated at the inner membranes of wild type and ∆pldA expressing MlaFK55L . Cultures of A . baumannii ATCC 17978 were grown in M63 media containing 5 mM sodium acetate and 4 mM MgCl2 to OD600 0 . 4 , then washed and resuspended in media containing 5 mM 2-13C sodium acetate ( Cat . No . CLM-381–0 , Cambridge Isotope Laboratories , Inc . ) . Membrane fractions were isolated from both wild type and mla mutant A . baumannii at simultaneous time points , and GPL were extracted and assessed using previously established LC-MS/MS methods with additional MRM values to account for the increased m/z ratios of 13C-labeled GPL . MRMs were selected to account for PG and PE having acyl chains of either C16:0/16:0 , C16:0/18:1 , and C18:1/18:1 as these were determined by total ion scan MS to be the predominant species of PG and PE GPL . Pulse experiments were performed at least twice for each mutant .	Gram-negative bacteria are a large group of single-celled organisms that share a typical external envelope . This casing is formed of an inner and an outer membrane , which have different structures and properties . The outer membrane lets nutrients penetrate inside the cell , but blocks out other compounds , such as antibiotics . It is made of a complex assembly of molecules , including glycerolphospholipids ( GPL ) that are produced inside the cells . Very little is known about how this external shield is created and maintained . For example , it was still unclear how GPLs were exported through the inner membrane to the outer one . To investigate these questions , Kamischke et al . exposed a species of Gram-negative bacteria to a molecule that is normally blocked by the outer membrane . If the outer membrane is not working properly , the compound can cross it and the cell turns blue . Kamischke et al . then introduced genetic changes at random locations in the genomes of the bacteria . If colonies became blue , this meant that the mutations had happened in a gene essential for the outer membrane . Sequencing these blue bacteria revealed 58 genes involved in keeping the outer membrane working properly . Amongst them , four genes coded for a transport machine , the Mla system , which allowed GPLs to cross the inner membrane and reach the outer membrane . The experiments also showed that a working Mla system was required for bacteria to survive antibiotics . Certain dangerous Gram-negative bacteria are now resistant to many drugs , having evolved unique envelopes that keep antibiotics at bay . By learning more about the outer membrane , we may be able to create new treatments to bypass or to disable this shield , for example by targeting the Mla system .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2019	The Acinetobacter baumannii Mla system and glycerophospholipid transport to the outer membrane
In order to assess the contribution of a central clock in the hypothalamic suprachiasmatic nucleus ( SCN ) to circadian behavior and the organization of peripheral clocks , we generated forebrain/SCN-specific Bmal1 knockout mice by using floxed Bmal1 and pan-neuronal Cre lines . The forebrain knockout mice showed >90% deletion of BMAL1 in the SCN and exhibited an immediate and complete loss of circadian behavior in constant conditions . Circadian rhythms in peripheral tissues persisted but became desynchronized and damped in constant darkness . The loss of synchrony was rescued by light/dark cycles and partially by restricted feeding ( only in the liver and kidney but not in the other tissues ) in a distinct manner . These results suggest that the forebrain/SCN is essential for internal temporal order of robust circadian programs in peripheral clocks , and that individual peripheral clocks are affected differently by light and feeding in the absence of a functional oscillator in the forebrain . The suprachiasmatic nucleus ( SCN ) in the hypothalamus is the primary regulator of daily rhythms of behavior and physiology in mammals ( Meijer and Rietveld , 1989; Ralph et al . , 1990; Sujino et al . , 2003; Welsh et al . , 2010 ) , yet the majority of tissues in the body possess autonomous cellular oscillators ( Yamazaki et al . , 2000; Abe et al . , 2002; Nagoshi et al . , 2004; Welsh et al . , 2004; Yoo et al . , 2004; Abraham et al . , 2005 ) . The circadian clock in the SCN as well as in other mammalian cells is composed of interacting positive- and negative-transcriptional and post-translational feedback loops involving Clock and Bmal1 transcription factors and their target genes , Period ( Per1 , 2 , and 3 ) and Cryptochrome ( Cry1 and 2 ) ( King et al . , 1997; Gekakis et al . , 1998; Hogenesch et al . , 1998; Kume et al . , 1999; Lee et al . , 2001; Lowrey and Takahashi , 2004 , 2011; Huang et al . , 2012; Mohawk et al . , 2012 ) . Extensive evidence suggests that the SCN controls not only behavioral rhythms but also the circadian programs of peripheral tissues ( Yoo et al . , 2004; Guo et al . , 2005 , 2006; Saini et al . , 2013; Yamaguchi et al . , 2013 ) . However , much remains unclear regarding the relationship between the central clock and peripheral oscillators , which can be entrained by many different stimuli ( Stokkan et al . , 2001; Buhr et al . , 2010; Saini et al . , 2012 ) . To clarify the contribution of a central clock to the circadian rhythms of peripheral tissues , we sought to find a way to disable the molecular oscillators in the brain without affecting the genetic components in the rest of the body . Traditionally , a functional role of a gene has been investigated by a gene targeting strategy , which eliminates the gene from embryonic stem cells so that it is inactivated ubiquitously in mice . While this approach provides a powerful method to study the function of genes in vivo , it cannot be applied to genes that affect developmental or metabolic processes crucial for survival . Moreover , even if the knockout mice survive to adulthood , some lines suffer from systemic conditions and hence require special handling in experiments . For instance , although a global knock out of Bmal1 ( Bmal1−/− ) completely eliminates circadian rhythms ( Bunger et al . , 2000 ) , the animals suffer from morbid conditions including arthropathy ( Bunger et al . , 2005 ) , sterility ( Alvarez et al . , 2008; Boden et al . , 2010 ) , defects in glucose homeostasis ( Rudic et al . , 2004; Lamia et al . , 2008; Marcheva et al . , 2010 ) , premature aging , and decreased lifespan [their average lifespan is 37 weeks ( Kondratov et al . , 2006; Sun et al . , 2006 ) ] . Possibly due to the pleiotropic effects of Bmal1 , results of studies such as food entrainment behavior with Bmal1−/− mice have been controversial ( Fuller et al . , 2008; Mistlberger et al . , 2008; Fuller et al . , 2009; Pendergast et al . , 2009; Storch and Weitz , 2009; Mistlberger et al . , 2009a , 2009b; Mieda and Sakurai , 2011 ) . To circumvent these problems , conditional knockout techniques utilizing Cre recombinase and loxP sequences have been applied widely in the nervous system ( Gavériaux-Ruff and Kieffer , 2007; Taniguchi et al . , 2011 ) . However , until recently , Cre drivers that can efficiently recombine a floxed allele in the majority of SCN neurons have not been identified ( Mieda and Sakurai , 2011; Musiek et al . , 2013 but see Husse et al . , 2011 ) . In this study , we examined the ability of a forebrain-specific Cre driver to excise floxed alleles of Bmal1 , a critical component of the molecular oscillator , specifically from brain regions including the SCN . The resulting knockout mice were devoid of BMAL1 expression in the forebrain/SCN but not in other peripheral tissues . Unlike Bmal1 global knockout mice , Bmal1 forebrain knockouts did not display detectable defects in reproduction , activity , body weight , or life span , thereby providing an ideal model to study circadian physiology in the absence of a central clock . These genetically engineered mice are better experimentally controlled than mice that receive SCN lesions , which is a classic method to induce arrhythmic behavior in rodents ( Meijer and Rietveld , 1989 ) . With SCN lesions , precise removal of the SCN tissue varies from mouse to mouse and requires laborious confirmation of the excision . In addition , ablation of the SCN destroys neuronal networks and communication pathways , rendering the network responses and mechanisms incapacitated . With forebrain-specific Bmal1 knockout mice , we examined the impact of defunct brain clocks on the organization of peripheral oscillators as well as their response to environmental signals such as light and food . Cre-mediated excision of Bmal1 in the forebrain/SCN confers complete arrhythmicity to circadian behavior . In contrast to the abolition of rhythms in the SCN , all peripheral tissues in these mice sustained circadian rhythmicity; however , they lacked phase coordination and normal amplitude in constant darkness . The synchrony and the oscillatory amplitude of peripheral rhythms in these mice were completely rescued by exposure to light dark cycles but not fully in time-restricted feeding experiments . The feeding cues appear to control the phase expression of the peripheral clocks in a tissue-specific way and in a distinct manner from light . These results demonstrate that an intact central clock plays an essential role in driving normal circadian behavior as well as synchronized and robust circadian programs in peripheral clocks and that light and feeding act on individual peripheral clocks differently in the absence of a functional oscillator in the SCN . Bmal1 ( also known as Arntl , Mop3 ) is one of the essential components of the molecular oscillator , unique in that the single knockout of Bmal1 can confer arrhythmicity both at the behavioral and molecular levels ( Bunger et al . , 2000 ) . To generate a forebrain-specific knockout of circadian rhythms , we crossed floxed Bmal1 ( Bmal1fx/fx ) mice ( Johnson et al . , 2014 ) to Camk2a::iCreBAC mice ( CamiCre or Cre ) ( Casanova et al . , 2001 ) and produced CamiCre+; Bmal1fx/fx ( that we denote as ‘Brain’ Knockout , BKO , defined as a forebrain-specific knockout ) as well as Bmal1fx/fx ( Fx/Fx ) , CamiCre+ ( Cre ) , and CamiCre+; Bmal1fx/+ ( Het ) control mice . Camk2a::iCreBAC mice were used because their Cre expression is faithful to the endogenous Camk2a gene expression pattern: it is neuron-specific , forebrain-enriched ( Casanova et al . , 2001 ) , and expressed highly in the SCN as observed in Cre-dependent fluorescence reporter mice , CamiCre+; tdTom+ ( Figure 1A–E ) . Although there are more than 37 different Camk2a::Cre transgenic lines reported in Mouse Genome Informatics database ( http://www . informatics . jax . org/searchtool/Search . do ? query=Camk2a-cre&page=featureList ) , the majority do not appear to express well in the SCN ( e . g . , http://connectivity . brain-map . org/transgenic/experiment/81162458 ) , likely because of shorter promoter regulatory sequences and position effects . The other advantage of this Cre driver is that its Cre expression is post-natal and starts at P3 ( Casanova et al . , 2001 ) , thus reducing potential developmental effects . Bmal1fx/fx mice were generated by inserting loxP sites into the introns surrounding exon 4 , which codes for the DNA-binding bHLH domain ( Johnson et al . , 2014 ) . Upon Cre recombination , exon 4 is deleted in BKO brain tissues that express Camk2a::iCre . The predicted BMAL1 sequence encounters a STOP codon after a stretch of 18 non-native residues following native coding exon 3 . The resulting protein is 91 amino acids long ( vs 626 amino acids of native BMAL1 ) and is almost identical to the truncated BMAL1 protein of Bmal1−/− mice ( Bunger et al . , 2000 ) ; both proteins contain no known functional domains and do not accumulate to a detectable level . 10 . 7554/eLife . 04617 . 003Figure 1 . CamiCre+; Bmal1fx/fx mice are forebrain/SCN-specific Bmal1 knockouts . ( A ) Dorsal view of a CamiCre-; tdTom+ control brain . Bright field . Scale bar: 1 mm . ( B ) Dorsal view of a CamiCre+; tdTom+ brain . Bright field . ( C ) Ventral view of ( B ) . ( D ) Fluorescence image of a coronal brain section containing the SCN from a CamiCre+; tdTom+ mouse . ( E ) Confocal image of the SCN showing tdTomato expression in the CamiCre+; tdTom+ mice . Scale bar: 50 µm . ( F , G ) Coronal brain sections at the level of the SCN , prepared from mice sacrificed every 4 hr starting at CT2 following 2 days of DD , were hybridized in situ to examine Bmal1 ( F ) and Per2 ( G ) mRNA levels in CamiCre+; Bmal1fx/fx ( BKO ) mice ( n = 3 , except n = 4 for CT10 , 14 , 18 ) and Bmal1fx/fx ( Fx/Fx ) control littermates ( same as BKO , except n = 1 for CT18 Bmal1 and n = 3 for CT18 Per2 ) . Below each coronal brain section is a close-up of the SCN . ( H , I ) Optical density graphs of the in situ time course data displayed in ( F ) and ( G ) . Shown are mean ± SEM , with significant effect of genotype in ( H ) [F1 , 46 = 40 . 7 , p < 0 . 0001] and ( I ) [F1 , 44 = 1148 , p < 0 . 0001] by GLM ANOVA . Tukey–Kramer multiple comparison post-tests ( p ≤ 0 . 05 ) showed that in contrast to Fx/Fx littermates , neither Bmal1 nor Per2 mRNA is rhythmic in BKOs . ( J ) Immunohistochemistry for BMAL1 on SCN-containing coronal sections of Fx/Fx ( n = 3 ) , BKO ( n = 3 ) , and Bmal1−/− ( KO , n = 1 ) mice sacrificed at ZT16 . Captured with an 20× objective . ( K , L ) Western blot analysis of BMAL1 in BKOs sacrificed at ZT16 . ( K ) Western blot of various tissues in Fx/Fx ( n = 2 ) , BKO littermate ( n = 2 ) , and KO ( n = 1 ) mice . ( L ) Western blot of frontal cortex , cerebellum , and liver in Cre ( n = 2 ) , Fx/Fx ( n = 2 ) , BKO ( n = 3 ) , and KO ( n = 2 ) mice . F . CTX: frontal cortex , Cer: cerebellum . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 00310 . 7554/eLife . 04617 . 004Figure 1—figure supplement 1 . Weights of Bmal1 brain knockout mice are similar to controls at 2 months and 4–6 months of age . All body weights were measured in live , wheel-naïve animals . Numbers of male mice weighed at 2 months ( = 54- to 62-day old ) and 4–6 months ( = 120- to 195-day old ) , respectively: Cre ( n = 25 , 35 ) , Het ( n = 13 , 15 ) , Fx/Fx ( n = 98 , 37 ) , and BKO ( n = 55 , 12 ) . Numbers of female mice weighed at 2 months and 4–6 months , respectively: Cre ( n = 28 , 39 ) , Het ( n = 23 , 35 ) , Fx/Fx ( n = 84 , 54 ) , and BKO ( n = 54 , 29 ) . Shown are the mean weight ±SD . A significant effect of genotype was found using GLM ANOVA , followed by Tukey–Kramer multiple comparison post-tests . The difference between Cre and Fx/Fx mice ( p ≤ 0 . 05 ) is likely due to the difference in the genetic background ( CamiCre+ is generated in FVB background and floxed Bmal1 in 129sv ) . Importantly , there was no significant difference between BKO males and control males at 4–6 months , and BKO females did not weigh less than any control group at either time point . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 00410 . 7554/eLife . 04617 . 005Figure 1—figure supplement 2 . Additional in situ analysis of Bmal1 brain knockout mice . ( A , B ) Coronal brain sections at the level of the SCN . Below each coronal brain section is a close-up of the SCN . ( A ) Mice were sacrificed at CT18 and CT6 following 2 days of DD . Sections were hybridized in situ to examine Bmal1 mRNA levels in Cre ( n = 5 at CT6 and 4 at CT18 ) , Fx/Fx ( n = 7 at CT6 and 7 at CT18 ) , Het ( n = 6 at CT6 and 6 at CT18 ) , BKO ( n = 4 at CT6 and 4 at CT18 ) , and Bmal1−/− mice ( KO , n = 1 ) . ( B ) Mice were sacrificed at CT10 and CT22 following 2 days of DD . Sections were hybridized in situ to examine Per2 mRNA levels in Cre ( n = 7 at CT10 and 6 at CT22 ) , Fx/Fx ( n = 10 at CT10 and 8 at CT22 ) , Het ( n = 7 at CT10 and 7 at CT22 ) , and BKO ( n = 9 at CT10 and 9 at CT22 ) mice . ( C , D ) Optical density of Bmal1 ( C ) and Per2 ( D ) mRNA in the SCN . Shown are mean ± SEM , with significant effect of genotype in ( C ) [F3 , 42 = 217 . 91 , p < 0 . 0001] and ( D ) [F3 , 62 = 28 . 00 , p < 0 . 0001] by GLM ANOVA . ( C ) Tukey–Kramer multiple comparison post-tests ( p ≤ 0 . 05 ) showed that BKO mice had significantly lower Bmal1 mRNA levels than all three control groups at both CT6 and CT18 . At both time points Het mice had significantly less Bmal1 mRNA than the other two controls but more mRNA than BKOs ( p ≤ 0 . 05 ) . ( D ) BKO mice had significantly lower levels of Per2 mRNA than all three control groups at CT10 but not at CT22 ( p ≤ 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 005 In contrast to global Bmal1 knockouts , the reproduction , longevity , and health of BKO mice were as robust as that of wild-type mice . The fertility of both males and females was indistinguishable from wild-type . At 2 months and 4–6 months of age , the body weights of BKO were found to be nearly identical to Fx/Fx , Cre , and Het mice ( Figure 1—figure supplement 1 ) . In order to assess the deletion of Bmal1 in the SCN , we examined the circadian mRNA expression pattern of Bmal1 and Per2 , a key Bmal1 target gene , by in situ hybridization after 2 days of constant darkness ( DD ) . At all seven time points taken over a period of 24 hr , BKOs showed a low level of Bmal1 and Per2 mRNA in the SCN as well as in the entire coronal brain section ( Figure 1F , G ) . In more detailed analysis , control mice ( Cre , Fx/Fx ) showed intense mRNA expression of Bmal1 in the SCN at its peak time ( CT18 ) as well as circadian fluctuation ( CT18 vs CT6 ) ( Figure 1F , H; Figure 1—figure supplement 2 ) . Accordingly , Per2 mRNA was also robustly expressed in the SCN in a circadian manner ( Figure 1G , I; Figure 1—figure supplement 2 ) . Het mice showed reduction in the Bmal1 mRNA abundance , yet cycling of Per2 mRNA was retained in the SCN ( Figure 1—figure supplement 2 ) , indicating that 50% wild-type Bmal1 mRNA is sufficient to sustain Per2 expression in the SCN . In contrast , in BKO mice , Bmal1 mRNA was reduced ∼90% in the SCN ( Figure 1F , H; Figure 1—figure supplement 2 ) , and Per2 mRNA cycling was significantly attenuated ( ∼20% WT levels at CT10 , see Figure 1G , I ) . To confirm that no BMAL1 protein was expressed in the SCN of BKOs , we performed immunohistochemistry in the SCN at Zeitgeber Time ( ZT ) 16 . In contrast to the intense BMAL1 staining in the SCN of Fx/Fx controls , BMAL1 staining in the BKO and global Bmal1 knockout SCN was reduced to background levels ( Figure 1J ) . To further characterize the tissue specificity of the BKOs , we used Western blot analyses to measure BMAL1 protein levels in various brain and body tissues at ZT16 ( Figure 1K , L ) . In contrast to Fx/Fx and Cre controls , BMAL1 ( ∼69 kD ) was detected only faintly in BKO frontal cortex and olfactory bulb ( Figure 1K ) , likely due to glial and non-neuronal expression of BMAL1 since Camk2a is not expressed in glial cells ( Lin et al . , 1987 ) . Levels of BMAL1 in the cerebellum of BKOs were similar to Fx/Fx and Cre control mice ( Figure 1L ) . Outside of the brain , like Fx/Fx mice but unlike global Bmal1−/− mice , BKOs showed equivalent BMAL1 expression levels in the liver , spleen , kidney , heart , and lung ( Figure 1K ) . Taken together , these results confirm the forebrain specificity of our Bmal1 conditional knockouts . To assess whether forebrain/SCN Bmal1 is necessary for normal circadian behavior , we examined circadian rhythms of locomotor activity in Cre , Fx/Fx , Het , and BKO mice in a 12 hr:12 hr light:dark ( LD ) cycle followed by DD and constant light ( LL ) conditions ( Figure 2 ) . In both DD and LL , control mice ( Cre , Fx/Fx , and Het ) expressed robust circadian rhythms of activity ( Figure 2A , C , D ) . The locomotor activity of Het mice was indistinguishable from that of the other controls . In contrast to controls , Bmal1 brain knockouts exhibited no circadian rhythm of activity in DD or in LL ( Figure 2B–D ) . Out of 31 BKO mice , none had a period in the circadian ( 18–30 hr ) range in DD . In LL , only 2 out of 21 mice had a period in the circadian range ( 24 . 1 and 20 . 9 hr ) . However , they showed extremely low amplitudes of circadian activity ( 0 . 025 and 0 . 019 , respectively , fraction of power in the circadian range , where 1 . 0 represents the total power in the normalized power spectrum ) , and spectral analysis using Fast Fourier Transform ( FFT ) did not detect periodicity in the circadian range in any of the BKOs under either DD or LL conditions ( Figure 2D ) . 10 . 7554/eLife . 04617 . 006Figure 2 . Complete loss of circadian rhythmicity of Bmal1 brain knockout mice in constant darkness and constant light . ( A , B ) Representative actograms of daily wheel-running activity of Cre , Fx/Fx , Het , and BKO mice . Activity records were double plotted , with each day being represented beneath and also to the right of the preceding day . Horizontal black and white bars at the top of each actogram represent lights off and on , respectively . Mice were housed in LD , released into DD for 4 weeks , returned to LD for 2 weeks , released into LL for 4 weeks , and then returned to LD . GLM ANOVA and Tukey–Kramer multiple comparison post-tests were used to compare genotypes tested . BKOs showed no significant periodicity in DD or LL . ( C–E ) Period ( C ) , amplitude of circadian rhythm ( D ) , and activity levels ( E ) in Cre mice ( n = 28 for DD , 12 for LD , 12 for LL ) , Fx/Fx mice ( n = 31 for DD , 18 for LD , 18 for LL ) , Het mice ( n = 31 for DD , 13 for LD , 15 for LL ) , and BKO mice ( n = 31 for DD , 16 for LD , 21 for LL ) . Bar graphs show mean ± SEM . ( C ) Free-running period was determined using χ2 periodogram for days 1–28 of DD or LL . A significant effect of genotype on period was found for both DD [F3 , 120 = 164 , 319 , p < 0 . 0001] and LL [F3 , 65 = 20 , 429 , p < 0 . 0001] . In DD , Cre mice were found to have a slightly longer period than the other two control groups ( p ≤ 0 . 05 ) . In LL , Het mice were found to have a shorter period than the other two control groups ( p ≤ 0 . 05 ) . Nevertheless , all three control groups showed a free-running period in both DD and LL similar to that reported for WT mice . ( D ) DD and LL amplitude of circadian rhythm represented by FFT in the circadian range . A significant effect of genotype on amplitude was found for both DD [F3 , 120 = 150 , p < 0 . 0001] and LL [F3 , 65 = 23 . 72 , p < 0 . 0001] , with BKO mice having significantly lower amplitude ( p ≤ 0 . 05 ) in both DD and LL . ( E ) Total daily DD , LD , and LL activity levels , in wheel revolutions per 24 hr . BKO activity levels were found to be significantly higher than those of controls in DD [F3 , 119 = 9 . 00 , p < 0 . 0001] ( p ≤ 0 . 05 ) and LL [F3 , 65 = 5 . 84 , p = 0 . 0014] ( p ≤ 0 . 05 ) but not in LD . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 006 Unexpectedly , total daily activity of BKOs was significantly higher than that of control mice both in DD and LL ( Figure 2E ) . Previous studies have found that the locomotor activity of Bmal1−/− mice is decreased dramatically below wild-type levels ( Bunger et al . , 2000 ) and that transgenic BMAL1 expression in skeletal muscle rescues this diminished activity ( McDearmon et al . , 2006 ) . Taken together , these results suggest that a peripheral defect in BMAL1 is responsible for the low activity levels that had been reported in global Bmal1−/− mice . In BKOs ( which do not have this peripheral defect so that locomotion is normal ) , the absence of a central circadian gating mechanism , which normally suppresses activity during the subjective daytime ( Davis and Menaker , 1980 ) , likely contributes to an increase in overall levels of activity . One of the notable differences between BKO and SCN-lesioned mice is that BKOs display apparent entrainment to light/dark cycles ( probably because BKOs are spared from surgical interruption of retinal inputs to the SCN which usually occurs in lesion experiments ) ( Figure 2 ) . However , the activity onset of BKOs in LD was variable and on average earlier than controls . As a result , daytime activity level was higher in BKOs ( 15 . 58 ± 1 . 8% of total daily activity ) compared to the three control groups ( 3 . 81 ± 0 . 96% , average of the three control groups ± SEM ) ( Figure 3—figure supplement 1 ) . Yet it was unclear whether the pre-dark activity of Bmal1 brain knockouts around the light/dark transition was due to poor entrainment of a residual circadian pacemaker or to a defect in masking . In order to test whether Bmal1 brain knockouts entrain to light , we used a skeleton photoperiod ( LDLD 1:10:1:12 , Figure 3A–E ) . The activity patterns of Cre , Fx/Fx , and Het mice during the skeleton photoperiod were virtually indistinguishable from the activity patterns in LD 12:12 ( Figure 3A ) . In contrast , none of the 10 BKO mice used in this study entrained normally to the skeleton photoperiod ( Figure 3B ) : 3 BKO mice showed activity that resembled their activity in DD ( Figure 3B , left ) , 2 BKO mice displayed a bimodal activity pattern ( Figure 3B , right ) , and 5 BKO mice exhibited activity that was intermediate between these two phenotypes ( Figure 3B , middle ) . Unlike control mice , BKO mice distributed their running activity equally between the subjective day and the subjective night of the skeleton photoperiod ( Figure 3C , D ) . In fact , controls spent 97 . 9% of their activity during the skeleton night , compared to only 52 . 5% for BKO mice ( Figure 3E ) . The amount of activity that BKOs spent during skeleton day vs night was not significantly different . This type of abnormal entrainment to skeleton photoperiods is reminiscent of the entrainment behavior of pinealectomized sparrows , which behave as a population of circadian oscillators that entrain with two opposite phase relationships to the skeleton photoperiod ( Takahashi and Menaker , 1982 ) . Therefore , we conclude that entrainment in Bmal1 brain knockout mice is abnormal under skeleton photoperiods , consistent with the idea that the coherence of phase during entrainment may be compromised . Overall this suggests that the periodicity seen in LD 12:12 in BKOs arises from entrainment of a population of driven oscillators in which phase control is labile . 10 . 7554/eLife . 04617 . 007Figure 3 . Both entrainment and masking are abnormal in Bmal1 brain knockout mice . ( A , B ) Representative double-plotted actograms of daily wheel-running activity of Cre , Fx/Fx , Het , and BKO mice . Mice were housed in LD , released into DD for 4 weeks , returned to LD for 2 weeks , introduced to a skeleton photoperiod ( LDLD 1:10:1:12 ) for 4 weeks , returned to LD for 2 weeks , and then switched to an ultradian cycle ( LD 3 . 5:3 . 5 ) for 3 weeks . Shown are days 1–24 of the skeleton photoperiod and days 3–15 of LD 3 . 5:3 . 5 . ( C , D ) Average activity profiles of Fx/Fx ( n = 12 ) and BKO littermates ( n = 10 ) during LDLD 1:10:1:12 . ( C ) Average profiles for individual mice; ( D ) the ensemble average for the entire group . Each data point represents counts per minute averaged across a 6-min bin ( ±SEM for D ) . ( E ) Total activity counts ( mean ± SEM ) during 10 hr of subjective day ( Sk . Day ) and 10 hr of subjective night ( Sk . Night ) of LDLD 1:10:1:12 . In contrast to Fx/Fx littermates , BKOs are equally active during skeleton day vs skeleton night . p < 0 . 0001 by t-test . ( F ) Masking of BKO ( n = 14 ) and Fx/Fx littermates ( n = 9 ) during specific 0 . 5 hr light bins of LD 3 . 5:3 . 5 . Total activity ( mean ± SEM ) during each 30 min light bin is divided by the total of that bin plus the corresponding dark bin . Thus the first data point for each genotype represents the distribution of activity between the first 30 min of light and the first 30 min of dark . There was a significant interaction between the genotype and time using Repeated Measures GLM ANOVA [F1 , 160 = 22 . 3 , p < 0 . 0001] . Specifically , days 1 , 6 , and 7 were found to be significantly different by Tukey–Kramer multiple comparison post-tests ( p ≤ 0 . 05 ) . ( G ) Representative double-plotted actograms of daily wheel-running activity of Fx/Fx ( left ) and BKO ( right ) littermates . These are the same mice as those shown in ( A , middle panel ) and ( B , middle panel ) , respectively . Light phases are indicated in yellow to show the structure of the LD 3 . 5:3 . 5 cycle as well as to help visualize the occurrence of wheel-running activity under this schedule . Note that , after 1 week of this cycle , initial phase relationships are regained . ( H ) Representative amplitude power spectra of Fx/Fx and BKO mice , from FFT analyses performed on the same activity records shown in ( G ) . The highest peaks for both genotypes were in the 7-hr range , corresponding to the LD 3 . 5:3 . 5 schedule . The second-highest peak for the controls was in the circadian ( 18–30 hr ) range ( left ) . None of the BKOs was found to have a period in the circadian range ( right ) . ( I ) Group averages of amplitudes in the 7-hr range and in the circadian range for the same Fx/Fx and BKO mice as in ( F ) . Compared to controls , BKOs had significantly higher amplitude in the 7-hr range ( *p < 0 . 0001 by t-test ) and significantly lower amplitude in the circadian range ( p < 0 . 0060 by t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 00710 . 7554/eLife . 04617 . 008Figure 3—figure supplement 1 . Masking in Bmal1 brain knockout mice . ( A ) Daytime activity levels during LD , as percentage of total daily activity , in Cre ( n = 12 ) , Fx/Fx ( n = 18 ) , Het ( n = 13 ) , and BKO ( n = 16 ) mice . A significant effect of genotype on daytime activity was found by GLM ANOVA [F3 , 58 = 12 . 43 , p < 0 . 0001] , with BKO mice having significantly higher daytime activity by Tukey–Kramer multiple comparisons post-test ( p ≤ 0 . 05 ) . Combined , controls allocate an average of 3 . 81 ± 0 . 96% of their total activity to running during the daytime . In contrast , 15 . 58 ± 1 . 8% of BKO wheel-running activity occurs during the light phase . ( B–E ) Shown are days 3–15 of a 21-day LD 3 . 5:3 . 5 cycle; these are the same days as those shown in Figure 3A , B . ( B , C ) Representative single-plotted actograms of daily wheel-running activity of ( B ) Fx/Fx and ( C ) BKO littermates . These are the same mice as those shown in Figure 3A ( middle panel ) , 3B ( middle panel ) , and 3G . Actograms are plotted modulo tau ( 25 . 2 hr for B ) to help visualize the circadian ‘beating’ , that is the gaps in activity in controls ( B ) caused by the suppressive effects of the light phases of the LD 3 . 5:3 . 5 cycle on running . A circadian component of wheel-running activity is clearly evident in controls ( B ) but absent in BKOs ( C ) . ( D , E ) Representative actograms of wheel-running activity of Fx/Fx ( D ) and BKO littermates ( E ) , plotted on a 7 hr time scale ( x axis ) . Y-axis shows successive 7 hr spans for days 3–15 of this cycle . The gaps in activity in some of the dark periods in Fx/Fx controls represent times when the circadian cycle is specifying rest even though lights are off ( see Figure 3G , left panel ) . To the right of each actogram is an activity profile for that individual mouse , averaged over 13 days . Each data point represents counts per minute averaged for each mouse across a 6-min bin . The boxed area corresponds to the first one-third of all the light phases of the LD 3 . 5:3 . 5 schedule . Substantial BKO wheel-running activity is seen during this interval , especially the first 30 min , during which BKOs fail to mask altogether ( see Figure 3F ) . Horizontal black and white bars represent lights off and on , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 008 Since the skeleton photoperiod and LD photoperiod results both indicated some impairment of masking in BKOs , we next tested whether negative masking ( the ability of light to acutely suppress locomotor activity ) in Bmal1 brain knockouts was defective . To assess masking in BKO mice , we used an LD 3 . 5:3 . 5-hr light–dark cycle ( Redlin et al . , 2005 ) ( Figure 3B , F , G ) . Mice are unable to entrain to LD 3 . 5:3 . 5 cycles , thus permitting the masking and the entraining effects of light to be distinguished . Maintaining this schedule for a week ensures the testing of all phases of a circadian cycle ( Figure 3G ) . On average , BKOs exhibited significantly less activity during the light portions of LD 3 . 5:3 . 5 ( 4 . 68 ± 0 . 71% of total activity vs 9 . 56 ± 1 . 83% for Fx/Fx littermates , p < 0 . 032 by t-test ) . Fx/Fx mice display more activity during the light portions of LD 3 . 5:3 . 5 because the circadian clock is promoting activity despite the presence of light ( Figure 3G , left ) . While Fx/Fx mice have an underlying circadian component of activity under LD 3 . 5:3 . 5 , BKOs have none ( Figure 3G–I; Figure 3—figure supplement 1 ) . Aside from this difference , when we examined the time course of the masking response , it became apparent that the masking of BKOs is delayed compared to controls ( Figure 3F; a score of 50% indicates failure to mask ) . Thus BKOs failed to mask during the first 30 min of the light phases . Plotting actograms on a 7-hr time scale aids in visualization of this masking impairment in BKOs ( Figure 3—figure supplement 1 ) . These entrainment and masking studies led us to conclude that BKOs lack an endogenous clock but also exhibit some impaired masking . To analyze the temporal organization of peripheral oscillators at an organismal level , we crossed the BKOs to PER2::LUC mice ( Yoo et al . , 2004 ) and monitored the bioluminescence signals from the SCN and other tissues in the body . Compared with the robust oscillation of the control SCN , the SCN of BKOs exhibited blunted expression patterns with low-amplitude but detectable rhythms , which were close to 24 hr ( 23 . 54 ± 0 . 29 hr in Fx/Fx control SCN vs 22 . 99 ± 2 . 36 hr in BKO SCN for mean period length ± SD , Figure 4A ) . These rhythms are likely to be derived from glial cells and/or residual neurons in which Cre-mediated excision of floxed Bmal1 was incomplete ( see Figure 1J ) or possibly from residual stochastic oscillations that we have observed in the SCN of global Bmal1 knockouts ( Ko et al . , 2010 ) . Rhythmic expression in the dorsomedial hypothalamus ( DMH ) in BKOs was also significantly attenuated . In contrast to the SCN and DMH , peripheral tissues retained persistent circadian rhythms ( Figure 4A ) , consistent with the normal expression of BMAL1 in peripheral tissues ( Figure 1K ) . 10 . 7554/eLife . 04617 . 009Figure 4 . Real-time reporting of circadian expression of PER2::LUC in the forebrain/SCN knockout mice . ( A ) Representative records of bioluminescence reporting of circadian expression from various tissues in Fx/Fx and BKO mice . Tissues were prepared from mice in LD . PMT counts are plotted against the LD cycle of day 1 . Shown are 10 days of continuous recording after explant preparation , except for the SCN for which medium was changed on day 12 . ( B , C ) Period plots of various tissues harvested from Fx/Fx control ( open circle ) and BKO ( dark circle ) mice in LD ( B ) and DD ( C ) . The sample size is indicated on right . Shown are mean period ± SD . p < 0 . 01 , **p < 0 . 0001 by t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 009 These observations prompted us to use the forebrain-specific knockout mice to re-investigate the relationship between the brain's central pacemaker and circadian oscillations of peripheral tissues . Previous studies have addressed a similar question by lesioning the SCN ( Yoo et al . , 2004; Tahara et al . , 2012; Saini et al . , 2013 ) . Unlike SCN-lesioned mice , however , BKOs preserve intact neural networks and show normal or enhanced levels of behavioral activity in LD and in constant conditions . This new analysis allows us to determine whether deleting BMAL1 ( thus inactivating the molecular clock ) in the forebrain , without destroying the structure of the SCN , is sufficient to affect the organization of peripheral rhythms . Age-matched pairs of mice ( Fx/Fx littermate control and BKO ) were maintained in DD for more than 30 days ( 30–44 days ) before harvesting eight different tissues for real-time reporting of circadian gene expression . As a light synchronized control group , tissues were also collected from mice in LD 12:12 . Consistent with previous studies ( Yoo et al . , 2004; Tahara et al . , 2012 ) , mean period length was variable yet characteristic from tissue to tissue ( Figure 4B , C ) . The difference of the mean periods between the Fx/Fx control and BKO mice , however , was not significant either in LD or DD , except for the SCN and DMH ( Figure 4B , C ) . To analyze the temporal organization of the rhythms of each tissue , we constructed a phase map by plotting the peak on the second day in the luminometry recording . Fx/Fx control mice displayed tightly clustered phases in peripheral rhythms in both LD and DD ( Figure 5A , B , E ) . On the other hand , while BKOs exhibited relatively coherent rhythms in peripheral tissues in LD ( Figure 5C , E; Figure 5—source data 1A ) , their phases were significantly dispersed among animals in DD ( Figure 5D , E; Figure 5—source data 1A ) . In order to measure the degree of phase coherence of peripheral clocks within individual animals , the circular variance of peak bioluminescence of all tissues in each mouse was compared ( Figure 5—source data 1B , C ) . Statistical analysis ( Figure 5F , G ) showed that BKO cultures from the DD condition have a significantly wider phase distribution among organs ( Figure 5G , p = 0 . 0008 by Mann–Whitney test ) . This result indicates that a loss of phase coordination occurred in peripheral clocks within individual BKO animals when placed in DD . In contrast , we found that BKOs displayed well-phased rhythms from tissue to tissue when maintained in LD ( Figure 5F , N . S . by Mann–Whitney test ) , suggesting that the internal phase synchrony is still maintained even in the absence of a functional master pacemaker when the animals are exposed to LD cycles . 10 . 7554/eLife . 04617 . 010Figure 5 . Phase analysis of circadian expression of various peripheral tissues from the forebrain/SCN knockout mice . ( A ) Phase map of circadian rhythms of various peripheral tissues from Fx/Fx control mice in LD ( n = 15 , except for pituitary n = 14 ) . The peak phases ( or averaged peak phases where more than two pieces were prepared from the same tissue ) in the second cycle were plotted against the LD cycle of the day when explant cultures were prepared . Tissues from the same animal are connected by colored lines with matched symbols . ( B ) Phase map of Fx/Fx control mice in DD ( n = 15 ) . The peak phases ( or averaged peak phases ) in the second cycle were plotted against the predicted onset of activity ( =CT12 ) of the day when the explant cultures were prepared . ( C ) Phase map of BKOs in LD ( n = 15 , except for pituitary n = 14 ) . ( D ) Phase map of BKOs in DD ( n = 15 ) . The phases were mapped against the predicted onset of activity ( = CT12 ) of paired Fx/Fx control mice . ( E ) Circular plots of peak bioluminescence rhythms in the SCN and peripheral tissues presented in ( A–D ) . A circle represents a 24-hr clock , and peak phases of bioluminescence rhythms of individual tissues were calculated as angles and plotted as colored symbols outside the circle . Each tissue is denoted by the same color and symbol scheme . The direction of the arrow indicates mean phase angle , with the length of the arrow expressing the strength of phase clustering . The sample size , mean phase angle , and circular variance for each dataset are summarized in Figure 5—source data 1A . ( F , G ) Scattered plot of circular variance in each individual mouse . Degree of variance among the peak phase values of pituitary , liver , kidney , heart , lung , and spleen in each individual mouse was calculated and expressed as circular variance ( see Figure 5—source data 1B , C ) and compared between Fx/Fx controls and BKOs in LD ( F ) and DD ( G ) . The bar is ±SD . A significant effect was found between Fx/Fx control and BKO mice in DD ( p = 0 . 0008 by Mann–Whitney test ) but not in LD ( p = 0 . 0627 by Mann–Whitney test ) . N . S . = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 01010 . 7554/eLife . 04617 . 011Figure 5—source data 1 . ( A ) Summary of statistical analysis of circular plots presented in Figure 5 . ( B ) Statistical analysis of circular variance for peak bioluminescence in individual mice in LD . ( C ) Statistical analysis of circular variance for peak bioluminescence in individual mice in DD . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 011 In addition to period and phase , we also estimated amplitude because there was a trend for oscillatory amplitudes in peripheral tissues from BKOs to be lower than that of controls ( Figure 6—figure supplement 1 ) . To compare across different experiments , all time-series data were adjusted for PMT background counts and PMT gain , before applying FFT-NLLS computation as developed previously ( Izumo et al . , 2006 ) . Only normalized amplitudes were compared in this study . For the SCN and DMH in which Bmal1 is deleted from the majority of neurons , only residual amplitude was detected ( Figure 6A , B ) . Likewise , the amplitude of the pituitary was lower in BKOs from LD , which was likely due to the partial deletion of BMAL1 ( see Figure 1K ) . This decrease was further pronounced when BKO pituitary was harvested from DD ( Figure 6A , B ) . For all other peripheral tissues , the relative amplitude was significantly decreased when a light-driven signal was removed from BKOs in DD , but the severe reduction was not observed in BKOs from LD ( Figure 6A , B , except for liver ) . These damped rhythms , however , regained oscillation after receiving a stimulatory medium change ( Figure 6—figure supplement 1 ) , which suggests the possibility that phase desynchrony is taking place within individual tissues of BKOs . 10 . 7554/eLife . 04617 . 012Figure 6 . Decreased amplitude in peripheral tissues of BKOs . ( A , B ) Normalized amplitude of bioluminescence rhythms from Fx/Fx control ( open bar ) vs BKO ( dark bar ) in LD ( A ) and DD ( B ) . Mean relative amplitude ±SD are shown . The sample size is the same as in Figure 4B , C . p < 0 . 05 , p < 0 . 01 , p < 0 . 0001 by t-test . ( C ) Representative records of real-time monitoring of circadian expression of heart tissues in Fx/Fx control ( upper ) and BKO ( lower ) mice maintained in DD . ( D ) Representative frames of bioluminescence imaging with grids over each heart tissue . ( E ) Heat maps of the brightest 200 time-series data beginning with the strongest signals in the grids shown in ( D ) . ( F , G ) Linear traces of the top 50 of time-series data from the Fx/Fx ( F ) and BKO ( G ) heart tissue shown in ( E ) . The Y-axis was expanded for the BKO sample in the most right graph . ( H ) Normalized amplitude quantified from the top 50 of time-series data shown in ( F ) and ( G ) . p < 0 . 0001 by t-test . ( I ) Circular plots of peak phase values of the top 50 time-series data shown in ( F ) for Fx/Fx and ( G ) for BKO tissues . Degree of variance was compared between the two samples by bootstrapping simulation ( *p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 01210 . 7554/eLife . 04617 . 013Figure 6—figure supplement 1 . Real-time monitoring of circadian expression in forebrain/SCN knockout mice from DD . Representative records of bioluminescence reporting of circadian expression from Fx/Fx and BKO mice maintained in DD . Tissues were harvested before the predicted onset of activity ( =CT12 ) of Fx/Fx control mice . Shown are 10 days of continuous recordings followed by medium change . Quantification of period , phase , and relative amplitude of these samples are shown in Figures 4C , 5 , 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 013 To distinguish whether the decreased amplitude of BKO tissues is due to phase desynchrony between the cells or to lowered amplitude within the cells , we performed bioluminescence imaging on a peripheral BKO tissue from DD . Heart tissue was chosen for this experiment , because the heart showed both significant phase dispersion and severe amplitude reduction in DD ( Figure 5D; Figure 6B , C ) . In this imaging , bioluminescence signals from heart tissue were insufficient to identify single cells in the field of view . Therefore , we developed a grid method to allow extraction of time-series data in a narrow area of the target-imaging sample ( Figure 6D ) . The size of the grid was narrowed down to 40 µm × 40 µm . Circadian rhythms in the brightest 200 grids were aligned in the order of expression intensity in a heat map ( Figure 6E ) . Consistent with the organotypic recording by LumiCycle luminometry ( Figure 6C ) , the heart cells from a wild-type mouse exhibited robust oscillation , while those from a BKO displayed damped rhythms with broader phase distribution . The damped rhythms were also clear in a linear graph in which the top 50 time-series were plotted ( Figure 6F , G ) . Amplitude analysis of these 50 time-series data showed that the oscillatory amplitude was significantly reduced within the grids in the BKO heart sample ( Figure 6H ) . Furthermore , their phases were significantly dispersed ( Figure 6I ) . These results suggest that the amplitude decrease in the BKO tissue is caused by a combined effect of amplitude reduction and phase desynchrony at the cellular level . Because nutrient signals can provide a potent entraining cue for peripheral circadian oscillators ( Green et al . , 2008; Bass and Takahashi , 2010; Mohawk et al . , 2012 ) , we investigated whether scheduled food restriction ( FR ) could restore circadian temporal organization of peripheral oscillators in forebrain-specific Bmal1 knockouts . First , to test the effect of forebrain Bmal1 on food-anticipatory activity ( FAA ) , we subjected BKOs to a schedule of FR under both LD and DD conditions ( Figure 7A ) . 10 . 7554/eLife . 04617 . 014Figure 7 . Intact FAA in Bmal1 brain knockout mice . ( A ) Schedule of a gradual temporal FR protocol in LD ( upper ) or DD ( lower ) . Horizontal black and white bars represent lights off and on , respectively . Gray bars represent food availability . Since abrupt shifts to FR can have a high morbidity in mice , a gentle temporal FR paradigm was used ( FR ramp ) , decreasing the duration of daily food availability from constant to a final 4 hr window over the course of 5 days . Because free-running in DD can obscure FAA in control mice , we obtained DD baseline activity and then switched mice to 14 days of LD ( not shown ) . We then transferred our mice directly from LD into DD FR following an LD food ramp ( lower ) . It should be noted that once gradual FR began , the time of onset of food availability was the same each day . ( B , F ) Representative double-plotted actograms of daily wheel-running activity of 3 Fx/Fx control littermates and 3 BKO mice during ad lib feeding and under subsequent FR during LD ( B ) or DD ( F ) . The boxed area toward the left side of each actogram indicates the daily interval of food availability under FR and yellow areas indicate time of lights-on during LD 12:12 . After 5 days of gradually decreasing food availability , the final food availability window was ZT/CT6–10 . For clarity , the 5-day FR ramp is not included in the boxed area ( B ) . ( C , G ) Mean locomotor activity profiles of Fx/Fx littermates ( n = 7 ) and BKO mice ( n = 14 ) under ad lib feeding ( left ) and during FR days 8–14 ( right ) under LD ( C ) or DD ( G ) . Each data point represents counts per minute averaged for each genotype across a 6-min bin ( ±SEM ) . The dashed boxed area ( left ) indicates , for comparison , the daily interval corresponding to subsequent food availability . The solid boxed area ( right ) indicates the daily interval of food availability under FR . ( D , H ) Time course of the development of FAA in Fx/Fx controls ( n = 7 ) and BKO mice ( n = 14 ) during LD FR ( D ) or DD FR ( H ) . FAA is plotted as the total number of activity counts ( mean ± SEM ) allocated to a 6-hr time interval prior to mealtime , ZT/CT0–6 ( D , H ) . ( E ) Number of hours by which FAA preceded daily meal times in Fx/Fx ( n = 7 ) and BKO mice ( n = 14 ) during LD FR . Wheel-running activity profiles were averaged for each individual during stable FR ( as in C ) , and the average time of onset of FAA was determined as the time before food availability at which FAA rose to its half-maximum value ( mean ± SEM ) . ( I ) Quantification of FAA under DD conditions in BKO mice ( n = 14 ) . Plotted is fold-change of wheel-running activity ( counts per minute , mean ± SEM ) in each mouse for CT0–6 ( window of FAA ) compared with CT10–24 ( the rest of the day except for the window of food availability ) . During FR , increased locomotor activity during CT0–6 compared with CT10–24 was highly significant ( p = 0 . 0001 by paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 01410 . 7554/eLife . 04617 . 015Figure 7—figure supplement 1 . Lack of Bmal1 expression in the DMH of Bmal1 brain knockout mice . ( A , B ) Food consumed during FR was significantly lower in BKOs compared to controls using Repeated Measures GLM ANOVA for both LD conditions [F1 , 291 = 43 . 40 , p < 0 . 0001] ( A ) and DD conditions [F1 , 293 = 15 . 47 , p = 0 . 0009] ( B ) . Food consumed on days 12–14 of LD FR ( A ) represents the average food consumed over those 3 days and is plotted in triplicate to enable comparison between LD and DD food consumed . ( C , D ) Coronal brain sections at the level of the DMH , prepared from mice under ad lib feeding at CT6 and CT18 following 2 days of DD , were hybridized in situ to examine Bmal1 exon 4 ( C ) and Per2 ( D ) mRNA levels in Fx/Fx control littermates ( n = 6 for CT6 , n = 7 for CT18 ) and BKO mice ( n = 6 for CT6 , n = 8 for CT18 ) . On the right of each coronal brain section is a close-up of the DMH . ( E , F ) Optical density of Bmal1 ( E ) and Per2 ( F ) mRNA in the DMC . Shown are mean ± SEM , * significant effect of genotype in ( E ) [F1 , 26 = 208 . 31 , p < 0 . 0001] but not in ( F ) using GLM ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 015 Both qualitatively and quantitatively , mice lacking forebrain Bmal1 exhibited clear food-anticipatory behavior in LD that was similar to that of control mice ( Figure 7B–E ) . Notably , the mean FAA profiles of BKOs were enhanced in DD during temporal food restriction ( Figure 7C , G ) , and the total daily activity levels in BKOs during FR were significantly higher ( almost two-fold higher ) than those of controls , despite the observation that BKOs exhibited a trend to eat less during FR ( Figure 7—figure supplement 1 ) . Surprisingly , FAA profiles showed that , during FR in DD , BKOs had extended activity for up to 15 hr before food availability , while controls allocated most of their activity to the 6-hr window of FAA ( Figure 7F , G ) . Another study has found similarly longer FAA in Bmal1-deficient mice ( Takasu et al . , 2012 ) . To further characterize the food-anticipatory behavior , we measured the time course over which FAA emerged in BKOs and controls after the start of food restriction . We found FAA development to be similar in the two genotypes under both LD and DD conditions . FAA appeared after about 2 days and reached a plateau after about 4 days in both genotypes ( Figure 7D , H ) . The mean time by which FAA anticipated the daily onset of food ( phase angle with respect to food presentation ) was not significantly different between the two genotypes during FR in LD ( Figure 7E ) . During FR in DD , the random activity in BKOs partially obscured their FAA; however , FAA was still recognizable above the baseline both in individual records ( Figure 7F ) and in the population activity profile ( Figure 7G ) . This activity change was statistically significant ( Figure 7I ) . Therefore , behavioral responses to temporal food restriction in BKOs were enhanced in DD . Previously , the DMH was thought to be a possible candidate site of a food-entrainable oscillator ( FEO ) that regulates feeding behavior [ ( Gooley et al . , 2006; Mieda et al . , 2006 ) but see ( Landry et al . , 2006 , 2007; Moriya et al . , 2009; Acosta-Galvan et al . , 2011; Landry et al . , 2011 ) ] . To examine Bmal1 deletion and Per2 gene expression at this anatomical site , we conducted in situ analysis on the DMH . We collected brains at CT6 and CT18 on the third day of ad lib feeding in DD . Controls showed high expression levels of Bmal1 mRNA in the DMH , albeit with no circadian fluctuation . On the other hand , BKOs had no significant Bmal1 mRNA in the DMH at either time point ( Figure 7—figure supplement 1 ) . Despite a complete lack of Bmal1 in the DMH , BKOs exhibited normal FAA , indicating that Bmal1 in the DMH is not necessary for food entrainment . Surprisingly , BKOs showed Per2 expression levels comparable to those of controls ( Figure 7—figure supplement 1 ) . Both genotypes displayed intense staining in the ventromedial portion of the compact nucleus of the DMH ( DMC , Figure 7—figure supplement 1 ) , and the difference between peak and trough Per2 mRNA was not significant for either genotype . This result implies that transcriptional regulation of Per2 in the DMH involves activation pathways ( e . g . , cAMP response element ( CRE ) and serum response element ( SRE ) mediated pathways ) in addition to CLOCK:BMAL1 as seen previously ( Travnickova-Bendova et al . , 2002; Gerber et al . , 2013 ) . It has been shown that rhythms of peripheral tissues can be dissociated from the control of the central clock in the SCN and be re-entrained to a new phase by temporal feeding restriction ( Damiola et al . , 2000; Stokkan et al . , 2001 ) . The ability of the peripheral tissues to respond to feeding signals was further demonstrated in vivo ( Tahara et al . , 2012; Saini et al . , 2013 ) , with an observation that phase changes take place faster in the liver when the SCN is surgically lesioned . To assess the impact of feeding cues on the synchrony of peripheral rhythms , we subjected BKO; PER2::LUC mice to a restricted feeding schedule in the absence of light signals ( Figure 8A ) . The duration of constant darkness was the same as the desynchronization experiment ( Figure 5 ) . During these >30 days , mice were fed ad libitum for the first 2 weeks , and after 5 days of a food ramp , temporal food restriction was performed for 2 weeks before harvesting the tissues for real-time gene expression recording . 10 . 7554/eLife . 04617 . 016Figure 8 . Effects of restricted feeding on circadian rhythms of peripheral tissues . ( A ) Schedule of a temporal FR protocol in DD for real-time circadian reporting assay . Horizontal black and white bars representing lights off and on for LD 12:12 are shown above as a reference for a daily feeding schedule . Gray bars represent food availability . After entrainment to LD 12:12 , mice were released into and maintained in DD for a total of >30 days . In DD , mice were fed ad libitum for the initial 2 weeks and underwent a 5-day FR ramp of gradually decreasing food availability to reach the final food availability window of 4 hr ( from ZT/CT6–10 ) for 2 weeks of FR . Tissues were harvested at 2 hr after removal of the food . ( B , C ) Representative double-plotted actograms of 2 Fx/Fx control ( B ) and 2 BKO mice ( C ) during 2 weeks of ad lib feeding followed by the 5-day FR ramp and subsequent 2 weeks of FR in DD . The boxed area toward the left side of each actogram indicates the daily interval of food availability under FR ( ZT/CT6–10 ) . ( D ) Phase map of circadian rhythms of various peripheral tissues from Fx/Fx controls in DD + FR ( n = 11 ) . The peak phases ( or averaged peak phases ) were plotted against the feeding time ( green bar ) . Tissues from the same animal are connected by colored lines with matched symbols . ( E ) Phase map of ( D ) was converted to CT for the harvest time . ( F ) Phase map of BKOs in DD + FR ( n = 11 ) . The peak phases ( or averaged peak phases ) were plotted against the feeding time ( green bars ) . ( G ) Circular plots of peak bioluminescence rhythms in the SCN and peripheral tissues presented in ( D–F ) . The mapping strategy is the same as in Figure 5E . The sample size , mean phase angle , and circular variance for each dataset are summarized in Figure 8—source data 1A . ( H ) Scattered plot of circular variance in each individual mouse . Degree of variance among the peak phase values of pituitary , liver , kidney , heart , lung , and spleen in an individual mouse was calculated and expressed as circular variance ( see Figure 8—source data 1B ) and compared between Fx/Fx controls and BKOs in DD + FR . The bar is ±SD . Note that the degree of variance is not changed by phase data conversion from ZT to CT . A significant effect was found between Fx/Fx control and BKO mice ( p = 0 . 0002 by Mann–Whitney test ) . ( I ) Normalized amplitude of bioluminescence rhythms from Fx/Fx control ( open bar ) vs BKO ( dark bar ) in DD + FR . Mean relative amplitude ±SD are shown . The sample size is the same as in ( D , F ) . *p < 0 . 001 , *p < 0 . 0001 by t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 01610 . 7554/eLife . 04617 . 017Figure 8—source data 1 . ( A ) Summary of statistical analysis of circular plots presented in Figure 8 . ( B ) Statistical analysis of circular variance for peak bioluminescence in individual mice under restriction feeding . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 017 Consistent with a previous report ( Storch and Weitz , 2009 ) , Fx/Fx control mice exhibited both free-running activity and FAA in constant darkness ( Figure 8B ) . The peripheral tissues in these mice displayed variable phase relationships with respect to the feeding time ( Figure 8D ) . However , when the peak phases were converted to the circadian time of the animal at harvest , the phases in a majority of tissues examined ( except for pituitary and liver ) became markedly clustered ( Figure 8E , G; Figure 8—source data 1A ) , suggesting that the circadian rhythms in many peripheral tissues in wild-type mice are still dictated by the signals from the intact central clock . On the other hand , the peripheral tissues of BKOs showed distinct changes in phase expression patterns in response to restricted feeding cues ( Figure 8F ) . Although BKOs exhibited prolonged pre-feeding activity in DD ( Figure 7F; Figure 8C ) , only the liver and kidney showed tightly clustered phases ( Figure 8G; Figure 8—source data 1A ) . Other tissues ( heart , lung , spleen ) were variable in phase , though their phase distribution was different either from that found in an ad libitum condition ( see also Figure 5D ) or from those in Fx/Fx ( Figure 8G , Figure 8—source data 1A ) . These results suggest that feeding cues differentially entrain peripheral clocks; FR strongly synchronizes circadian clocks in the liver and kidney but has weaker effects on other peripheral tissues such as the pituitary , heart , lung , and spleen . To our surprise , with the exception of the liver , oscillatory amplitude of peripheral rhythms of BKOs remained lower than that of Fx/Fx control mice even after 2 weeks of FR ( Figure 8I ) . Lower amplitude is not related to the state of phase synchrony , because relative amplitude of kidney still remained low even though FR synchronized its phase . This indicates that the damped rhythms were not completely restored by food signals . To compare ad libitum and FR conditions directly , we performed a ‘meta-analysis’ on phase and amplitude within Fx/Fx and BKOs , respectively ( Figure 9A–D ) . Statistical comparison between ad libitum in DD ( BKO from DD ) and FR in DD ( BKO from DD + FR ) confirmed tissue specificity of phase entrainment by feeding ( Figure 9; Figure 9—source data 1A , B ) . This result is in striking contrast to the case in which BKOs were exposed to LD cycles , which synchronized all peripheral rhythms examined even in the absence of a functioning SCN ( Figure 9B , D; Figure 9—source data 1B ) . Nevertheless , both internal synchrony and oscillatory amplitude in peripheral tissues were maintained most optimally when an intact SCN was present ( Figure 5; Figure 6A; Figure 8H ) . Taken together , these results indicate that , while individual tissues can be entrained by light/dark cycles and feeding schedules independently , the central clock plays an important role in sustaining coherence among peripheral tissues and robust circadian programs at the organismal level . 10 . 7554/eLife . 04617 . 018Figure 9 . Analysis of circadian rhythms of peripheral tissues under different conditions . ( A , B ) Comparison of peak bioluminescence rhythms in the peripheral tissues in Fx/Fx ( A ) and BKO ( B ) mice under LD , DD , and FR conditions . Circular plots are as presented in Figure 5E and Figure 8G . Watson-Williams F-test and bootstrap analysis were performed to compare the mean phase angle ( μ ) and the variance ( V , distribution of peak phase values ) between two circular data , respectively . Statistical comparison was summarized in Figure 9—source data 1A , B ) . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 . N . S . = not significant . ( C , D ) Comparison of normalized amplitude of bioluminescence rhythms from Fx/Fx ( C ) and BKO ( D ) mice under LD , DD , and FR conditions . Each bar graph is as presented in Figure 6A , B and Figure 8I . Mean relative amplitude ±SD are shown . p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 by ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 01810 . 7554/eLife . 04617 . 019Figure 9—source data 1 . ( A ) Summary of statistical comparison of peak phases from Fx/Fx mice under different conditions . ( B ) Summary of statistical comparison of peak phases from BKO mice under different conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 04617 . 019 Bmal1 is the only component of the mammalian circadian clock whose sole deletion generates a complete loss of circadian rhythms in mice ( Bunger et al . , 2000 ) . Therefore , elimination of Bmal1 from a target tissue provides an invaluable model to study the behavioral and physiological consequences of abolishing circadian rhythms in a tissue-specific manner . In this study , we succeeded in removing Bmal1 specifically from the forebrain using a Camk2a::iCreBAC line ( Casanova et al . , 2001 ) . Notably , this CamiCre driver was able to efficiently recombine and delete our target floxed gene in the entire SCN . These forebrain/SCN-specific Bmal1 knockouts exhibited a total loss of circadian behavioral rhythmicity , and their circadian phenotype in DD and LL is similar to that of global Bmal1 knockouts ( Bunger et al . , 2000 ) . Thus , Bmal1 in the forebrain is essential for normal circadian activity rhythms . However , a Cre driver that can express more specifically in the SCN would be necessary to confirm the dominance of the SCN over other sites of the brain . BKOs displayed apparent entrainment to LD cycles; however , additional experiments revealed two deficits in these mice . First , entrainment in a skeleton photoperiod was abnormal with a bimodal pattern . A similar pattern is found in pinealectomized sparrows in which a population of oscillators entrains with two different and opposite phase angles ( Takahashi and Menaker , 1982 ) . Second , there is a delay in masking behavior to light under LD 3 . 5:3 . 5 which is similar to that seen in Clock mutants ( Redlin et al . , 2005 ) . Thus , the variable phase angle of entrainment in BKOs on LD 12:12 is consistent with weaker synchronization of residual oscillators as well as a small deficit in masking . Despite a complete loss of circadian behavioral rhythmicity , Bmal1 forebrain-specific knockout mice are as healthy as wild-type mice , consistent with our earlier results using tissue-specific rescue of Bmal1−/− mice ( McDearmon et al . , 2006 ) . These results suggest that Bmal1's contribution to the circadian phenotype seen in Bmal1−/− mice is distinct from its role in the maintenance of normal activity levels , body weight , reproduction , and longevity . For instance , the skeletal muscle defects seen in Bmal1−/− mice ( Andrews et al . , 2010 ) are likely a causative factor in their low activity . Muscle-rescued Bmal1−/− mice exhibit improvement of body weight , activity , and longevity but not of arthropathy ( McDearmon et al . , 2006 ) . Our findings as well as others' suggest that the effects of BMAL1 can be dissociated between brain and other tissues including the reproductive system . While the effect of Bmal1 deletion is local , the impact of disabling the brain's clock on peripheral circadian organization is significant . Previous studies have investigated the effect of SCN ablation on peripheral rhythms by making SCN lesions and recording circadian gene expression of peripheral tissues either ex vivo ( Yoo et al . , 2004 ) or in vivo ( Tahara et al . , 2012; Saini et al . , 2013 ) . Here , we used a Cre-mediated genetic excision approach to remove a critical component of the molecular oscillator from the intact SCN network . This strategy demonstrates that internal synchrony and high amplitude rhythms are lost in the absence of a functional master oscillator in the forebrain/SCN and supports previous reports ( Yoo et al . , 2004; Tahara et al . , 2012; Saini et al . , 2013 ) . Nevertheless , further studies are needed to elucidate the functional consequences of damped circadian rhythms in each peripheral organ . On the other hand , the dispersed phases in BKOs were not observed when mice were placed in an LD condition ( Figure 5C , E ) . Given that the BKO's circadian behavior is driven by light from retinal pathways presumably acting via the SCN , this demonstrates that cyclic photic input ( Figure 3 ) is sufficient to sustain coherent temporal synchrony of peripheral oscillators . Alternatively , the effects of light could also be mediated by activity ( wheel-running ) -associated stimuli driven by LD as another source of systemic signals impinging on peripheral clocks analogous to that found previously ( Kornmann et al . , 2007; Hughes et al . , 2012 ) . BKOs offer an advantage for investigating the effects of a disabled master clock in the brain because they do not suffer from the morbid phenotypes observed with a global loss of Bmal1 . Using the forebrain Bmal1-deficient animals , we were able to conduct food restriction experiments under completely standard conditions . In contrast to global Bmal1 knockouts , altered experimental conditions such as enriched liquid food ( Storch and Weitz , 2009 ) , LD 18:6 ( Pendergast et al . , 2009 ) , or food placed on the cage bottom , and gentle handling to prompt feeding ( Fuller et al . , 2008 ) were not required for these mice . Our results show that BKOs indeed displayed normal FAA as other studies using global Bmal1 knockouts have reported ( Mistlberger et al . , 2008; Pendergast et al . , 2009; Storch and Weitz , 2009 ) , demonstrating that BMAL1 in the forebrain is not essential for food entrainment behavior . Specifically , our Bmal1 in situ results in the DMH indicate that the DMH is not necessary for FAA as reported previously ( Landry et al . , 2006 , 2007; Moriya et al . , 2009; Acosta-Galvan et al . , 2011; Landry et al . , 2011 ) . In our experiments , BKOs tend to exhibit intact food-anticipatory behavior under constant darkness conditions . However , overall activity of BKO's circadian behavior was also significantly elevated in DD , which makes it difficult to distinguish whether the elevated pre-feeding activity is due to enhanced FAA or to the effect of overall activity increase . Nevertheless , these results suggest that an inhibitory system ( Davis and Menaker , 1980 ) suppresses the total activity of both circadian behavior and FAA . It remains to be determined what factors underlie suppression of the activity and how their regulation in the SCN and other brain regions is altered by the deletion of BMAL1 . Previous work has shown that restricted feeding can alter the phase of peripheral rhythms without affecting the phase of the SCN ( Damiola et al . , 2000; Stokkan et al . , 2001 ) . While peripheral oscillators can be uncoupled from the central clock in the SCN , recent studies have suggested that SCN-derived signals counteract the feeding cues for rapid phase-shifting of peripheral clocks ( Saini et al . , 2013 ) and that phase-shifting rates are variable among different tissues ( Damiola et al . , 2000; Yamazaki et al . , 2000 ) . Consistent with these observations , peripheral tissues in our control mice displayed variable phases relative to the FR schedule when conducted in constant darkness ( Figure 8D ) . In these free-running conditions , the circadian activity rhythms of the mice were not affected by FR; and as a consequence , the phase relationships between the FR schedule and circadian phase could be dissociated . When control mice in DD under ad libitum conditions were compared to control mice in DD under FR using circadian phase as a marker , the liver , kidney and heart significantly shifted their mean circular phase in response to feeding , showing an effect of FR on these tissues [Figure 9A , compare Fx/Fx from DD with Fx/Fx from DD + FR ( in CT ) ; Figure 9—source data 1A] . However , the phases of the peripheral rhythms were significantly clustered with the phase of the activity rhythm ( Figure 8E ) , strongly suggesting that signals from the SCN continue to dictate the expression of peripheral phases in wild-type mice despite the expression of FAA behavior . Indeed , only the liver was strongly clustered in phase with FR in DD ( Figure 8G , Fx/Fx from DD + FR ) compared to the circadian phase of activity [Figure 8G , Fx/Fx from DD + FR ( in CT ) ] in which all peripheral tissues were clustered in phase . Thus we see a rather complex picture in control mice in which FR exerts tissue-specific effects on peripheral organ circadian phase and coherence . In contrast , in the absence of a functional pacemaker in the forebrain , liver and kidney clocks shifted to a distinct phase in response to food restriction and became synchronized ( Figure 8F , G; Figure 9B ) , which supports the results of an in vivo study using SCN-lesioned animals ( Saini et al . , 2013 ) . We found , however , that other tissues such as heart , lung , and spleen did not entrain to food restriction even after 2 weeks of treatment ( Figure 9B; Figure 9—source data 1A ) . Although the precise mechanism is not known , the lower degree of synchrony of these tissues could be due to the longer duration of pre-feeding behavioral activity which could mobilize additional ( conflicting ) signals . At the same time , these tissues showed distinct phase shifts in response to feeding ( Figure 9B; Figure 9—source data 1B ) . Circular phase variance analysis indicates that internal coherence among different organs within an individual mouse still remained disrupted ( Figure 8H ) . Therefore , these results suggest that , while synchrony of individual tissues such as liver and kidney can be maintained independently by a feeding schedule , synchrony among peripheral tissues requires central clocks in the forebrain/SCN . The effects of FR were different from the effects of LD cycles ( Figure 5; Figure 9B ) , which sustained both phase synchrony and oscillatory amplitude in peripheral clocks in BKOs ( Figure 9 ) . Notably , both phase angle ( Figure 5E; Figure 5—source data 1A ) and phase coherence among peripheral tissues ( Figure 5F ) were similar to those in Fx/Fx control mice which possess an intact central clock . We speculate that , while feeding cues can antagonize SCN-derived signals , LD cycles appear to act in concert with the SCN or signals arising from the SCN . Although detailed pathways through which the SCN controls the peripheral clocks remain unknown , recently , SRF was identified as a transcription factor that regulates gene expression in peripheral tissues in response to systemically oscillating signals ( Gerber et al . , 2013 ) . It will be interesting to explore how these signals affect circadian gene expression in vivo to determine the mechanisms by which the SCN communicates with peripheral oscillators . Our experiments reveal the diverse behavioral and peripheral physiological consequences of inactivating a forebrain clock . They pave the way for further exploring the complex relationship between central and peripheral clocks . All mice were housed under LD 12:12 unless otherwise noted . Camk2a::iCreBAC mice ( CamiCre ) ( MGI:2181426 ) were kindly provided by Dr G . Schutz ( Casanova et al . , 2001 ) . For a Cre reporter mouse , Rosa26-CAG-LSL-tdTomato ( Ai14 , kindly provided by Dr Hongkui Zeng ) was used ( Madisen et al . , 2010 ) . Bmal1fx/fx mice ( Johnson et al . , 2014 ) are homozygous for a Bmal1 allele which has loxP sites inserted into the introns surrounding exon 4 . Bmal1fx/fx mice ( 129SvJ × C57BL/6J N3 backcross ) were crossed to CamiCre to produce Bmal1fx/fx ( Fx/Fx ) , CamiCre+;Bmal1fx/+ ( Het ) , and CamiCre+; Bmal1fx/fx ( BKO ) mice . CamiCre mice were mated to their siblings ( FVB/N × C57BL/6J N3 backcross ) to generate CamiCre hemizygous controls . Bmal1−/− global knockout mice ( Bunger et al . , 2000 ) were 129SvJ × C57BL/6J N14 backcross congenic animals . For the real-time reporting assay , PER2::LUC mice ( Yoo et al . , 2004 ) were crossed to Bmal1fx/fx to produce CamiCre+; Bmal1fx/fx; Per2Luc/+ and its control littermate , Bmal1fx/fx; Per2Luc/+ . For the bioluminescence imaging experiments , Per2Luc mice were homozygous . For all experiments , male and female mice were used in balanced ratios; mice were 2–7 months old for behavior experiments and 3–11 months old for bioluminescence recordings and age-matched across groups , except for FR ( males only were used for behavior experiments ) , where mice were 1 . 9–3 . 6 months old at the start of the experiment . All animal studies and Materials and methods were in accordance with Northwestern University and UT Southwestern Medical Center guidelines for animal care and use . The following sets of genotyping primers were used: CamiCre: iCre-PCR-F 5′-TCTGATGAAGTCAGGAAGAACC-3′ and iCre-PCR-R 5′-GAGATGTCCTTCACTCTGATTC-3′ ( amplified a 400 bp product ) . PCR reactions were carried out in a final volume of 30 µl buffer consisting of 5 µl Promega 5× Colored buffer , 2 . 7 µl 25 mM MgCl2 , 0 . 38 µl 20 µM each primer , 3 µl 2 mM dNTPs , 3 µl NaOH-extracted tail genomic DNA , and 0 . 3 µl Promega Go Taq ( 1 U/rx; Promega , Madison , WI ) . PCR conditions were: 5 min at 95°C , followed by 30 cycles of 1 min at 94°C , 2 min at 58°C , and 2 min at 72°C , followed by 8 min at 72°C . The reaction products were analyzed on a 2% agarose gel . Bmal1fx/fx: OL5436 5′- CCC TGA ACA TGG GAA AGA GA -3′ and OL6013 5′- ATT CAC CTT TTG GGG AGG AC -3′ ( floxed band: 360 bp , WT band: 310 bp ) . PCR reactions were carried out in a final volume of 25 µl buffer consisting of 5 µl Promega 5× Colored buffer , 3 µl 25 mM MgCl2 , 1 µl 20 µM each primer , 2 . 5 µl 2 mM dNTPs , 3 µl NaOH-extracted tail genomic DNA , and 0 . 3 µl Promega Go Taq ( 1 . 25 U/rx ) . PCR conditions were: 5 min at 95°C; followed by 37 cycles of 30 s at 95°C , 30 s at 60°C , and 30 s at 72°C , followed by 5 min at 72°C . Ai14: Ai14F 5′- TAC GGC ATG GAC GAG CTG TAC AAG TAA -3′and Ai14R 5′- CAG GCG AGC AGC CAA GGA AA -3′ ( amplified a 517 bp product ) . Bmal1−/− and PER2::LUC mice were genotyped as previously described ( Bunger et al . , 2000; Yoo et al . , 2004 ) . To record the rhythm of locomotor activity , adult mice ( at least 8-week old ) were individually housed in activity wheel-equipped cages under LD 12:12 for at least 10 days . Locomotor activity was recorded and analyzed using ClockLab software ( Actimetrics , Wilmette , IL ) as previously described ( Vitaterna et al . , 1999; Siepka and Takahashi , 2005; McDearmon et al . , 2006; Vitaterna et al . , 2006 ) . Fluorescent lights ( 300–600 lux inside the cage ) were used for behavior experiments . Mice were transferred into DD for 4 weeks , returned to LD for 2 weeks , released into LL for 4 weeks , and then returned to LD for at least 2 weeks . For the entrainment experiment , mice were placed in LD 12:12 for at least 10 days , released into DD for 4 weeks , followed by LD for 2 weeks , before they were introduced to a skeleton photoperiod ( LDLD 1:10:1:12 ) for 4 weeks . The mice were returned to LD for 2 weeks and then switched to an ultradian cycle ( LD 3 . 5:3 . 5 ) for 3 weeks . For the food restriction experiment , mice were placed in wheel-equipped cages under LD 12:12 for 2 weeks , then underwent a 5-day FR ramp of gradually decreasing food availability ( Figure 7A ) , followed by 14 days of FR in LD . They were returned to ad lib feeding under LD for 2 weeks , DD for 2 weeks , and LD for 2 weeks . Then the mice underwent a 5-day FR ramp after which they were released into DD for 14 days of FR in DD . On the first day of the FR ramp , food was removed at ZT/CT18 and 12 hr later food was returned . On each successive day , food was removed 2 hr earlier until food was available for 4 hr , from ZT/CT6–10 ( Figure 7A ) . This was the final food availability window for 14 days of FR . A regular mouse diet ( 5K52; Lab Diet , Richmond , IN; contains 19 . 3% protein and 6 . 2% fat , or 2918; Irradiated Global Diet , Teklad Diets , Madison , WI; contains 18 . 6% protein and 6 . 2% fat ) was used during FR experiments . Food pellets were weighed daily for each mouse before and after food availability to monitor food consumption . Individual mean daily activity profiles were computed by ClockLab; group average profiles were made using Microsoft Excel . For the bioluminescence recording experiment , mice were initially entrained in LD 12:12 for a minimum of 6 days ( 6–12 days ) , as the light-on time was delayed for 6 hr , before they were released into DD for 2 weeks . Then the mice underwent a 5-day FR ramp as above , after which 4-hr FR was conducted from ZT/CT6–10 for 2 weeks ( Figure 8A ) . On the final day of 2 weeks of FR , tissues were harvested for bioluminescence recordings at 2 hr after the food was withdrawn ( ZT/CT12 ) . Wheel-running activity was recorded and analyzed essentially as previously described ( Vitaterna et al . , 2006 ) . The free-running periods in DD and LL were calculated from the entire 28-day interval by using χ2 periodogram analysis ( Clocklab software , Actimetrics ) . The amplitude of circadian rhythm was analyzed using the fast Fourier transform ( FFT ) , which estimates the relative power of approximately 24 hr period rhythm in comparison with all other periodicities in the time-series ( Shimomura et al . , 2001 ) . The power spectral densities for frequencies ranging from 0 to 1 cycle/hr were determined and normalized to a total power ( area under the curve ) of 1 . 0 . The peak in the circadian range ( 18- to 30-hr period or 0 . 033–0 . 055 cycles/hr ) of the relative power was determined for each animal for comparison . If no significant periodicity in the 18–30 hr range was detected by FFT , the free-running period was not scored . For analysis of total daily activity , the total number of wheel revolutions per day was averaged in LD ( last 10 days of initial LD interval ) , DD , and LL ( 28 days ) using the same days described above for free-running period and amplitude analysis . Effects of genotype were analyzed by a Generalized Model ( GLM ) ANOVA by using NCSS ( Kayesville , UT ) , with Tukey–Kramer multiple comparison post-tests for pairwise comparisons . Two-sample t-tests were used to analyze subjective day vs night activity during skeleton photoperiod and amplitudes during LD 3 . 5:3 . 5 , and Repeated Measures GLM ANOVA with Tukey–Kramer multiple comparison post-tests was used to compare the time course of masking between genotypes during days 3–15 of the 21-day LD 3 . 5:3 . 5 cycle . FFT was used to analyze amplitudes in the circadian as well as 7-hr range during LD 3 . 5:3 . 5 . To analyze FR data , Repeated Measures GLM ANOVA with Tukey–Kramer multiple comparison post-tests was used to compare the effect of genotype , time , and interaction between the two on FAA ( absolute counts ) , FAA ( % of total activity ) , weights , total daily activity , and food intake . Two-sample t-test and paired t-test , respectively , were used to calculate FAA onset and activity fold-change from ad lib to FR . For the time course collection of brain tissue for in situ hybridization , after 2 weeks of entrainment in LD , animals were released into DD for 2 days and tissues were harvested on the third day of DD . Brains were rapidly dissected and frozen on dry ice and stored at −80°C until further processing . Standard in situ hybridization procedures were used as previously described ( Sangoram et al . , 1998 ) . Alternate 20-µm thick coronal sections were collected from each brain from the rostral to the caudal end of the SCN or DMH ( 24–32 total , 12–16 for each probe ) by using a Leica CM3050S cryostat . Sections were thaw-mounted onto SuperFrost plus microscope slides ( VWR , Radnor , PA ) , then stored at −80°C until all sectioning was completed . Alternate sections were hybridized to a combination oligoprobe against mBmal1 exon 4 and to a riboprobe against Per2 . For the Bmal1 combination oligoprobe , three 37 to 50-mer oligonucleotide probes ( IDT , Coralville , IA ) were radiolabeled at the 3′ ends with 33P via terminal I deoxynucleotidyl transferase ( Gibco/Invitrogen , Life Technologies , Grand Island , NY ) and used in equal proportions: Mop3Ex4Probe1 5′-AACTGTTCATTTTGTCCCGACGCCTCTTTTCAATCTGACTGTGGGCCTCC-3′ , Mop3Ex4Probe2 5′-CCTGGACATTGCATTGCATGTTGGTACCAAAGAAGCCAATTCATC-3′ , Mop3Ex4Probe3 5′-CTGAACAGCCATCCTTAGCACGGTGAGTTTATCTAAC-3′ . Templates for the anti-sense and sense Per2 probes were PCR-generated by using the following primers: Per2-insitu-f 5′-ACG AGA ACT GCT CCA CGG GAC-3′ , Per2-insitu-r 5′-ACA GCC ACA GCA AAC ATA TCC GC-3′ . PCR products were cloned into a TA cloning vector ( Invitrogen ) . The T7 promoter from the TA cloning vector was used to generate both anti-sense and sense probes . All probes were labeled with 33P . Quantitation of the autoradiogram signal was performed by using NIH ImageJ 1 . 34s software ( NIH , Bethesda , MD ) and normalized to radioactive standards as described previously ( Vitaterna et al . , 1999 ) . The optical density ( OD ) of individual SCN or DMC was normalized by subtracting the OD of an area of identical size in the lateral hypothalamus or the magnocellular nucleus of lateral hypothalamus , respectively , from the same side ( left or right ) and section . Normalized values from three sections near the middle ( anterior–posterior ) SCN or caudal DMH ( at the level of DMC ) were used to calculate an average for each brain . Effects of genotype , time , or their interactions were analyzed by GLM ANOVA , with Tukey–Kramer multiple comparison post-tests for pairwise comparisons . For Westerns blots , tissues were harvested at ZT16 from mice in LD 12:12 . Tissues were homogenized by a Polytron homogenizer in a buffer containing 44 . 6 mM Tris–HCl , 5 . 5 mM Tris–Base , 154 mM NaCl , 29 . 8 mM sodium pyrophosphate , 20 mM glycerol 2-phosphate , 50 mM NaF , 1 µg/ml aprotinin , 1 µg/ml leupeptin , 1 mM PMSF , 1 mM sodium orthovanadate , 1 mM EDTA , 1 mM p-nitrophosphate , 0 . 1% SDS , 0 . 5% sodium deoxycholate , and 1% NP-40 . Homogenates were centrifuged at 15 , 000×g for 20 min at 4°C . Supernatants were collected and protein concentration was estimated using Bio-Rad DC Protein Assay according to the manufacturer's instructions . Total protein ( 10 µg ) was diluted 1:1 with Laemmli sample buffer and resolved on a 10% SDS-polyacrylamide gel by electrophoresis . Thereafter , proteins were electrotransferred onto a GE Healthcare PVDF transfer membrane . The membranes were blocked with PBST ( PBS + 0 . 1% Tween-20 ) containing 5% non-fat powdered milk for 1 hr and then incubated with the rabbit polyclonal anti-MOP3 antibody ( 1:6 , 250 , generated in Dr Bradfield's lab and is available at NB100-2288 , Novus Biologicals LLC , Littleton , CO ) , followed by anti-rabbit IgG secondary antisera horseradish peroxidase ( 1:1000; PI-1000 , Vector Laboratories , Burlingame , CA ) . Proteins were visualized with a chemiluminescence detection system ( ECL Western blotting detection analysis system; Amersham Pharmacia , GE Healthcare Bio-Sciences , Pittsburgh , PA ) and with subsequent exposure to autoradiographic films . Brain tissues for immunohistochemistry were harvested at ZT16 under LD 12:12 . Animals were anesthetized with ketamine/xylazine/saline cocktail ( 10 mg/ml ketamine , 10 mg/ml xylazine ) at 0 . 01 ml/g body weight or with sodium pentobarbital at 120 mg/kg body weight and then perfused intracardially with 25 ml of 4% paraformaldehyde ( pH 7 . 4 , Sigma-Aldrich , St Louis , MO ) in 63 . 4 mM phosphate buffer ( pH 7 . 5 ) . The brains were removed and post-fixed for 2 hr at 4°C in 4% paraformaldehyde in 63 . 4 mM phosphate buffer and transferred to 20% sucrose/phosphate buffer overnight . For immunohistochemistry , 50-µm coronal sections were collected through the entire SCN using a Leica cryostat and processed free-floating . Sections were incubated with the rabbit polyclonal anti-MOP3 ( BMAL1 ) antibody ( 1:1000 ) followed by anti-rabbit IgG secondary biotinylated antisera ( 1:200; BA-1000 , Vector Laboratories ) and visualized using Vectastain ABC Kit and 0 . 5 mg/ml diaminobenzidine , 0 . 01% hydrogen peroxide , and 0 . 03% NiCl2 in 0 . 05 M Tris ( pH 7 . 2 ) . Sections were mounted onto gelatin-coated microscope slides using aqueous mounting medium ( 3:1 glycerol/phosphate buffer ) , cover-slipped , and imaged with a Leica DM-RB upright microscope using Openlab software in the Biological Imaging Facility at Northwestern University . For the whole brain imaging , a Zeiss Stereo Discovery Microscope V12 was used under a bright field . Fluorescence images were taken using an Olympus MVX10 ( 1× ) and a Zeiss LSM 510 confocal microscope ( 20× objective ) . To monitor real-time reporting of PER2::LUC oscillations , tissues were harvested after cervical dislocation between ZT/CT10 . 5–13 . Coronal sections of the brain were sliced at 300-μm thickness by a vibratome in ice-cold Hank's balanced salt solution ( HBSS , Invitrogen ) . Each individual SCN was dissected out and cultured on a Millicell organotypic membrane ( PICM ORG50 , Millipore , Billerica , MA ) in a 35 mm tissue culture dish containing 1 . 2 ml DMEM ( 90-013-PB , Mediatech , Manassas , VA ) , supplemented with 2% B27 ( Invitrogen ) , 10 mM Hepes ( pH 7 . 2 ) , 4 mM L-glutamine , 0 . 035% sodium bicarbonate , 25 units/ml penicillin , 25 μg/ml streptomycin , and 0 . 1 mM luciferin ( L-8240 , Biosynth AG , Staad , Switzerland ) . Other tissues were processed as previously described ( Yamazaki and Takahashi , 2005 ) and cultured as above . The bioluminescence was monitored by a light-tight 32-channel LumiCycle luminometry ( Actimetrics , Wilmette , IL ) maintained at 36°C and was recorded at an interval of 10-min continuously for a minimum of 8 days followed by medium change to confirm the viability of the samples . All data were analyzed essentially as described previously ( Izumo et al . , 2006 ) . Briefly , raw data of bioluminescence records were corrected for background counts and PMT gain . For animals harvested from a DD condition ( 30–44 days in DD ) , activity onset of each Fx/Fx control mouse was set as CT12 . The time-series data were detrended by a 24-hr running average and then subjected to FFT-NLLS analysis to calculate periods and phases . The initial 20 hr of data were trimmed because they contain an acute effect of explant preparation . Data beyond 192 hr before medium change were also trimmed except for one group ( 1 Fx/Fx control and 1 BKO ) , which was affected by an electrical shutdown on day 6 . The amplitude was computed at a 36–60 hr range and normalized as previously described ( Izumo et al . , 2006 ) . For all data , the calculations were averaged where duplicated samples were prepared from the same tissue within an individual mouse , and the averaged value was used to map or calculate for all mice collected . Samples with RelAmp Error ≥0 . 84 ( background counts ) or out of a circadian ( 17–31 hr ) range ( 24 hr ± 30% of 24 hr ) in the periodicity were excluded from the analysis . A total of 1106 LumiCycle files were processed and used in this study . To construct a phase map , smoothed time-series were used to determine the peak on the second day . Phase angles and circular variances of circular plots were computed using Oriana ( Kovach Computing Services , Wales , UK ) . The variance of the peak phases between two groups was compared using bootstrap analysis ( Sato et al . , 2007 ) . For each iteration of the bootstrap , the difference of the variance between the two groups was calculated by random sampling with replacement from the data within the each of the two groups . Following 20 , 000 iterations , ninety-five percent confidence intervals were used to determine whether the differences were significantly different from zero . For real-time bioluminescence imaging analysis , age-matched pairs of mice ( Fx/Fx control and BKO ) were placed in DD for >30 days before harvesting the tissues . The tissue collection was repeated for a total of three pairs of mice . The tissues were processed and prepared in the same manner as in the luminometry recording , except that a glass bottom culture dish ( MatTek , Ashland , MA ) was used . The sealed culture was placed onto the stage chamber maintained at 36°C inside the LV200 Bioluminescence Imaging System ( Olympus America , Irving , TX ) . The sample was imaged using a 10× objective lens with 10-min exposure time at 25-min interval for 7 days . The imaging files were converted to linear time-series by quantifying the signals on grid matrices . The heat map was generated using the top 200 time-series data starting with the strongest signals . The top 50 of these time-series data were analyzed for period , phase , and relative amplitude by FFT-NLLS as described above . The phases calculated by FFT-NLLS were further converted to the actual time and divided by each circadian period to normalize period differences . Statistical analysis was conducted between paired mice as described above .	Jet lag is a common experience when flying long distance . This disorientating phenomenon occurs when our internal ‘body clock’ remains set to the time zone where the plane departed and fails to reset to the new local time . Our internal clock actually consists of a series of clocks—each of which is based upon groups of genes that are switched on and off at different times of the day and night . There is a master clock in our brain and a series of peripheral clocks in our other organs and tissues . The master clock is thought to coordinate the peripheral clocks , which in turn control the fluctuating activity of a specific organ in response to the time of day . To further investigate the master clock , a typical approach has been made to disable it by deleting the genes for its components . But some of these deletions can cause abnormalities in mice and some are lethal . To get around these problems , Izumo , Pejchal et al . have devised a way to delete a molecular component of the master clock only in the mouse's brain . Izumo , Pejchal et al . used this approach to specifically disable the mouse's master clock and , unlike mice that completely lack the Bmal1 gene , mice with the brain-specific deletion were as healthy and lived as long as normal mice . A molecular probe was used to monitor the peripheral clocks in different organs and tissues of these mutant mice , and revealed that , without a working master clock , the peripheral clocks were no longer synchronized . Izumo , Pejchal et al . found that the lost synchrony could be partially restored by training the mice to adapt to cycles of light and dark and feeding schedules . Following on from the work of Izumo , Pejchal et al . , one of the next challenges is to understand how the master clock communicates with the peripheral clocks in different organs and tissues around the body .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Differential effects of light and feeding on circadian organization of peripheral clocks in a forebrain Bmal1 mutant
Across many studies , animals with enhanced synaptic plasticity exhibit either enhanced or impaired learning , raising a conceptual puzzle: how enhanced plasticity can yield opposite learning outcomes ? Here , we show that the recent history of experience can determine whether mice with enhanced plasticity exhibit enhanced or impaired learning in response to the same training . Mice with enhanced cerebellar LTD , due to double knockout ( DKO ) of MHCI H2-Kb/H2-Db ( KbDb−/− ) , exhibited oculomotor learning deficits . However , the same mice exhibited enhanced learning after appropriate pre-training . Theoretical analysis revealed that synapses with history-dependent learning rules could recapitulate the data , and suggested that saturation may be a key factor limiting the ability of enhanced plasticity to enhance learning . Optogenetic stimulation designed to saturate LTD produced the same impairment in WT as observed in DKO mice . Overall , our results suggest that the recent history of activity and the threshold for synaptic plasticity conspire to effect divergent learning outcomes . The prospect that learning might be enhanced by enhancing synaptic plasticity has captured the imagination of many , fueled by reports of super cognition in transgenic mice with enhanced synaptic plasticity ( McConnell et al . 2009; Tang et al . , 1999; Lee and Silva , 2009; Huh et al . , 2000; Ito , 2002 ) . Synaptic plasticity is a fundamental property of neural circuits , hence its enhancement has the potential to enhance a wide range of brain functions , and benefit a wide range of patients . It could accelerate the recovery of sensory , motor , or cognitive function after a stroke or other injury; counter the decline of learning and memory in the elderly; or be used in conjunction with behavioral therapy to enhance drug rehabilitation , treatment for post-traumatic stress disorder , speech therapy , or other kinds of rehabilitation . Rapid progress in our understanding of plasticity at the synaptic level is providing targets for drug development and other molecular strategies for enhancing synaptic plasticity ( Lee and Silva , 2009 ) . However , optimism for this approach has been tempered by the observation that the enhancement of synaptic plasticity can , in some cases , impair learning ( Migaud et al . , 1998; Uetani et al . , 2000; Hayashi et al . , 2004; Cox et al . , 2003 ) . The finding that manipulations to enhance synaptic plasticity can either enhance or impair learning has been reported for different brain regions , and for both associative LTP and LTD ( Migaud et al . , 1998; Uetani et al . , 2000; Hayashi et al . , 2004; Cox et al . , 2003; Takeuchi et al . , 2008; Koekkoek et al . , 2005; McConnell et al . 2009 ) . Yet , there have been no experimental tests of why enhanced synaptic plasticity can have these opposite effects at the behavioral level . Moreover , despite extensive theoretical study of how enhanced plasticity can impair memory ( the plasticity-stability dilemma; Toulouse et al . , 1986; Carpenter and Grossberg , 1987; Amit and Fusi , 1994; Amit and Fusi , 1992; Gerrow and Triller , 2010; Frey and Morris , 1997; Reymann and Frey , 2007; Clopath et al . , 2008; Barrett et al . , 2009; Redondo and Morris , 2011 ) , there has been little theoretical treatment of how enhanced plasticity could impair learning . Hence , the principles governing the learning outcome under conditions of enhanced plasticity have remained elusive , as have the principles for promoting enhanced learning . This fundamental gap in our understanding of how enhanced synaptic plasticity functions in the context of an intact neural circuit is limiting the application of synaptic plasticity enhancers in patients who could potentially benefit from this approach . We combined experiment and theory to address this conceptual gap . We measured learning in mice deficient in molecules of the class-I major histocompatibility molecule ( MHCI ) complex , which have enhanced synaptic plasticity in multiple brain regions , including the cortex , hippocampus , thalamus , and cerebellum ( Huh et al . , 2000; Syken et al . , 2006; McConnell et al . 2009; Lee et al . , 2014 ) . We focused on the cerebellum to take advantage of the known links between synaptic plasticity and cerebellum-dependent oculomotor learning . In the cerebellum , classical MHCI H2-Kb ( H2-Kb ) and MHCI H2-Db ( H2-Db ) are highly expressed in Purkinje cells , and double-knockout mice , MHCI H2-Kb/H2-Db−/− ( KbDb−/−; Vugmeyster et al . , 1998; Schott et al . , 2003; referred to as double knockout ( DKO ) here ) , exhibit enhanced associative LTD at the parallel fiber-Purkinje cell synapses ( pf-Pk LTD ) ( McConnell et al . 2009 ) . For many years , pf-Pk LTD was widely considered to be the mechanism of all cerebellum-dependent learning ( Ito , 2002 ) ; however , recent evidence from animals with disrupted pf-Pk LTD suggests it contributes selectively to certain forms of cerebellum-dependent learning and not others ( Boyden et al . , 2006; Hansel et al . , 2006; Titley et al . , 2010; Schonewille et al . , 2011; Aiba et al . , 1994; Shibuki et al . , 1996; Endo et al . , 2009; Feil et al . , 2003; Lee et al . , 2009; Li et al . , 1995; Miyata et al . , 2001 ) . We leveraged a set of closely related oculomotor learning tasks with different dependence on pf-Pk LTD to analyze how enhanced pf-Pk LTD functions in an intact circuit . We tested the ability of DKO mice to adaptively modify their vestibulo-ocular reflex ( VOR ) . The VOR is an eye movement response to a vestibular stimulus , which functions to stabilize images on the retina during head motion . In wild type mice , motor learning can adaptively increase or decrease the amplitude of the VOR to improve image stabilization . Previous work has suggested that LTD contributes selectively to VOR learning when training to increase the amplitude of the VOR is done using high-frequency ( ≥1 Hz ) visual-vestibular stimuli , and much less so , if at all , when VOR learning is tested with other training paradigms ( Boyden et al . , 2006; Hansel et al . , 2006; Titley et al . , 2010; Schonewille et al . , 2011; Aiba et al . , 1994; Shibuki et al . , 1996; Endo et al . , 2009; Feil et al . , 2003; Lee et al . , 2009; Li et al . , 1995; Miyata et al . , 2001 ) . We found that mice with enhanced pf-Pk LTD exhibited the same , specific VOR learning deficit . DKO mice were significantly impaired in learning to increase the amplitude of the VOR when training was done using 1 Hz visual-vestibular stimuli ( Figure 1D , solid bars; Figure 1—source data 1 ) . However , as previously reported in LTD-impaired mice ( Boyden et al . , 2006; Schonewille et al . , 2011 ) , there was no significant impairment of learning to increase the VOR when training was done with lower frequency visual-vestibular stimuli of 0 . 6 Hz ( Figure 1—figure supplement 1 , top ) , or when learning to decrease the VOR at either training frequency ( Figure 1E , solid bars , Figure 1—figure supplement 1 , bottom ) . Baseline performance of the VOR and visually-driven oculomotor behaviors were indistinguishable between DKO and WT mice ( Figure 1—figure supplements 2–3 ) , suggesting an impairment of the learning mechanism itself , rather than other sensory or motor deficits . 10 . 7554/eLife . 20147 . 003Figure 1 . Rescue of H2-Db expression in adult Purkinje cells rescues learning impairment in DKO mice with enhanced cerebellar plasticity . ( A ) Circuit for VOR learning . Vestibular input drives eye movements via a direct pathway through the vestibular nuclei ( VN ) , and a side-loop through the granule cells ( GC ) , parallel fibers ( PF ) , and Purkinje cells ( Pk ) of the cerebellar flocculus . The climbing fiber ( CF ) input to the Purkinje cells from the inferior olive ( IO ) carries visual signals , and can trigger LTD in the parallel fiber-to-Purkinje cell synapses ( blue arrow ) , which is enhanced in mice deficient in the Class-I major histocompatibility molecules H2-Kb and H2-Db ( KbDb−/−; referred to as double knockout , DKO ) . A lentiviral construct expressing H2-Db under the control of the Purkinje cell-specific L7 promoter was injected bilaterally into the flocculi of adult mice ( see Materials and methods for details ) . ( B ) RT-PCR confirmed the presence of H2-Db mRNA in the cerebellar flocculus of DKO mice injected with the L7::H2-Db-T2A-GFP virus ( lane 4 ) . Lane 1: Positive control , WT ( thalamus ) ; Lane 2: Negative control , DKO ( spleen ) ; Lane 3: DKO ( flocculus ) infected with CMV::H2-Db-HA; Lane 4: DKO ( flocculus ) infected with L7::H2-Db-T2A-GFP . Ladder in the left lane . ( Lane 3 , CMV::H2-Db-HA , is a positive control for detection of H2-Db expression in cerebellum , but because it was not restricted to Purkinje cells , it was not used further in this study; GAPDH was not loaded; full details in Materials and methods ) . ( C ) Floccular Purkinje cells of DKO mice infected with L7::H2-Db-T2A-copGFP virus ( white arrowheads ) and stained with anti-copGFP immunohistochemistry . Molecular layer ( ML ) , Purkinje cell layer ( PkL ) , Granule cell layer ( GCL ) . ( D ) Training to increase the VOR . Left , A vestibular stimulus was paired with oppositely directed visual stimulus motion . Right , DKO mice ( solid red ) were impaired on VOR-increase learning compared to wild type mice ( WT; solid black ) ( p=0 . 004 , t ( 38 ) = 3 . 08 ) . Virally-mediated expression of H2-Db in Purkinje cells of the adult cerebellar flocculi ( L7::H2–Db ) rescued the learning deficit in DKO mice ( hatched red; p=0 . 0005 , t ( 37 ) = 3 . 81 vs DKO without virus , solid red ) , so that they learned as well as WT mice injected with the same virus ( hatched black; n . s . p=0 . 65 , t ( 22 ) = 0 . 46 ) and better than DKO mice that received virus expressing only GFP ( L7::GFP , open red , p=0 . 03 , t ( 25 ) = 2 . 290 ) . Virally-mediated expression of H2-Db had no significant effect on learning in the WT mice ( hatched vs . solid black , n . s . p=0 . 70 , t ( 23 ) = 0 . 39 ) , and expression of GFP had no effect in DKO mice ( open vs solid red; n . s . p=0 . 26 , t ( 34 ) = 1 . 13 ) or WT mice ( open vs . solid black; n . s . p=0 . 79 , t ( 24 ) = 0 . 26 ) . Mean ± s . e . m . In this and all figures , numbers in bars indicate n = number of animals . ( E ) Training to decrease the VOR . Left , A vestibular stimulus was paired with a visual stimulus that moved with the head . Right , VOR-decrease learning in DKO mice ( solid red ) was not significantly different from WT ( solid black ) ( n . s . p=0 . 08 , t ( 36 ) = 1 . 86 ) . Expression of H2-Db had no significant effect on VOR-decrease learning in DKO mice ( hatched vs . solid red; n . s . p=0 . 43 , t ( 33 ) = 0 . 79 ) , and was not different from mice that received control virus expressing only GFP ( hatched vs . open red; n . s . p=0 . 29 , t ( 25 ) = 1 . 08 ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00310 . 7554/eLife . 20147 . 004Figure 1—source data 1 . Rescue of H2-Db expression in adult Purkinje cells rescue learning impairment in DKO mice with enhanced cerebellar plasticity . Data show VOR learning , after thirty minutes of either VOR-increase training or VOR-decrease training , for both DKO and WT mice in three different conditions ( without virus , with L7::H2-Db virus , or control L7::GFP virus ) . The eye movement response to the vestibular stimulus alone , i . e . the VOR , was measured in total darkness , and learning was calculated as the percentage change in the VOR gain after training relative the baseline VOR gain measured before training . Positive values indicate increase VOR learning and negative values indicate decrease VOR learning corresponding to the respective training stimuli . Each number corresponds to the learning in an individual animal . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00410 . 7554/eLife . 20147 . 005Figure 1—figure supplement 1 . DKO mice were selectively impaired on high-frequency VOR-increase learning . Top , DKO mice ( solid red ) were impaired on VOR-increase learning compared to WT mice ( solid black ) only when training and testing were performed using vestibular and visual stimuli at a stimulus frequency of 1 . 0 Hz ( *p=0 . 004 , t ( 38 ) = 3 . 08 ) , but not 0 . 6 Hz ( p=0 . 33 , t ( 20 ) = 1 . 00 ) . Bottom , VOR-decrease learning in DKO mice was not significantly different from WT at either frequency ( 0 . 6 Hz , p=0 . 43 , t ( 12 ) = 0 . 88; 1 . 0 Hz , p=0 . 08 , t ( 36 ) = 1 . 86 ) . Data for training and testing at 1 . 0 Hz reproduced from Figure 1D , E . Mean ± s . e . m . Number in bars represent n = number of animals . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00510 . 7554/eLife . 20147 . 006Figure 1—figure supplement 2 . Baseline oculomotor performance was normal in DKO mice . ( A ) Gain of the eye movement responses during performance of the VOR in the dark ( VOR: 0 . 6 Hz , p=0 . 27 , t ( 22 ) = 1 . 12; 1 . 0 Hz , p=0 . 62 , t ( 31 ) = 0 . 49 ) , in the presence of the visual-vestibular training stimuli used to induce VOR-increase learning ( VOR-increase training: 0 . 6 Hz , p=0 . 45 , t ( 22 ) = 0 . 63; 1 . 0 Hz , p=0 . 60 , t ( 38 ) = 0 . 53 ) , in the presence of the visual-vestibular training stimuli used to induce VOR-decrease learning ( VOR-decrease training: 0 . 6 Hz , p=0 . 17 , t ( 12 ) = 1 . 44; 1 . 0 Hz , p=0 . 09 , t ( 19 ) = 1 . 79 ) , and during the optokinetic response to visual stimulus motion with the head stationary ( OKR: 1 . 0 Hz , p=0 . 84 , t ( 12 ) = 0 . 20 ) . The baseline , pre-training eye movement performance of the DKO mice ( red bars ) was indistinguishable from WT mice ( black bars ) , suggesting there was no sensory or motor deficit that could account for the VOR-increase learning impairment in DKO mice . Mean ± s . e . m . ( B ) Eye movement phase ( relative to the vestibular stimulus for VOR , VOR-increase , and VOR-decrease training and relative to the visual stimulus for OKR ) , during the same stimuli as in ( A ) was also normal in DKO mice ( VOR: 0 . 6 Hz , p=0 . 57 , t ( 22 ) = 0 . 57; 1 . 0 Hz , p=0 . 28 , t ( 31 ) = 1 . 10; VOR-increase training: 0 . 6 Hz , p=0 . 78 , t ( 22 ) = 0 . 27; 1 . 0 Hz , p=0 . 32 , t ( 38 ) = 1 . 00; VOR-decrease training: 0 . 6 Hz , p=0 . 61 , t ( 12 ) = 0 . 52; 1 . 0 Hz , p=0 . 94 , t ( 19 ) = 0 . 08; OKR: 1 . 0 Hz , p=0 . 51 , t ( 16 ) = 0 . 67 ) . Positive ( negative ) values indicate phase lag ( lead ) . Mean ± s . e . m . ( C ) Peak retinal-slip velocity ( image motion on the retina ) was not significantly different between DKO and WT mice during VOR-increase training , confirming that the two genotypes experienced similar sensory errors during training ( 0 . 6 Hz , p=0 . 75 , t ( 18 ) = 0 . 33; 1 . 0 Hz , p=0 . 57 , t ( 37 ) = 0 . 57 ) , although there was a difference in learning at 1 Hz ( Figure 1D ) . Mean ± s . e . m . ( D ) Phase of retinal slip relative to the vestibular stimulus was not significantly different between DKO mice and WT controls during VOR-increase training ( 0 . 6 Hz , p=0 . 78 , t ( 18 ) = 0 . 28; 1 . 0 Hz , p=0 . 70 , t ( 37 ) = 0 . 39 ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00610 . 7554/eLife . 20147 . 007Figure 1—figure supplement 3 . Control for eye movements during training . Subsampling was performed on the data from Figure 1D to test whether the impaired learning phenotype in DKO mice could result from the trend ( n . s . , p=0 . 60 , t ( 38 ) = 0 . 53 ) for the DKO mice to have a lower mean eye movement gain during the visual-vestibular training stimuli used to induce learning ( see Figure 1—figure supplement 2A , fourth pair of bars from left , VOR Increase Training 1 . 0 Hz ) . Data from Figure 1D were subsampled by alternately dropping data from the DKO mouse with the smallest eye movements during training , and from the WT mouse with the largest eye movements during training , until the trend for the DKO mice to have a smaller mean gain during training was eliminated , which was achieved after dropping data from two DKO mice and one WT mouse ( left ) . Even after controlling for eye movement performance during training , DKO mice ( red ) exhibited a significant deficit relative to WT ( black ) in VOR-increase learning ( right; p=0 . 003 , t ( 35 ) = 3 . 22 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 007 The impaired learning phenotype in DKO mice could be attributed to the loss of H2-Db expression in Purkinje cells , the post-synaptic site of pf-Pk LTD . Although H2-Kb and H2-Db expression is not exclusive to the cerebellar Purkinje cells , virally-mediated rescue of H2-Db expression specifically in Purkinje cells of the flocculus ( Figure 1A–C ) , the cerebellar region necessary for VOR learning , of adult global DKO mice , was sufficient to rescue their impaired VOR-increase learning ( Figure 1D , hatched bars ) . Thus , rescue of H2-Db expression in adult neurons can restore normal function , even in animals that developed in the absence of this molecule , indicating that the role of MHCI molecules is not confined to the developing nervous system , but actively regulates plasticity in adults as well . Expression of H2-Db had no effect on VOR-decrease learning ( Figure 1E , hatched bars ) , which is insensitive to perturbations of pf-Pk LTD ( Boyden et al . , 2006; Hansel et al . , 2006; Schonewille et al . , 2011 ) . The similarity of the learning deficit in mice with enhanced pf-Pk LTD to that previously reported in mice with impaired pf-Pk LTD ( Boyden et al . , 2006 ) suggested the possibility of a similar underlying cause . We hypothesized that in both cases , the behavioral deficit could reflect the unavailability of pf-Pk LTD during training . In particular , the lower induction threshold for pf-Pk LTD in the DKO mice ( McConnell et al . 2009 ) could allow normal basal activity in the circuit to aberrantly recruit LTD and deplete the pool of LTD-eligible synapses . Thus , the capacity for LTD could be exhausted , i . e . , saturated , even before the start of training , rendering the circuit unable to support new learning that depends on pf-Pk LTD ( Figure 2A ) . 10 . 7554/eLife . 20147 . 008Figure 2 . Elevated LTD before training impairs LTD-dependent learning . ( A ) Saturation hypothesis to explain impaired learning with enhanced synaptic plasticity . Top , In naïve WT mice , at the start of training , synapses are presumably available ( white synaptic spines ) to selectively undergo associative synaptic plasticity ( long-term depression , LTD; blue spines ) during training , thereby supporting normal learning . Bottom , In DKO mice , the lower induction threshold for LTD could enable spontaneous activity in the circuit to aberrantly recruit LTD at a random subset of spontaneously active synapses before training , thereby depleting the pool of synapses eligible to undergo LTD , and preventing normal learning . Behavioral pre-training ( orange arrow ) restores the capacity for LTD-dependent learning in the DKO mice ( Figure 3 ) . We tested whether LTD saturation and impairment of LTD-dependent learning can be induced in WT mice with climbing fiber stimulation ( cyan arrow; Figure 2B ) . ( B ) Climbing fiber stimulation in WT mice before VOR training recapitulates the learning impairment in the DKO mice . Optogenetic stimulation of climbing fibers for 30 min , to induce pf-Pk LTD in the flocculus of WT mice , blocked subsequent VOR-increase learning ( solid cyan trace; p=0 . 03 , F ( 1 , 10 ) = 5 . 912 , two-factor repeated measures ANOVA , CF stim n = 6 , Sham n = 6 ) but had no effect on VOR-decrease learning ( dashed cyan trace; n . s . p=0 . 68 , F ( 1 , 5 ) = 0 . 20 ) relative to sham stimulation controls in animals that did not express ChR2 in the climbing fibers ( black ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00810 . 7554/eLife . 20147 . 009Figure 2—source data 1 . Elevated LTD before training impairs LTD-dependent learning . Climbing fiber stimulation in WT mice before VOR training recapitulates the learning impairment in the DKO mice . Data show time course of VOR learning in WT mice during 30 min of normal visual-vestibular VOR training following 30 min of optical stimulation to either optogenetically stimulate climbing fibers ( CFs ) or provide sham controls ( light only in WT without ChR2 expressed in the CFs ) . Learning was calculated as the percentage change in VOR gain measured after each 10 min block of visual-vestibular training relative to the baseline VOR gain measured immediately before visual-vestibular training . Positive values indicate increase VOR learning and negative values indicate decrease VOR learning corresponding to the respective training stimuli . Each row of 4 numbers within the condition columns corresponds to the time course of learning in an individual animal . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 00910 . 7554/eLife . 20147 . 010Figure 2—figure supplement 1 . Climbing fiber stimulation did not permanently impair VOR-increase learning . ( A ) Circuit diagram of the VOR , showing optogenetic stimulation of the climbing fiber inputs to floccular Purkinje cells in the cerebellar flocculus . Expression of ChR2 in climbing fibers was achieved with injections of adeno-associated virus ( AAV ) carrying ChR2 under the CaMKIIα promoter ( CaMKIIα-ChR2 ( H134R ) -EYFP ) into the inferior olive ( IO ) . Blue light stimulation was delivered directly to the cerebellar flocculus to activate climbing fibers relevant to VOR learning . CF: climbing fiber , IO: inferior olive , PF: parallel fiber , GC: granule cell , Pk: Purkinje cell , VN: vestibular nuclei . ( B ) Thirty minutes of pre-training with climbing fiber stimulation did not significantly affect the VOR at the start of VOR-increase ( circles/solid traces ) or VOR-decrease ( triangles/dashed traces ) training . Blue: mice with ChR2 expression in the climbing fibers ( n = 6 ) . Black: Sham stimulation control mice experiencing the same illumination of the cerebellar flocculus during pre-training but not expressing ChR2 in the climbing fibers ( n = 6 ) . During pre-training with optogenetic climbing fiber stimulation , animals were restrained with their head stationary in the dark . At 10 min intervals , the optogenetic stimulation was briefly interrupted to test the VOR . All groups exhibited a temporary decrease in the VOR during the pre-training period , but this effect did not depend on climbing fiber stimulation ( blue vs . black traces ) . Mean ± s . e . m . ( C ) Climbing fiber stimulation immediately before training impaired VOR-increase learning ( Figure 2B ) , but the same animals exhibited normal learning in response to the same visual-vestibular VOR-increase training stimuli several days later when trained without additional climbing-fiber stimulation pre-training ( blue circles/solid trace ) compared to sham stimulation control animals not expressing ChR2 in the climbing fibers ( black circles/solid trace; ANOVA , p=0 . 79 , F ( 1 , 70 ) = 0 . 074 ) . Thus , climbing-fiber stimulation had no long-lasting adverse effects , but rather created a temporary state of the circuit that was unresponsive to VOR-increase training ( Figure 2B ) . VOR-decrease learning was also normal when tested several days after climbing fiber stimulation ( blue triangles/dashed trace; ANOVA , p=0 . 83 , F ( 1 , 88 ) = 0 . 048 compared to sham controls without ChR2 shown in black triangles/dashed trace ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 01010 . 7554/eLife . 20147 . 011Figure 2—figure supplement 2 . Non-specific LTD may have no immediate effect on behavior , yet deplete the pool of synapses available to support LTD-dependent learning . For the VOR , non-specific LTD induced by climbing fiber stimulation or the lower threshold for LTD in DKO mice has no effect on the VOR amplitude ( Figure 1—figure supplement 2A and Figure 2—figure supplement 1B ) , yet could saturate LTD in the synapses that support VOR increase learning . This would occur if the LTD is induced both in synapses whose depression contributes to VOR-increase learning , and in additional synapses whose depression could negate the effects of LTD in the first subset . Two specific possibilities are illustrated . ( A ) There could be some pf-Pk synapses whose depression causes an increase in VOR amplitude ( up arrow ) and others whose depression causes a decrease in VOR amplitude ( down arrow ) . The latter may not normally undergo LTD in response to the VOR-decrease training paradigms used in our study ( since there is no apparent effect of LTD enhancement or impairment on the learning induced by such training; Figure 1E , solid bars , Figure 1—figure supplement 1 , bottom , and Boyden et al . , 2006 ) . Nevertheless , when they do undergo LTD through a non-specific induction process such as climbing fiber stimulation or a higher spontaneous LTD rate in DKO mice , this would oppose the effects of LTD in the synapses whose depression tends to increase the gain of the VOR , to yield no net change in the gain of the VOR ( top ) . ( B ) There could be some pf-Pk synapses whose depression causes an increase in VOR amplitude ( up arrow ) and others whose depression is irrelevant for the VOR , followed by homeostatic normalization . Top: If too many synapses undergo LTD , there could be a homeostatic process , such as synaptic rescaling or an increase in dendritic excitability , that resets all synaptic weights to values close to their original value , without resetting the LTD mechanism or the capacity for additional LTD , so that these synapses remain LTD-ineligible ( blue shading ) . Thus , the VOR would not be altered , but the LTD mechanism would not be available to support learning . Bottom: If LTD is induced selectively in the synapses that contribute to VOR-increase learning , the weight of these specific synapses remain depressed , and hence the gain of the VOR increased , even after homeostatic normalization restores the summed weight across all synapses to the original value . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 011 To assess whether the saturation of LTD could produce a motor learning phenotype like that observed in the DKO mice , we conducted stimulation experiments . In other brain areas , direct stimulation of the relevant circuits to induce and saturate plasticity has been shown to occlude or impair subsequent learning . We used a similar approach to test whether saturation of LTD can produce the selective impairment of high-frequency VOR-increase learning observed in the DKO mice . Climbing-fiber stimulation is known to induce LTD in simultaneously active pf-Pk synapses ( Crepel and Jaillard , 1991; Ekerot and Kano , 1985; Ito and Kano , 1982 ) . Therefore , we opotogenetically stimulated the climbing fiber input to the cerebellar flocculus to elevate the level of pf-Pk LTD in WT mice prior to VOR training ( Figure 2A , cyan arrow ) . Climbing fibers were optogenetically stimulated for 30 min ( 250 ms trains of three 2 ms light pulses , repeated every 1 s ) while the mouse was head-restrained in the dark without visual or vestibular stimuli . During normal VOR learning , climbing fiber activation is thought to induce LTD selectively in those pf-Pk synapses activated by the visual and vestibular stimuli used to induce learning . In contrast , optogenetic climbing fiber stimulation delivered in the absence of such stimuli should induce LTD randomly in spontaneously active pf-Pk synapses . This non-specific LTD-induction procedure did not affect the amplitude of the VOR , measured after climbing fiber stimulation ( Figure 2—figure supplement 1B ) . This is consistent with the normal baseline VOR amplitude in DKO mice ( Figure 1—figure supplement 2A ) , and in wild type mice after lesions of the flocculus ( Rambold et al . , 2002; Koekkoek et al . , 1997; Katoh et al . , 2005 ) . Together , these observations indicate that non-specific manipulations of the flocculus are not sufficient to have a coordinated effect on the VOR behavior , and that LTD only increases the amplitude of the VOR if it is induced selectively in the appropriate subset of pf-Pk synapses . Nevertheless , if non-specific LTD depleted the pool of LTD-eligible synapses , it could impair subsequent LTD-dependent learning ( Figure 2—figure supplement 2 ) . Accordingly , VOR-increase learning was impaired after climbing fiber stimulation ( Figure 2B , CF Stim , cyan vs . Sham Stim , black; Figure 2—source data 1 ) . Sham stimulation controls exhibited VOR-increase learning , confirming that disrupted VOR-increase learning was specific to stimulation of climbing fibers , rather than reflecting nonspecific , optical or mechanical perturbation of the circuit . Notably , climbing fiber stimulation before training did not perturb subsequent VOR-decrease learning , which also relies on the cerebellar flocculus ( Koekkoek et al . , 1997; Rambold et al . , 2002 ) but is insensitive to disruptions of pf-Pk LTD ( Boyden et al . , 2006 ) ( Figure 2B , dashed traces ) . The specificity of the learning impairment in WT mice after climbing fiber stimulation indicates that elevated levels of pf-Pk LTD prior to training can produce a phenotype like that observed in the DKO mice ( Figure 1D ) . If elevated pf-Pk LTD prior to training is contributing to the learning impairment in the DKO mice , then any procedure that reverses pf-Pk LTD might reset the synapses to a state more capable of supporting LTD-dependent learning . Pf-Pk LTD can be actively reversed by post-synaptic LTP of the same synapses ( Lev-Ram et al . , 2003 ) , providing a cellular mechanism for reversing LTD saturation . Moreover , behavioral VOR-decrease training has been shown to rapidly reverse any evidence of prior VOR-increase learning ( Boyden and Raymond , 2003 ) , suggesting that it rapidly reverses any pf-Pk LTD or other plasticity induced during VOR-increase learning . Therefore , we tested whether VOR-decrease training could put the VOR circuit of DKO mice into a state more capable of supporting LTD-dependent VOR-increase training . Mice were given thirty minutes of VOR-decrease pre-training immediately before VOR-increase training . In the DKO mice , this pre-training significantly enhanced subsequent VOR-increase learning relative to DKO mice that did not receive the pre-training ( Figure 3A , compare red traces and bars in middle vs . left panels; Figure 3—source data 1 ) . Notably , pre-training had the opposite effect on WT mice , impairing subsequent VOR-increase learning ( Figure 3A , compare black traces in middle vs . left panels ) . Since the pre-training enhanced learning in the DKO mice , but impaired learning in the WT mice , the DKO mice learned better than the WT after pre-training . Hence , the pre-training not only reversed the learning impairment , but also revealed a capacity for enhanced learning in the DKO mice relative to WT mice ( Figure 3A , compare black vs . red traces in middle panel ) . 10 . 7554/eLife . 20147 . 012Figure 3 . Behavioral pre-training reveals enhanced learning in mice with enhanced LTD . ( A ) The same VOR-increase training procedure induced dramatically different learning outcomes in the DKO mice with different pre-training procedures ( p=0 . 01 , F = 5 . 153 , ANOVA ) . Left , Without pre-training , DKO mice with enhanced pf-Pk LTD were impaired on VOR-increase learning ( *p=0 . 002 , F ( 1 , 38 ) = 11 . 08 , two-factor repeated measures ANOVA; WT n = 16 , . DKO n = 24 , ) . Middle , Pre-training with an associative VOR-decrease paradigm that was not significantly different between the genotypes ( dotted lines , p=0 . 19 , F ( 1 , 29 ) = 1 . 79; WT n = 12 , DKO n = 19 ) reversed the learning impairment in DKO mice ( red ) so that they learned more than WT ( black ) during subsequent VOR-increase training ( p=0 . 02 , F ( 1 , 29 ) = 5 . 95; WT n = 12 , DKO n = 19 ) . Right , Pre-training with a vestibular stimulus alone decreased the VOR gain comparably between the two genotypes ( dotted line , p=0 . 30 , F ( 1 , 17 ) = 1 . 25; WT n = 6 , DKO n = 7 ) , but there was no improvement of subsequent VOR-increase learning in the DKO mice relative to WT mice ( p=0 . 13 , F ( 1 , 11 ) = 2 . 70; WT n = 6 , DKO n = 7 ) . In DKO mice , VOR-increase learning was better after associative VOR-decrease pre-training compared with no pre-training ( *p=0 . 005 , Fischer’s LSD ) or vestibular-only pre-training ( p=0 . 03 ) ( compare red bar graphs and learning curves ) . In contrast , in WT mice , VOR-increase learning was worse after associative VOR-decrease pre-training compared with no pre-training ( p=0 . 037 , Fischer’s LSD ) or vestibular only pre-training ( p=0 . 049 ) ( compare black learning curves ) . Learning is plotted on the same scale in each plot , and aligned on the values at the start of VOR-increase training for DKO mice . Mean ± s . e . m . ( B ) Virally-mediated rescue of H2-Db expression in floccular Purkinje cells ( L7::H2-Db , left ) eliminated the enhanced VOR-increase learning in DKO mice after associative VOR-decrease pre-training ( compare with middle panel of A ) , so that learning was indistinguishable from WT mice injected with the same virus ( VOR-increase learning , p=0 . 98 , F ( 1 , 22 ) = 0 . 0004; VOR-decrease pre-training , p=0 . 53 , F ( 1 , 22 ) = 0 . 40; two-factor repeated measure ANOVA; WT n = 9; DKO n = 15 ) . The enhanced VOR-increase learning phenotype was present in DKO mice that received control virus expressing only GFP ( L7::GFP , right , p=0 . 05 , F ( 1 , 18 ) = 4 . 29; WT n = 9 , DKO n = 11 ) although the VOR-decrease pre-training itself was not significantly different between the two genotypes ( p=0 . 20 , F ( 1 , 18 ) = 1 . 75; WT n = 9 , DKO n = 11 ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 01210 . 7554/eLife . 20147 . 013Figure 3—source data 1 . Behavioral pre-training reveals enhanced learning in mice with enhanced LTD . The same VOR-increase training procedure induced dramatically different learning outcomes in the DKO mice with different pre-training procedures . Data show time course of VOR learning in WT and DKO mice during 30 min of different pre-training conditions ( no pre-training , VOR-decrease pre-training , and vestibular only pre-training ) followed by 30 min of normal VOR-increase training . Data are separated by pre-training condition and by whether mice received virus for rescue expression of H2-Db in Purkinje cells ( without virus , with L7::H2-Db virus , or control L7::GFP virus ) . Learning was calculated as the percentage change in the VOR gain after each block of 10 min training relative the baseline VOR gain measured before any training occurred . Negative values indicate decrease VOR learning and positive values indicate increase VOR learning . However , in this case with pre-training , a reduction in negative values also indicates increase VOR learning . Each row of numbers within the condition columns corresponds to the time course of learning in an individual animal . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 01310 . 7554/eLife . 20147 . 014Figure 3—figure supplement 1 . Control for efficacy of VOR-decrease pre-training . Subsampling was performed on the data from Figure 3A , middle to test whether the enhanced learning phenotype in DKO mice could result from the trend ( n . s . , p=0 . 19 , F ( 1 , 29 ) = 1 . 79; WT n = 12 , DKO n = 19 ) for the DKO mice to undergo a smaller mean decrease in VOR gain during the VOR-decrease pre-training . Data from Figure 3A were subsampled by alternately dropping data from the DKO mice with the smallest decrease in VOR gain during pre-training ( as measured at the end of the 30 min of pre-training ) , and from WT mice with the largest decrease in VOR gain during pre-training , until the trend for the DKO mice to have a smaller mean decrease in VOR-gain during pre-training was eliminated . Even after controlling for the efficacy of the VOR-decrease pre-training , there was a trend for DKO mice ( red ) to learn more than WT ( black ) during subsequent VOR-increase training ( p=0 . 07 , F ( 1 , 26 ) = 3 . 68; WT n = 11 , DKO n = 17 ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 01410 . 7554/eLife . 20147 . 015Figure 3—figure supplement 2 . Normal retention of VOR-decrease learning in DKO mice . DKO mice showed normal induction and retention of VOR-decrease learning in response to the VOR-decrease training protocol that was used for behavioral pre-training in Figure 3 . A retention ratio was calculated as the learned percentage change in the VOR ( relative to the pre-training baseline ) measured after a 10 min retention period , divided by the percentage change in the VOR measured immediately after training . A retention ratio of 1 . 0 would represent perfect retention of VOR-decrease learning and values less than 1 . 0 would represent forgetting during the 10 min retention period . DKO mice ( solid red bars ) exhibited no additional forgetting of VOR-decrease learning ( p=0 . 31 , t ( 5 ) = 1 . 11; DKO n = 4 , WT n = 3 ) that could explain the enhanced VOR-increase learning that was observed during the same 10 min period after VOR-decrease training ( Figure 3A , middle panel ) . The retention of VOR-decrease learning was also stable in animals injected either with virus to rescue H2-Db specifically in floccular Purkinje cells ( L7::H2-Db , hatched bars , p=0 . 275 , t ( 11 ) = 1 . 148 , WT n = 5 , DKO n = 8 ) or with control virus expressing only GFP ( L7::GFP , open bars , p=0 . 99 , t ( 11 ) = 0 . 013 , WT n = 6 , DKO n = 7 ) ( Figure 1E ) . During the 10 min retention period , mice were restrained in the dark with their head stationary . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 01510 . 7554/eLife . 20147 . 016Figure 3—figure supplement 3 . No enhanced learning phenotype was observed in DKO mice when tested using lower visual-vestibular stimulus frequencies . When testing and training were conducted using visual and vestibular stimuli at a frequency of 0 . 6 Hz , pre-training with the VOR-decrease paradigm ( dotted lines ) had no effect on subsequent VOR-increase learning in DKO mice ( red ) compared with WT mice ( black ) ( ANOVA , p=0 . 10 , F ( 1 , 84 ) = 2 . 761 , WT n = 7 , DKO n = 7 ) , in contrast to the enhanced VOR-increase learning after pre-training that was observed when the training and testing were done at a stimulus frequency of 1 Hz ( Figure 3A , middle panel ) . Mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 016 The enhanced learning in the DKO mice relative to WT mice after pre-training could not be explained by differences in the efficacy of VOR-decrease learning ( Figure 3—figure supplement 1 ) , nor could it be explained by more rapid forgetting of the effects of pre-training in the DKO mice , because VOR-decrease learning was retained normally ( Figure 3—figure supplement 2 ) . Also , the enhanced learning was unmasked specifically by pre-training with the associative VOR-decrease training paradigm , which paired visual and vestibular stimuli . Simply decreasing the VOR amplitude with a non-associative , habituation paradigm , which presented the vestibular stimulus alone , was not sufficient to unmask the enhanced learning ( Figure 3A , right , Vestibular Only Pre-training ) . Thus , only the appropriate pre-training experience can put the circuit of the DKO mice into a state that enables their enhanced synaptic plasticity to support enhanced learning ( Figure 2A , orange arrow ) . The enhanced learning phenotype of the DKO mice had features in common with the impaired learning phenotype in these mice . First , the enhanced learning phenotype , like the impaired learning phenotype ( Figure 1D , solid bars , Figure 1—figure supplement 1 , top ) , was only observed when training was done using high-frequency ( 1 . 0 Hz ) visual-vestibular stimuli ( Figure 3A , middle panel ) , but not a slightly lower stimulus frequency of 0 . 6 Hz ( Figure 3—figure supplement 3 ) . In addition , the enhanced learning phenotype , like the impaired learning phenotype in the DKO mice , reverted to WT phenotype after virally-mediated expression of H2-Db in the Purkinje cells of adult DKO mice ( compare Figure 3B , left panel , with Figure 3A , middle panel , and Figure 1D , hatched bars ) . The commonality of these features suggests that both the enhanced and impaired learning phenotypes of DKO mice share a common mechanism , involving the same set of synapses . The experimental results above raise the possibility that the learning impairment of the DKO mice relative to WT mice could result from the saturation of LTD . We conducted computational modeling studies to further assess the plausibility of this hypothesis , and , more generally , to develop an understanding of the properties of synaptic plasticity that can support both enhanced and impaired learning , depending on recent experience . First , we examined how the competition between two opposing factors determines the learning outcome . One factor is the enhanced intrinsic propensity for synapses to undergo LTD , which alone would enhance learning . The other factor is the depletion of the number of synapses eligible for LTD , which alone would impair learning . Second , we characterized the properties of synaptic models that could reproduce the opposite effects of behavioral pre-training , namely impairing learning in WT mice , and enhancing learning in DKO mice . The modeling generates predictions about the essential features of synapses that could support both the impaired and enhanced learning outcomes we observed empirically . Moreover , it provides general insights about how the prior history of activity in the circuit interacts with the threshold for plasticity to determine whether learning is impaired or enhanced . We adopted a theoretical framework ( Fusi et al . , 2005; Fusi and Abbott , 2007a; Lahiri and Ganguli , 2013 ) previously used to study memory , to compare different synaptic models . Each model incorporated three key experimental findings: ( 1 ) the selective contribution of pf-Pk LTD to VOR-increase and not VOR-decrease learning ( Boyden et al . , 2006 ) ; ( 2 ) the observation that pf-Pk LTD is easier to induce in the DKO mice ( i . e . , the ‘lower threshold’ for LTD induction McConnell et al . 2009 ) ; and ( 3 ) the ability of VOR-decrease training to reverse the effects of VOR-increase training ( Boyden and Raymond , 2003 ) . The contribution of pf-Pk LTD to VOR-increase learning was modeled by using the rate at which synapses transitioned to a depressed state during training as the measure of VOR-increase learning ( Ito , 1972 ) . Training to increase the VOR was modeled by increasing the rate of LTD ‘events’ , which can be thought of as the rate at which patterns of neural activity occur with the potential to induce LTD ( see Appendix for details ) . The lower threshold for pf-Pk LTD in the DKO mice was modeled as an increase in the probability that an LTD ‘event’ ( i . e . , near-coincident parallel fiber and climbing fiber activation ) will actually induce LTD , which we will refer to as the ‘intrinsic’ LTD rate . In vitro , Pf-Pk LTD can be actively reversed by post-synaptic LTP of the same synapses ( Lev-Ram et al . , 2003 ) . Therefore , the ability of VOR-decrease training to reverse the effects of VOR-increase training was modeled by increasing the rate of LTP events at the pf-Pk synapses during VOR-decrease training ( Boyden and Raymond , 2003 ) . However , we note that the mechanism of VOR-decrease learning is not currently known . Moreover , known asymmetries in VOR-increase and VOR-decrease learning ( Boyden and Raymond , 2003; Kimpo et al . , 2005 ) suggest that the mechanism of VOR decrease learning is not simply pf-Pk LTP . Therefore , we do not attempt to model the decrease in VOR gain itself , but to merely capture the effects of VOR-decrease training on the pf-Pk synapses , and , more specifically , its effect on the availability of pf-Pk synapses to undergo LTD during VOR-increase learning . We implemented synaptic models with different numbers of potentiated and depressed states and different probabilities of transitioning between states ( Montgomery and Madison , 2002; Petersen et al . , 1998 ) , and compared their ability to reproduce our empirical observations of VOR learning in wild type and DKO mice ( Figure 4A ) . Specific synaptic models were considered for their analytical tractability and prevalence in theoretical treatment . In all of the models , the lower threshold for LTD in the DKO mice interacted with the rate of spontaneous LTD events caused by basal activity in the circuit ( basal rate of parallel fiber and climbing fiber coactivation ) to bias the initial distribution of synapses towards the depressed state ( s ) prior to learning ( Figure 4B , D , top right , blue bars ) . Neurons in the depressed states were ineligible or less eligible to undergo additional LTD . Thus , for DKO mice , the outcome of VOR-increase training depended upon a competition between two opposing forces: ( 1 ) an enhanced intrinsic propensity for eligible synapses to undergo LTD , which alone would enhance learning; and ( 2 ) depletion of synapses eligible for LTD , i . e . , saturation of LTD , which alone would impair learning . 10 . 7554/eLife . 20147 . 017Figure 4 . Synaptic models with amplified saturation effects and stubborn synaptic states account for learning in mice with normal and enhanced LTD . ( A ) Four empirical comparisons constrain the models . Left , Empirical results replotted from Figure 3 , with all curves aligned to the start of VOR-increase training , P values can be found in the legend for Figure 3 . Right , Less than and greater than symbols ( < and > ) indicate which mice exhibited greater VOR-increase learning . In all panels of Figure 4: red , DKO mice; black , WT mice; solid lines , no pre-training; dashed lines , with VOR-decrease pre-training . ( B ) A binary synapse model with a strong synaptic state ( orange ) and a weak state ( blue ) . Synapses transition between the two states at the rate of depression ( blue curved arrow ) and potentiation ( orange curved arrow ) . The fraction of synapses in each state prior to VOR-increase learning is indicated by blue and orange bars . VOR-increase learning is measured by the decrease in synaptic weights during training . For DKO mice , the rate of depression was higher than WT , reflecting the lower threshold for LTD , ( thick blue arrow ) , hence a greater fraction of the synapses were in the weak state ( blue bars ) prior to any VOR training . VOR-decrease pre-training ( bottom panels ) increased the fraction of synapses in the strong , LTD-eligible state ( orange ) in both WT and DKO mice . Center , The binary synapse model predicts enhanced learning in DKO vs . WT mice without pre-training ( solid red vs solid black trace ) and enhanced learning in WT mice with vs . without pre-training ( dashed vs solid black trace ) , in contradiction to the empirical results in A ( green brackets and green Ø ) . ( C ) The pooled resource model . Left , The probability of synaptic depression varied with the level of a shared resource that was depleted by the occurrence of depression at other synapses . Right , This model fails to account for the impaired learning in WT mice after pre-training ( dashed black vs . solid black; green bracket ) . ( D ) The serial synaptic model with multiple strong ( orange ) and weak ( blue ) states , but only two values of synaptic strength , can account qualitatively for the effects of enhanced LTD and pre-training on learning ( compare center panel with A ) . Before training , the synapses were strongly biased towards the weak state in the DKO mice , reducing the fraction of LTD-eligible synapses ( blue arrowheads ) , and impairing learning relative to WT ( solid red vs . solid black ) , as observed empirically . VOR-decrease pre-training shifted the bias towards the strong states ( bottom panels ) . In DKO mice , this increased the fraction of LTD-eligible synapses ( blue triangle ) , and enhanced learning ( dashed red ) . In WT mice , pre-training biased the synapses to be too deep into the chain of potentiated states , so that the fraction of LTD-eligible synapses was reduced ( blue triangle ) and learning impaired ( dashed black ) . ( E ) The non-uniform multistate model . Left , Each state is of varying strength from strong ( orange ) to weak ( blue ) , and the transition probabilities between states decay exponentially the further the state is from the center . Right , This model qualitatively reproduced all of the empirical observations of learning . DOI: http://dx . doi . org/10 . 7554/eLife . 20147 . 017 Classical models of LTP and LTD did not reproduce our observation of impaired learning with enhanced plasticity . We tested a simple binary synapse model ( Figure 4B ) , and a more generalized linear multistate model with multiple synaptic strengths ( see Appendix ) . These models encapsulate classical notions of synaptic plasticity as straightforward changes in synaptic strength , to a maximal or minimal bound . However , one can show mathematically , that for all values of the parameters of these models , the enhanced intrinsic LTD rate dominates the saturation effect , at least for the initial phase of learning . Thus , these models incorrectly predict that enhanced plasticity would enhance VOR-increase learning ( Figure 4B , solid red vs . black trace , green bracket; see Appendix for an analytical solution and predictions of models for longer time scales ) . Thus , classical models of synaptic plasticity could not readily account for the behavioral results observed empirically . To predict impaired learning with enhanced plasticity , a mechanism to amplify the effect of depleting the synapses eligible for LTD was required . We first considered a synaptic model in which LTD driven by spontaneous activity in the circuit would not only deplete the LTD-eligible pool , but also retard LTD in the remaining LTD-eligible synapses ( Figure 4C ) , as one might expect , for example , if a protein necessary for LTD induction was present in a cell in limited quantities . This resource-depletion model reproduced the impaired learning of DKO mice , however , it failed to predict the empirical observation of impaired learning after pre-training in WT mice ( Figure 4C , dotted black vs . solid black trace , green bracket ) . To account for this latter observation , the synaptic architecture had to include ‘stubborn’ synaptic states whereby too many LTD-reversing events can impair the capacity for subsequent LTD . One model that possesses both essential properties of amplified saturation effects and stubborn synaptic states is a serial model ( Leibold and Kempter 2008; Ben Dayan Rubin and Fusi , 2007 ) with only two different synaptic weights , but with each weight associated with multiple , internal states ( Figure 4D ) . Enhancing LTD in these models leads to an exponential distribution over synaptic states , which strongly depletes the pool of synapses available to express LTD . This exponential distribution of synapses ( Figure 4D , top right ) can account for the impaired learning phenotype in the DKO mice by providing sufficient depletion of LTD-eligible synapses by spontaneous basal activity to overwhelm the higher intrinsic LTD rate in the remaining LTD-eligible synapses . In this model , pre-training in the DKO mice reversed this saturation bias ( Figure 4D , bottom right vs . top right ) , allowing the higher intrinsic LTD rate to dominate . In contrast , WT mice started with many LTD-eligible synapses , but pre-training pushed synapses deep into the chain of potentiated states , thereby reducing their ability to undergo a subsequent transition to a depressed state ( Figure 4D , bottom left , orange bars ) . Notably , the model predicts that with extended VOR-increase training , the advantage conferred on the DKO mice relative to WT mice by pre-training should disappear ( see Appendix ) . Other synaptic models in which the capacity for a synaptic weight to change depends on the history of prior plasticity events could also account for our empirical observations . Models with such metaplasticity include the cascade model ( Fusi et al . , 2005 ) ( see Appendix ) , and a multistate model with multiple synaptic strengths and lower transition probabilities for the deeper states ( Figure 4E ) . Given appropriate parameters , these models are capable of reproducing all of the qualitative learning outcomes observed experimentally ( Figure 4A , right ) , in contrast to the classical , binary or linear multistate models of plasticity , which are unable to do so for any choice of parameters . This successful class of models illustrates general principles about how the enhancement of plasticity at a given synapse can contribute to both impaired and enhanced learning , depending on the recent history of activity . In essence , under conditions where the recent activity leaves many synapses of WT mice in the labile , LTD-eligible states , enhancing plasticity tends to push the synapses out of these LTD-eligible states before training and thus impairs learning despite enhanced plasticity ( compare Figure 4D , top-left with top-right ) . Under conditions where the recent activity leaves many synapses of WT mice in the ‘stubborn , " potentiated states , enhancing plasticity can push them into the labile , LTD-eligible states , and enhance subsequent learning ( compare Figure 4D , bottom-left with bottom-right ) . A history-dependent alteration in the capacity to undergo additional plasticity has been documented experimentally at some synapses in the hippocampus ( Montgomery and Madison , 2002; Petersen et al . , 1998 ) . Such history-dependence in the plasticity of cerebellar synapses , required by our model to explain our observed behavioral phenotypes , constitutes a key prediction that can be tested in future empirical investigations of the synaptic physiology . Our results provide new insights about how enhanced synaptic plasticity can yield either enhanced or impaired learning , and begin to identify the factors that favor enhanced versus impaired learning when synaptic plasticity is enhanced . Although it was previously known that enhanced plasticity can have these opposite effects on behavior , these divergent results were obtained by different labs , using different learning tasks that depend on different brain regions , and different lines of mice with enhanced plasticity . Here , we established that the same individual animals with enhanced plasticity can respond to the same behavioral training with either enhanced or impaired learning , depending on the recent history of experience . Thus , the capacity for new learning is determined by a dynamic interplay between the threshold for synaptic plasticity and the recent history of activity . Classical models of synaptic plasticity , in which an LTP event simply increases the synaptic strength and an LTD event simply decreases the synaptic strength , do not readily explain our behavioral data . We showed that in such models , enhancing plasticity led to enhanced learning , across the entire parameter space , and independent of the history of previous learning experiences . A critical , theoretical ingredient required to account for our data is a history-dependent form of synaptic plasticity ( Montgomery and Madison , 2002; Petersen et al . , 1998; Fusi et al . , 2005 ) in which repeated LTD changes the internal state of the synapse into a less labile state . Then enhanced LTD can deplete the pool of labile synapses capable of supporting further learning , leading to impaired learning despite enhanced plasticity . Thus , the similar learning deficits in mice with enhanced pf-Pk LTD ( Figure 1D ) and impaired pf-Pk LTD ( Boyden et al . , 2006; Hansel et al . , 2006 ) could reflect a similar underlying cause , namely the unavailability of pf-Pk LTD during training ( Figure 2A ) . More generally , our results suggest a new hypothesis for why enhanced plasticity can impair learning . Such impairments have generally been attributed to an over-recruitment of the enhanced plasticity mechanism at inappropriate synapses during training corrupting the memory trace ( Migaud et al . , 1998; Koekkoek et al . , 2005; Martin et al . , 2000 ) . Our results raise the alternative possibility that enhancing the plasticity mechanism necessary for learning might lead , instead to its under-recruitment at appropriate synapses during learning , as a result of saturation ( Figure 2A ) . The possibility that synaptic plasticity can be saturated has long been recognized ( Martin et al . , 2000 ) . Behavioral training paradigms that induce learning can occlude the subsequent induction of synaptic plasticity in brain slices ( Schreurs et al . , 1997; Pascoli et al . , 2012; Rioult-Pedotti et al . , 1998 ) , and strong , artificial stimulation of neural activity in vivo to saturate a plasticity mechanism has been shown to impair subsequent learning in animals with normal synaptic plasticity ( Martin et al . , 2000; Moser et al . , 1998 ) , as we found for cerebellar climbing fiber stimulation ( Figure 2B ) . Despite its consideration in these contexts , saturation has not been identified previously as a factor that could limit the ability of enhanced synaptic plasticity to enhance learning . Our results provide initial experimental evidence for this hypothesis . In particular , our results suggest that in mice with a low threshold for associative synaptic plasticity , saturation of the plasticity and occlusion of further learning may occur , not only in response to the patterns of elevated neural activity that can induce saturation in WT animals , but also in response to the normal , ongoing , basal levels of activity in a circuit , in the absence of any training or neural stimulation ( compare Figure 1D and Figure 2B ) . Thus , enhanced synaptic plasticity , in the form of a lower threshold for induction , can be opposed by a tendency for plasticity to saturate , which , in turn , limits the capacity for new learning . The capacity for new learning could be decreased or increased by manipulations that altered activity in the VOR circuit for a few tens of minutes . In WT mice , 30 min of elevated climbing-fiber activity induced a state that prevented subsequent learning ( Figure 2B , CF Stim , solid cyan ) . Recovery from this saturation also occurred over a timescale of tens of minutes; although VOR-increase learning was profoundly suppressed after climbing fiber stimulation , there was evidence for recovery of learning during the last few minutes of the 30 min training session ( Figure 2B , solid blue trace at 30 min ) , and full recovery a few days later ( Figure 2—figure supplement 1C ) . Likewise , the natural manipulation of circuit activity caused by 30 min of behavioral pre-training was apparently sufficient to reverse saturation in the DKO mice and transform their learning impairment into a learning enhancement ( Figure 3A ) . Thus , if the recent neural activity is appropriately patterned , rapid recovery from saturation is possible , creating the potential for enhanced plasticity to support enhanced learning . This dependence on the recent history of activity may explain the difference between the results from DKO mice in vivo versus in vitro . In vitro , the levels of spontaneous activity are much lower than in vivo , hence any saturation that is present in vivo would rapidly decay in vitro , revealing enhancement rather than saturation of pf-Pk LTD ( McConnell et al . , 2009 ) . The dependence of learning on the recent history of activity may also explain the observation of either impaired or enhanced learning on behavioral tasks that depend on different parts of the cerebellum ( McConnell et al . , 2009 ) , which have different levels of spontaneous activity ( Zhou et al . , 2014 ) , and different dependence of LTD induction on the patterns of neural activation ( Suvrathan et al . , 2016 ) . In sum , in both DKO and WT mice , the capacity for new learning was highly dependent on the recent history of activity in the circuit over the previous tens of minutes . Our findings reinforce the idea that synaptic plasticity and learning are not isomorphic: one cannot predict the learning outcome from the synaptic properties alone . Rather , the capacity for new learning is determined by a dynamic interplay between the threshold for synaptic plasticity and the recent history of activity . A better understanding of this interaction is of great clinical significance , with the potential to guide the treatment of a wide range of patients who could benefit from enhanced neural plasticity , such as those recovering from brain injury . Our results suggest that synaptic plasticity enhancers may be most effective if combined with strategies for controlling basal levels of neural activity . In particular , suppression of neural activity before training may prime enhanced learning under conditions of enhanced associative plasticity . In addition , our finding that the appropriate behavioral pre-training can unmask enhanced learning in mice with enhanced synaptic plasticity ( Figure 3A ) raises the possibility that behavioral therapy could provide an alternative to drugs in patients with pathologically altered synaptic plasticity ( Koekkoek et al . , 2005; Yashiro et al . , 2009; Baudouin et al . , 2012 ) . A better understanding of how the threshold for synaptic plasticity affects function has the potential to influence many areas of neuroscience . Synaptic plasticity plays a role in nearly all brain functions , from the most basic sensory processing to the highest cognitive functions , and from early development through aging ( Hübener and Bonhoeffer , 2014; Greenwood , 2007; Chen and Tonegawa , 1997; Meredith , 2015 ) . To get traction on the broad issue of how enhanced synaptic plasticity influences learning , we harnessed the analytical power of the relatively simple and well-characterized vestibular and oculomotor systems , and leveraged manipulations informed by specific knowledge about signaling and plasticity in those systems . Yet the finding that the enhancement of synaptic plasticity can result in either enhanced or impaired learning has been reported for many different brain regions , and for both associative LTP and LTD . Therefore , saturation should be considered as a factor that could limit the ability of enhanced plasticity to enhance learning in neural circuits throughout the brain . One can speculate that in each brain area , neural circuits have evolved to optimize the threshold for plasticity to delicately balance the need to prevent inappropriate inputs from triggering and saturating plasticity , while allowing the appropriate inputs to drive learning . All experimental procedures were approved by the Administrative Panel on Laboratory Animal Care at Stanford University under the animal care and use committee ( IACUC ) Protocol #9143 , titled ‘Vestibular and Visual Control of Eye Movements in Mice’ . All mice were housed on a reversed 12 hr light/12 hr dark cycle , and experiments were conducted during the animals’ dark cycle . After implant surgery for behavioral experiments , mice were single-housed in individual cages . All other mice were maintained in group housing of up to five animals per cage . MHCI H2-Kb/H2-Db−/− ( KbDb−/−; referred to as double knockout ( DKO ) ) mice on a C57BL/6 genetic background ( Vugmeyster et al . , 1998; Schott et al . , 2003 ) were maintained as a homozygous breeding colony . Age-matched C57BL/6 WT mice ( RRID:IMSR_JAX:000664 ) were purchased from Jackson Laboratory . Lentivirus was used to drive expression of H2-Db in cerebellar Purkinje cells . H2-Db ( Garstka et al . , 2007 ) was cloned into the BamHI site in the fourth exon of the L7/pcp-2 gene ( Zhang et al . , 2001 ) . L7::H2-Db or L7::GFP ( Oberdick lab ) was cloned into a pCDH-EF1-MCS-T2A-copGFP lentiviral vector backbone ( System Biosciences , Mountain View , CA; where EF1 is the promoter for elongation factor 1α , MCS is the multiple cloning site , T2A is a self-cleaving peptide that allows expression of multiple proteins from a single transcript , and copGFP is a green fluorescent protein used as a reporter ) , using SwaI restriction sites , with the EF1 promoter removed to generate the following constructs: L7::H2-Db-T2A-copGFP , L7::H2-Db- ( stop ) -T2A-copGFP , and L7::GFP-T2A-copGFP . In the L7::H2-Db-T2A-copGFP construct , the T2A enables separate expression of H2-Db and copGFP proteins , and is indicated as L7:: H2-Db/GFP . Because a few amino acid residues from the T2A would be left on H2-Db expressed from the L7::H2-Db-T2A-copGFP virus , a second construct was designed , L7::H2-Db- ( stop ) -T2A-copGFP , which used the stop codon to allow H2-Db expression without any additional amino acid residues from the T2A , and without expression of copGFP . There was no significant behavioral difference between DKO mice injected with L7::H2-Db-T2A-copGFP ( n = 7 ) and L7::H2-Db- ( stop ) -T2A-copGFP ( n = 8 ) ( VOR-increase learning at 1 . 0 Hz: p=0 . 23 ) ; both viruses rescued the learning deficit in the DKO mice , therefore the data from these two groups were pooled and indicated as L7::H2-Db . The L7::GFP-T2A-copGFP expressed GFP and copGFP , as a negative control for the L7 promoter , and is indicated as L7::GFP . All three L7 constructs were sequenced and packaged by the Neuroscience Gene Vector and Virus Core at the Stanford School of Medicine . Lentiviruses were produced by polyethyleneimine-mediated transfection of 293 T cells with four separate plasmids encoding HIV-1 gag-pol , HIV-1 rev , VSV-G envelope , and the HIV-1 based genome vector . Virus-containing culture media was harvested 24 and 48 hr post-transfection , filtered through a 0 . 45 μm filter and concentrated by ultracentrifugation . Concentrated virus was stored in single-use aliquots at −80°C . A fourth virus , CMV::H2-Db-HA , provided a positive control for detection of H2-Db in the cerebellum in initial validation experiments , but because its expression was not restricted to Purkinje cells , it was not used further in this study . An HA tag was fused to the c-terminus of H2-Db by PCR . The CMV promoter was PCRed from a pcDNA3 vector ( Thermo Fisher Scientific ) . CMV::H2-Db-HA was then cloned into a pCDH lentivector backbone ( System Biosciences ) . The construct was sequenced for verification and packaged into lentivirus by the Neuroscience Gene Vector and Virus Core at the Stanford School of Medicine . Two to three weeks after virus injection , the cerebellar flocculi of two DKO mice injected with L7::H2-Db-T2A-copGFP virus were dissected for mRNA analysis by RT-PCR . Thalamus from one WT control and spleen from one DKO mouse were used as positive and negative control samples , respectively . Primers for H2-Db were designed to detect exon 2 and exon 3 regions of H2-Db . RNA was extracted from each sample using RNAqueous-4PCR ( Ambion , Life Technologies , NY ) and cDNA was synthesized using the iScript cDNA Synthesis Kit ( Bio-Rad ) . PCR products were evaluated by gel electrophoresis to confirm the presence of PCR products of predicted size of ~250 bp . H2-Db primers: Sense- 5’CAAGAGCAGTGGTTCCGAGTGAG-3’; Antisense- 5’CTTGTAATGCTCTGCAGCACCACT-3’ . Reactions for RT-PCR were carried out as previously described ( Lee et al . , 2014 ) using 1 ug of cDNA as a template ( 5 min at 95°C followed by 40 cycles ( 30 s at 95°C , 30 s , at 60°C , 30 s , 72°C ) ) ( Veriti 96-well Thermal Cycler , Applied Biosystems ) . Glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) was used as internal control . GAPDH primers: Sense- 5’ATTGTCAGCAATGCATCCTGC-3’ Antisense- 5’AGACAACCTGGTCCTCAGTGT-3’ . The quality of cDNAs was confirmed by genotyping PCR reactions as described previously ( Lee et al . , 2014 ) using 0 . 5 ~ 1 µg of cDNA as template and the samples containing genomic DNAs were discarded . Two weeks after virus injection , three DKO mice injected in the flocculus with L7::H2-Db-T2A-copGFP were deeply anesthetized and immediately perfused with 0 . 1M PBS followed by 4% paraformaldehye ( w/v ) in PBS . The brain was removed and post-fixed for 2 hr at 20°C . After fixation , the brain was placed in 30% sucrose ( w/v ) in PBS solution overnight at 4°C . Then brain was then embedded in OTC ( Sakura Fine Tek ) and frozen for cryosectioning . Coronal sections of 20 um were made through the cerebellar flocculi . Slices were then incubated for 1 hr at 20°C with blocking solution containing 10% normal goat serum and 1% BSA in PBS with 0 . 3% Triton-X ( PBST ) , and then overnight at 4°C with primary antibodies diluted in blocking solution ( anti-TurboGFP , Evrogen AB513 ) . Slices were then washed three times with PBST and incubated for 1 hr at 20°C with secondary antibody ( Alexa Fluor 488 goat anti rabbit , Invitrogen A2153 ) . Fluorescence images were taken using a Nikon Eclipse E800 microscope . During each behavioral experiment , the head of the mouse was immobilized by attaching the implanted head post to a restrainer . The restrainer was attached to a turntable ( Carco Electronics , Pittsburgh , PA ) , which delivered a vestibular stimulus by rotating the mouse about an earth-vertical axis . Visual stimuli were delivered by moving an optokinetic drum made of white translucent plastic with black and white vertical stripes , each of which subtended 7 . 5° of visual angle . The optokinetic drum was back-lit by fiber-optic lights . Horizontal eye position was measured using the eye coil method , and sampled at a rate of either 500 or 1000 Hz . Eye velocity was calculated by differentiating eye position measurements obtained from the eye coil . Any data segment containing a saccade or motion artifact was excluded from the analysis and then a sinusoid was fit to remaining data to extract the amplitude and phase of the eye velocity response to vestibular or visual stimuli . The VOR was measured as the eye-movement response to a sinusoidal vestibular stimulus ( rotation of the head about an earth-vertical axis , 1 . 0 Hz or 0 . 6 Hz , ±10°/s peak velocity ) in complete darkness ( i . e . , with no visual inputs ) , in 40 s blocks . The VOR gain was calculated as the ratio of the eye-to-head movement amplitudes . The optokinetic reflex ( OKR ) was measured as the eye-movement response to an optokinetic visual stimulus ( 1 . 0 Hz , ±10°/s peak velocity ) delivered with the head stationary , and the gain of the OKR was calculated as the ratio of eye-to-visual stimulus amplitudes . Training to increase the VOR gain consisted of pairing a ± 10°/s sinusoidal vestibular stimulus with oppositely directed ±10°/s sinusoidal optokinetic drum rotation , such that a VOR gain of 2 would be required to stabilize the image on the retina ( Figure 1D ) . Training or pre-training to decrease the VOR gain consisted of pairing a ± 10°/s sinusoidal vestibular stimulus with ±10°/s sinusoidal rotation of an optokinetic drum ( visual stimulus ) in the same direction as the head , such that the visual stimulus was stationary relative to the mouse and required a VOR gain of zero to stabilize the image on the retina ( Figure 1E ) . Vestibular-only pre-training consisted of 1 Hz , ±10°/s sinusoidal vestibular stimulation in the dark ( i . e . , no visual stimulus ) . Training and pre-training were each conducted in three ten-minute blocks . Before and after each block , the eye movement response to the vestibular stimulus alone , i . e . , the VOR , was measured in total darkness , and learning was calculated as the change in VOR gain after each block of training . The frequency of the visual-vestibular training stimuli and the vestibular testing stimuli were either 1 . 0 Hz or 0 . 6 Hz . Experimenters running the behavioral experiments were blind to the genotype of the mice . Individual mice were used for multiple behavioral experiments ( subjected to different training and testing conditions ) , separated by at least two days , with no specific randomization in the order of experiments . If the longevity of the eye coil and head post implants allowed , some mice underwent the same type of behavioral experiment more than once , in which case , the results from the replications were averaged for that animal and a single averaged value was used in the group analysis . In some experiments , animals were subjected to 30 min of optogenetic climbing fiber stimulation prior to visual-vestibular VOR-increase or VOR-decrease training . Experiments were performed on WT mice injected in the inferior olive with virus carrying ChR2 , and sham control mice that were not injected with any virus , but underwent that same surgical procedure of eye coil , head post and cannula implantation for behavioral testing . In both the experimental and sham control groups , optical stimulation was delivered unilaterally to the cerebellar flocculus through the implanted cannula via a 200 um optical fiber connected to a blue laser ( 473 nm , Laserglow ) . The optical stimulation consisted of 250 ms trains of 3 pulses , repeated every 1 s , with each pulse 2 ms in duration at an intensity of ≤3 mW . The optical stimulation was delivered in 10 min blocks over the course of 30 min while the animal was head-restrained and stationary in the dark . The gain of the VOR was assessed before and after each block of optical stimulation . VOR training began within 2 min of the end of the 30 min of optical stimulation , using vestibular-visual stimulus pairing to either increase or decrease the gain of the VOR . Adequate sample size was determined based on previous experiments of VOR behavior ( Boyden and Raymond , 2003; Boyden et al . , 2006 ) and optogenetic stimulation ( Nguyen-Vu et al . , 2013 ) in mice , as borne out the by results . Statistical analyses were performed using Microsoft Excel and Graphpad Prism ( RRID:SCR_002798 ) . Data are presented as means ± s . e . m unless otherwise indicated . The Kolmogorov-Smirnov test was used to test for normality , and Bartlett’s multiple sample test was used to determine equal variance . Unpaired Student's t tests ( 2-sided ) were used to compare groups . When the time course data were compared , two-factor repeated measure ANOVA was used to test for a significant difference between groups . A Fischer’s LSD post-hoc test was used only when there was a significant difference between groups . For all tests , p<0 . 05 was considered to be statistically significant . We used a computational approach to determine the essential features of synaptic models that could account qualitatively for our central empirical observations: ( 1 ) without pre-training , enhanced plasticity impairs learning; ( 2 ) pre-training rescues learning in mice with enhanced plasticity; ( 3 ) pre-training impairs learning in WT mice; and ( 4 ) with appropriate pre-training , mice with enhanced plasticity learn faster than WT ( Figure 4 , left , reproduced from Figure 3A ) . In all cases , the contribution of pf-Pk LTD to VOR learning was modeled by measuring the initial rate of VOR-increase learning with the rate at which synapses transitioned to a depressed state during training ( Ito , 1972 ) . This rate was determined by three factors: ( 1 ) fdep , the rate of candidate LTD events , or the pattern of neural activity with the potential to induce LTD ( i . e . , near simultaneous activation of cerebellar parallel fibers and climbing fibers ) ; ( 2 ) qdep , the intrinsic plasticity rate , which corresponded to the threshold for LTD induction , or the probability that a candidate LTD event would cause an eligible synapse to transition to a depressed state; and ( 3 ) pstrong , the number of synapses eligible to transition to a depressed state . The lower induction threshold in DKO mice was modeled as an increase in the intrinsic plasticity rate , qdep ( MHC ) > qdep . Training to increase the VOR was modeled as an increase in the rate of candidate LTD events , fdep → fdep + Δf . Instead of explicitly modeling VOR-decrease learning , because its mechanism ( s ) are still unknown , we only modeled the component that reverses VOR-increase as an increase in the rate of LTP events: fpot → fpot + Δf . For illustration , we chose fpot = fdep = ½ in the absence of VOR training , though none of our results depend upon this balanced rate of candidate LTP and LTD events . We compared synaptic models with different numbers of potentiated and depressed states and different probabilities of transitioning between states . See Appendix for details . Computational models were simulated using custom MATLAB ( RRID:SCR_001622 ) code ( MATLAB R2013b , The MathWorks Inc . , Natick , Massachusetts ) . The code for simulating the computational models is publicly available .	All animals can learn from their experiences . One of the main ideas for how learning occurs is that it involves changes in the strength of the connections between neurons , known as synapses . The ability of synapses to become stronger or weaker is referred to as synaptic plasticity . High levels of synaptic plasticity are generally thought to be good for learning , while low levels of synaptic plasticity make learning more difficult . Nevertheless , studies have also reported that high levels of synaptic plasticity can sometimes impair learning . To explain these mixed results , Nguyen-Vu , Zhao , Lahiri et al . studied mice that had been genetically modified to show greater synaptic plasticity than normal mice . The same individual mutant animals were sometimes less able to learn an eye-movement task than unmodified mice , and at other times better able to learn exactly the same task . The main factor that determined how well the mice could learn was what the mice had experienced shortly before they began the training . Nguyen-Vu et al . propose that some experiences change the strength of synapses so much that they temporarily prevent those synapses from undergoing any further changes . Animals with these “saturated” synapses will struggle to learn a new task , even if their brains are normally capable of high levels of synaptic plasticity . Notably , even normal activity appears to be able to put the synapses of the mutant mice into a saturated state , whereas this saturation would only occur in normal mice under a restricted set of circumstances . Consistent with this idea , Nguyen-Vu et al . showed that a specific type of pre-training that desaturates synapses improved the ability of the modified mice to learn the eye-movement task . Conversely , a different procedure that is known to saturate synapses impaired the learning ability of the unmodified mice . A future challenge is to test these predictions experimentally by measuring changes in synaptic plasticity directly , both in brain slices and in living animals . The results could ultimately help to develop treatments that improve the ability to learn and so could provide benefits to a wide range of individuals , including people who have suffered a brain injury or stroke .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	A saturation hypothesis to explain both enhanced and impaired learning with enhanced plasticity
Many bacteria communicate with kin and coordinate group behaviors through a form of cell-cell signaling called acyl-homoserine lactone ( AHL ) quorum sensing ( QS ) . In these systems , a signal synthase produces an AHL to which its paired receptor selectively responds . Selectivity is fundamental to cell signaling . Despite its importance , it has been challenging to determine how this selectivity is achieved and how AHL QS systems evolve and diversify . We hypothesized that we could use covariation within the protein sequences of AHL synthases and receptors to identify selectivity residues . We began by identifying about 6000 unique synthase-receptor pairs . We then used the protein sequences of these pairs to identify covariation patterns and mapped the patterns onto the LasI/R system from Pseudomonas aeruginosa PAO1 . The covarying residues in both proteins cluster around the ligand-binding sites . We demonstrate that these residues are involved in system selectivity toward the cognate signal and go on to engineer the Las system to both produce and respond to an alternate AHL signal . We have thus demonstrated that covariation methods provide a powerful approach for investigating selectivity in protein-small molecule interactions and have deepened our understanding of how communication systems evolve and diversify . Quorum sensing ( QS ) is a widespread form of cell-cell signaling that bacteria use to coordinate the production of public goods including toxins , antibiotics , bioluminescence , and secreted enzymes ( Waters and Bassler , 2005; Whiteley et al . , 2017 ) . Many Proteobacteria ( Case et al . , 2008 ) and Nitrospirae ( Mellbye et al . , 2017 ) employ a form of QS based on acyl-homoserine lactone ( AHL ) signals . AHL QS systems consist of two proteins: a LuxI-type signal synthase and a LuxR-type receptor ( Figure 1A ) . The signal synthase produces an AHL from S-adenosyl methionine ( SAM ) and an acyl-acyl carrier protein for some LuxI-type synthases or an acyl-coenzyme A substrate for others ( Schaefer et al . , 2008; Figure 1B ) . AHL signals can freely diffuse through cell membranes ( Kaplan and Greenberg , 1985; Pearson et al . , 1999 ) and at low cell density the QS system is ‘off’ . At high cell density , the signal accumulates and binds the LuxR-type receptor which is a cytosolic transcription factor that regulates gene expression in response to signal binding . AHL signals share a conserved lactone core , but vary in the acyl moiety which can be a fatty acid ranging from 4 to 20 carbons long , with potential oxidation on the C3 carbon and varying degrees of unsaturation , or can have an aromatic or branched structure ( Rajput et al . , 2016 ) . This variability in the acyl portion of the signal confers selectivity to the system . Typically , a LuxI-type synthase produces a primary AHL to which its paired LuxR-type receptor selectively responds ( Aframian and Eldar , 2020 ) . Selectivity is critical to cell signaling in order to avoid undesired cross-talk or spurious outputs ( Laub , 2016 ) . In the case of QS , selectivity ensures bacteria cooperate only with kin cells . Despite its importance , we know little about how QS systems achieve selectivity or how they evolve and diversify to use new signals . Although the conserved amino acids essential for synthase and receptor activity are well described ( Parsek et al . , 1997; Zhang et al . , 2002 ) , residues that dictate selectivity are often different from those that are required for activity ( Collins et al . , 2005 ) . Due to the low amino acid sequence identity between LuxI/R homologues , it has been difficult to determine how QS systems discriminate between various AHL signals ( Fuqua et al . , 1996 ) . We hypothesized that we could use covariation patterns to identify QS selectivity residues . Such methods have been used to identify amino acid residues that interact with each other within proteins and between proteins that physically bind one another ( Aakre et al . , 2015; Ovchinnikov et al . , 2014; Skerker et al . , 2008 ) . Here , we endeavored to expand these methods to assess the interaction between AHL synthases and receptors . While AHL synthases and receptors do not physically interact , they interact indirectly via binding to a shared cognate signal and are believed to coevolve to maintain this shared signal recognition ( Aframian and Eldar , 2020 ) . Phylogenetic analyses also support coevolution of synthases and receptors ( Gray and Garey , 2001; Lerat and Moran , 2004 ) . We therefore hypothesized that we could apply covariation methods in a novel way to identify amino acid residues that covary between QS synthases and receptors , and further , that the covarying residues would be those responsible for signal selectivity . We used a statistical method , GREMLIN ( Kamisetty et al . , 2013 ) , to measure covariation within the sequences of AHL synthase-receptor pairs and mapped the covarying residues onto the LasI/R QS system of Pseudomonas aeruginosa PAO1 . By engineering substitutions in the top-scoring residues identified by GREMLIN , we demonstrate that they are indeed important for signal selectivity and , further , that these residues can be used to rationally engineer LasI/R to produce and respond to a non-native signal . We thus provide a proof of principle for a new use of covariation methods to investigate selectivity in non-physical protein-protein interactions and at the same time identify determinants of QS signal selectivity . To begin our analysis , we gathered select protein sequences for known synthase-receptor pairs ( Supplementary file 1A ) and used these sequences to search the European Nucleotide Archive ( ENA ) database ( Amid et al . , 2019 ) from the European Bioinformatics Institute and the Integrated Microbial Genomes and Microbiomes ( IMG/M ) database ( Chen et al . , 2021 ) from the Joint Genome Institute ( JGI ) for additional synthase-receptor pairs . The genes for synthase-receptor pairs are frequently co-located on the genome , and organisms can harbor more than one complete QS system ( Fuqua et al . , 1996 ) . To increase the likelihood of identifying true pairs , we required that the two genes be separated by no more than two coding sequences . A total of 6360 non-identical pairs were identified . We further discarded pairs that were more than 90% identical to another pair , resulting in 3489 representative AHL synthase-receptor pairs . We aligned these sequences to the LasI/R QS system from P . aeruginosa PAO1 . Not only is P . aeruginosa a clinically important pathogen , the Las system is well studied and crystal structures have been solved for both LasI ( Gould et al . , 2004 ) and LasR ( Zou and Nair , 2009 ) , making this a particularly useful model system for our studies . We connected the sequences of the synthase and the receptor from each pair and used GREMLIN to analyze covariation patterns in these sequences . We applied average product correction ( APC ) to the GREMLIN covariance coefficients , a common technique shown to improve the accuracy of coevolution analyses ( Buslje et al . , 2009 ) . An overview of our workflow is shown in Figure 2—figure supplement 1 . We performed the same analysis by aligning the synthase-receptor pairs to the LuxI/R system from Vibrio fischeri MJ11 . The top-ranking coevolving residue pairs overlap significantly between the LasI/R and LuxI/R systems ( 62 . 5% in common among the top 0 . 05% residue pairs ) ( Figure 2—figure supplement 2A–D ) . We integrated the analyses for the LasI/R and LuxI/R systems by using the higher score for each residue pair and the top 10 residue pairs are shown in Figure 2A and Figure 2—figure supplement 3A . As a control , we randomly paired the synthases and receptors from different species and reanalyzed them using GREMLIN . While top-scoring covarying residues had a minimal GREMLIN score ( with APC ) of 0 . 09 , the highest score from the randomized control was 0 . 08 ( Figure 2B and Figure 2—figure supplement 3B ) . This control provides a guideline for our analysis; residues with a GREMLIN score ( with APC ) above or near the maximal score for the randomized control are likely to be meaningful . For both LasI and LasR , the top-scoring covarying residues cluster around the ligand-binding pocket . For LasR , the top-scoring residues map exclusively to the ligand-binding domain with an average distance of 5 . 0 Å from the co-crystalized native ligand N-3-oxo-dodecanoyl-L-homoserine lactone ( 3OC12-HSL ) ( Figure 2C ) . In contrast , the top residues identified in the randomized control are scattered throughout LasR , including three residues in the DNA-binding domain , and are an average distance of 18 Å from 3OC12-HSL ( Figure 2D ) . In LasI , the top-scoring covarying residues cluster around the hydrophobic pocket thought to bind the fatty acyl substrate ( Figure 2E ) and are an average distance of 3 . 7 Å from an acyl substrate modeled into the LasI structure ( Gould et al . , 2004 ) . As with LasR , the residues identified in the randomized control are scattered throughout LasI , with many of the residues exposed to solvent ( Figure 2F ) . The randomized control residues in LasI are over three times further from the fatty acyl substrate , mean distance = 12 Å , compared to the covarying residues . Due to their location near the ligand-binding pockets , several of the covarying residues have been previously studied in various LasI/R homologues . Encouragingly , many of these residues have been reported to be important for protein activity and , in some cases , for selectivity . We have summarized several of these studies in Supplementary files 1B-C . To determine whether residues identified by GREMLIN are involved in LasR selectivity , we introduced substitutions into a selection of the top-scoring amino acids , G38 , R61 , A127 , S129 , and L130 , focusing on common natural variants at each position ( Supplementary file 1D ) . By expressing LasR in Escherichia coli and measuring its activity against a previously reported panel of 19 AHL signals ( Figure 3—figure supplement 1; Wellington and Greenberg , 2019 ) , we were able to quickly prioritize variants for further study . The majority of our LasR variants retained the ability to respond to AHLs and had an altered selectivity profile when compared to wild type ( Figure 3 ) . For example , LasRG38V responds to N-3-oxo-tetradecanoyl-L-homoserine lactone ( 3OC14-HSL ) and N-3-oxo-hexadecanoyl-L-homoserine lactone ( 3OC16-HSL ) but not 3OC12-HSL , while LasRL130F is more subtly altered and responds more strongly than wild type to N-3-oxo-octanoyl-L-homoserine lactone ( 3OC8-HSL ) and N-3-oxo-decanoyl-L-homoserine lactone ( 3OC10-HSL ) , consistent with a previous study of this amino acid substitution ( McCready et al . , 2019 ) . We , and others , have previously demonstrated that compared to native activity , QS receptor sensitivity and selectivity can be altered when the receptor gene is expressed in E . coli or overexpressed in P . aeruginosa ( Moore et al . , 2015; Wellington and Greenberg , 2019 ) . We therefore engineered several mutations in lasR on the P . aeruginosa PAO-SC4 chromosome to study LasR activity in the native context . P . aeruginosa PAO-SC4 is an AHL synthase-null mutant which we use here to measure LasR activity in response to exogenously provided AHL signals . We measured LasR activity by using a transcriptional reporter in which the promoter of the LasR-regulated gene rsaL controls gfp expression ( Wellington and Greenberg , 2019 ) . This provides a direct measure of LasR activity as a transcriptional activator . The lasR mutations largely had the same effect on activity and selectivity in P . aeruginosa as they did when lasR was expressed in E . coli ( Figure 4—figure supplement 1A ) . Consistent with its role in forming a water-mediated hydrogen bond with the C3 oxygen of 3OC12-HSL , and with previous studies ( Collins et al . , 2006; Gerdt et al . , 2015 ) , LasR R61 mutants were less responsive to oxo-substituted AHLs , but maintained wild type or better levels of activation by unsubstituted AHLs ( Figure 4 and Figure 4—figure supplement 1A ) . On the other hand , LasRA127L had an increased sensitivity to numerous signals ( Figure 4 ) . LasRA127M was also more strongly activated than wild type by multiple signals , as was LasRL130F ( Figure 4—figure supplement 1A ) . Residue A127 interacts with 3OC12-HSL near the middle of its fatty acyl tail . The A127M and A127L substitutions may increase signal binding through increased hydrophobic contacts with the acyl chain . Residue L130 is near the homoserine lactone core of the bound signal . The L130F substitution results in structural changes that potentially increase LasR stability , thereby broadening its selectivity ( McCready et al . , 2019 ) . The amino acid substitutions also affect the sensitivity of LasR to 3OC12-HSL ( Figure 4E and Figure 4—figure supplement 1B ) . Interestingly , two of our variants are more sensitive to 3OC12-HSL than wild-type LasR . LasRA127L is roughly threefold more sensitive and LasRL130F is twofold more sensitive . This increased sensitivity comes at the cost of decreased selectivity for both of these mutant proteins . In fact , many of our single amino acid variants displayed reduced selectivity compared to wild-type LasR ( Figure 3 and Figure 4—figure supplement 1A ) . The observed changes in LasR sensitivity and selectivity could be due to multiple factors including altered receptor affinity for a signal , altered protein stability , or both . The stability of LuxR-type receptors is intertwined with signal binding . QS systems are subject to elaborate control , a key component of which is that QS receptors , including LasR , are unstable and insoluble in the absence of bound signal ( Oinuma and Greenberg , 2011; Sappington et al . , 2011 ) . Thus , changes in signal affinity usually lead to changes in the amount of soluble receptor present in a cell ( McCready et al . , 2019 ) . Conversely , changes to the expression level of the receptor can lead to altered signal sensitivity and selectivity ( Wellington and Greenberg , 2019 ) . To determine whether receptor stability was affected in our variants , we assessed the abundance of soluble LasR in P . aeruginosa by immunoblotting in a selection of variants with altered 3OC12-HSL sensitivity: LasRA127L , LasRL130F , and LasRR61L . Comparing these variants to wild type , there were not substantial changes in the abundance of soluble LasR , though LasRR61L was somewhat less abundant than wild type ( Figure 4—figure supplement 2 ) . Similar to LasR , we focused our LasI amino acid alterations on the top-scoring positions: L102 , T142 , T145 , and L157 ( Supplementary file 1E ) . We expressed wild-type or mutated lasI on a low copy number plasmid in the AHL synthase-null P . aeruginosa PAO-SC4 and extracted AHLs produced by these bacteria from culture fluid . While bioassays are commonly used for the detection of AHLs ( Chu et al . , 2011 ) , they suffer from multiple drawbacks . In particular , bioassays are not equally sensitive to all AHLs and typically cannot be used to determine which AHLs are produced and in what ratio . To screen our LasI variants for altered activity and selectivity , we developed a thin layer chromatography ( TLC ) method based on our existing high-performance liquid chromatography ( HPLC ) radiotracer assay ( Schaefer et al . , 2018 ) . In this method , the C1 position in the homoserine lactone ring is labeled with 14C . The label is incorporated into AHLs at a ratio of one 14C per AHL molecule . This results in unbiased detection of all AHLs produced . While the established method uses HPLC to separate and detect AHLs one sample at a time , we can run nine samples per TLC plate , resulting in a more high-throughput assay . Using our TLC method , we confirmed that lasI directs the synthesis of the same primary product whether it is expressed on a plasmid or from the chromosome ( Figure 5—figure supplement 1A ) . HPLC analysis of matched extracts confirmed that the major LasI product observed by TLC is 3OC12-HSL ( Figure 5—figure supplement 1B ) . As expected , an empty vector control did not produce detectable AHLs , nor did we detect radioactivity in a media-only control . We then screened the activity of each lasI mutant by TLC ( Figure 5—figure supplement 1C–E ) . Several mutants produced little or no detectable AHLs , but others , such as LasIT145S , appeared to produce more 3OC12-HSL than wild-type LasI . These changes in selectivity and rate of synthesis may be due to biochemical changes in LasI or to altered protein expression or stability . We analyzed select extracts by both TLC and HPLC and found that the results were consistent between the two methods , further validating the TLC method . Based on our TLC results , we selected one variant with altered selectivity , LasIL157W , for further study by HPLC . By TLC , LasIL157W produces three signals: 3OC12-HSL and two signals of unknown identity ( Figure 5—figure supplement 1D ) . Using HPLC , we found that LasIL157W produces equal amounts of two 14C-AHLs consistent with 3OC10-HSL and 3OC8-HSL along with a lesser amount of 3OC12-HSL ( Figure 5 ) . Residue L157 is located near the bottom of the acyl-binding pocket where it likely interacts with the end of the 12-carbon acyl chain in 3OC12-HSL . The L157W substitution could decrease the volume of the pocket , improving affinity for shorter substrates . These findings demonstrate that the covarying residues identified by GREMLIN influence LasI activity and selectivity , and that a single amino acid substitution is sufficient to significantly alter LasI selectivity . In general , multiple amino acid changes are required to generate a protein with orthogonal selectivity ( Aakre et al . , 2015; Collins et al . , 2006; Skerker et al . , 2008 ) . In non-QS proteins , altered selectivity has been engineered by swapping the covarying residues in one homolog to the identities in another ( Aakre et al . , 2015; Skerker et al . , 2008 ) . Here , we seek to ‘rewire’ LasI/R to use an orthogonal signal . We targeted the MupI/R system from Pseudomonas fluorescens NCIMB 10586 , which uses the signal 3OC10-HSL ( Hothersall et al . , 2011 ) . MupI and MupR share 52% and 39% identity with LasI and LasR , respectively . MupI/R is among the systems closest in sequence identity to LasI/R that use a signal other than 3OC12-HSL . LasR and MupR differ at eight covariation sites in the ligand-binding domain with a GREMLIN score ( with APC ) > 0 . 08 ( Figure 6—figure supplement 1A ) . LasR modified to contain all eight substitutions was inactive . However , there were several intermediate variants that displayed an increased response to 3OC10-HSL . We identified three amino acid substitutions that are sufficient for this increased sensitivity: L125F , A127M , and L130F ( Figure 6A , Figure 6—figure supplement 1B ) . LasRL125F , A127M , L130F is over 20-fold more sensitive to 3OC10-HSL than wild-type LasR . The L125F substitution appears to be the primary driver of this altered selectivity ( Figure 6B–C and Figure 6—figure supplement 2A ) . LasR L125 is located in the 3OC12-HSL binding pocket , where it interacts with the end of the signal’s acyl tail . The L125F substitution may decrease the size of the binding pocket , improving hydrophobic interactions with shorter acyl chains . All ‘MupR-like’ LasR variants responded to 3OC12-HSL with similar sensitivity to wild-type LasR when expressed in E . coli ( Figure 6—figure supplement 2B–C ) . While 3OC10-HSL is the native signal for the MupI/R system , the sensitivity of MupR to other signals is unknown . To compare the activity of our LasR variants with that of MupR , we developed a transcriptional reporter of MupR activity in E . coli . To do this , we used the promoter of mupI to control gfp expression in the plasmid pPROBE-GT . LuxR-type receptors often positively regulate their paired synthase gene ( Ng and Bassler , 2009 ) . Searching the promoter region upstream mupI , we identified an inverted repeat centered at −64 . 5 relative to the start codon of mupI that has high similarity ( 12/20 base pairs ) to the LasR-binding site upstream the LasR-regulated gene rsaL ( Figure 6—figure supplement 3A; Whiteley and Greenberg , 2001 ) . It is therefore plausible that MupR regulates mupI . We created an arabinose-inducible mupR expression vector in the plasmid pJN105 and introduced it along with pPROBE-PmupI into E . coli . If MupR activates transcription from the mupI promoter , this reporter strain should express gfp in the presence of 3OC10-HSL . We validated the reporter by measuring GFP fluorescence in the presence or absence of arabinose ( to induce mupR expression ) and/or 3OC10-HSL ( Figure 6—figure supplement 3B ) . While there is some leaky expression of mupR from the pJN105 vector , indicated by increased fluorescence in the presence of 3OC10-HSL and absence of arabinose , the reporter strain requires both arabinose and 3OC10-HSL for maximal fluorescence . Using this reporter we found that MupR , when expressed in E . coli , is equally sensitive to 3OC10-HSL and 3OC12-HSL ( Figure 6—figure supplement 3C ) . Thus , our ‘MupR-like’ LasR variant mimics the activity of MupR in this context . To confirm our findings in the native context , we engineered mutations into lasR on the chromosome of the AHL synthase-null P . aeruginosa PAO-SC4 . We found that wild-type LasR has minimal 3OC10-HSL activity , whereas the ‘MupR-like’ LasRL125F , A127M , L130F is much more sensitive to 3OC10-HSL and responds to concentrations as low as 1 µM . As in E . coli , the ‘MupR-like’ variant maintains its 3OC12-HSL activity in P . aeruginosa ( Figure 6—figure supplement 4A–C ) . To determine whether the ‘MupR-like’ LasR variant stimulates social behaviors , we assessed QS-dependent protease production by plating P . aeruginosa on casein agar . In this assay , QS-regulated protease production is required for cell growth , and high levels of protease result in a zone of clearing around the colony and a white zone of partially degraded casein at the periphery of the clearing ( Chen et al . , 2019 ) . AHL synthase-null P . aeruginosa PAO-SC4 grows poorly on casein agar with 3OC10-HSL or with no signal added , but grows well and produces protease in response to 3OC12-HSL . In contrast , P . aeruginosa PAO-SC4 LasRL125F , A127M , L130F grows on casein agar with either 3OC10-HSL or 3OC12-HSL , indicating that 3OC10-HSL stimulates protease production in the ‘MupR-like’ variant ( Figure 6D ) . LasI differs from MupI at five high-scoring covariation residues: LasI M125 , T145 , M152 , V159 , and N181 ( Figure 7—figure supplement 1A ) , the first three of which line the LasI acyl-binding pocket ( Figure 7A , Figure 7—figure supplement 1B ) . Swapping these three residues for their MupI identities resulted in a synthase that has substantially altered selectivity . LasIM125I , T145S , M152L produces about twofold more 3OC10-HSL than 3OC12-HSL . The M125I substitution alone was sufficient to relax LasI’s selectivity , resulting in a synthase that produces roughly equal amounts of 3OC10-HSL and 3OC12-HSL ( Figure 7B , Figure 7—figure supplement 2A–C ) . LasI M125 is located near the C9 carbon of the acyl chain of a substrate modeled into the LasI structure ( Gould et al . , 2006 ) . Changes to this residue could obstruct binding of substrates with longer acyl chains while increasing affinity for shorter substrates . As a comparison , we measured the activity of mupI expressed in P . aeruginosa , and found it produces 9:1 3OC10-HSL:3OC12-HSL ( Figure 7B and Figure 7—figure supplement 2D ) . All single and double ‘MupI-like’ LasI variants retained AHL synthase activity , but only those that contain the M125I substitution displayed increased 3OC10-HSL production relative to 3OC12-HSL ( Figure 7—figure supplement 2E ) . Despite decades of study , it has been challenging to determine how AHL QS systems distinguish between signals . We hypothesized that we could identify covariation patterns in AHL QS systems and that these patterns would illuminate residues important for signal selectivity . Using a novel application of GREMLIN to analyze the sequences of 3489 unique AHL QS systems , we identified amino acids that strongly covary between AHL synthases and receptors . The top-scoring residues in our analysis cluster near the ligand-binding pockets for both proteins and are more than three times closer to the signal molecule compared to top-scoring residues in a randomized control . We focused our study on P . aeruginosa LasI/R . Through targeted alterations in the top-scoring covarying residues , we demonstrate that these amino acids are indeed determinants of signal selectivity . We have thus validated a new application of covariation analysis for proteins that interact indirectly and not through direct binding to one another . This use of covariation analysis for non-physical protein-protein interactions may be useful for other systems in which proteins are connected by small molecules , for example , metabolic pathways . Additionally , these strong covariation results further support the view that AHL synthases and receptors coevolve . For both the synthase , LasI , and the receptor , LasR , a single amino acid substitution is sufficient to significantly alter selectivity . Interestingly , our amino acid substitutions also revealed that LasR is not optimized to be as sensitive as possible to its native 3OC12-HSL signal . The increase in sensitivity of specific variants came at the cost of decreased selectivity , which suggests that QS systems may evolve to balance these two properties . Furthermore , increased sensitivity to the native signal may lead to premature activation of the QS regulon , which would likely decrease fitness ( Darch et al . , 2012 ) . The products and behaviors regulated by QS , such as pyocyanin and protease production , are complexly regulated not only by LasI/R , but also by other factors including additional QS receptors ( Brint and Ohman , 1995; Ochsner et al . , 1994 ) . The variants generated in our study provide us with the tools to assess the impact of LasI/R sensitivity and selectivity on QS timing and gene expression level and , ultimately , on these more complex social phenotypes . We also demonstrated that we can use covarying residues to rationally engineer a QS system to produce and respond to a non-native signal . By changing the covarying residues in LasI/R to their MupI/R identities , we improved the sensitivity of LasR to 3OC10-HSL over 20-fold and increased the production of 3OC10-HSL by LasI roughly 15-fold . For both the synthase and receptor , a single amino acid substitution was the primary driver of the altered selectivity . This was surprising given the low sequence identity between LasI/R and MupI/R . These findings suggest new QS systems might evolve with relative ease . Further , the ability to engineer QS selectivity could be beneficial to synthetic biology where AHL signaling is a powerful tool to build biological circuits ( Davis et al . , 2015 ) . Notably , there has not been a previous attempt to simultaneously engineer both an AHL synthase and receptor to use a non-native signal . Though we were able to substantially increase the 3OC10-HSL activity of LasI/R , our variants retained their native 3OC12-HSL activity . We have thus generated a promiscuous system with broadened selectivity . Similarly , a directed evolution study of the AHL receptor LuxR found that it evolves through promiscuous intermediates ( Collins et al . , 2005 ) . This has also been observed in other systems , such as toxin-antitoxin systems ( Aakre et al . , 2015 ) . Proteins tend to evolve through broadly active intermediates before gaining new specificity . In this way , the system maintains functionality en route to altered selectivity . QS systems appear to follow these same trends . Further work is needed to fully ‘rewire’ LasI/R to exclusively use a non-native signal and to enable the rational engineering of QS systems to use any signal of choice . One limitation we faced is a lack of close LasI/R homologs with known signals . The identification of a more closely related system to LasI/R may provide a better starting point for engineering altered selectivity . It has been demonstrated that ‘supporting’ residues , that is , residues within a protein that covary with the selectivity residues , may indirectly impact selectivity by influencing the orientation of selectivity residues ( Aakre et al . , 2015 ) . Thus , given the large differences in sequence identity between the Las and Mup systems , there are likely other residues that must be changed to fully swap selectivity . Supporting residues may be a tractable starting point . Alternatively , our ‘Mup-like’ variants could serve as the basis for identifying QS proteins with altered selectivity using saturating mutagenesis and/or in vitro evolution . Such studies would illuminate additional determinants of selectivity and potentially uncover ‘rules’ of signal binding . The variants reported in this study serve as a foundation for future work on signal sensitivity and selectivity and the rational engineering of QS systems . Collectively , our results provide insight into AHL QS selectivity and will help us predict signal selectivity in newly identified QS systems , in metagenomes , and in naturally occurring QS variants such as those found in clinical isolates . More broadly , we have gained insight into how AHL QS systems evolve and diversify and have validated a new use of covariation analysis for investigating protein-ligand selectivity in coevolving proteins that are connected by a small molecule . Starting from 24 pairs of manually curated QS synthases and receptors ( Supplementary file 1A ) , we searched for homologs in complete bacterial genomes using BLAST ( e-value <0 . 01 ) ( Altschul et al . , 1990 ) . We filtered the BLAST hits by sequence identity ( >30% ) to the query and the alignment coverage ( query coverage >0 . 75 and hit coverage >0 . 75 ) , and the filtered hits were aligned by Clustal Omega ( Sievers and Higgins , 2021 ) . We selected the LasI/R system from P . aeruginosa PAO1 as the target and mapped the multiple sequence alignments ( MSA ) to the target system . We built sequence profiles from the MSA with HMMER ( Eddy , 2009 ) and hmmbuild for the QS synthases and receptors , respectively . The sequence profiles were then used to search against the ENA database ( Amid et al . , 2019 ) and the IMG/M database ( Chen et al . , 2021 ) from JGI using HMMER hmmsearch . A total of 149 , 837 and 5 , 046 , 620 homologs were found in these databases for the QS synthase and receptor , respectively . Because the synthases and receptors of the known QS systems frequently locate near each other in the genome , we kept synthase-receptor gene pairs that are separated by no more than two other open reading frames in the genome or contig . A total of 14 , 980 synthase-receptor gene pairs were identified and they represent 6360 non-identical QS systems . In another attempt , we carried out the same procedure using the LuxI/R system from V . fischeri MJ11 as the target system . A similar number of QS systems were identified . We connected the synthase and receptor protein sequences for each QS system we found in the databases and derived the alignments between these QS systems to the target QS system ( LasI/R ) from the hmmsearch result . We filtered the MSA for synthase-receptor pairs by sequence identity ( maximal identify for remaining sequences ≤90% ) and gap ratio in each sequence ( maximal gap ratio ≤25% ) . We applied GREMLIN to analyze the covariation in the MSA ( Kamisetty et al . , 2013 ) , and the GREMLIN coefficients were normalized using APC ( Buslje et al . , 2009 ) as we described previously ( Ovchinnikov et al . , 2014 ) . The GREMLIN coefficients after APC were used as measures for covariation signals between synthase and receptor amino acid residues . As a control , we connected each synthase sequence with a randomly selected receptor sequence and performed the covariation analysis in the same way . We mapped the top-scoring covarying residues in the LasI/R system onto the crystal structures for each protein . Reported distances between residues and ligands are the shortest distance between any non-hydrogen atoms . For LasR , distances were calculated using PDB 6V7X . For LasI , distances were calculated using a LasI structure with 3-oxo-C12-acyl-phosphopantetheine modeled into the acyl-binding pocket ( Gould et al . , 2004 ) . Reported distances for LasI are between residues and the acyl portion of the modeled substrate . Bacterial strains and plasmids are listed in the key resources table . Unless otherwise specified , P . aeruginosa and E . coli were grown in lysogeny broth ( LB ) ( 10 g tryptone , 5 g yeast extract , 5 g NaCl per liter ) buffered with 50 mM 3- ( N-morpholino ) propanesulfonic acid ( MOPS ) ( pH 7 ) ( LB/MOPS ) or on LB agar ( LB plus 1 . 5% Bacto agar ) ( Wellington and Greenberg , 2019 ) . Liquid cultures were grown at 37°C with shaking . For radiotracer TLC experiments , P . aeruginosa was grown in Jensen’s medium with 0 . 3% glycerol ( Schaefer et al . , 2018 ) . Casein agar was made using minimal broth plus 1% sodium caseinate ( casein broth ) and 1 . 5% agar as previously reported ( Chen et al . , 2019 ) . For plasmid selection and maintenance , antibiotics were used at the following concentrations: P . aeruginosa , 30 μg/mL gentamicin ( Gm ) and 150 μg/mL carbenicillin ( Cb ) ; E . coli 10 μg/mL Gm and 100 μg/mL ampicillin ( Ap ) . BD Difco Pseudomonas Isolation Agar was prepared according to manufacturer’s directions and supplemented with 100 μg/mL Gm as needed . Where needed for gene expression , L-arabinose ( 0 . 4% w/v ) was added . All chemicals and reagents were obtained from commercial sources . AHLs were dissolved either in dimethyl sulfoxide ( DMSO ) or in ethyl acetate ( EtAc ) acidified with glacial acetic acid ( 0 . 01% v/v ) . AHLs in DMSO were used at ≤1% of the final culture volume and AHLs dissolved in EtAc were dried on the bottom of the culture vessel prior to addition of the bacterial culture . DMSO or acidified EtAc was used as a vehicle control where appropriate . pJN-lasI and pJN-RBSmupI were constructed using E . coli-mediated DNA assembly ( Kostylev et al . , 2015 ) . Briefly , for pJN-lasI , lasI was amplified from P . aeruginosa PAO1 genomic DNA ( gDNA ) using primers lasI-pJN-F and lasI-pJN-R ( Supplementary file 1F ) . pJN105 was amplified using the reverse complement of these primers . The resulting PCR products were treated with the restriction enzyme DpnI to remove the parent template . Both PCR products were then used to transform E . coli ( NEB 5-alpha ) . The resulting constructs were confirmed by Sanger sequencing . For pJN-RBSmupI , we began by amplifying mupI from P . fluorescens Migula ( ATCC 49323 ) gDNA using primers mupI-F and mupI-R . We then used primers mupI-pJN-F and mupI-pJN-R to amplify the mupI PCR product and used the reverse complement of these two primers to amplify pJN-RBSlasI . The resulting PCR products were treated the same as for pJN-lasI . We constructed pJN-RBSlasI using restriction digestion . The lasI gene , including its upstream ribosomal binding site ( RBS ) , was amplified from P . aeruginosa PAO1 gDNA using primers RBS-lasI-F and lasI-pJN-R . pJN-lasI and the RBS-lasI PCR product were digested using NheI and SacI , gel or column purified , respectively , ligated by T4 DNA ligase , and transformed into NEB 5-alpha . The resulting constructs were confirmed by Sanger sequencing . Plasmids were introduced into E . coli by using heat shock and were introduced into P . aeruginosa by electroporation . Point mutations were introduced to lasI and lasR on pJN-lasI and JNL or pEXG2-lasR , respectively , using site directed mutagenesis by PCR . Primers were designed to amplify each plasmid while introducing the desired mutation ( s ) . The resulting PCR products were treated with DpnI and were then used to transform NEB 5-alpha . Plasmids from the resulting colonies were screened for the desired mutations by Sanger sequencing . To mutate lasR on the P . aeruginosa PAO-SC4 chromosome , E . coli S17-1 was used to deliver pEXG2-lasR containing various lasR mutations to PAO-SC4 via conjugation and potential mutants were isolated as previously described ( Kostylev et al . , 2019 ) . All mutations were confirmed by PCR amplification of lasR from the genome followed by Sanger sequencing . pJN105-mupR was created by amplifying mupR from P . fluorescens gDNA using the primers mupR-F and mupR-R , which add homology to pJN105 , including an RBS . pJN-RBSmupI was amplified with the reverse complement of these two primers to generate the vector backbone with homology to mupR . The resulting PCR product was treated with DpnI and then both PCR products were purified using a Monarch PCR and DNA Cleanup Kit ( NEB ) . The two fragments were ligated using Gibson assembly , then transformed into E . coli NEB 5-alpha . pPROBE-PmupI was constructed by amplifying the mupI promoter ( −300 to +42 ) from P . fluorescens gDNA using the primers PmupI-F and PmupI-R . This PCR product was cleaned up with a Monarch kit , then amplified with PmupI-pPR-F and PmupI-pPR-R to add homology to the pPROBE-GT vector . pPROBE-GT was amplified with the reverse complement of these two primers . The vector PCR was treated with DpnI , then both PCR products were purified and used to transform E . coli NEB 5-alpha . The resulting constructs were confirmed by Sanger sequencing . LasR activity was measured in E . coli containing pJNL and pPROBE-PrsaL or in P . aeruginosa PAO-SC4 containing pPROBE-PrsaL using previously reported methods ( Wellington and Greenberg , 2019 ) . Briefly , overnight-grown cultures were diluted 1:100 and grown back to log-phase . For E . coli , cultures were grown to an optical density at 600 nm ( OD ) of 0 . 3 , treated with L-arabinose ( 0 . 4% ) , and incubated with AHLs for 4 hr . MupR activity was measured in E . coli harboring pJN105-mupR and pPROBE-PmupI using this same protocol . For P . aeruginosa , cultures were grown to an OD between 0 . 05 and 0 . 3 , were diluted to an OD of 0 . 01 , and then incubated with AHLs for 16–18 hr . LasR activity was measured as GFP fluorescence ( excitation 490 nm , emission 520 nm , gain 50 ) using a Synergy H1 microplate reader ( Biotek Instruments ) . Activity measurements were normalized by dividing by OD600 and subtracting background values ( fluorescence per OD600 for cultures incubated with vehicle control ) . Half maximal effective concentrations , EC50 , were calculated using GraphPad Prism . The relative levels of soluble LasR in P . aeruginosa PAO-SC4 with unmarked lasR mutations were assessed using published methods ( Schuster and Greenberg , 2007 ) . Briefly , cultures were grown overnight in LB/MOPS , diluted 1:100 in LB/MOPS containing 2 µM 3OC12-HSL , and grown to an OD600 of 2 . Cells were collected by centrifugation at 4°C and suspended in LasR purification buffer ( 25 mM Tris-HCl pH 7 . 8 , 150 mM NaCl , 1 mM ethylenediaminetetraacetic acid , 1 mM dithiothreitol , 0 . 5% Tween-20 , 10% glycerol , 2 µM 3OC12-HSL ) . The cell suspensions were sonicated and the resulting lysates were subjected to ultracentrifugation at 55 , 000 rpm for 30 min at 4°C . Protein concentrations were determined by NanoDrop and normalized samples were separated by SDS-PAGE . The separated proteins were transferred to a PVDF membrane which was treated with polyclonal antibodies against LasR ( Covance; 1:1000 dilution ) . Proteins were detected using a secondary anti-rabbit horseradish peroxidase IgG and chemiluminescent substrate . Cultures of P . aeruginosa PAO1ΔrhlI or of P . aeruginosa PAO-SC4 with pJN-empty or with wild-type or mutated pJN-lasI were grown overnight in Jensen’s medium with 0 . 3% glycerol . Overnight cultures were used to inoculate fresh medium ( 1% v/v ) . When the OD reached 0 . 5 , lasI expression was induced with arabinose ( 0 . 4% ) and 1 . 1 mL cultures were incubated with 1 . 1 μCi/mL L-[1-14C]-methionine ( 14C-methionine , American Radiolabeled Chemicals ) for 90 min ( Schaefer et al . , 2018 ) . Cells were pelleted by centrifugation and 1 mL of supernatant fluid was extracted twice with 2 mL acidified EtAc . The extracts were dried under N2 and resuspended in 15 μL acidified EtAc . 5 μL of each extract was spotted on an aluminum backed C18-W-silica TLC plate ( Sorbtech ) . AHLs were separated using 70% methanol in water , then the TLC plate was dried and exposed to a phosphor screen for at least 16 hr . Phosphor screens were imaged with a Sapphire Biomolecular Imager ( Azure Biosystems ) . To confirm TLC findings , select extracts were dried , suspended in methanol , and analyzed by C18-reverse-phase HPLC using a previously reported method ( Schaefer et al . , 2018 ) . For better detection of AHLs , we slightly modified the radiolabeling protocol detailed above , modeling it after a previously published method ( Leadbetter and Greenberg , 2000 ) . Cultures of P . aeruginosa PAO-SC4 with wild-type or mutated pJN-RBSlasI were grown overnight in LB/MOPS . Overnight cultures were used to inoculate 5 mL LB/MOPS ( 1% v/v ) . After 2 hr , lasI expression was induced with arabinose ( 0 . 4% ) and cultures were grown to OD 0 . 7 . Cells were centrifuged at 5000 rpm for 10 min , and pellets were suspended in 1 . 1 mL phosphate buffered saline with 10 mM glucose . After shaking incubation at 37°C for 10 min , 1 . 1 μCi 14C-methionine was added to the cell suspension . Cell suspensions were incubated with radiolabel for 2 hr , after which cells were pelleted by centrifugation and 1 mL supernatant fluid was extracted twice with 2 mL acidified EtAc . Radiolabeled AHLs were dried under N2 and suspended in methanol . One-third of each extract was analyzed by reverse-phase HPLC using a gradient of 10–100% methanol-in-water ( Schaefer et al . , 2018 ) . Casein agar was used to evaluate protease production as previously described ( Chen et al . , 2019 ) . Cultures were grown in LB/MOPS overnight , diluted to an OD600 of 0 . 1 , and then 3 µL was spotted on casein agar plates ( 60 mm × 15 mm ) containing AHLs or DMSO as a vehicle control . Colony growth and casein proteolysis around the colony were assessed after 48 hr at 37°C .	Communication is vital in any community and it is no different for bacteria . Some of the microbes living in bacterial communities are closely related to one another and can help each other survive and grow . They do this by releasing chemical signals that coordinate their behaviors , including those that are damaging to the infected host . A key aspect of this coordination is knowing how many related bacteria are present in a given environment . In a process known as quorum sensing , the bacteria release a chemical signal which neighboring sibling bacteria detect and respond to . The larger the population of bacteria , the more the signal accumulates . At a certain threshold , the signal activates the genes needed to trigger a coordinated action , such as producing toxins or antibiotics . Many bacteria communicate using acylhomoserine lactone signaling systems , which involve different signals depending on the species of bacteria . But it is unclear how this diversity evolved , and how bacteria can distinguish between signals from related and unrelated bacterial cells . To understand this , Wellington Miranda et al . used computational techniques to analyze how the proteins responsible for acylhomoserine lactone signaling coevolved . The analysis identified specific parts of these proteins that determine which signal will be produced and which will trigger a bacterium into action . Wellington Miranda et al . then used these insights to engineer the bacteria Pseudomonas aeruginosa to produce and respond to a signal that is naturally made by another bacterial species . These computational methods could be used to analyze other proteins that have coevolved but do not physically interact . Within the area of quorum sensing , this approach will help to better understand the costs and benefits of signal selectivity . This may help to predict bacterial interactions and therefore behaviors during infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2021	A covariation analysis reveals elements of selectivity in quorum sensing systems
The precise pattern of motor neuron ( MN ) activation is essential for the execution of motor actions; however , the molecular mechanisms that give rise to specific patterns of MN activity are largely unknown . Phrenic MNs integrate multiple inputs to mediate inspiratory activity during breathing and are constrained to fire in a pattern that drives efficient diaphragm contraction . We show that Hox5 transcription factors shape phrenic MN output by connecting phrenic MNs to inhibitory premotor neurons . Hox5 genes establish phrenic MN organization and dendritic topography through the regulation of phrenic-specific cell adhesion programs . In the absence of Hox5 genes , phrenic MN firing becomes asynchronous and erratic due to loss of phrenic MN inhibition . Strikingly , mice lacking Hox5 genes in MNs exhibit abnormal respiratory behavior throughout their lifetime . Our findings support a model where MN-intrinsic transcriptional programs shape the pattern of motor output by orchestrating distinct aspects of MN connectivity . Breathing is a fundamental motor behavior required for life . In mammals , specialized circuits have evolved to support robust and efficient breathing to cope with changing metabolic demands . While respiratory rhythmogenesis occurs within a small neuronal kernel in the pre-Bötzinger complex , a large contingent of respiratory circuitry is dedicated to transforming this rhythm into precisely patterned motor output for efficient muscle contraction ( Feldman et al . , 2013 ) . Phrenic Motor Column ( PMC ) neurons are a critical node in these circuits , as they integrate multiple descending and local inputs to mediate rhythmic contraction of the diaphragm muscle , which is essential for driving airflow into the lungs during inspiration ( Greer , 2012 ) . The establishment of phrenic motor neuron ( MN ) identity relies on the intersection of transcription factor-based programs acting along the dorsoventral and rostrocaudal axes of the spinal cord during development ( Chaimowicz et al . , 2019; Edmond et al . , 2017; Philippidou et al . , 2012 ) . However , it is not known whether these MN-intrinsic transcriptional programs are required for the generation of patterned respiratory motor output . While most neural circuits undergo significant maturation at postnatal stages , the circuits that underlie breathing not only need to be functional at birth , but are also critically required in utero . Fetal contractions of the diaphragm muscle are essential for lung and diaphragm development and defects in these contractions result in lung hypoplasia and central sleep apneas ( Harding , 1997 ) . Accordingly , inputs that drive the activation of phrenic MNs must be established prior to birth , indicating that hardwired transcriptional programs likely control major aspects of this connectivity . Because inspiratory drive is relatively weak during embryonic stages , coupling of phrenic MNs through gap junctions is critical to maintain early diaphragm contractions ( Greer and Funk , 2005 ) . Therefore , tight clustering of PMC neurons is essential for proper development of the respiratory system , but the molecular mechanisms that control phrenic MN clustering and topography are largely unknown . In addition to establishing robust embryonic diaphragm contractions , PMC clustering may also be important for the selective establishment of presynaptic inputs . Interestingly , unlike the majority of MNs in the spinal cord , phrenic MNs appear to lack monosynaptic sensory input from Ia proprioceptive afferents and extensive inputs from spinal cord interneurons ( Wu et al . , 2017 ) . Instead , the majority of their monosynaptic inputs ( ~70% ) are derived from the rostral Ventral Respiratory Group ( rVRG ) , a group of mostly glutamatergic neurons in the brainstem that transmit the excitatory drive to activate MNs in a rhythmic pattern generated in the pre-Bötzinger complex ( Wu et al . , 2017 ) . Inhibition also contributes to shaping phrenic motor output . Inspiratory-phase inhibitory inputs onto phrenic MNs rapidly modulate MN excitability to act as a gain control for motor output and control motor neuron synchronization within respiratory bursts ( Parkis et al . , 1999 ) . Synchronous inspiratory MN firing on a short time scale is thought to enhance inspiratory muscle activation , thus ensuring robust and efficient breathing ( Bou-Flores and Berger , 2001 ) . Importantly , firing of phrenic MNs in an unpatterned manner increases the risk of phrenic MN adaptation , diaphragm muscle fatigue , and respiratory failure ( Martin-Caraballo et al . , 2000 ) . Despite the critical role of inhibition in patterning phrenic MN activity , the molecular determinants that underlie the establishment of inhibitory inputs onto phrenic MNs have not been identified . Here , we show that Hox5 transcription factors are required for robust and efficient breathing . Hox5 loss renders mice particularly vulnerable to respiratory dysfunction in the first two weeks of life , suggesting that Hox5 mutations may contribute to early life respiratory conditions . We also show that Hox5 proteins establish phrenic MN clustering and topography through the regulation of a network of cell adhesion molecules . We find that a subset of cadherins are specifically expressed in phrenic MNs and that loss of cadherin function through conditional disruption of downstream β/γ-catenin signaling leads to phrenic MN cell body disorganization and dendrite displacement . MN-specific deletion of Hox5 genes results in a selective loss of inhibitory inputs to PMC neurons and a dramatic change in the activation pattern of phrenic MNs . Our results demonstrate that Hox5 transcription factors determine phrenic MN topography and connectivity to generate robust breathing behaviors . We previously showed that two Hox5 paralogs , Hoxa5 and Hoxc5 ( collectively referred to as Hox5 genes ) are required for the early development and survival of phrenic MNs and the innervation of the diaphragm ( Landry-Truchon et al . , 2017; Philippidou et al . , 2012 ) . Mice lacking both Hox5 genes in MNs ( Hoxa5 flox/flox; Hoxc5-/-; Olig2::Cre , referred to as Hox5MNΔ mice ) die at birth due to respiratory defects ( Philippidou et al . , 2012 ) . While the neonatal lethality of Hox5MNΔ mice underscores the critical requirement for Hox5 genes in respiration , it had prevented us from examining the role of Hox5 proteins in respiratory behaviors and functional connectivity at postnatal stages . To examine the role of Hox5 genes in breathing behaviors and phrenic MN output over time we utilized an alternative mouse model . We generated mice in which a single Hox5 gene , Hoxa5 was selectively deleted from MNs by crossing a conditional Hoxa5 allele ( Tabariès et al . , 2007 ) to Olig2::Cre mice ( Hoxa5MN∆ ) ( Dessaud et al . , 2007 ) . Hoxa5MNΔ mice were viable , and we therefore introduced the Hoxa5MNΔ mutant allele into a Hoxc5 heterozygous background . Mice lacking Hoxa5 specifically from MNs and a single copy of Hoxc5 ( Hoxa5 flox/flox; Hoxc5+/-; Olig2::Cre , referred to as Hoxa5MNΔ; c5het mice ) exhibit a 60% reduction in total diaphragm motor innervation ( Figure 1—figure supplement 1a–b ) . Importantly , around 50% of Hoxa5MNΔ; c5het mice survive to adulthood , enabling us to examine how MN-specific Hox5 loss impacts respiration and phrenic MN output . In order to assess breathing in Hoxa5MNΔ; c5het mice , we utilized unrestrained whole body flow-through plethysmography ( Figure 1a ) . We found that adult ( P80 ) Hoxa5MNΔ; c5het mice show a 40% decrease in tidal volume ( the amount of air inhaled during a normal breath ) , accompanied by a compensatory increase in respiratory frequency ( Figure 1b–c , Figure 1—figure supplement 2a–b ) . The increased frequency allows Hoxa5MNΔ; c5het mice to breathe in an equal volume of air per minute ( minute ventilation ) as control animals ( Figure 1c , Figure 1—figure supplement 2c ) , albeit at the cost of expending more energy . We next submitted mice to a hypercapnic challenge ( 5% CO2 , Figure 1d ) . In control ( Hoxa5 flox/flox; Hoxc5+/- ) mice , exposure to 5% CO2 results in an increase in tidal volume and breathing frequency . Hoxa5MNΔ; c5het mice only slightly increase their tidal volume during hypercapnia , resulting in a striking 30% decrease in ventilation and a diminished capacity to respond to respiratory challenges ( Figure 1e–f , Figure 1—figure supplement 2d–f ) . Notably , the ability to overcome hypercapnic and hypoxic conditions is particularly important during perinatal stages , as a compromised response to respiratory insults could increase susceptibility to sudden infant death syndrome ( SIDS ) ( Kinney et al . , 2009 ) . We recorded breathing in unrestrained control and Hoxa5MNΔ; c5het mice at different ages to temporally characterize respiratory dysfunction in the absence of Hox5 genes . We found that tidal volume is consistently decreased in Hoxa5MNΔ; c5het pups as early as P3 ( Figure 1g ) ; however , the compensatory increase in breathing frequency does not develop until approximately 2 weeks after birth ( Figure 1h ) . Thus , perinatal Hoxa5MNΔ; c5het mice exhibit a severe ( 50–60% ) reduction in ventilation , even at rest in normal air , for the first two weeks of life ( Figure 1i ) . Consistent with this , we find that approximately 50% of neonatal Hoxa5MNΔ; c5het mice perish within a week after birth . Altogether , our plethysmography results demonstrate that Hox5 genes are required for the emergence of proper respiratory behavior . Importantly , we have identified a critical temporal window during which Hoxa5MNΔ; c5het mice are especially vulnerable to respiratory dysfunction , similar to perinatal susceptibility to SIDS . What are the molecular underpinnings of respiratory dysfunction in Hoxa5MNΔ; c5het mice ? While the reduction in diaphragm innervation likely contributes to tidal volume changes , we wanted to identify additional fundamental properties of phrenic MNs that could contribute to respiratory circuit formation and function that are altered in Hoxa5MNΔ; c5het mice . Phrenic MNs form a tightly-packed neuronal cluster at cervical levels of the spinal cord ( C3-C5 ) . This clustering is critical for the proper development of the respiratory system because it facilitates recruitment of motor units through electrical coupling in the embryo to compensate for weak inspiratory drive ( Greer and Funk , 2005 ) . In addition , the stereotypical orientation of phrenic dendrites likely enables their targeting by premotor respiratory neurons . However , the mechanisms that control PMC clustering and dendritic topography have not been established . To examine whether phrenic MN organization and dendritic orientation are altered in Hoxa5MNΔ; c5het mice , we injected the lipophilic dye diI into phrenic nerves at e18 . 5 . diI diffuses along the phrenic nerve to label both PMC cell bodies in the spinal cord and axons innervating the diaphragm . To ensure that we labeled the full extent of the PMC we only analyzed mice in which all diaphragm projections were labeled ( Figure 2—figure supplement 1a–b ) . In control mice , retrogradely-labelled PMC neurons are found in close proximity to each other , with no distance in between , however in Hoxa5MNΔ; c5het mice phrenic MNs lose their stereotypical clustering organization ( Figure 2a , Figure 2—figure supplement 1c–d ) . To quantitate the impact of Hox5 loss on PMC organization , we established a clustering index , representing the percentage of PMC neurons in contact with each other ( 1 = 100% , see Materials and methods ) . We found a significant reduction in the clustering index of Hoxa5MNΔ; c5het mice , indicating that Hox5 genes control phrenic MN clustering ( Figure 2b ) . To verify that the MN disorganization observed upon Hox5 deletion was not due to the progressive loss of PMC neurons by e18 . 5 ( Figure 2—figure supplement 1e ) , we assessed PMC clustering at earlier developmental time points . We observed a similar disorganization of phrenic MNs , identified by the expression of the MN-specific transcription factor Isl1/2 and the phrenic-specific transcription factor Scip , in Hoxa5MNΔ; c5het mice at e12 . 5 ( Figure 2—figure supplement 2a–d ) . We also introduced the Hox5MNΔ allele ( Hoxa5 flox/flox; Hoxc5-/-; Olig2::Cre ) into a Bax mutant background to prevent programmed cell death ( Knudson et al . , 1995; Philippidou et al . , 2012 ) . Since Bax deletion circumvents phrenic MN loss in the absence of Hox5 genes , we performed this analysis in Hox5MNΔ mice to define the impact of complete Hox5 loss of function on PMC organization . Deletion of Bax , both in control and in Hox5MNΔ ( Hox5MNΔ; Bax-/- ) mice , dramatically increased the number of PMC neurons . However , phrenic MNs still appeared to lose their tight clustering in Hox5MNΔ; Bax-/- mice , demonstrating that Hox5 genes drive a program that controls PMC organization and clustering independently of phrenic MN survival ( Figure 2—figure supplement 2e–h ) . In addition to the loss of phrenic MN clustering , we also observed a remarkable redistribution of phrenic MN dendrites in Hoxa5MNΔ; c5het mice after diI injection . While in control mice phrenic dendrites project in two major directions , dorsolateral and ventromedial , in Hoxa5MNΔ; c5het mice the most dorsal dendritic projections are lost and dendrites appear defasciculated ( Figure 2c ) . To quantitate the change in dendritic orientation , we established a grid separating the area proximal to phrenic cell bodies in eight squares and measured the percentage of labelled dendrites in each square ( Figure 2c–e , Figure 2—figure supplement 1g–h , see Materials and methods ) . We found a significant decrease in dorsolateral dendrites in Hoxa5MNΔ; c5het mice , as well as an increase in the number of dendrites approaching the midline ( Figure 2c–e ) . While in control mice phrenic dendrites rarely cross the midline , in Hoxa5MNΔ; c5het mice we find that a number of dendrites traverse the midline and continue to grow contralaterally , despite dendrites covering less area overall ( Figure 2f–h , Figure 2—figure supplement 1f ) . Our data indicate that Hox5 genes control phrenic MN dendritic topography , which likely contributes to their presynaptic targeting . In order to understand how Hox5 transcription factors regulate phrenic MN topography and clustering , we analyzed the transcriptome after Hox5 loss using RNA sequencing ( RNA-seq ) , taking advantage of Hb9::GFP mice , which selectively express GFP in MNs ( Wichterle et al . , 2002 ) . Using fluorescence activated cell sorting ( FACS ) , we isolated RNA from GFP-positive MNs from the cervical spinal cord ( C3-C6 , which encompasses the entire PMC ) of control ( Bax-/- ) and Hox5MNΔ; Bax-/- mice at e12 . 5 ( Figure 3a ) . We performed RNA-seq on isolated GFP+ MNs in Bax-/- and Hox5MNΔ; Bax-/- mice to identify gene expression changes independently of phrenic MN cell death that begins at e12 . 5 . Since Bax deletion circumvents phrenic MN loss in the absence of Hox5 genes , we performed this analysis in Hox5MNΔ mice to maximize gene expression changes . Our analysis identified 837 genes that were differentially expressed between Bax-/- and Hox5MNΔ; Bax-/- mice ( p<0 . 05 , Figure 3b–c ) . Gene ontology analysis revealed that the majority of downregulated genes were implicated in processes relevant to nervous system development , including neuron projections and dendritic and synapse development ( Figure 3d ) . Since Hox5 genes control PMC clustering and dendritic orientation , we reasoned that cell adhesion molecules ( CAMs ) might be good candidate Hox5 downstream effectors . Therefore , we performed in situ hybridization to validate whether CAMs identified in our RNA-seq ( Figure 3—figure supplement 1 ) showed Hox5-dependent PMC-specific expression . We found that ALCAM , Edil3 , cdh9 , Ptprt , Lsamp and Negr1 were highly and specifically expressed in the PMC at e12 . 5 ( Figure 4a ) . A subset of these CAMs were previously established as phrenic-specific markers ( ALCAM , Edil3 , cdh9 and Negr1 ) , while our analysis also identified novel PMC genes ( Ptprt and Lsamp ) ( Machado et al . , 2014 ) . We further found that these phrenic-specific CAMs require Hox5 proteins for their expression , as they were downregulated in Hox5MNΔ; Bax-/- mice ( Figure 4a ) . This downregulation was further recapitulated in both Hoxa5MNΔ; c5het mice and Hoxa5MNΔ; c5het; Bax-/- mice ( Figure 4—figure supplement 1a ) , indicating that a single copy of Hoxc5 is insufficient to induce PMC-specific CAM expression . Our results suggest that Hox5 proteins regulate PMC clustering and position through controlling the expression of a network of downstream cell adhesion molecules . Based on our RNA-seq analysis and validation , we identified cdh9 as the PMC-specific CAM that was most downregulated after Hox5 deletion ( Figure 3—figure supplement 1 ) . In the spinal cord , cadherin function is required for the segregation and clustering of limb-innervating MNs into nuclear structures called pools , however the role of cadherins in respiratory motor neurons has not been examined ( Price et al . , 2002 ) . We asked whether cadherins might play a role in PMC organization . First , we wanted to define the full repertoire of cadherin expression in the PMC . We performed in situ hybridization for all type I and type II cadherins and found that cdh2 , 6 , 9 , 10 , 11 and 22 were expressed in the PMC at e13 . 5 , while the rest of the family members were either expressed in other MN subtypes or not found in the spinal cord ( Figure 4—figure supplement 1b–c ) . To further quantify cadherin expression in phrenic MNs , we performed fluorescence in situ hybridization for PMC-enriched cadherins combined with immunofluorescence for the phrenic marker Scip . We found that all PMC cadherins were uniformly expressed in the majority of phrenic MNs ( >90% ) at e13 . 5 ( Figure 4b–c , Figure 4—figure supplement 1d ) . At cervical levels of the spinal cord , cdh9 and 10 expression appears to be restricted to phrenic MNs , while cdh2 , 6 , 11 and 22 are broadly expressed in all MN populations ( Figure 4b–c ) . Our data establish a comprehensive combinatorial cadherin code that uniquely defines PMC neurons . The highly specific PMC cadherin expression pattern suggests that cadherins could have important functions in phrenic MNs . In order to assess the role of classical cadherins in PMC development , we eliminated their function by inactivating β- and γ-catenin specifically in MNs using an Olig2::cre promoter ( β-catenin flox/flox;γ-catenin flox/-;Olig2::cre , referred to as βγ-catMNΔ mice ) ( Figure 5a ) . β- and γ-catenin are obligate intracellular factors required for cadherin-mediated cell adhesive function and are necessary for the organization of limb-innervating motor pools ( Demireva et al . , 2011 ) . The strategy of inactivating β/γ-catenin in MNs circumvents potential redundancy that can arise through the expression of multiple cadherins in the PMC and allows us to establish a cadherin requirement in PMC development before dissecting individual cadherin function . Single β- or γ-catenin mutants exhibited normal phrenic MN numbers , cell body position and clustering ( Figure 5—figure supplement 1b–c ) , indicating that disruption of Wnt signaling through β-catenin inactivation does not affect PMC topography . Joint inactivation of β- and γ-catenin , however , and disruption of cadherin signaling , led to a marked disorganization and loss of phrenic MN clustering ( Figure 5b–c ) . We find that the PMC clustering index is significantly reduced in βγ-catMNΔ mice at e13 . 5 ( Figure 5c ) . We also observed a ~30% reduction in the number of Scip+ MNs that settle in the ventral spinal cord in βγ-catMNΔ mice ( Figure 5—figure supplement 1b ) , partly due to the failure of a subset of PMC neurons to migrate away from the midline ( Figure 5—figure supplement 1d ) . Next , we examined whether cadherins also play a role in PMC dendritic orientation . Since βγ-catMNΔ mice die around e14 . 5-e15 . 5 ( Demireva et al . , 2011 ) , we performed diI phrenic nerve injections at e14 . 5 . Similar to Hoxa5MNΔ; c5het mice , we found a change in dendritic orientation and loss of the dorsal-most dendrites in βγ-catMNΔ mice ( Figure 5d–e , Figure 5—figure supplement 1e ) . Unlike in Hoxa5MNΔ; c5het mice however , phrenic dendrites do not cross the midline in βγ-catMNΔ mice , suggesting that multiple pathways are acting downstream of Hox5 proteins to dictate precise phrenic dendritic orientation . Despite having a striking effect on PMC dendrites , joint inactivation of β- and γ-catenin did not affect phrenic axon growth or guidance , as diaphragm innervation appears normal ( Figure 5—figure supplement 1a ) , indicating that Hox5 genes control phrenic axon and dendrite development through distinct molecular programs . Our results demonstrate that cadherins are required in phrenic MNs downstream of Hox5 genes for proper clustering and dendritic orientation . Our in vivo plethysmography data ( Figure 1 ) provided an overview of how the entire respiratory system , including sensory feedback and possible compensatory changes due to hypoxia and hypercapnia , responds to Hox5 gene deletions . To determine whether the loss of Hox5-dependent transcriptional programs specifically affects the activity of phrenic MNs in response to circuitry intrinsic to the brainstem and spinal cord , we performed suction electrode phrenic nerve recordings from isolated brainstem-spinal cord preparations that exhibit fictive breathing ( Figure 6a ) ( Cregg et al . , 2017 ) . We examined whether loss of Hox5 genes results in changes in phrenic MN activity at P4 , a timepoint at which respiratory bursts are especially rhythmic and robust . Despite the changes in PMC clustering and dendritic topography , no significant changes in phrenic nerve burst frequency , rhythmicity , or duration were observed in Hoxa5MNΔ; c5het mice ( Figure 6b–c ) , indicating that the brainstem circuits that drive inspiratory bursts , located within the pre-Bötzinger complex , are intact and able to transmit excitatory drive to phrenic MNs via the rVRG . Our results indicate that parameters that reflect inspiratory/expiratory balance on a long time scale , such as frequency and burst duration , are largely Hox5-independent . However , respiratory efficiency also relies on MN activity on a shorter time scale ( i . e . 10–100 ms ) , and the precise temporal pattern of phrenic MN firing during inspiratory bursts is necessary for forceful diaphragm contractions ( van Lunteren and Sankey , 2000 ) . We found that MN bursts from control ( Hoxa5 flox/flox; Hoxc5+/- ) mice exhibited a highly reproducible firing pattern , with inspiratory bursts comprised of periods of activity interspersed with periods with no unit activity ( Figure 6d ) . These silent periods progressively lengthen throughout the burst . In addition , periods of activity are dominated by large amplitude rhythmic oscillations that occur at approximately 30 Hz ( Figure 6d ) , generated by the summations of multiple phrenic MN units firing in the same temporal pattern . Interestingly , these 30 Hz oscillations are matched to the fusion frequency of the diaphragm muscle , which is the frequency of firing at which the diaphragm is tonically and maximally contracted ( Martin-Caraballo et al . , 2000 ) . Thus , patterned firing of phrenic MNs close to the fusion frequency promotes highly efficient diaphragm contraction . Remarkably , this pattern in inspiratory bursts is abolished in Hoxa5MNΔ; c5het mice ( Figure 6d ) . Inspiratory bursts exhibit near continuous firing after loss of Hox5 genes , and thus the silent periods within the burst are largely lost ( Figure 6e ) . In addition , the large amplitude rhythmic 30 Hz activity was eliminated , suggesting that the firing of phrenic MNs in Hoxa5MNΔ; c5het mice is no longer constrained to occur at specific times but is instead distributed throughout the burst , thus reducing the compound action potentials seen in control mice . Power spectrum analysis to resolve the recording into its component frequencies indicated a decrease in 30 Hz activity with a concomitant broad spectrum increase in higher frequencies ( Figure 6f–g ) . Importantly , firing of phrenic MNs in an unpatterned manner that does not correlate with the diaphragm fusion frequency provides no additional contractile benefit , and in fact may increase the risk of phrenic MN adaptation , diaphragm muscle fatigue , and respiratory failure ( Martin-Caraballo et al . , 2000 ) . We confirmed these changes were not due to a reduction in phrenic MN number , as blocking apoptosis by introducing the Bax deletion into Hoxa5MNΔ; c5het mice did not rescue the phrenic MN firing pattern ( Figure 6—figure supplement 1a–d ) . In addition , bursts from Hoxa5MNΔ; c5het mice at the time of birth ( P0 ) display a similar change in motor output ( Figure 6—figure supplement 2a–d ) , suggesting that Hox5 genes function during embryonic development to shape phrenic MN activity at birth . Collectively , these data show that Hox5-dependent transcriptional programs are required for shaping the pattern of phrenic MN output and confining firing to frequency oscillations that promote efficient diaphragm contraction while preventing muscle failure . Phrenic MN firing pattern and synchronicity are thought to be generated in part by inhibitory synaptic transmission that modulates firing in response to excitatory drive ( Bou-Flores and Berger , 2001; Marchenko and Rogers , 2009 ) . We therefore sought to determine whether a loss of inhibitory synaptic transmission underlies the changes in firing present in Hoxa5MNΔ; c5het mice . We performed unilateral local microinjections of the GABAA receptor antagonist picrotoxin and the glycine receptor antagonist strychnine into the ventral spinal cord at C3-C6 , the location of the PMC . Injection of picrotoxin and strychnine in control mice resulted in a firing pattern indistinguishable from that seen in Hoxa5MNΔ; c5het mice ( Figure 7a ) . Disinhibited control phrenic MNs fired throughout the burst with reduced periods of no activity ( Figure 7b ) . Power spectra analysis revealed a decrease in 30 Hz activity with a concomitant broad increase in high frequency activity ( Figure 7c–d ) . Picrotoxin and strychnine injections into Hoxa5MNΔ; c5het mice had little effect on phrenic MN firing ( Figure 7a ) . The ability to convert the firing pattern of control mice into one similar to Hoxa5MNΔ; c5het mice by disinhibition locally on the PMC , and the fact that disinhibition had little effect on Hoxa5MNΔ; c5het phrenic MN firing , implies that Hoxa5MNΔ; c5het mice have lost the inhibitory synaptic transmission which is important for generating this pattern . We explored whether we could detect any anatomical alterations in inhibitory synaptic inputs by performing synaptic puncta quantitation . We counted perisomatic inhibitory synapses , as defined by apposition of the presynaptic marker GAD67 and the postsynaptic marker gephyrin , on phrenic MNs in control and Hoxa5MNΔ; c5het mice ( Figure 7e–f ) . Hoxa5MNΔ; c5het mice exhibited a 20% reduction in inhibitory synapse number as compared to control ( Figure 7g ) . Our results show that functional phrenic MN output is altered in the absence of Hox5 genes , likely due to loss of a subset of inhibitory inputs that act to pattern motor output . Together , our data support a model where Hox5-dependent transcriptional programs shape the pattern of respiratory output by establishing inhibitory inputs onto phrenic MNs . The establishment of clustered neuronal nuclei is a conserved and prominent organizational feature of the CNS and is thought to be critical for neural circuit assembly ( Jessell et al . , 2011 ) . Phrenic MN clustering serves an additional function . The generation of fetal breathing movements is required for lung and diaphragm maturation; however , descending inspiratory drive is relatively weak during embryonic development ( Greer , 2012 ) . Electrical coupling between tightly clustered phrenic MNs facilitates the recruitment of multiple motor units to compensate for weak inputs and generate adequate synchronous motor drive to the diaphragm ( Greer and Funk , 2005 ) . While electrical coupling is not observed in mature phrenic MNs , when maximum motor unit recruitment becomes inefficient and is no longer desirable , it is especially beneficial during the development of the respiratory system ( Greer et al . , 1999 ) . Therefore , phrenic MN clustering likely serves a dual function during development , enabling both electrical coupling and premotor targeting . Our results indicate that Hox5 transcription factors regulate a molecular program that defines PMC position and clustering . Our RNA-seq analysis revealed a number of cell adhesion molecules that are downregulated in the absence of Hox5 genes . CAM expression also appears to distinguish phrenic MNs from other MN populations , alluding to MN subtype-specific CAM functions ( Machado et al . , 2014 ) . Comprehensive analysis of cadherin expression identified a combinatorial PMC cadherin code ( cdh2 , 6 , 9 , 10 , 11 and 22 ) . Since our strategy of inactivating β/γ-catenin in MNs eliminated all cadherin signaling , we cannot definitively establish whether the full complement of PMC-specific cadherins is necessary for efficient phrenic MN clustering . Expression of single or a small subset of cadherins may be insufficient to endow PMC neurons with a unique identity for self-recognition , and multiple cadherin expression might be necessary for their segregation from limb-innervating and axial MNs found at the same levels . PMC cadherins belong to all three specificity groups identified recently on the basis of their heterophilic interactions , further restricting the likelihood that PMC neurons will interact with other MN populations through heterophilic interactions between cadherins of the same subgroup ( Brasch et al . , 2018 ) . While the function of the type I cadherin , N-cadherin is required for clustering of all MN populations , additional type II cadherin divergent expression among MN populations may be required for specific MN subtype clustering ( Dewitz et al . , 2019 ) . We also demonstrate that Hox5 genes are required for the correct orientation of PMC dendrites . Phrenic MN dendrites form tightly fasciculated bundles that adopt a distinct ventromedial and dorsolateral orientation during development ( Allan and Greer , 1997 ) . By late embryonic stages ( e18 . 5 ) , phrenic dendrites are restricted to the ipsilateral side and rarely cross the midline . Loss of Hox5 genes leads to a loss of stereotypical dendritic organization and an increased crossing to the contralateral site . Loss of cadherin function leads to similar dendritic reorganization and loss of dorsolateral dendrites , arguing that cadherins are key effectors of Hox5 action . However , cadherin loss does not increase the number of dendrites that cross the midline , pointing to additional mechanisms acting downstream of Hox5 proteins . Notably , while cadherin signaling is critical for dendritic growth and orientation , β/γ-catenin inactivation does not impair phrenic MN axon growth or diaphragm innervation , suggesting that Hox5 genes control two independent molecular programs that regulate axonal and dendritic growth respectively , to coordinate the wiring specificity of phrenic MNs . The requirement for cadherins in establishing both cell body and dendritic topography suggests a prominent cadherin role in PMC presynaptic connectivity . Interestingly , a subpopulation of neurons in the pre-Bötzinger complex also expresses cdh9 , which could indicate a broad role for cadherins is establishing synaptic connectivity throughout respiratory circuits ( Yackle et al . , 2017 ) . How are phrenic MNs specifically targeted by respiratory premotor populations while eschewing inputs from other descending neurons , locomotor-related interneurons , and sensory afferents directed to other nearby MN populations ? MN identity , established by early transcriptional programs , is emerging as a critical determinant of MN connectivity ( Dasen , 2017 ) . Despite eradicating phrenic MN identity through Hox5 deletion , which led to loss of PMC topography and downregulation of PMC-specific CAMs , descending excitatory inputs to PMC neurons appear to be largely unperturbed , as we still observe regular phrenic MN bursting in isolated brainstem-spinal preparations . The persistence of excitatory inputs could indicate that Hox5-dependent precise PMC topography is not necessary for these populations to synapse on phrenic MNs or that multiple redundant mechanisms have evolved to maintain this connection that is extremely critical for life . Loss of Hox5 genes appears to selectively impact the establishment of inhibitory inputs onto phrenic MNs , suggesting distinct requirements for PMC targeting by individual premotor populations . How do Hox5 genes influence PMC connectivity ? In sensory-motor circuits , the correct positioning of MNs appears to be critical for their targeting by sensory axons ( Sürmeli et al . , 2011 ) . In addition , it has recently been reported that spatial features of the MN dendritic tree , such as the angle of interaction with approaching axons , can also act as a determinant of their connectivity with sensory neurons ( Balaskas et al . , 2019 ) . Here , we demonstrate that Hox5 genes determine PMC cell body and dendritic topography through the induction of cadherin expression . Do cadherins solely function to position phrenic MNs and dendrites at the right place during development or do they have additional roles as molecular recognition cues in presynaptic targeting ? In the retina , cadherins control the topography of axonal and dendritic arbors of synaptic partners to place them in close proximity and enable synaptogenesis ( Duan et al . , 2014; Duan et al . , 2018 ) . In the hippocampus however , cadherins influence synaptic fidelity and potentiation without overtly affecting cell morphology , pointing to position-independent roles in synaptic connectivity ( Basu et al . , 2017 ) . Similarly , mutations in transcription factors that alter molecular identity but not cell body position dramatically reconfigure MN inputs , indicating that MN position is unlikely to be the only critical parameter for MN connectivity ( Hinckley et al . , 2015; Machado et al . , 2015 ) . Loss of inhibitory inputs onto PMC neurons likely results from both positional and molecular changes resulting from MN-specific Hox5 deletion . Defining the explicit contribution of topography to phrenic MN connectivity will require decoupling PMC position from Hox5-mediated transcriptional programs that also control the molecular determinants of phrenic identity . The major function of phrenic MNs is to efficiently contract the diaphragm muscle , and as such , MNs could potentially function to merely execute complex computations occurring in upstream brainstem respiratory circuits . Consistent with this idea , recent monosynaptic viral-based retrograde tracing of phrenic MN inputs revealed that the major PMC projection arises from excitatory rVRG neurons that propagate the respiratory rhythm generated by the pre-Bötzinger complex ( Wu et al . , 2017 ) . However , in addition to this excitation , phrenic MNs receive multiple modulatory inputs , including serotonergic and cholinergic inputs , indicating at least some degree of computation transforms rhythmic signals into appropriate motor patterns at the MN level . Phrenic MNs also receive substantial descending inhibitory inputs and we observed an abundance of inhibitory synapses on phrenic MN cell bodies . While we did not observe a complete loss of inhibitory synapses in Hoxa5MNΔ; c5het mice , it is likely that only a subset of phrenic MN inhibition is dedicated to patterning motor neuron activity within inspiratory bursts . Alternatively , while we still observe synaptic puncta on phrenic cell bodies , a number of these synapses may be non-functional , as cadherins are also required for synaptic organization ( Yamagata et al . , 2018 ) . What is the source of this inhibition and how does it contribute to shaping phrenic MN output ? Rabies-virus mediated retrograde tracing revealed a population of PMC-projecting inhibitory neurons within the rVRG ( Wu et al . , 2017 ) . Excitatory and inhibitory rVRG neurons are activated concurrently , such that inhibition is in phase with excitatory inputs generating inspiration ( Parkis et al . , 1999 ) . This inspiratory-phase inhibition synchronizes MN output on a short time scale , and this oscillation frequency is thought to match the frequency that produces maximal muscle force to generate robust diaphragm contractions ( Bou-Flores and Berger , 2001; Parkis et al . , 2003; Sebe et al . , 2006 ) . Our results demonstrate that Hox5 genes are required for establishing these inhibitory inputs onto phrenic MNs , revealing how early transcriptional programs contribute to phrenic MN patterned output . Loss of Hox5 genes results in pronounced defects in breathing behaviors , including reductions in tidal volume and inability to respond to respiratory challenges such as hypercapnia . There are likely multiple contributing factors to this respiratory dysfunction , including the loss of diaphragm innervation , the reduction in phrenic MN numbers and the inefficient activation of the remaining phrenic MNs due to the loss of electrical coupling and inhibitory inputs ( Figure 7—figure supplement 1 ) . Differences in tidal volume are somewhat mitigated with age , reflecting the decrease in motor unit recruitment during quiet breathing with the maturation of the respiratory system ( Greer et al . , 1999 ) . At birth , the vast majority of phrenic MNs are recruited at rest , making the impact of phrenic MN loss on tidal volume more pronounced . As motor units mature , recruitment of a small subset of phrenic MNs is sufficient to generate efficient diaphragm contractions , partly compensating for phrenic MN loss . However , upon a respiratory challenge such as hypercapnia , no additional motor units are available to be recruited in Hoxa5MNΔ; c5het mice , leading to severe decreases in minute ventilation and compromising the hypercapnia response . In Hoxa5MNΔ; c5het mice there is a gradual compensation for tidal volume reduction by increasing breathing frequency at rest . However , this compensation does not occur until 2 weeks of age , likely reflecting the maturation of central chemosensory regions that provide respiratory feedback ( Putnam et al . , 2005 ) . This indicates that Hox5 mutations render mice particularly vulnerable to respiratory failure during the first 2 weeks of life , and consistent with this we observe increased perinatal lethality in Hoxa5MNΔ; c5het mice , reminiscent of SIDS . While much attention has been focused on identifying deleterious genetic variants that impair CO2-sensing populations , such as serotonergic neurons , as causal to SIDS , another possibility is that gene variants causing defects in phrenic MN connectivity and function may be an alternative risk factor for neonatal lethality ( Kinney et al . , 2009; Rand et al . , 2013; Van Norstrand and Ackerman , 2010 ) . As GWAS studies are becoming increasingly common , they may in the future reveal that mutations in Hox5 transcription factors , or downstream cell adhesion molecules , also lead to respiratory dysfunction in humans and are a genetic risk factor for SIDS . The loxP-flanked Hoxa5 ( Tabariès et al . , 2007 ) , β-catenin ( Brault et al . , 2001 ) , and γ-catenin ( Demireva et al . , 2011 ) alleles , Hoxc5 mutant strains ( McIntyre et al . , 2007 ) , Olig2::cre ( Dessaud et al . , 2007 ) , Hb9::GFP ( Wichterle et al . , 2002 ) , γ-catenin-/- ( Ruiz et al . , 1996 ) , and Bax-/- ( Knudson et al . , 1995 ) lines were generated as previously described and maintained on a mixed background . Mouse colony maintenance and handling was performed in compliance with protocols approved by the Institutional Animal Care Use Committee of Case Western Reserve University . Mice were housed in a 12 hr light/dark cycle in cages containing no more than five animals at a time . In situ hybridization and immunohistochemistry were performed as described ( Philippidou et al . , 2012 ) on tissue fixed for 2 hr in 4% PFA and cryosectioned at 16 μm ( 12 μm for synaptic puncta quantitation ) . In situ probes were generated from e12 . 5 cervical spinal cord cDNA libraries using PCR primers with a T7 RNA polymerase promoter sequence at the 5’ end of the reverse primer . All probes generated were 600–1000 bp in length . The sequences used for the PCR primers , probe length , and additional BLAST results verifying specificity of the cadherin probes is located in the attached Supplementary file 1 . Wholemounts of diaphragm muscles were stained as described ( Philippidou et al . , 2012 ) . Diaphragm staining was performed using either neurofilament/synaptophysin primary antibodies ( for mice without Hb9::GFP ) or with GFP primary antibodies ( for mice with Hb9::GFP ) . The following antibodies were used: goat anti-Scip ( 1:5000; Santa Cruz Biotechnology , RRID:AB_2268536 ) , mouse anti-islet1/2 ( 1:1000 , DSHB , RRID:AB_2314683 ) ( Tsuchida et al . , 1994 ) , rabbit anti-neurofilament ( 1:1000; Synaptic Systems , RRID:AB_887743 ) , rabbit anti-synaptophysin ( 1:250 , Thermo Fisher , RRID:AB_10983675 ) , rabbit anti-GFP ( 1:1000 , Invitrogen , RRID:AB_221570 ) , α-bungarotoxin , Alexa Fluor 555 conjugate ( 1:1000; Invitrogen , RRID:AB_2617152 ) , goat anti-ChAT ( 1:200 , Millipore , RRID:AB_2079751 ) , mouse anti-GAD67 ( 1:500 , Millipore , RRID:AB_2278725 ) , and mouse anti-gephyrin ( 1:3000 , Synaptic Systems , RRID:AB_2232546 ) . Images were obtained with a Zeiss LSM 800 confocal microscope and analyzed with Zen Blue and ImageJ ( Fiji ) . Diaphragm innervation was quantified using the simple neurite tracer plugin in ImageJ . For synaptic puncta quantitation performed at P10 , phrenic motor neurons were identified by determining the proper rostrocaudal level using surrounding Hoxa5 and Hoxc8 expression . High resolution individual synaptic puncta were imaged using Zeiss Airyscan . For labeling of phrenic motor neurons , crystals of carbocyanine dye , DiI ( Invitrogen , #D3911 ) were pressed onto the phrenic nerves of eviscerated embryos , and the embryos were incubated in 4% paraformaldehyde at 37°C in the dark for 2 weeks for e14 . 5 embryos and 4–5 weeks for e18 . 5 embryos . Subsequently , spinal cords were dissected , embedded in 4% low melting point agarose ( Invitrogen ) and sectioned using a Leica VT1000S vibratome at 100 to 150 μm . To calculate the clustering index for PMC neurons , we utilized two complementary approaches . For experiments with membrane staining ( diI tracing ) , the number of retrogradely-traced neurons that were in contact with at least one other labelled neuron was counted and divided by the total number of traced neurons to calculate a clustering index . A clustering index of 1 indicates that all retrogradely-traced MNs were clustered . For experiments with nuclear protein staining , we connected all PMC nuclei ( Scip+Isl1/2+ ) to their nearest neighbor to form a perimeter of the PMC . We then counted the number of phrenic MNs and the area occupied by the PMC . Clustering index was defined as the number of PMC neurons/1000 μm2 . For the analysis of dendritic orientation in Hoxa5MNΔ; c5het mice , we superimposed a grid over phrenic MN cell bodies spanning 200 μm in each direction . We then used Fiji ( ImageJ ) to calculate the fluorescent intensity in each square and divided this by the sum of the total fluorescent intensity to calculate the percentage of dendrites in each area . For βγ-catMNΔ mice , since MN cell bodies were more dispersed , we only analyzed dorsal dendrites by calculating the fluorescent intensity of dendrites projecting dorsal to cell bodies divided by the total intensity . C3-C6 cervical spinal cords were dissected from 3 control ( Bax-/- ) and 3 Hox5MNΔ; Bax-/- mutant embryos in a Hb9::GFP background at e12 . 5 and motor neurons were sorted by fluorescence–activated cell sorting ( FACS ) on a Sony iCyt cell sorter . RNA was extracted using the PicoPure RNA isolation system ( Arcturus , #KIT0204 ) with RIN >8 via Tapestation analysis ( Agilent ) . rRNA depleted libraries were prepared from 10 to 20 ng of total RNA using the KAPA stranded RNA-seq kit with Riboerase ( KAPA , #KK8483 ) and amplified by 15 cycles of PCR . Paired-end , 150 bp sequencing was performed on the Illumina HiSeq 2500 and generated a total of 58–94 million reads in each direction per sample after filtering . Read quality was assessed using FASTQC ( https://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) and adapters were trimmed using Trim Galore ( https://www . bioinformatics . babraham . ac . uk/projects/trim_galore/ ) . Filtered and trimmed reads were aligned to the mouse genome ( GRCm38 . p5 ) using TopHat v2 . 1 . 0 ( Kim et al . , 2013 ) . Gene counts were obtained using htseq-count ( Anders et al . , 2015 ) . Differentially expressed genes were considered significant if p<0 . 05 due to the dilution factor from the lack of genetic tools available to specifically sort PMC neurons . Plots were created in R ( v 3 . 5 ) . The R package pheatmap was used to generate the hierarchical clustered heatmap using row-scaled values of differentially expressed genes with p<0 . 05 . Function enrichment was performed using the R package gProfileR with FDR < 0 . 05 . The resulting GO terms were simplified based on similarity using REViGO ( Supek et al . , 2011 ) . Electrophysiology was performed as previously described ( Cregg et al . , 2017 ) . Mice were cryoanesthetized and rapid dissection was carried out in 22–25°C oxygenated Ringer’s solution . The solution was composed of 128 mM NaCl , 4 mM KCl , 21 mM NaHCO3 , 0 . 5 mM NaH2PO4 , 2 mM CaCl2 , 1 mM MgCl2 , and 30 mM D-glucose and was equilibrated by bubbling in 95% O2/5% CO2 . The hindbrain and spinal cord were exposed by ventral laminectomy , and phrenic nerves exposed and dissected free of connective tissue . A transection at the pontomedullary boundary rostral to the anterior inferior cerebellar artery was used to initiate fictive inspiration . Electrophysiology was performed under continuous perfusion of oxygenated Ringer’s solution in a rostral to caudal direction to prevent drug diffusion to the brainstem during local injection . Suction electrodes were attached to phrenic nerves just proximal to their arrival at the diaphragm . For local injections , we used the following drugs: picrotoxin ( PTX ) ( GABAA receptor antagonist , 1 . 25 mM , Tocris Bioscience , #1128 ) and strychnine hydrochloride ( Strych ) ( glycine receptor antagonist , 1 . 25 mM , Sigma , #S8753 ) dissolved in Ringer’s solution with trypan blue for visualization . The signal was band-pass filtered from 10 Hz to 3 kHz using Grass amplifiers , amplified 5 , 000-fold , and sampled at a rate of 50 kHz with a Digidata 1440A ( Molecular Devices ) . Data were recorded using AxoScope software ( Molecular Devices ) and analyzed in Spike2 ( Cambridge Electronic Design ) . Burst duration , percent of burst time with no activity , and power spectra were computed from five bursts per mouse , while burst frequency was determined from 5 min of recording time per mouse . Traces analyzed for percent of burst time with no activity and power spectral analysis were of similar amplitude . In injection experiments , bursts were analyzed 8–10 min after injection . For power spectra , relative power is defined as the absolute power for that frequency bin divided by the sum of the absolute power over all frequency bins from 10 Hz – 400 Hz . Control mice for electrophysiology experiments were all Hoxa5 flox/flox; Hoxc5+/- . Conscious , unrestrained adult ( P80 ) mice were placed in a whole body , flow through plethysmograph chamber ( emka ) attached to a differential pressure transducer ( emka ) . The chamber was filled with normal air ( 79% nitrogen , 21% oxygen ) , and flow was maintained at 0 . 75 L/min per chamber for all gas mixtures . Mice were acclimated for 1 hr in normal air , and then a 5% CO2 mixture ( 74% nitrogen , 21% oxygen , 5% carbon dioxide ) was introduced to the chamber for 15 min , after which the mice were removed . Experiments were performed in pairs , with each pair consisting of one littermate control and one experimental mouse of the same sex and approximate weight . Thirty seconds of resting breaths were analyzed using iox2 software ( emka ) near the end of the acclimation period in normal air , and another thirty seconds of resting breaths were analyzed 10 to 15 min after the introduction of 5% CO2 . Mice were directly observed to identify resting breaths . Each mouse was recorded on three consecutive days and the values were averaged together to reduce variability . Data are presented as fold control , where the control is the matched littermate in normal air . For neonatal plethysmography , we modified syringes to use as chambers , as smaller chambers increase signal detection . Littermate pups were recorded in normal air every 3–4 days from P3 to P17 , and 30–50 breaths were analyzed . Control mice for plethysmography experiments were all Hoxa5 flox/flox; Hoxc5+/- . For all experiments a minimum of three embryos per genotype , both male and female , were used for all reported results unless otherwise stated . Genotypes for control mice in Hox5 experiments include cre negative Hoxa5 flox/flox; Hoxc5+/- and Hoxa5 flox/flox; Hoxc5+/+ mice . Genotypes for control mice in catenin experiments include cre negative β-catenin flox/flox; γ-catenin flox/- and β-catenin flox/flox; γ-catenin flox/+ mice . Data are presented as box and whisker plots with each dot representing data from one mouse unless otherwise stated . Small open squares in box and whisker plots represent the mean . P-values were calculated using unpaired , two-tailed Student’s t test . p<0 . 05 was considered to be statistically significant , where p<0 . 05 , p<0 . 01 , p<0 . 001 , and **p<0 . 0001 . The RNA-seq data reported in this paper is uploaded to the NCBI Gene Expression Omnibus under accession number GSE138085 .	In mammals , air is moved in and out of the lungs by a sheet of muscle called the diaphragm . When this muscle contracts air gets drawn into the lungs and as the muscle relaxes this pushes air back out . Movement of the diaphragm is controlled by a group of nerve cells called motor neurons which are part of the phrenic motor column ( or PMC for short ) that sits within the spinal cord . The neurons within this column work together with nerve cells in the brain to coordinate the speed and duration of each breath . For the lungs to develop normally , the neurons that control how the diaphragm contracts need to start working before birth . During development , motor neurons in the PMC cluster together and connect with other nerve cells involved in breathing . But , despite their essential role , it is not yet clear how neurons in the PMC develop and join up with other nerve cells . Now , Vagnozzi et al . show that a set of genes which make the transcription factor Hox5 control the position and organization of motor neurons in the PMC . Transcription factors work as genetic switches , turning sets of genes on and off . Vagnozzi et al . showed that removing the Hox5 transcription factors from motor neurons in the PMC changed their activity and disordered their connections with other breathing-related nerve cells . Hox5 transcription factors regulate the production of proteins called cadherins which join together neighboring cells . Therefore , motor neurons lacking Hox5 were unable to make enough cadherins to securely stick together and connect with other nerve cells . Further experiments showed that removing the genes that code for Hox5 caused mice to have breathing difficulties in the first two weeks after birth . Although half of these mutant mice were eventually able to breathe normally , the other half died within a week . These breathing defects are reminiscent of the symptoms observed in sudden infant death syndrome ( also known as SIDS ) . Abnormalities in breathing occur in many other diseases , including sleep apnea , muscular dystrophy and amyotrophic lateral sclerosis ( ALS ) . A better understanding of how the connections between nerve cells involved in breathing are formed , and the role of Hox5 and cadherins , could lead to improved treatment options for these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2020	Phrenic-specific transcriptional programs shape respiratory motor output
Reproductive synchrony generally occurs in various group-living animals . However , the underlying mechanisms remain largely unexplored . The migratory locust , Locusta migratoria , a worldwide agricultural pest species , displays synchronous maturation and oviposition when forms huge swarm . The reproductive synchrony among group members is critical for the maintenance of locust swarms and population density of next generation . Here , we showed that gregarious female locusts displayed more synchronous sexual maturation and oviposition than solitarious females and olfactory deficiency mutants . Only the presence of gregarious male adults can stimulate sexual maturation synchrony of female adults . Of the volatiles emitted abundantly by gregarious male adults , the aggregation pheromone , 4-vinylanisole , was identified to play key role in inducing female sexual maturation synchrony . This maturation-accelerating effect of 4-vinylanisole disappeared in the females of Or35-/- lines , the mutants of 4-vinylanisole receptor . Interestingly , 4-vinylanisole displayed a time window action by which mainly accelerates oocyte maturation of young females aged at middle developmental stages ( 3–4 days post adult eclosion ) . We further revealed that juvenile hormone/vitellogenin pathway mediated female sexual maturation triggered by 4-vinylanisole . Our results highlight a ‘catch-up’ strategy by which gregarious females synchronize their oocyte maturation and oviposition by time-dependent endocrinal response to 4-vinylanisole , and provide insight into reproductive synchrony induced by olfactory signal released by heterosexual conspecifics in a given group . Reproductive synchrony , characterized by a pronounced temporal clustering of births , estrus , or mating , widely occurs in the animal kingdom , especially in group-living species ( Ims , 1990 ) . Several prominent cases are best known for their extreme manifestations , for example , sea turtle oviposition , firefly flashing , and fish spawning , involving a mass of individuals with the same reproductive state at certain time windows ( Buck and Buck , 1968; Harrison et al . , 1984; Kelly and Sork , 2002 ) . Reproductive synchrony may offer adaptive advantages for group-living species , such as predation swamping and inbreeding avoidance ( Janzen , 1971 ) . Therefore , understanding how reproductive cycle is synchronized among individuals would provide insight to the biological flexibility in group-living animals . Reproductive synchrony is a complex process that requires the integration of extra- and endo-signals to coordinate the timing of reproductive cycles between individuals in a group ( Kobayashi et al . , 2002; Dey et al . , 2015 ) . In fact , intra-group variation in developmental status can be induced by many factors , including different nutrition , temperature , and order of eclosion ( Ward and Webster , 2016 ) , which essentially makes synchronous reproduction between all members an apparent improbability . Social interaction is considerably critical for triggering reproductive synchrony of individuals in group-living species ( French and Stribley , 1985; Ims and Steen , 1990; Jovani and Grimm , 2008 ) . A well-known example is the Whitten effect which is induced by the presence of males in rodents , ewes , and monkey ( Vandenbergh , 1967; Cahill et al . , 1974; Gattermann et al . , 2002 ) . Various kinds of signals , odors , touch , or voice can act as social clues to underpin synchronization with reproduction ( Rekwot et al . , 2001; Kobayashi et al . , 2002; Noguera and Velando , 2019 ) . Endo-signals , such as hormone release , gene expression , and epigenetic modification , have also been suggested to be involved in these interaction processes ( Engel et al . , 2016; Noguera and Velando , 2019 ) . However , the mechanisms by which social cue/hormone interaction synchronizes the reproductive cycles of individuals within local breeding groups remain largely unknown . Locusts often form large swarms with hundreds to thousands of individuals , regarded as one of the most extraordinary examples of coordinated behavior ( Ariel and Ayali , 2015; Buhl and Rogers , 2016 ) . Depending on population density , locusts display striking phenotypic plasticity , with a cryptic solitarious phase and an active gregarious phase ( Wang and Kang , 2014 ) . Gregarious locusts , compared to solitarious conspecifics , show much higher synchrony in physiological and behavioral events , such as egg hatching and sexual maturation , as well as synchronous feeding and marching behaviors ( Norris , 1954; Uvarov , 1977 ) . Reproductive synchrony in gregarious locusts provides benefits for individuals in several aspects , such as more favorable microenvironment , lower risk of predation , efficiently forging , as well as more encounters with mates , therefore ensures high-density conditions for the next generation , and is essential for maintenance of locust swarm ( Beekman et al . , 2008 , Maeno et al . , 2021 ) . Some sort of vibratory stimulus , maternal microRNAs , and SNARE protein play important roles in the egg-hatching synchrony of gregarious locusts ( Chen et al . , 2015; He et al . , 2016; Nishide and Tanaka , 2016 ) . It has been revealed that the presence of mature male adults has effectively accelerating effects on synchrony of sexual maturation of immature male and female conspecifics in two locust species , Schistocerca gregaria and Locusta migratoria ( Norris , 1952; Loher , 1997; Guo and Xia , 1964; Norris and Richards , 1964 ) . The accelerating effects of several prominent volatiles released by gregarious mature males in male maturation have been examined in the desert locust . Four volatile pheromones ( benzaldehyde , veratrole , phenylacetonitrile [PAN] , and 4-vinylveratrole ) have significantly stimulatory effects on sexual maturation of male adults , with PAN having the most pronounced effect ( Mahamat et al . , 1993; Assad et al . , 1997 ) . However , how conspecific interactions affect female sexual maturation remain unclear and the pheromones those contribute to maturation synchrony of females have not been determined so far . In this study , we investigated mechanisms underlying sexual maturation synchrony of female adults in the migratory locust by comparing phase-related maturation patterns of females using multi-disciplinary studies , including physiology , chemical ecology , genomics , and gene manipulation . Unexpectedly , we found that aggregation pheromone , 4-vinylanisole , induced sexual maturation synchrony of female adult locusts . Our results highlight a parsimonious role of olfactory cues in the formation of locust swarms by triggering aggregation behavior and sexual maturation synchrony . We first investigated whether there was a difference of reproduction synchrony between gregarious and solitarious female locusts by determining the distribution of first oviposition date . The curve of the first oviposition time of gregarious females was much narrower than that of solitarious females ( 60% decrease in the standard deviation [SD] , Figure 1A ) , implying that the first reproductive cycle was more consistent among gregarious female individuals . As an essential premise of egg laying , sexual maturation states , indicated by the length of terminal oocyte relative to the final mature size , were then measured . The development of terminal oocyte increased with ages of female adults of both phases . Gregarious female adults displayed more uniform and rapid patterns than that of solitarious females after 4 days post adult eclosion ( PAE 4 days ) ( Figure 1B and Figure 1—figure supplement 1 ) . These results indicate that gregarious female adults display significant sexual maturation synchrony and higher maturation rate . We next investigated whether conspecific interactions can induce sexual maturation synchrony of female adults ( Figure 1—figure supplement 2A and B ) . We found that the maturation synchrony of terminal oocytes of females was significantly retarded by the removal of male adults in gregarious phase ( Figure 1C ) . However , raised with either solitarious female or male did not affect sexual maturation of solitarious females ( Figure 1D ) . The exposure of odor blends from gregarious male adults significantly advanced the maturation synchrony of gregarious females and solitarious females , whereas no effects were observed when exposed to the background air , female odors , or odors from solitarious males ( Figure 1E and F and Figure 1—figure supplement 2C , D ) . To further explore the roles of olfactory cues in females’ sexual maturation process , we examined the performance of loss-of-function mutants of olfactory receptor co-receptor gene ( Orco-/- ) established by CRISPR/Cas9 , which display significant olfactory deficiency ( Li et al . , 2016 ) . The best-fit normal curve of the first oviposition date was much wider in Orco-/- females than in wild-type ( WT ) females , when they were reared together with gregarious males ( with 63% increase in the SD , Figure 1G ) . When reared together with gregarious male adults or exposure to their odor blends , the sexual maturation of Orco-/- females was less synchronous than that of WT females ( Figure 1H , I and Figure 1—figure supplement 2E , F ) . Thus , olfactory signals from gregarious male adults are essential for triggering the synchrony of female sexual maturation in migratory locusts . To identify key active compounds that can promote sexual maturation synchrony of female locusts , we compared the volatile emission dynamics of gregarious male adults , gregarious female adults , and solitarious male adults from PAE 1 to 8 days . In total , 14 chemicals were identified in the volatiles released by male adults ( Figure 2A ) . After PAE 4 days , only five compounds displayed considerably higher abundance in gregarious male adults , compared to that released by gregarious female adults and solitarious male adults , which showed no accelerating effects on female sexual maturation synchrony . Thus , these five kinds of gregarious male adult-abundant volatiles , including PAN , guaicol , 4-vinylanisole ( 4-VA ) , vertrole , and anisole ( Figure 2A and Figure 2—figure supplement 1 ) , might serve as candidate pheromones for female maturation acceleration . We exposed gregarious young female locusts for 6 days after fledging to different synthetic blends of those five compounds ( PAN , guaicol , 4-VA , vertrole , and anisole ) . The full blend of five components was effective in promoting the synchrony of oocyte development . Only the omission of 4-VA , but not other four compounds , from the full blend lost the accelerating effects on sexual maturation synchrony of gregarious females ( Figure 2B ) . Moreover , the exposure to 4-VA can induce similar effects on female sexual maturation synchrony to the full blend ( Figure 2C ) . In addition , the accelerating effects of 4-VA on maturation synchrony displayed a dose-threshold pattern , with an effective concentration more than 100 ng ( Figure 2D and Figure 2—figure supplement 2 ) . We further examined the performance of Or35-/- females that cannot sense 4-VA ( Guo et al . , 2020 ) . Compared to WT females , the best-fit normal curve of the first oviposition date of Or35-/- females was much wider ( 52% increase in the SD , Figure 2E ) . The sexual maturation in Or35-/- females was more uneven than that of WT females when they were reared together with gregarious males ( Figure 2F ) or exposed to the odors of gregarious males ( Figure 2—figure supplement 3 ) . Moreover , the synchronous effects of 4-VA completely disappeared in Or35-/- females ( Figure 2G ) . In fact , intra-group variation generally exists in the maturation period of female locusts due to differences in nymph experience , nutrition , and fledging time ( Uvarov , 1977 ) . We hypothesized that there is a differential effect of 4-VA on the maturation rate for female individuals with distinct developmental statuses to achieve maturation synchrony . To test this , we determined the accelerating effects of 4-VA on young females at three different ages after fledging: PAE 1–2 days , 3–4 days , and 5–6 days , respectively . We found that female maturation synchrony was significantly enhanced only when young females were treated by 4-VA at PAE 3–4 days , while did not change at PAE 1–2 days or PAE 5–6 days , indicating the time window of 4-VA action on sexual maturation of females ( Figure 3A ) . Moreover , we compared the effects of gregarious males with different ages on female maturation . The maturation synchrony of females was significantly enhanced by gregarious males aged at PAE 3–4 days and PAE 5–6 days ( Figure 3—figure supplement 1 ) , which could release more 4-VA ( Figure 2—figure supplement 1 ) . By contrast , rearing together with the fifth instar of gregarious males and male adults aged at PAE 1–2 days did not significantly affect the maturation synchrony of gregarious female adults ( Figure 3—figure supplement 1 ) . Gene expression profiles in fat body tissue have been demonstrated to correlate tightly with the sexual maturation of female locusts ( Guo et al . , 2014 ) . Therefore , we further evaluated the time window effects of 4-VA on female sexual maturation at the transcriptomic level . Through RNA-seq , we verified that the gene expression profiles of fat body displayed more remarkable changes in female adults exposed to 4-VA at PAE 3–4 days ( 1700 differentially expressed genes [DEGs] ) than those at PAE 1–2 days ( 505 DEGs ) and at PAE 5–6 days ( 582 DEGs ) ( Figure 3—figure supplement 2 ) . Meanwhile , Kyoto encyclopedia of genes and genomes enrichment analysis showed that there were more signal pathways affected by 4-VA treatment at PAE 3–4 days than PAE 1–2 days and PAE 5–6 days . Notably , genes related to energy metabolism , such as retinol metabolism , glycerolipid metabolism , pyruvate metabolism , as well as fatty acid biosynthesis , which play essential roles in ovary development , were significantly activated by 4-VA treatment at PAE 3–4 days ( Figure 3—figure supplement 2 ) . Thus , PAE 3–4 days should be a critical time window for 4-VA-induced acceleration of female sexual maturation . To explore the regulatory mechanism underlying the time window effects of 4-VA on the sexual maturation synchrony of female locusts , we examined the performance of major signaling pathways involved in the sexual maturation of female locusts . First , we determined whether females display time-dependent electrophysiological response to 4-VA by performing electroantennography ( EAG ) and single sensilla response ( SSR ) experiments . We found that 4-VA-induced EAG and SSRs of female adults displayed obvious dose-dependent effects ( Figure 3B and Figure 3—figure supplement 3 ) . However , there was no difference of EAG and SSRs of females among the ages of PAE 2 days , PAE 4 days , and PAE 6 days , although LmigOr35 expression levels were dynamic during ovary development ( Figure 3B and Figure 3—figure supplements 3 and 4 ) . These results suggested that peripheral olfactory perception may not be involved in the time window effects of 4-VA . We then compared gene expression profiles of two main neuroendocrinal tissues , the brain and corpus cardiacum-corpora allatum ( CC-CA ) complex , between the controls and 4-VA-exposed females at PAE 3–4 days . Notably , gene expression profiles in CC-CA significantly changed upon 4-VA treatment , with 290 DEGs , much more than that in the brain ( 89 DEGs ) ( Figure 3C and Figure 3—figure supplement 5 ) , implying the molecular and physiological activities in CC-CA might be remarkably affected by 4-VA stimuli . Moreover , a series of DEGs of CC-CA involved in juvenile hormone ( JH ) metabolism were enriched . There was significantly higher expression of genes related to JH synthesis but lower expression of genes associated with JH degradation ( Figure 3C and D and Figure 3—figure supplement 6 and Supplementary file 1 ) . These results indicated a potential role of JH signaling pathway in mediating the effects of 4-VA at PAE 3–4 days . We therefore tested whether 4-VA exposure can affect the hemolymph JH titer in immature females . As expected , the JH titer was significantly elicited by 4-VA exposure in females aged at PAE 3–4 days , rather than PAE 1–2 days or PAE 5–6 days ( Figure 3E ) . Similarly , the expression levels of vitellogenin ( Vg ) , a key downstream component of JH signaling triggering ovary development in locusts ( Song et al . , 2014 ) , were prominently increased in fat body and ovary of females aged at PAE 3–4 days upon 4-VA stimuli ( Figure 3F–H ) . By comparison , JH titer did not significantly change in Or35-/- females exposed to 4-VA , contrast to over twofold increase in WT females ( Figure 4A ) . Similar patterns were observed for the expression levels of Vg in fat body ( Figure 4B and Figure 4—figure supplement 1A ) and ovary ( Figure 4C ) . To verify the critical roles of the JH/Vg pathway in mediating the effect of 4-VA , we further carried out rescue experiments by the injection of JH analog ( methoprene ) in Or35-/- females . Methoprene-injected Or35-/- females displayed more uniform sexual maturation ( Figure 4D ) . Meanwhile , the expression levels of Vg in fat body and ovary significantly increased in methoprene-injected Or35-/- females ( Figure 4E and F , and Figure 4—figure supplement 1B ) . Results of rescue experiments on WT females indicated that inhibition of JH synthesis by destroying CA using precocene I blocked the 4-VA-accelarated female sexual maturation and Vg expression , which could be recovered by JH III application after precocene treatment ( Figure 4G–I ) . These results provide clear evidence that the JH/Vg signaling pathway can mediate the time-dependent accelerating effects of 4-VA on sexual maturation synchrony in female locusts . Our current study demonstrates conclusively that aggregation pheromone , 4-VA , acts to promote female maturation synchrony in locusts . The pheromone is abundantly released from gregarious male adults and speeds up oocyte development of females aged at PAE 3–4 days through activating JH synthesis and vitellogenesis ( Figure 5 ) . Our findings highlight a ‘catch-up’ strategy of reproductive synchrony by a time window effect combined with extra- and endo-signals in group-living animals . Reproduction synchrony involves consistence in maturation , mating , and egg laying , among which sexual maturation synchrony serves as the most foundational step for oviposition uniformity ( Hassanali et al . , 2005 ) . Extremely high energy cost for female reproduction could restrict migration to pre- , post- , or inter-oviposition period in locusts , thus have crucial effects on collective movement of local populations ( Min et al . , 2004 ) . Given this , a balance of sexual maturation timing among female members presents an essential subject for maintenance of locust swarms . We here demonstrated that young female adults reared with older gregarious male adults show faster and more synchronous sexual maturation in the migratory locust , supporting the accelerate role of crowding in sexual maturation of females ( Guo and Xia , 1964; Norris and Richards , 1964 ) . Together with the accelerating effects on immature male sexual maturation induced by older gregarious male adults reported previously ( Torto et al . , 1994; Mahamat et al . , 2011 ) , young adults of both sexes lived in gregarious conditions prefer more synchronous maturation than individuals reared in solitary . The consistent maturation in both sexes will greatly reduce intra- and inter-sexes competitions for mate selection and thus ensures reproductive synchronous in whole locust populations . We demonstrated that a single minor component ( 4-VA ) of the volatiles abundantly released by gregarious male adults is sufficient to induce the maturation synchrony of female adults . By comparison , four volatiles ( benzaldehyde , veratrole , PAN , and 4-vinylveratrole ) showed stimulatory effects on male maturation ( Mahamat et al . , 2011 ) . Thus , there might exist a sex-dependent action modes of maturation-accelerating pheromones: multi-component pheromones for males and single active component for females , possibly due to different selective pressures between two sexes in response to social interaction . Further exploration will be performed to confirm this hypothesis by determining whether 4-VA has maturation-accelerating effects on male adults in the migratory locust in future . We prefer that 4-VA acts as a critical multi-functional pheromone for the formation of large locust swarms . Earlier , we have demonstrated that 4-VA is mainly released by gregarious nymphs and male adults ( Wei et al . , 2017 ) and can induce strong attraction behavior of both gregarious and solitarious phases ( Guo et al . , 2020 ) , indicating its releaser pheromone role in in keeping locust individuals living together . Meanwhile , the present study shows that 4-VA , acting as a primer pheromone , promotes the maturation synchrony of young female adults , which might facilitate simultaneous oviposition and egg hatching to reduce the predation risk of an individual via the dilution effect ( Ward and Webster , 2016 ) . Thus , a dual role of 4-VA , including both primer and releaser pheromones , could be proposed in triggering the formation of locust swarms . The maintenance and coordination of locust swarming require elaborate communication mechanisms behind the interaction among individuals ( Pener and Simpson , 2009; Wang and Kang , 2014 ) . It is likely an effective and optimized strategy of group-living animals to use a single chemical pheromone to elicit both behavioral and endocrine responses in conspecifics ( Rekwot et al . , 2001 ) . The action of 4-VA displays a remarkable context-dependent manner , such as phase- , sex- , dose- , and time-dependent , reflecting physiological adaption of locusts to the highly dynamic nature of population density . A dose-dependent manner was found for the maturation synchrony effect of 4-VA . We find that only gregarious males aged after PAE 3 days have the accelerating effects on female maturation synchrony , which may be attributed to their significantly increased 4-VA content during adult development . Although gregarious nymphs ( the fifth instar ) and female adults can release relatively small amount of 4-VA ( Wei et al . , 2017 ) , they did not promote female maturation based on our current results . Thus , the accelerating effects on female maturation synchrony induced by gregarious male adults may depend largely on 4-VA content they released . The ineffectiveness of the fifth nymphs and females in maturation acceleration of female adults may due to their low 4-VA content under efficient threshold . In fact , the fifth nymphs have been shown to display inhibiting effects on male maturation in S . gregaria ( Assad et al . , 1997 ) . Therefore , the mechanisms underlying pheromone-mediated sexual maturation may differ between different locust species . Recently , 4-VA has been identified in the volatiles released by male adults of S . gregaria ( Torto et al . , 2021 ) , whether this volatile has maturation-accelerating effect in this locust species needs further validation . Our results reveal that JH signaling pathway presents as the critical endocrinal factor mediating the accelerating effect of 4-VA on female maturation . This finding is consistent with the role of JH as the major gonadotropin modulating Vg biosynthesis in the fat body and its uptake by the growing oocytes in the migratory locust ( Jindra et al . , 2013; Guo et al . , 2014; Song et al . , 2014 ) . It is also supported by the significance of CA ( a major JH biosynthesis tissue ) in pheromone-induced maturation process in the desert locust ( Odhiambo , 2009 ) . Interesting , it has been suggested that the release of the maturation-accelerating pheromone by adult males is under the control of CA ( Loher , 1997 ) . Thus , there should be a complex feedback interaction between 4-VA and JH signaling pathway . Extensive studies have established the central roles of JH signaling in mediating the effects of social interactions on reproduction in different kinds of insect species , including eusocial insects ( Robinson and Vargo , 1997; Korb , 2015 ) , the burying beetle , Nicrophorus vespilloides ( Engel et al . , 2016 ) , the German cockroach , Blattella germanica ( Uzsák and Schal , 2012 ) , and so on . Such an interaction between social clues and internal hormonal signals that coordinates ovary development is also common among group-living vertebrates ( Drickamer , 1977; McClintock , 1978; Berger , 1992 ) . We demonstrate that 4-VA stimulates sexual maturation of young females within a distinct developmental time window . Compared to the females aged at PAE 1–2 days and 5–6 days , the females aged at PAE 3–4 days were more sensitive to 4-VA stimuli . This point was strongly supported by several lines of evidence from temporal-dependent comparisons of oocyte development , gene expression profiles , JH titer , as well as Vg biosynthesis . It has been shown that JH titers , Vg expression , the size of terminal oocytes , dramatically increased at PAE 3–4 days , implies the PAE 3–4 days is an essential time window for JH-regulated ovary development in female locusts ( Luo et al . , 2017; Wu et al . , 2018 ) . The finding that 4-VA accelerates maturation of less-developed females rather than more-developed females supports a ‘catch-up’ model in achievement of female maturation synchrony in locusts . We find that the release of 4-VA by gregarious males continuously increased after adult eclosion , with maximal 4-VA release at PAE 8 days . The age of maximal 4-VA production outwardly seems to be unmatched with the sensitive developmental stage to 4-VA of females ( PAE 3–4 days ) . In insects , it is very common for males to mature earlier than females ( Alonzo , 2013 ) . In the locust , male adults also display earlier sexual maturation for several days , compared to females . In given locust population , individuals successively emerge to adults in a couple of days . Therefore , age-dependent increase in 4-VA release in gregarious male adults presents a persistent stimulus for less-developed young female adults , and thus maximizes maturation synchrony of female locusts , which could reduce male competitions for mate selection . Peripheral and central neural sensitivity to olfactory clues have been demonstrated to vary with developmental stages or physiological statuses ( Guo et al . , 2011; Gadenne et al . , 2016 ) . Given this , sensory processing sensitivity or JH biosynthesis activity might be involved in the stage-dependent sensitivity of females to 4-VA stimuli . However , peripheral olfactory neuron might not be involved in the stage-specific sensitivity to 4-VA stimuli , because we did not detect significant changes of peripheral electrophysiological response during female ovary development . A possible explanation is that signaling factors responsible for JH synthesis might be turned on specifically at Mid-PAE of female locusts upon 4-VA stimuli , such as GPCRs and transcription factors ( Bendena et al . , 2020 ) . Although there are only a few DEGs in the brain of females exposed to 4-VA , we cannot exclude the involvement of the central nerve system pathway by other regulatory mechanisms , for example , neurotransmitter release , or post-transcription regulation ( Nouzova et al . , 2018 ) . Further studies should elucidate detailed mechanisms of the linking between 4-VA and JH biosynthesis in female locusts . In summary , we revealed a catch-up strategy of female reproductive synchrony in locust swarms , whereby 4-VA acts as a maturation-accelerating pheromone hastening less-developed females through triggering JH biosynthesis . Our findings provide novel insight into the mechanisms underlying individual interaction during aggregation in group-living animals . All insects used in experiments were reared in the same locust colonies at the Institute of Zoology , Chinese Academy of Sciences , Beijing , China . Briefly , gregarious locusts were reared in cages ( 30 cm × 30 cm × 30 cm ) with 800–1000 first-instar insects per cage in a well-ventilated room . Solitarious locusts were individually raised in a ventilated cage ( 10 cm × 10 cm × 25 cm ) . All locusts were cultured under the following conditions: a L14:D10 photoperiod , temperature of 30°C ± 2°C , relative humidity of 60% ± 5% , and a diet of fresh greenhouse-grown seedlings and bran . All insects used for sexual maturity determination were virgin females . The ovary of tested females was dissected and placed in locust saline , and the terminal oocytes were isolated . The lengths of terminal oocytes were photographed and measured under the Leica DFC490 stereomicroscope ( Leica , Germany ) . Given that the maximum length of terminal oocytes in gregarious females is much longer than that in solitarious females ( Chen et al . , 2015 ) , the sexual maturity was presented as the length of terminal oocyte relative to the maximum length . Individuals between 0 and 24 hr after adult molting are referred as PAE 1 day adults , with each subsequent day representing an additional 24 hr period . Given that both gregarious and solitarious locusts begin to mate at PAE 6 days , the first oviposition time was recorded after PAE 6 days . For gregarious locusts , 10 females and 10 males at PAE 6 days were placed in a cage ( 30 cm × 30 cm × 30 cm ) . The females were individually marked , and their first oviposition times were recorded by collecting egg pods every 4 hr per day after mating . Females those laid new eggs could be easily distinguished by much thinner abdomen with white foam around ovipositor . For solitarious locusts , each female was reared together with a single male at PAE 6 days , and the first oviposition time was recorded by collecting egg pods every day after mating . To ensure the consistency of mating age in gregarious and solitarious locusts , females that did not successfully mate within 24 hr after paired rearing were excluded in both phases . The distribution curve of the first oviposition time was calculated based on data collected from all females . Given that the difference of female sexual maturation synchrony between gregarious and solitarious phases appeared at PAE 6 days , the lengths of terminal oocytes of virgin females were detected at PAE 6 days after each treatment in subsequent experiments . For the stimulation of gregarious females , 10 gregarious females were reared with 10 gregarious males or 10 gregarious females in a same cage ( 15 cm × 15 cm × 10 cm ) from PAE 1 to 6 days ( Figure 1—figure supplement 2A ) . For the stimulation of solitarious females , one solitarious female was reared with one solitarious male or one solitarious female in a same cage ( 15 cm × 15 cm × 10 cm ) from PAE 1 to 6 days ( Figure 1—figure supplement 2B ) . The ovaries of treated females were dissected in locust saline and the lengths of terminal oocytes were measured as described above . To determine the effect of locust volatiles on female sexual maturation , virgin female adults were separately reared with females or male adults by a breathable partition . For gregarious phase , 10 gregarious females were reared with 10 gregarious males or 10 gregarious females in a breathable partition cage ( 15 cm × 15 cm × 10 cm ) from PAE 1 to 6 days ( Figure 1—figure supplement 2C ) . For solitarious phase , one solitarious female was reared with 10 gregarious males or 10 solitarious males in a breathable partition cage ( 15 cm × 15 cm × 10 cm ) from PAE 1 to 6 days ( Figure 1—figure supplement 2D ) . The ovaries of treated females were dissected in locust saline and the lengths of terminal oocytes were measured as described above . The volatiles of gregarious male adults , gregarious female adults , and solitarious male adults at PAE 1 , 2 , 4 , 6 , 8 days were collected by solid phase microextraction ( SPME ) for 30 min following our previously study ( Wei et al . , 2017 ) . In detail , a fiber ( PDMS/DVB 65 μm ) was introduced into a glass jar ( 10 . 5 cm high ×8 . 5 cm internal diameter ) to absorb odors . The SPME volatiles collected from an empty glass jar for 30 min served as the control . Eight biological replicates were performed for each treatment . The fibers with absorbed odors were subjected to chemical analyses with GC-MS/MS . A Bruker GC system ( 456-GC ) coupled with a triple quadrupole ( TQ ) mass spectrometer ( Scion TQ MS/MS , Inc , German ) equipped with an DB-1MS column ( 30 m × 0 . 25 mm ID ×0 . 25 μm film thickness , Agilent Technologies ) was used to quantify the volatile compounds in the SPME samples . Bruker chemical analysis MS workstation ( MS Data Review , Data Process , version 8 . 0 ) was used to analyze and process the data . Mixed samples consisting of standard compounds in different dosages ( 0 . 1 , 1 , 5 , 10 , and 20 ng/μl ) were used as external standards to develop the standard curves to quantify the volatiles . The same thermal program and Multiple Reaction Monitoring ( MRM ) method were used for standard compound detection . For mixture treatment , 10 gregarious virgin females were stimulated by the mixed odor blend ( the concentrations of PAN , guaiacol , 4-VA , vertrole , and anisole were 1 , 10 , 3 , 2 , 3 μg/μl , respectively ) or paraffin oil from PAE 1 to 6 days . In detail , a breathable vial containing the mixture or paraffin oil was placed with 10 virgin females in a cage for 6 days . The vial was replaced by newly diluted compounds every day . The 4-VA treatment assay was performed by the same method , and the dose of 4-VA used is 100 ng/μl , and the concentration of 4-VA released was measured as 3 . 1–40 ng/0 . 5 hr within 24 hr exposure . To determine the time-dependent effect of 4-VA , control females were treated by paraffin oil from PAE 1 to 6 days . In parallel , paraffin oil was placed by 4-VA at PAE 1–2 , 3–4 , 5–6 days , respectively . The ovaries were dissected and the lengths of terminal oocytes were measured as described above . The brains , CC-CA , and fat body of females were dissected and stored immediately in liquid nitrogen for further experiments . An aliquot of odor was dissolved in paraffin oil ( w/v ) and loaded with 10 μl on a 5 × 40 mm filter paper strip ( Whatman ) , which was placed inside a Pasteur pipette . This odor was used on subsequent EAG assay . Hexane was tested as negative controls . The antennae of the adult locusts were cut at the bases of the flagella and distal antennal . Segments were cut off 2 mm and then fixed between two electrodes with electrode gel Spectra 360 ( Parker , Orange , NJ ) . The EAG signals were amplified , monitored , and analyzed with the EAG-Pro software ( IDAC4 , Syntech , the Netherlands; EAG software v2 . 6c ) . A continuous air flow of 30 ml/s was produced by a stimulus controller ( Syntech CS-05 ) . Stimulation duration was 1 s and the intervals were 1 min . The blank was applied at the start and end of the stimulation series . The average EAG amplitude was subtracted from that of the blank . SSRs were recorded and analyzed , and stimuli were prepared as previously described ( Li et al . , 2016 ) . The locust was placed in a plastic tube 1 cm in diameter , and its head and antennae were fixed with dental wax . A tungsten wire electrode was electrolytically sharpened by 10% NaNO2 . The recording electrode was inserted into the bottom of the sensilla through a micromanipulator ( Narishige , Japan ) . The reference electrode was inserted into the eye . Recording electrodes were connected to amplifiers ( IDAC4 , Syntech , the Netherlands ) . The frequency variation of each pulse at 0 . 2 s was calculated by using automatic frequency meter software . Signals were recorded for 10 s , starting 1 s before stimulation . The preparation is held in a humidified continuous air flow delivered by the Syntech Stimulus controller ( CS-55 model , Syntech ) at 1 . 4 l/min . Chemical substances as SSR stimulants included mineral oil as the blank , which was used to dilute the 4-VA at 1 , 10 , 100 , 1000 , 10 , 000 , 100 , 000 , 1 , 000 , 000 ng/μl , respectively . A piece of filter paper ( Whatman , UK ) was placed in a 15 cm Pasteur glass tube and 10 μl of volatile solution was added to the filter paper . Responses were calculated by counting the number of action potentials 1 s after stimulation . Total RNA from different tissues were extracted using the TRIzol reagent ( Invitrogen TRIzol Reagent , Cat . 15596026 ) and treated with DNase I following the manufacturer’s instructions . For RNA-seq , three independent replicates were performed for each sample . The RNA-seq data reported here have been deposited in the Genome Sequence Archive ( Genomics , Proteomics & Bioinformatics 2017 ) in National Genomics Data Center ( Nucleic Acids Res 2020 ) , Beijing Institute of Genomics ( China National Center for Bioinformation ) , Chinese Academy of Sciences , under accession number CRA003038 that are publicly accessible at https://bigd . big . ac . cn/gsa . cDNA libraries were prepared according to Illumina’s protocols . Raw data were filtered , corrected , and mapped to locust genome sequence using TopHat2 software . The number of total reads was normalized by multiple normalization factors . Transcript levels were calculated using the reads per kb million mapped reads criteria . The differences between the test and control groups were represented by p values . DEGs were detected by using edgeR package with significance levels at p < 0 . 05 . Principal component analysis ( PCA ) was accomplished using the princomp and pca functions . Enrichment analysis of the Gene Ontology ( GO ) was carried out based on an algorithm presented by GOstat . For qPCR , cDNA was reverse transcribed with 2 μg of total RNA using M-MLV Reverse Transcriptase ( Promega , Madison , WI ) . The relative mRNA levels of targeting genes were quantified by Real Master Mix Kit ( Tiangen ) with LightCycler 480 instrument ( Roche ) . Melting curve analysis was performed to confirm the specificity of amplification . The primers used for qPCR were presented in Supplementary file 2 . Ovaries and fat body of tested females were collected and homogenized in TRIzol reagent ( 5 individuals/sample , 6 biological repeats/treatment ) . Total protein was extracted following manufacturer’s instructions . Total protein ( 100 μg ) were separated by gel electrophoresis and then transferred onto polyvinylidene difluoride membranes ( Millipore ) . Non-specific binding sites on the membranes were blocked with 5% bovine serum albumin . The blots were incubated with the primary antibodies ( rabbit anti-Vg serum , 1:500 , Beijing Protein Innovation Co . , Ltd . , BPI ) in TBST overnight at 4°C . After incubation , the membranes were washed , incubated with anti-rabbit IgG secondary antibody ( 1:5000 ) ( EASYBIO , China ) for 1 hr at room temperature , and then washed again . Protein bands were detected by chemiluminescence ( ECL kit , CoWin ) . The antibodies were stripped from the blots , re-blocked , and then probed with an anti-GAPDH antibody ( 1:5000 ) ( Wang et al . , 2013 ) . Protein bands were detected by chemiluminescence ( ECL kit , Thermo Scientific ) . The intensities of the Western blot signals were quantified using densitometry . Twenty microliters of hemolymph were added to a 1 . 5 ml tube with 100 μl of 70% methanol and thoroughly mixed . Then , 200 μl of hexane was added to the solution and thoroughly mixed again . The mixture was centrifuged at 5000 rcf for 10 min at 4°C . Then , 150 μl of supernatant was placed into a new tube , and the JH precipitate was dried by nitrogen . The JH precipitate was dissolved in 50% methanol , mixed by vortexing , and centrifuged at 13 , 000 rpm for 10 min at 4°C . JH III in the supernatant was detected using the rapid resolution liquid chromatography system ( ACQUITY UPLC I-Class , Waters , Milford , MA ) . An ACQUITY UPLC BEH C18 column ( 50 × 2 . 1 mm , 1 . 7 μm ) was used for LC separation . The autosampler was set at 10°C , using gradient elution with 0 . 1% formic acid methanol as solvent A and 0 . 1% formic acid water as solvent B . The flow rate was set at 0 . 2 ml/min . Mass spectrometry detection was performed on an AB SCIEX Triple Quad 4500 ( Applied Biosystems , Foster City , CA ) with an electrospray ionization source ( Turbo Ionspray ) . The detection was performed in positive electrospray ionization mode . The [M + H] of the analyte was selected as the precursor ion . The quantitation mode was MRM mode using the mass transitions ( precursor ions/product ions ) . The MRM ( m/z ) of JH III was 267 . 2/235 . 2 . Data acquisition and processing were performed using AB SCIEX Analyst 1 . 6 Software ( Applied Biosystems ) . For rescue experiment in Or35 mutants , the active JH analog , S- ( + ) -methoprene ( Santa Cruz Biotech Dallas , TX ) was topically applied to the pronotum of locusts ( 150 µg per locust ) from PAE 3 to 4 days according to previously published work ( Song et al . , 2013; Wu et al . , 2016 ) , and acetone was used as the control . Meanwhile , the treated females were stimulated with 4-VA . The treated females were dissected at PAE 6 day , and the lengths of terminal oocytes were measured as previously described . The ovaries and fat bodies of females at PAE 4 days were dissected and stored immediately in liquid nitrogen for further Western blot experiments . For rescue experiment in WT females , precocene I ( Sigma ) dissolved in acetone ( 100 μg/μl ) was added to the dorsal neck membrane of locusts ( 500 μg/locust ) within 12 hr after eclosion to inactive the corpora allata . JH III dissolved in acetone ( 20 μg/μl ) was topically applied at 100 μg per locust aged at PAE 3 day to restore the JH activity . All females were treated by 4-VA at PAE 3–4 days . The treated females were dissected at PAE 6 day , and the lengths of terminal oocytes were measured , the mRNA level and protein level of Vg were detected to validate the effect of JH in 4-VA-accelerated sexual maturation and vitellogenesis . For the measurement of oviposition time and sexual maturation , individuals were randomly allocated into experimental group and control group , and no restricted randomization was applied . The data that do not meet normal distribution was excluded for the analysis of sexual maturity , mRNA levels , protein levels , as well as JH titer measurement . The distribution of the first oviposition time and the consistency of sexual maturation ( represented by the length of terminal oocytes ) were analyzed using Levene’s test according to previous studies ( Rohner et al . , 2013; He et al . , 2016 ) . The mean value of the first oviposition time between two groups was analyzed using Student’s t-test . One-way ANOVA followed by Tukey’s post hoc test was used for multi-group comparisons . All data were statistically analyzed using GraphPad Prism 5 software and SPSS 17 software . All experiments were performed with at least three independent biological replicates .	Since 2019 , a plague of flying insects known as migratory locusts has been causing extensive damage to crops in East Africa . Migratory locusts sometimes live a solitary lifestyle but , if environmental conditions allow , they form large groups containing millions of individuals known as swarms that are responsible for causing locust plagues . Locusts are able to maintain such large swarms because they can aggregate and synchronize . When they live in swarms , individual locusts produce odors that are sensed by other individuals in the group . For example , an aggregation pheromone , called 4-vinylanisole , is known to help keep large groups of locusts together . However , it is less clear how odors synchronize the reproductive cycles of the females in a swarm so that they are ready to mate with males and lay their eggs at the same time . To address this question , Chen et al . examined when female locusts reached sexual maturity after they were exposed to odors produced by other locusts living alone or in groups . The experiments found that only 4-vinylanisole , which was abundantly released by adult male locusts living in groups , stimulated female locusts to reach sexual maturity at the same time . This odor increased the levels of a hormone known as juvenile hormone in less-developed females to help them reach sexual maturity sooner . These findings demonstrate that when migratory locusts are living in swarms , male locusts promote the female locusts to reach sexual maturity at the same time by promoting less-developed females to ‘catch up’ with other females in the group . A next step will be to investigate the neural and molecular mechanisms underlying the ‘catch up’ effect induced by 4-vinylanisole .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "biochemistry", "and", "chemical", "biology" ]	2022	Aggregation pheromone 4-vinylanisole promotes the synchrony of sexual maturation in female locusts
Paired fins are a defining feature of the jawed vertebrate body plan , but their evolutionary origin remains unresolved . Gegenbaur proposed that paired fins evolved as gill arch serial homologues , but this hypothesis is now widely discounted , owing largely to the presumed distinct embryonic origins of these structures from mesoderm and neural crest , respectively . Here , we use cell lineage tracing to test the embryonic origin of the pharyngeal and paired fin skeleton in the skate ( Leucoraja erinacea ) . We find that while the jaw and hyoid arch skeleton derive from neural crest , and the pectoral fin skeleton from mesoderm , the gill arches are of dual origin , receiving contributions from both germ layers . We propose that gill arches and paired fins are serially homologous as derivatives of a continuous , dual-origin mesenchyme with common skeletogenic competence , and that this serial homology accounts for their parallel anatomical organization and shared responses to axial patterning signals . It was classically proposed that the paired fins of jawed vertebrates evolved by transformation of a gill arch – a theory stemming largely from Gegenbaur's ( Gegenbaur , 1878 ) interpretation of a shared anatomical ground plan between the gill arch and pectoral fin skeletons of cartilaginous fishes ( sharks , skates and rays ) ( reviewed by Coates , 1994; Coates , 2003 ) . In vertebrate embryos , the jaw , hyoid and gill arch skeleton ( or , in amniotes , their derivatives , the jaw , auditory ossicles and laryngeal skeleton ) arises from a series of transient , bilaterally paired pharyngeal arches that form on the sides of the embryonic head ( Gillis et al . , 2012a; Graham et al . , 2019 ) , while the paired fins or limbs of jawed vertebrates arise as buds that project from the embryonic trunk ( Tickle , 2015 ) . Cell lineage tracing studies in bony vertebrates ( Chai et al . , 2000; Jiang et al . , 2002; Couly et al . , 1993; Kague et al . , 2012 ) have revealed that the pharyngeal arch skeleton derives largely from the neural crest – a vertebrate-specific , multipotent cell population that undergoes epithelial-to-mesenchymal transition from the dorsal neural tube , and that gives rise to a plethora of derivatives , including skeletal and connective tissue lineages ( Green et al . , 2015 ) . The skeleton of paired appendages , on the other hand , derives from the lateral plate – a distinct mesodermal subpopulation that arose along the chordate stem ( Prummel et al . , 2019; Prummel et al . , 2020 ) . As shared embryonic origin has classically been regarded as a key criterion for serial homology ( discussed by Hall , 1995 ) , Gegenbaur’s gill arch hypothesis of paired fin origin was widely discounted , and is now generally deemed fundamentally flawed ( Coates , 2003 ) . Importantly , though , the distinct embryonic origins of the gill arch and paired fin skeletons may not hold true: mesodermal contributions to the posterior pharyngeal skeleton have been demonstrated in tetrapods , but are much less widely appreciated than those from neural crest . Cell lineage tracing using quail-chick chimaeras and viral labelling have revealed that the avian cricoid and arytenoid laryngeal cartilages derive from lateral mesoderm , and not neural crest ( Noden , 1986; Noden , 1988; Evans and Noden , 2006 ) – a finding that has since been corroborated by genetic lineage tracing experiments in mouse ( Tabler et al . , 2017; Adachi et al . , 2020 ) . Additionally , ablation ( Stone , 1926 ) and lineage tracing experiments ( Davidian and Malashichev , 2013; Sefton et al . , 2015 ) have revealed a mesodermal origin of the posterior basibranchial cartilage in axolotl . Currently , however , there are no mesodermal fate maps of the pharyngeal skeleton of fishes , and so it remains to be determined whether mesodermal contributions to the posterior pharyngeal skeleton are an ancestral feature of jawed vertebrates , and whether mesoderm is competent to give rise to gill arch cartilages – that is , the ancestral skeletal derivatives of the posterior pharyngeal arches , and Gegenbaur’s proposed evolutionary antecedent to paired fins . We sought to map the contributions of neural crest and mesoderm to the pharyngeal and paired fin endoskeleton in a cartilaginous fish , the little skate ( Leucoraja erinacea ) , as data from this lineage will allow us to infer ancestral germ layer contributions to the pharyngeal and paired fin skeletons , and to test the developmental potential of neural crest and mesodermal skeletal progenitors in a taxon that has retained an ancestral gill arch anatomical condition . We find that the gill arch skeleton of skate embryos receives contributions from both cranial neural crest and lateral mesoderm , revealing its dual embryonic origin . These findings point to an ancestral dual embryonic origin of the pharyngeal endoskeleton of jawed vertebrates , and to gill arches and paired appendages as serial derivatives of a dual-origin , neural crest- and mesodermally-derived mesenchyme with equivalent skeletogenic potential at the head-trunk interface . In the skate , neural tube closure begins at embryonic stage ( S ) 16 and is complete by S18 ( Ballard et al . , 1993 ) . in situ expression analysis of the gene encoding the conserved neural crest specifier Foxd3 reveals that by S18 , pre-migratory cranial neural crest cells are specified within the dorsal neural tube but are not yet undergoing epithelial-to-mesenchymal transition ( Figure 1A ) . At S18 , we can also recognize molecularly distinct lateral mesodermal populations , including tbx1-positive cranial paraxial mesoderm , or ‘head mesoderm’ ( Figure 1B ) , which is morphologically continuous with pitx2- and hand2-positive lateral plate mesoderm ( Figure 1C ) , and myf5-positive somitic and pre-somitic paraxial mesoderm ( Figure 1—figure supplement 1 ) . The clear spatial segregation and accessibility of these tissues in S18 skate embryos ( Figure 1D ) renders them amenable to fate mapping by labelling with lipophilic dyes – either by microinjecting the lumen of the neural tube ( to label pre-migratory neural crest cells ) , or by microinjecting mesodermal mesenchyme underneath the head ectoderm – and so we used this approach to directly test the contributions of these tissues to the pharyngeal and paired fin endoskeleton . To label skate cranial neural crest ( NC ) cells , we microinjected the lumen of the neural tube with CM-DiI at the hindbrain level . This resulted in very bright labelling at the point and time of injection ( Figure 2A ) , though analysis of embryos collected shortly post-injection in section reveals that the cells of the neural tube were labelled around its entire circumference , broadly , along the length of the hindbrain region ( Figure 2Ai ) . At five days post-injection , we observed abundant CM-DiI-labelled NC cells streaming into the pharyngeal arches ( Figure 2B ) , and at S31/32 ( ~8–10 weeks post-injection ) , we tested for NC contributions to cartilages throughout the pharyngeal skeleton . We have previously shown that the cartilaginous skeletal elements of embryonic skates can be readily identified , morphologically , in DAPI-stained vibratome or paraffin sections ( Figure 2—figure supplement 1 ) , and that labelling of early embryonic progenitors with lipophilic dyes is an effective way of mapping contributions to the cartilaginous endoskeleton ( Gillis et al . , 2013; Gillis and Hall , 2016; Gillis et al . , 2017; Criswell et al . , 2017; Criswell and Gillis , 2020 ) . While the extent of CM-DiI-labelling of skeletal derivatives is always greatly reduced , relative to the labelling of progenitor cells at the time of injection ( due to dilution of the CM-DiI label over several weeks of growth ) , positively-labelled cells are nevertheless unequivocally recognizable within the skeleton , due to the persistent brightness of the label . To add an additional level of stringency to our analysis , we only scored contributions to the skeleton consisting of clusters of two or more labelled cells , and contributions that were located in the centre of a skeletal element ( to avoid inadvertently scoring CM-DiI-labelled connective tissue abutting the cartilage ) . As embryonic cartilage is a homogeneous tissue , consisting of a single cell type ( the chondrocyte ) , we can therefore trace , with great certainty , the contributions of labelled progenitors to the differentiated cartilaginous endoskeleton . Using the approach outlined above , we readily observed clusters of NC-derived chondrocytes , for example , in the cartilage of the palatoquadrate ( Figure 2C ) and the epibranchial and branchial ray cartilages of the first gill arch ( Figure 2D ) . Overall , our analysis recovered NC contributions to major paired elements of the pharyngeal skeleton ( i . e . jaw , hyoid and gill arch elements ) and/or to the ventral midline cartilages across all labelled embryos ( n = 20/20 ) , but no contributions to the pectoral girdle ( Figure 2E; Supplementary file 1 ) . These findings are consistent with previous assessments of NC contribution to the pharyngeal and paired fin skeleton of zebrafish using genetic lineage tracing ( Kague et al . , 2012 ) . We next sought to complement our NC fate map with a test for mesodermal contributions to the pharyngeal and paired fin skeleton in the skate . To do this , we used sub-ectodermal microinjection of lipophilic dyes ( CM-DiI or SpDiOC18 ) to label lateral mesoderm at three positions – within the tbx1-expressing head mesoderm ( HM ) , at the boundary between HM and pitx2/hand2-expressing lateral plate mesoderm ( LPM ) , or exclusively within LPM ( Figure 1D ) – either alone ( Figure 3A ) , or in combination with neural crest labelling ( Figure 3B ) . We once again left labelled embryos to develop for ~8–10 weeks post-injection , and then scored the embryos for contributions to the skeleton , as described above . Embryos labelled within the HM at S18 showed little contribution to the pharyngeal skeleton ( labelled chondrocytes were recovered in gill arch cartilage of n = 1/10 labelled embryos; Supplementary file 1 ) , while in the collective majority of embryos labelled at the HM-LPM boundary ( n = 10/21 ) or within the LPM ( n = 14/17 ) , we observed substantial contributions to the pectoral girdle and fin skeleton ( Figure 3C ) . The ‘cardiopharyngeal field’ is a mesodermal territory that encompasses both the cranial paraxial and anterior lateral plate mesoderm , and that gives rise to pharyngeal arch ( branchiomeric ) musculature and the cardiovascular system ( Diogo et al . , 2015; Prummel et al . , 2020 ) . Accordingly , we observed extensive contributions from the skate HM , HM-LPM and LPM domains to the heart , blood vessels and pharyngeal arch musculature ( Figure 3—figure supplement 1; Supplementary file 1 ) . Remarkably , in many embryos labelled at the HM-LPM boundary ( n = 11/21 ) or within LPM ( n = 8/17 ) , we also recovered label-retaining chondrocytes in the skeleton of gill arches 1–5 . Mesodermally-derived chondrocytes were recovered within the epi- or ceratobranchial cartilages and branchial rays of gill arches 1–4 ( e . g . Figure 3D , E ) , as well as in the ceratobranchial of gill arch 5 , in close proximity to the label-retaining pectoral girdle and surrounding connective tissue ( Figure 3F – also , see Figure 3—figure supplement 2 for additional examples of mesoderm-derived label-retaining chondrocytes within the gill arch skeleton ) . Overall , our analysis recovered no mesodermal contributions to the mandibular or hyoid arch skeleton , but substantial mesodermal contributions to the paired cartilages of gill arches 1–5 , as well as to the pectoral girdle and fin skeleton ( Figure 3G; Supplementary file 1 ) . When considered alongside lineage tracing data from bony fishes , our findings allow us to infer an ancestral mesodermal contribution to the jawed vertebrate gill arch skeleton ( Figure 4A ) , with the transition from neural crest-derived to mesodermally-derive skeletogenic mesenchyme occurring gradually , and spanning the region of the posterior ( i . e . ancestrally gill-bearing ) pharyngeal arches ( Figure 4B ) . Taken together , our fate mapping experiments point to a neural crest origin of the mandibular and hyoid arch skeleton , a dual NC/mesodermal origin of the gill arch skeleton and an exclusively mesodermal origin of the pectoral fin skeleton in cartilaginous fishes ( Figure 4C ) . In light of the dual embryonic origin of the mammalian thyroid cartilage and exclusively mesodermal origin of the cricoid and arytenoid cartilages ( which are regarded as derivatives of the 4th and 6th pharyngeal arches ) , it is likely that boundaries of neural crest- and mesodermally-derived skeletogenic mesenchyme have shifted through vertebrate evolution . Our findings also have important implications for understanding the evolutionary origin of paired appendages . With waning support for Gegenbaur’s gill arch hypothesis , the lateral fin fold hypothesis of Balfour ( Balfour , 1881 ) , Thacher ( Thacher , 1877 ) and Mivart ( Mivart , 1879 ) emerged as the favoured scenario of paired fin origins . This hypothesis purports that paired fins originated from a continuous epithelial fold that flanked the trunk of the embryo , and that was subsequently segmented into distinct appendages at the pectoral and pelvic levels ( reminiscent of the origin of the 1st and 2nd dorsal fins from a continuous median fin fold in sharks ) . While palaeontological and embryological evidence for the existence of a lateral fin fold ( in phylogeny or ontogeny ) remains scant , there is evidence of shared molecular patterning mechanisms between dorsal median fins and paired appendages ( Freitas et al . , 2006; Dahn et al . , 2007; Letelier et al . , 2018 ) , and of the existence of broad zones of competence along the length of the trunk , from which ectopic fin/limbs or buds may be induced to form ( Cohn et al . , 1995; Kawakami et al . , 2001; Yonei-Tamura et al . , 2008 ) . From these observations , a scenario has emerged in which an established appendage patterning developmental module was co-opted , bilaterally , from the dorsal midline to the flank , giving rise to paired pectoral and pelvic appendages . We previously discovered shared , biphasic roles for Shh signalling in anteroposterior axis establishment and proliferative expansion of skeletal progenitors in the skate hyoid and gill arches and the tetrapod limb bud ( Gillis et al . , 2009; Gillis and Hall , 2016 ) , and we now show that these shared patterning functions transcend the germ layer origin of Shh-responsive skeletogenic mesenchyme ( i . e . neural crest alone in the hyoid arch , neural crest and lateral mesoderm in the gill arches and lateral mesoderm alone in the fin/limb bud ) ( Figure 4B ) . We propose that shared responses of hyoid , gill arch and limb skeletal elements to perturbations in Shh signalling – despite differences in the source of Shh in these organs ( i . e . the gill arch epithelial ridge and limb bud zone of polarizing activity – Riddle et al . , 1993; Gillis and Hall , 2016; Figure 4B ) – reflect a common underlying competence of gill arch and fin/limb skeletogenic mesenchyme to respond to these patterning signals , and serial homology of the skeletal derivatives of this mesenchyme . The zones of competence that underlie the origin of pectoral and pelvic appendages within the trunk could , accordingly , be extended rostrally to include zones of neural crest and mixed neural crest/lateral mesodermal contribution to the pharyngeal endoskeleton , and this , in turn , could account for the serial derivation of gill arches and paired appendages along the gnathostome stem . Indeed , reports of a fossil jawless vertebrate with gill arches extending down the length of the trunk ( Janvier et al . , 2006 ) further support the shared competence of pharyngeal and lateral trunk mesenchyme to give rise to both gill arch and fin/limb skeletal elements . It has been proposed that the neural crest acquired its skeletogenic potential by co-opting a chondrogenic gene regulatory network that arose , ancestrally , within mesoderm ( Meulemans and Bronner-Fraser , 2007; Cattell et al . , 2011 ) – a view that is further supported by the discovery of conserved molecular features of the developing neural crest and mesoderm-derived cartilages of vertebrates and the ( presumably mesoderm-derived ) cellular cartilages of some invertebrates ( Cole and Hall , 2004a; Cole and Hall , 2004b; Jandzik et al . , 2015; Tarazona et al . , 2016 ) . It is therefore to be expected that neural crest and mesodermal mesenchyme share fundamental molecular mechanisms of skeletogenesis . However , there is nevertheless a heterogeneity across mesenchymal subpopulations in their competence to respond to particular patterning cues . For example , in birds , specific regions of foregut endoderm are both necessary and sufficient for the specification of mandibular arch skeletal elements , but can only induce these elements to form from the neural crest mesenchyme that populates the mandibular arch ( and not from the mesenchyme of the more caudal pharyngeal arches – Couly et al . , 2002 ) . Conversely , quail-chick heterotopic transplantation experiments have shown that midbrain-derived neural crest mesenchyme is competent to give rise to the pleurosphenoid of the lateral braincase wall , even though this element typically derives exclusively from paraxial mesoderm ( Schneider , 1999 ) . Examples such as these point to more cryptic domains of skeletogenic mesenchyme , with distinct competencies , that do not necessarily align with germ layer boundaries . While the molecular basis of this mesenchymal regionalization may not be known , such regions of shared competence may be operationally defined using cell lineage tracing or transplantation experiments , and may be further tested for shared transcriptional features ( i . e . indicative of shared downstream effectors of common inductive cues , and the deployment of shared gene regulatory networks ) . We also propose that , on an evolutionary time scale , these regions of competence may be predisposed to the iterative deployment of developmental mechanisms , resulting in serial homology . Importantly , a competence-based hypothesis of gill arch-fin serial homology decouples the origin and evolutionary histories of gill arches/paired appendages as anatomical structures and the molecular mechanism that direct their patterning – i . e . it accounts for the former , but leaves the latter open to further discourse around the deep homology of appendage patterning mechanisms within vertebrates or , more broadly , metazoans ( Shubin et al . , 2009 ) . It is widely appreciated that , in animals , a relatively small number of developmental signalling pathways are used repeatedly , and in different combinations/contexts , to instruct the development of a great many embryonic tissues and organs . This , in turn , precludes the straightforward inference of homology of anatomical structures based on shared molecular patterning mechanisms ( Dickinson , 1995 ) . We argue that recognition of anatomical similarity due to common response to instructive cues within generative tissues , rather than focusing on the cues themselves , can allow us to bridge the gap between patterning mechanisms and morphology , and may provide a basis for inferring homology of morphology , even when considering structures that develop under the influence of upstream patterning mechanisms with complex and/or distinct evolutionary histories . Homology is a hierarchical concept , and two complex features ( e . g . organs ) – which arise within the context of an embryonic tissue , by deployment of a gene regulatory network operating downstream of an inductive or patterning cue – may be homologous at one biological level of organization , while simultaneously non-homologous at another ( Hall , 2003; Wagner , 2014 ) . While reconstructing the evolutionary history ( homology ) of individual genes or gene regulatory network nodes is becoming increasingly straightforward , meaningfully testing the homology of putatively distantly-related structures at the anatomical level – whether historical homologues across taxa , or serial homologues within a taxon – has , in many cases , lingered as problematic . Developmental competence , or the cell-autonomous property that imparts on tissues an ability to respond to external stimuli ( e . g . organizers and signalling centres ) ( Waddington , 1947 ) , may represent a tangible means of linking upstream molecular developmental mechanisms with ultimate anatomical readouts ( Spemann , 1915 ) . In light of the demonstrated lability of germ layer fates within the vertebrate skeleton ( Teng et al . , 2019 ) , we suggest that , in the case of anatomy , competence – which is inherently testable , either by natural ( i . e . evolutionary ) or laboratory experimentation – may supersede germ layer origin as a primary criterion of homology . L . erinacea eggs were obtained at the Marine Biological Laboratory ( Woods Hole , MA , USA ) and maintained in a flow-through seawater system at ~15°C to the desired developmental stage . Embryos for mRNA in situ hybridization were fixed in 4% paraformaldehyde in phosphate-buffered saline ( PBS ) overnight at 4°C , rinsed three times in PBS , dehydrated into 100% methanol and stored at −20°C . Embryos injected with CM-DiI and SpDiOC18 were fixed in 4% paraformaldehyde in PBS overnight at 4°C , rinsed three times in PBS , and stored in PBS + 0 . 02% sodium azide at 4°C . L . erinacea embryos were embedded in paraffin wax and sectioned at 8 µm thickness for mRNA in situ hybridization as previously described ( O'Neill et al . , 2007 ) . Wholemount and paraffin chromogenic mRNA in situ hybridization experiments for FoxD3 ( GenBank accession number MN478366 ) , Tbx1 ( GenBank accession number MT150581 ) , Pitx2 ( GenBank accession number MT150579 ) , Hand2 ( GenBank accession number MT150580 ) and Myf5 ( GenBank accession number MT150582 ) were performed as previously described ( O'Neill et al . , 2007 ) with modifications according to Gillis et al . , 2012b . Preparation and microinjection of CM-DiI and SpDiOC18 was carried out as previously described ( Gillis et al . , 2017; Criswell and Gillis , 2020 ) . After labelling , sealed eggs were returned to a flow-through seawater system at ~15°C to the desired developmental stage , and then euthanized using an overdose of tricaine ( 1 g/L in seawater ) prior to fixation . Labelled embryos to be analysed by vibratome sectioning were rinsed 3 × 5 min in PBS , embedded in 15% ( w/v ) gelatin in PBS and post-fixed in 4% paraformaldehyde in PBS for 4 nights at 4°C before sectioning at 100 µm on a Leica VT1000S vibratome . Sections were then DAPI-stained ( 1 µg/mL ) , coverslipped with Fluoromount-G ( Southern Biotech ) and imaged on an Olympus FV3000 confocal microscope . Labelled embryos to be analysed by paraffin histology were embedded and sectioned as previously described ( O'Neill et al . , 2007 ) .	A common way to evolve new body parts is to copy existing ones and to remodel them . In insects for example , the antennae , mouth parts and legs all follow the same basic body plan , with modifications that adapt them for different uses . In the late 19th century , anatomist Karl Gegenbaur noticed a similar pattern in fish . He saw similarities between pairs of fins and pairs of gills , suggesting that one evolved from the other . But there is currently no fossil evidence documenting such a transformation . Modern research has shown that the development of both gill and fin skeletons shares common genetic pathways . But the cells that form the two structures do not come from the same place . Gill skeletons develop from a part of the embryo called the neural crest , while fin skeletons come from a region called the mesoderm . One way to test Gegenbaur’s idea is to look more closely at the cells that form gill and fin skeletons as fish embryos develop . Here , Sleight and Gillis examined the gills and fins of a cartilaginous fish called Leucoraja erinacea , also known as the little skate . Sleight and Gillis labelled the cells from the neural crest and mesoderm of little skate embryos with a fluorescent dye and then tracked the cells over several weeks . While the fins did form from mesoderm cells , the gills did not develop as expected . The first gill contained only neural crest cells , but the rest were a mixture of both cell types . This suggests that fins and gills develop from a common pool of cells that consists of both neural crest and mesoderm cells , which have the potential to develop into either body part . This previously unrecognised embryonic continuity between gills and fins explains why these structures respond in the same way to the same genetic cues , regardless of what cell type they develop from . Based on this new evidence , Sleight and Gillis believe that Gegenbaur was right , and that fins and gills do indeed share an evolutionary history . While firm evidence for the transformation of gills into fins remains elusive , this work suggests it is possible . A deeper understanding of the process could shed light on the development of other repeated structures in nature . Research shows that animals use a relatively small number of genetic cues to set out their body plans . This can make it hard to use genetics alone to study their evolutionary history . But , looking at how different cell types respond to those cues to build anatomical features , like fins and gills , could help to fill in the gaps .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2020	Embryonic origin and serial homology of gill arches and paired fins in the skate, Leucoraja erinacea
Schistosomes infect hundreds of millions of people in the developing world . Transmission of these parasites relies on a stem cell-driven , clonal expansion of larvae inside a molluscan intermediate host . How this novel asexual reproductive strategy relates to current models of stem cell maintenance and germline specification is unclear . Here , we demonstrate that this proliferative larval cell population ( germinal cells ) shares some molecular signatures with stem cells from diverse organisms , in particular neoblasts of planarians ( free-living relatives of schistosomes ) . We identify two distinct germinal cell lineages that differ in their proliferation kinetics and expression of a nanos ortholog . We show that a vasa/PL10 homolog is required for proliferation and maintenance of both populations , whereas argonaute2 and a fibroblast growth factor receptor-encoding gene are required only for nanos-negative cells . Our results suggest that an ancient stem cell-based developmental program may have enabled the evolution of the complex life cycle of parasitic flatworms . Schistosoma flatworms infect 230 million people worldwide and cause ∼250 , 000 deaths per year ( van der Werf et al . , 2003 ) . These trematodes are transmitted through a life cycle that alternates between asexual and sexual generations in invertebrate intermediate and vertebrate definitive hosts , respectively ( Clark , 1974; Shoop , 1988 ) . The life cycle initiates as eggs are excreted from a mammalian host into freshwater , releasing ciliated , free-swimming larvae called miracidia that seek out and penetrate a snail intermediate host . Entry into the snail triggers a series of morphological , physiological , and biochemical transformations ( Basch and DiConza , 1974; Kawamoto et al . , 1989; Ludtmann et al . , 2009; Wu et al . , 2009; Parker-Manuel et al . , 2011 ) , followed by a clonal expansion of the larvae ( called sporocysts at this stage ) inside the snail host , ultimately producing thousands of infective cercariae ( Figure 1A ) ( Cheng and Bier , 1972; Ward et al . , 1988 ) . Mature cercariae then emerge from the snail into freshwater , burrow through the epidermis of mammalian hosts , migrate to species-specific niches in the host vascular system , develop to adulthood , and begin to reproduce sexually , thereby completing the life cycle . Thus , asexual amplification inside of the snail is vital for propagation of schistosomes . 10 . 7554/eLife . 00768 . 003Figure 1 . Germinal cells are detected throughout the asexual phase of the S . mansoni life cycle . ( A ) A schematic timeline of schistosome asexual amplification . ( B–C ) Maximum intensity projections of confocal stacks ( top ) and single optical slices ( bottom ) of a POPO-1 and SYTOX-Green co-stained miracidium ( B ) and a sporocyst 24 hr after in vitro transformation ( C ) . ( D ) Representative images of cells at metaphase ( M ) , anaphase ( A ) , and telophase ( T ) ( from left to right ) , captured in sporocysts 24 hr post-transformation . ( E–G ) Cryosections of the tentacle of a Biomphalaria glabrata snail showing a mother sporocyst ( perimeter highlighted by dashed line ) with daughter sporocysts packed inside ( 3 weeks post infection ) ( E ) ; an individual daughter sporocyst that has migrated to the digestive glands of a B . glabrata snail 6 weeks post infection ( F ) ; and cercarial embryos within a daughter sporocyst in the digestive glands of a B . glabrata snail 6 weeks post infection ( G ) ( staged after Cheng and Bier , 1972 ) . Actin is stained with phalloidin . Peanut agglutinin ( PNA ) visualizes acetabular glands and ducts of the cercariae . ( H ) A mature cercaria . The inset shows a magnified view of this animal’s head visualized with PNA and POPO-1 staining . Scale bars are 20 µm , except in ( E ) which is 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 003 A population of totipotent stem cells , historically called ‘germinal cells’ , is thought to underlie this unique intramolluscan amplification by undergoing multiple rounds of proliferation and de novo embryogenesis in the absence of fertilization ( Olivier and Mao , 1949; Cort et al . , 1954; Whitfield and Evans , 1983 ) . Early ultrastructural and histological studies recognized these cells by their stem cell-like morphology and rapid cycling kinetics ( Schutte , 1974; Pan , 1980 ) . In support of the totipotency of these germinal cells , serial transplantation of sporocysts into naive snail hosts led to continuous sporocyst propagation and cercarial production ( Jourdane and Théron , 1980 ) . These classic studies led to the model that division of these diploid presumptive totipotent stem cells in mother sporocysts produces progeny that are able to independently initiate the embryogenesis of daughter sporocysts ( Whitfield and Evans , 1983 ) . These daughter sporocysts , which are essentially sacs filled with germinal cells , can then produce more daughter sporocysts or infective cercariae in the same manner as they were generated themselves . This process represents ‘polyembryony’—during which multiple embryos are produced from the same zygote with no intervening gamete production . Thus , germinal cells appear to possess a unique developmental program , and it is unknown how they are specified , maintained , and regulated molecularly . In planarians , free-living flatworm relatives of schistosomes , a population of pluripotent stem cells called neoblasts can regenerate injured tissues and replenish a whole animal from a single cell ( Newmark and Sánchez Alvarado , 2002; Wagner et al . , 2011 ) . Guided by this knowledge , we recently identified a population of neoblast-like cells in adult Schistosoma mansoni ( Collins et al . , 2013 ) . These observations led us to hypothesize that germinal cells underlying schistosome asexual amplification may share a similar molecular program . Here , we show that the proliferating cells in sporocysts express many conserved stem cell genes . Using RNA interference ( RNAi ) we identify conserved regulators that are required to maintain the proliferative capacity of these cells . The similarity between these germinal cells in schistosome larvae , somatic stem cells in schistosome adults , and planarian neoblasts , links embryonic development and homeostatic tissue maintenance in these parasites; furthermore , it suggests that adaptation of an ancient stem cell developmental program may have enabled the evolution of complex trematode life cycles . Based on previous studies showing that germinal cells in schistosome larvae have a distinct morphology—high nuclear-to-cytoplasmic ratio , a large nucleolus , and cytoplasm densely packed with ribosomes ( Pan , 1980 ) —we reasoned that nucleic acid stains that preferentially label RNA could provide a means to label these cells specifically . Thus , we screened a number of nucleic acid-binding dyes and determined that POPO-1 clearly distinguishes the RNA-rich germinal cells from the other somatic cell types . In particular , POPO-1 strongly stains the nucleolus and cytoplasm of the germinal cells . This staining enabled us to track these cells through various stages of intramolluscan development both in vitro and in vivo ( Figure 1 ) . Consistent with previous work ( Pan , 1980 ) , we found 10–20 germinal cells in the posterior half of the body of free-swimming miracidia after hatching from the egg ( Figure 1B ) . In vitro transformation of miracidia into sporocysts triggered germinal cell proliferation ( Yoshino and Laursen , 1995; Ivanchenko et al . , 1999; Bixler et al . , 2001 ) : we observed these cells in mitosis as early as 24 hr post-transformation ( ∼0 . 2 mitoses per animal ) ( Figure 1C , D ) , which is consistent with the behavior of these cells in vivo after miracidia penetrate a snail host ( Schutte , 1974 ) . Following their long-term development in vivo , POPO-1 staining identifies a similar cell type throughout various stages of schistosome asexual development , including: developing daughter sporocysts in the mother ( Figure 1E ) ; motile daughters that have broken away from the mother and migrated to the snail digestive glands ( Figure 1F ) ; and cercarial embryos developing inside the daughter sporocysts ( Figure 1G ) . These POPO-1-labeled cells ultimately segregated into a compact cluster in cercariae ( Cheng and Bier , 1972 ) ( Figure 1H ) , presumably saved for the next stage of development after their penetration into mammalian hosts ( Dorsey et al . , 2002 ) . These observations reveal a morphologically homogeneous presumptive stem cell population that persists through the larval stages of the schistosome life cycle . In order to characterize these cells molecularly , we focused on the miracidium-to-mother sporocyst transition . To define this transition more precisely and examine the initiation of cell proliferation , at different times following in vitro transformation , we treated sporocysts with 5-ethynyl-2′-deoxyuridine ( EdU ) , a thymidine analogue that is incorporated into DNA during S-phase of the cell cycle ( Salic and Mitchison , 2008 ) . For these experiments , miracidia were transformed to mother sporocysts in vitro and co-cultured with an immortalized snail cell line ( Bge cells ) under hypoxic conditions ( Figure 2A ) ( Yoshino and Laursen , 1995; Ivanchenko et al . , 1999; Bixler et al . , 2001 ) . During the first 20 hr after transformation , few cells incorporated EdU . However , between 20–24 hr , we typically observed 1–2 clusters of EdU+ germinal cells , which are identified by their nucleolar and cytoplasmic POPO-1 staining . At later time points ( 48–64 hr ) , a large fraction of germinal cells in the sporocyst incorporated EdU , and all EdU+ cells exhibit a germinal cell-specific morphology . These cells were actively dividing as we routinely observed many mitotic figures in these samples ( Figures 1D and 2A ) . As a result of this massive proliferative burst , these germinal cells increased dramatically in number and came to occupy most of the sporocyst body , tripling sporocyst size during the first 3 days . These results suggest that a tightly regulated , sharp developmental transition begins at ∼20 hr post-transformation , when the first wave of proliferation starts . 10 . 7554/eLife . 00768 . 004Figure 2 . Life-cycle stage-specific transcriptional profiling to characterize germinal cell gene expression . ( A ) EdU labeling to detect proliferating cells at various time points following transformation . For these experiments sporocysts were co-cultured with Bge cells , a cell line derived from embryos of B . glabrata snails , to sustain the normal development of larval schistosomes . Arrows indicate cells weakly incorporating EdU at the early time point; arrowheads indicate proliferating Bge cells . Asterisks highlight cells in mitosis . In the absence of SYTOX-Green , POPO-1 also stains somatic nuclei , which are small and compact . Scale bars are 20 µm . ( B ) Transcripts expressed in sporocysts 48 hr post-transformation , ranked by abundance as measured by RNAseq . Orthologs shared between sporocyst-enriched genes and planarian neoblast-enriched transcripts are highlighted in cyan . ( C ) Relative expression levels of ago2-1 , vlg-3 , and nanos-2 during sporocyst development with respect to the expression in miracidia , measured by qPCR in biological triplicate . Error bars are standard deviations . p<0 . 01 . p<0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 00410 . 7554/eLife . 00768 . 005Figure 2—figure supplement 1 . Transcriptional profiling reveals genome-wide similarity between schistosome germinal cells and planarian neoblasts . ( A ) Histogram of fold changes of gene expression in sporocysts 48 hr post-transformation with respect to expression in miracidia , follows a log-normal distribution . The half standard deviation is used as the threshold to define differentially expressed transcripts . ( B ) Expression of sporocyst-specific genes that have homologs in neoblast-enriched transcripts is plotted logarithmically in fold change vs RPKM . A group of well-characterized conserved stem cell/neoblast markers , including ago2-1 , vlg-3 , and fgfrA , listed in Table 1 , is highlighted by open symbols . ( C ) Confocal sections of sporocysts 24 hr post-transformation showing the expression of various conserved neoblast genes which overlap with the expression of cell cycle-associated transcripts . From top to bottom: mago nashi homolog ( Smp_103470 ) , far upstream ( fuse ) binding protein ( Smp_044550 ) , nucleosome assembly protein ( nap , Smp_023530 ) , adenosylhomocysteinase ( ahcy , Smp_028440 ) , pescadillo related gene ( Smp_055800 ) , and p30 dbc protein ( Smp_193440 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 005 Since the miracidium-to-sporocyst transition is associated with both an increase in the number of proliferative germinal cells and a switch from a quiescent state to an actively proliferating state , we reasoned that the transcriptional profile of sporocysts would be enriched for mRNAs associated with germinal cells . Thus , we compared the gene expression profiles of miracidia and sporocysts 48 hr post-transformation using RNA sequencing ( RNAseq ) . Of the 10 , 852 predicted genes in the annotated S . mansoni genome ( Berriman et al . , 2009; Protasio et al . , 2012 ) , 6677 genes were detected in 48 hr sporocysts with an RPKM ( reads per kilobases of transcript per million mapped reads ) value above 1 ( Figure 2B ) ; 1662 of these genes were upregulated relative to miracidia ( Figure 2—figure supplement 1A and Supplementary file 1A ) . Besides the clonal expansion of germinal cells , the transition from miracidium-to-sporocyst is associated with numerous physiological and anatomical changes . Reasoning that germinal cells and planarian neoblasts may share a similar transcriptional signature , we compared our dataset with transcripts enriched in FACS-purified planarian neoblasts ( Önal et al . , 2012 ) to identify putative germinal cell-specific transcripts . Indeed , we noticed substantial overlap between these two datasets: of the 1662 genes upregulated in sporocysts , ∼30% of them ( 581 genes ) shared similarity ( e-value < e−10 ) with neoblast-enriched transcripts ( Figure 2—figure supplement 1 , and Supplementary file 1C ) . Conversely , more than ∼20% ( 864/4032 ) of neoblast-enriched transcripts shared similarity with sporocyst-enriched mRNAs . To better define the similarity between these datasets , we identified a total of 4749 1:1 schistosome-to-planarian orthologs using reciprocal BLAST comparisons ( e-value < e−10 ) ( Figure 2B ) . 1579 orthologs were enriched in planarian neoblasts; quite interestingly , 1525 of them ( 96 . 5% ) were expressed in sporocysts as well ( RPKM in sporocysts >1 ) . We examined the overlap between neoblast-enriched transcripts and sporocyst-enriched genes and identified 331 orthologs ( 20% of 1662 genes that are upregulated in sporocysts ) ( Figure 2B , and Supplementary file 1D ) . This list may contain a core set of genes that are essential for stem cell proliferation and maintenance . As a comparison , 12% ( 1248/10 , 125 ) of the remaining schistosome transcripts have orthologs in the planarian neoblast transcriptome . Given the differences in how these datasets were generated ( i . e . , FACS-purified cells from planarian vs whole schistosome sporocysts ) , together with the fact that the miracidium-to-sporocyst transformation is expected to result in transcriptional changes independent of germinal cell number , this degree of overlap seems likely to underestimate the similarity between the transcriptomes of germinal cells and planarian neoblasts . In addition to components of cell cycle and DNA repair machinery , examination of these overlapping gene sets identified many conserved factors associated with stem cell maintenance and germ cell development in diverse organisms ( Table 1 ) , including a pair of fibroblast growth factor receptors ( fgfr ) ( Ogawa et al . , 2002; Lanner and Rossant , 2010; Wagner et al . , 2012; Collins et al . , 2013 ) , three components of PRC2 ( polycomb repressive complex 2; suppressor of Zeste 12 , Sm-sz12 , enhancer of Zeste , Sm-ezh , and embryonic ectoderm development , Sm-eed ) ( Surface et al . , 2010; Wagner et al . , 2012; Önal et al . , 2012 ) , a p53 homolog ( Pearson and Sánchez Alvarado , 2010 ) , a bruno-like ( bruli ) RNA-binding protein ( Guo et al . , 2006 ) , and an argonaute2-like gene ( Gomes et al . , 2009; Rouhana et al . , 2010; Li et al . , 2011; Leonardo et al . , 2012 ) . Strikingly , 11/37 DEAD-box helicases ( DBHs ) in the S . mansoni genome are upregulated in sporocysts ( Supplementary file 1B ) , including three vasa/PL10 homologs ( Skinner et al . , 2012 ) that may play a similar role in flatworm parasites to that of vasa in other metazoans ( Juliano et al . , 2010; Tsai et al . , 2013 ) ( for consistency with previous studies [Shibata et al . , 1999; Ohashi et al . , 2007; Skinner et al . , 2012] , we will refer to these vasa/PL10 homologs as vasa-like genes [vlg] ) . Taken together , this comparison between parasitic and free-living flatworms suggests that proliferating germinal cells in sporocysts share some common features with planarian stem cells . Interestingly , this set of conserved germinal cell-enriched genes also contains several targets accessible with available small-molecule drugs ( e . g . , cyclophilin , adenosylhomocysteinase , transketolase ) ( Crowther et al . , 2010 ) , suggesting that these cells could serve as a potential vulnerable point to block schistosome propagation . 10 . 7554/eLife . 00768 . 006Table 1 . Expression in miracidia and sporocysts of schistosome homologs of planarian neoblast-enriched transcripts , measured by RNAseqDOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 006Gene nameRPKM ( miracidia ) RPKM ( sporocysts ) Fold changevasa-like ( vlg , Smp_068440 , 154320 , 033710 ) 20 . 6/46 . 1/272 . 498 . 5/79 . 2/423 . 14 . 8/1 . 7/1 . 6polo kinase ( Smp_009600 ) 43 . 0149 . 53 . 5fgfr ( Smp_175590 , 157300 ) 3 . 5/3 . 010 . 2/6 . 42 . 9/2 . 1 sz12 ( Smp_047720 ) 5 . 616 . 02 . 8 bruli ( Smp_041350 ) 5 . 515 . 32 . 8 Sedt8 ( Smp_055310 ) 1 . 94 . 82 . 6 egr ( Smp_094930 ) 3 . 07 . 42 . 5cyclin B ( Smp_082490 ) 79 . 7194 . 02 . 4 nlk ( Smp_074080 ) 7 . 518 . 02 . 4ago2-1 ( Smp_179320 ) 258 . 9537 . 52 . 1PCNA ( Smp_046500 ) 194 . 0395 . 02 . 0 inx ( Smp_141290 ) 102 . 6203 . 72 . 0 ezh ( Smp_078900 ) 8 . 115 . 51 . 9PHB ( Smp_075210 , 075940 ) 222 . 0/197 . 7421 . 0/290 . 81 . 9/1 . 5 pp32a ( Smp_010940 ) 641 . 71176 . 31 . 8 H2A ( Smp_086860 ) 240 . 3437 . 11 . 8 THOC ( Smp_005260 ) 219 . 6397 . 31 . 8egfr ( Smp_093930 , 165470 ) 6 . 8/7 . 012 . 1/10 . 01 . 8/1 . 4 CHD ( Smp_158050 ) 39 . 668 . 61 . 7tudor-like ( Smp_081570 ) 222 . 8367 . 81 . 7 H2B ( Smp_108390 ) 124 . 6206 . 51 . 7 ef-tu ( Smp_073500 ) 151 . 6243 . 61 . 6 HSP60 ( Smp_008545 ) 2051 . 53224 . 71 . 6 fhl ( Smp_048560 ) 2 . 03 . 11 . 5 eed ( Smp_165220 ) 32 . 746 . 31 . 4 junl ( Smp_067520 ) 5 . 57 . 71 . 4Expression of genes in bold is confirmed in this study with qPCR or FISH . In addition to their widely appreciated roles in germ cell specification , maintenance , and differentiation , a network of conserved post-transcriptional regulators ( e . g . , piwi/argonaute , vasa , and nanos ) are associated with multipotency in many metazoans ( Juliano et al . , 2010 ) . In diverse organisms , co-expression of these ‘germline genes’ specifies progenitor cells that can contribute to both soma and germline ( Juliano et al . , 2010 ) . Though piwi and vasa appear to be absent from schistosome genomes ( Gomes et al . , 2009; Collins et al . , 2013 ) , the S . mansoni genome has three argonaute homologs , two nanos homologs ( Figure 3—figure supplement 1 ) , and three vasa/PL10 homologs ( vlg ) ( Skinner et al . , 2012; Tsai et al . , 2013 ) . These genes , with the exception of Sm-nanos-1 ( Smp_057740 ) , are all expressed abundantly in sporocysts . Among them , Sm-ago2-1 , Sm-vlg-3 , and Sm-nanos-2 , are upregulated in sporocysts , but with different kinetics ( Figure 2C and Table 1; for clarity , the prefix ‘Sm’ will be omitted from gene names for the rest of the paper ) : ago2-1 increases steadily in expression throughout early sporocyst development; vlg-3 expression increases sharply following transformation , then plateaus; and nanos-2 mRNA expression does not increase until 4 days post-transformation . To examine whether these genes are expressed in germinal cells , we developed a method for whole-mount RNA fluorescence in situ hybridization ( FISH ) in sporocysts . Using this method , we confirmed that ago2-1 transcripts were enriched in germinal cells ( Figure 3A ) . These cells also expressed the cell cycle-regulated gene PCNA ( Figure 3A top ) . Following a 24 hr EdU pulse , >95% ( 2178/2282 ) of EdU+ cells expressed ago2-1 and ∼90% ( 1626/1815 ) of the ago2-1+ cells incorporated EdU ( Figure 3A , bottom ) . Similar observations were made with vlg-3 ( Figure 3B ) , suggesting that both ago2-1 and vlg-3 are expressed preferentially in proliferative germinal cells . 10 . 7554/eLife . 00768 . 007Figure 3 . ago2-1 and vlg-3 are expressed in proliferative germinal cells . ( A ) Top: confocal sections showing colocalization of ago2-1 and PCNA by FISH in sporocysts 24 hr post-transformation . Bottom: cells expressing ago2-1 incorporate EdU after a pulse 48–72 hr post-transformation . ( B ) vlg-3 is co-expressed with the cell cycle-associated transcript , H2A ( top ) , and vlg-3 is expressed in cells that incorporate EdU following a pulse at 48–72 hr post-transformation ( 1254 vlg-3+ EdU+/1362 EdU+ cells ) ( bottom ) . Arrowheads indicate proliferating Bge cells; asterisks indicate mitotic cells . Scale bars are 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 00710 . 7554/eLife . 00768 . 008Figure 3—figure supplement 1 . SmAGO2-1 is homologous to PIWI and AGO2 proteins , and SmNANOS2 is homologous to NANOS proteins . The alignments are shown with respect to homologs of Drosophila melanogaster ( Dm ) , Mus musculus ( Mm ) , and planarian Schmidtea mediterranea ( SMED ) proteins . ( A ) PIWI domains . ( B ) Nanos RNA-binding domains . Blue: conserved residues , red: divergent residues . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 008 By contrast , nanos-2 expression was detected only in a subset of ago2-1+/vlg-3+ cells ( Figure 4A , B ) . This observation suggests the existence of two distinct populations of germinal cells: nanos-2+ or nanos-2− cells . Monitoring the ratio of these two cell populations over time , we found the nanos-2− population expanded at a higher rate ( doubling time ∼24 hr ) than the nanos-2+ population ( Figure 4C ) . To examine if these cells possess differences in their cell cycle kinetics , we pulsed sporocysts with EdU for various lengths of time and determined the fraction of nanos-2+or nanos-2− germinal cells that were EdU+ ( Figure 4A , D ) . Following a 4-hr EdU pulse , ∼70% of the nanos-2− cells were EdU+ , whereas only ∼5% of the nanos-2+ cells were EdU+ . The fraction of nanos-2+ cells that were EdU+ increased to ∼40% with an 8-hr pulse , and to ∼75% with a 16-hr pulse . With either an 8-hr or 16-hr pulse , ∼95% of nanos-2− germinal cells were EdU+ . To rule out the possibility that EdU+/nanos-2+ cells are progeny of the more rapidly proliferating nanos-2− cells , we administrated a short 4-hr EdU pulse followed by a 12-hr chase period in the absence of EdU . We found that only 6% of nanos-2+ cells were EdU+ . Since a similar fraction of EdU+/nanos-2+ cells is observed immediately following a pulse ( Figure 4D ) , it is unlikely that nanos-2− germinal cells differentiate to produce nanos-2+ cells . Collectively , these observations suggest that nanos-2+ germinal cells have a longer cell cycle and enter S-phase less often than do nanos-2− germinal cells . This differential rate of proliferation likely explains why the upregulation of nanos-2 expression is delayed following transformation ( Figure 2C ) . 10 . 7554/eLife . 00768 . 009Figure 4 . ago2-1 , vlg-3 , and nanos-2 expression identifies heterogeneity in the germinal cell population . ( A ) FISH to detect nanos-2 and ago2-1 mRNA in EdU-labeled parasites . Relative to the nanos-2− germinal cells , nanos-2+ cells require longer time periods to incorporate EdU . Germinal cells are defined as ago2-1+ cells . The open arrowheads indicate nanos-2+ cells that are EdU− , whereas filled arrowheads point to those that are EdU+ . Times for EdU pulses are indicated in figures . ( B ) FISH to detect nanos-2 and vlg-3 mRNA . Scale bars are 20 µm . ( C ) The ratio between nanos-2− and nanos-2+ germinal cells increases with time after transformation . ( D ) Fractions of cells that incorporate EdU after pulses of various lengths . All EdU pulses start at 20 hr post-transformation , and the end times are indicated in the parentheses along the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 009 Motivated by the pivotal role of post-transcriptional regulation in various stem cell populations and during embryonic development ( Juliano et al . , 2010 ) , we characterized the functions of ago2-1 and vlg-3 using RNAi ( Boyle et al . , 2003; Rinaldi et al . , 2009 ) . For these experiments parasites were treated with double-stranded RNA ( dsRNA ) continuously while in the egg , after hatching , and during sporocyst development ( Figure 5A ) . Using this procedure we were able to achieve robust reductions ( >80% ) in both ago-2 and vlg-3 mRNA levels , as measured by quantitative real-time PCR ( qPCR ) . To assess the functions of these genes we monitored EdU incorporation in RNAi-treated parasites . Consistent with these genes regulating germinal cell proliferation , RNAi of either ago2-1 or vlg-3 resulted in significantly fewer EdU+ germinal cells following a 24-hr EdU pulse ( Figure 5B , C ) . Inhibition of either ago-2 or vlg-3 also resulted in a significant reduction in the levels of cell cycle-associated transcripts such as polo kinase and histone H2B , as well as germinal cell-associated genes ( ago2-1 , vlg-3 , and fgfrA ) ( Figure 5D ) . To resolve whether these effects were due to defects in cell proliferation or loss of the germinal cells , we examined the expression pattern of germinal cell markers by FISH after RNAi treatments . We observed that RNAi of vlg-3 resulted in the loss of both nanos-2+ and nanos-2− germinal cells ( Figure 5E ) , suggesting that it promotes the maintenance of the entire germinal cell population . In contrast to vlg-3 , reduction of ago2-1 levels resulted in loss of only nanos-2− germinal cells; nanos-2+ cells were still present ( Figure 5F ) . However , the absence of proliferation after ago2-1 RNAi ( Figure 5C ) suggests that the remaining nanos-2+ cells fail to proliferate . These observations are consistent with qPCR quantification , in which expression of nanos-2 was reduced after vlg-3 RNAi but retained after ago2-1 RNAi ( Figure 5D ) . 10 . 7554/eLife . 00768 . 010Figure 5 . vlg-3 and ago2-1 are required for germinal cell maintenance and proliferation . ( A ) Timeline for RNAi experiments; EdU was added to the culture at 20–40 hr post-transformation . ( B ) Average number of EdU+ nuclei per sporocyst in control , ago2-1 ( RNAi ) , and vlg-3 ( RNAi ) experiments . ( C ) Representative confocal stacks showing EdU incorporation after RNAi . Arrowheads indicate proliferating Bge cells; asterisk indicates an EdU+ mitotic cell . ( D ) Relative gene expression levels measured by qPCR for control and RNAi sporocysts , both in biological triplicate . The white bars indicate the genes targeted by RNAi . Error bars represent standard deviations . p<0 . 01 , p<0 . 001 , p<0 . 0001 ( t-test ) . ( E–F ) FISH to detect nanos-2 and ago2-1 expression after vlg-3 RNAi ( E ) , and nanos-2 and vlg-3 expression after ago2-1 RNAi ( F ) . In ( E ) , non-specific binding of probes to the Bge cells illustrates the background levels ( arrowheads ) . The expression patterns in control RNAi animals are unchanged . Scale bars are 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 010 Noting that ago2-1 and nanos-2 are also expressed in somatic stem cells in adult schistosomes ( Collins et al . , 2013 ) , we speculated that a similar network may regulate both adult stem cells and germinal cells . Indeed , fgfrA , which is essential for maintenance of adult stem cells , is also upregulated in sporocysts ( Figure 6A , Figure 2—figure supplement 1 , and Table 1 ) and its expression depends on either ago2-1 or vlg-3 , suggesting that fgfrA is present in germinal cells ( Figure 5D ) . To examine the function of fgfrA we performed RNAi experiments . Disruption of fgfrA mRNA blocked germinal cell proliferation , measured by reduced EdU incorporation ( Figure 6B ) and downregulation of cell cycle-associated transcripts ( Figure 6C ) . Similar to ago2-1 RNAi parasites , fgfrA RNAi also resulted in reduced ago2-1 and vlg-3 expression levels , while having no significant effect on the expression of nanos-2 ( Figure 6C ) . These results suggest a similar role for FGF signaling in controlling the proliferation of stem cells in both larval and adult schistosomes . 10 . 7554/eLife . 00768 . 011Figure 6 . fgfrA is required for germinal cell proliferation . ( A ) Relative expression levels of fgfrA during sporocyst development with respect to the expression in miracidia , measured by qPCR . ( B ) Average number of EdU+ nuclei per sporocyst in control and fgfrA ( RNAi ) parasites labeled at 20–40 hr post-transformation . ( C ) Relative gene expression levels measured by qPCR for control and fgfrA ( RNAi ) sporocysts . The white bar indicates efficient knockdown of fgfrA by RNAi . qPCR experiments were performed in biological triplicate . Error bars are standard deviations . p<0 . 001 , ***p<0 . 0001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00768 . 011 Comparison between sporocyst germinal cells described in this study and somatic stem cells recently identified in adult schistosomes ( Collins et al . , 2013 ) reveals significant molecular similarities , suggesting that these stem cells may persist throughout the entire schistosome life cycle . These germinal cells also exhibit many transcriptional and functional similarities with the neoblasts of free-living planarians . In planarians , PRC2 components are enriched in neoblasts , and are required for neoblast self-renewal and long-term maintenance ( Wagner et al . , 2012; Önal et al . , 2012 ) ; disruption of ago2 expression downregulates a specific set of miRNAs and blocks neoblast self-renewal ( Rouhana et al . , 2010; Li et al . , 2011 ) ; Smed-fgfr1 is expressed in a subset of planarian neoblasts; Smed-fgfr4 ( Wagner et al . , 2012 ) and Smed-bruli ( Guo et al . , 2006 ) are both enriched in neoblasts; and vasa is required for the proliferation and expansion of neoblasts ( Rouhana et al . , 2010; Wagner et al . , 2012 ) . Although schistosomes do not have a true vasa ortholog , the vasa/PL10 homolog vlg3 may have assumed a ‘vasa-like’ role in schistosome stem cell maintenance ( Tsai et al . , 2013 ) . Consistent with this idea , vasa/PL10 homologs in monogenean flatworms , which also lack vasa , are essential for germline development ( Ohashi et al . , 2007 ) . This conservation at both the cellular and molecular levels suggests an ancient role for these genes in regulating stem cell populations in flatworms . In light of these observations , it is plausible to suggest that neoblast-driven developmental programs inherited from their free-living ancestors may have enabled the evolution of complex trematode life cycles . These pluripotent cells may help the worms adapt successfully to obligate parasitism , by enabling both rapid expansion of an infective population and long-term tissue maintenance in the hostile environments within their intermediate and definitive hosts . Thus , future studies deciphering the evolutionary relationships between various neoblast-like cell populations are essential to understand the successful transmission and pathogenesis of various parasitic flatworms . Our data revealing the existence of two germinal cell subpopulations ( nanos-2+ and nanos-2− cells ) are reminiscent of observations in some other trematode species , in particular Echinostoma . In these parasites , proliferating germinal cells are present as two morphologically distinguishable populations ( Galaktionov and Dobrovolskij , 2003 ) . One of these populations , smaller in cell size , was speculated to be the most ‘undifferentiated’ cell type , whereas the other was thought to have more restricted developmental potential and enter embryogenesis directly . Although this morphological heterogeneity was not observed in schistosomes ( Pan , 1980 ) , our results uncovered molecular heterogeneity in these germinal cells . The observation that germinal cells expressing nanos-2 exhibit slower cell-cycle kinetics is consistent with the conserved role of nanos in lengthening the cell cycle by repressing mitotic transcripts in primordial germ cells from many animals ( Juliano et al . , 2010 ) . Along with our observations that nanos-2 is also expressed in germ cells in schistosome adult gonads ( Wang and Newmark , unpublished ) as well as adult somatic stem cells ( Collins et al . , 2013 ) , it is reasonable to expect the nanos-2+ cells may be more totipotent or germ cell-like , whereas the nanos-2− germinal cells may be more primed towards somatic fates . Given recent advances ( Dvořák et al . , 2010; Rinaldi et al . , 2012 ) , transgenic approaches to dissect the existence of such a developmental hierarchy may be possible in the future . Our results provide an initial molecular description of germinal cells in schistosome intramolluscan development . Based on the classic literature , the biology of these cells is quite unique: proliferative totipotent stem cells that directly undergo embryogenesis in the absence of fertilization . We find that , in spite of this unique developmental program , these cells possess a molecular signature similar to that of neoblasts in free-living flatworms , as well as stem cells from diverse organisms . This conserved molecular context opens access to understanding these cells , and may lead to strategies for intervening and blocking the transmission of the disease . S . mansoni ( strain: NMRI ) life cycle stages were provided by the Biomedical Research Institute ( BRI , Rockville , MD ) via the NIAID Schistosomiasis Resource Center through NIH-NIAID contract no . HHSN272201000005I . To purify S . mansoni eggs , livers from mice ( 6–8 weeks post infection ) were sterilized and digested in 5% clostridial collagenase ( Sigma , St . Louis , MO ) solution at 37°C for ∼20 hr ( Mann et al . , 2010 ) . The digested suspension was then forced through a 105-µm sieve , followed by repeated centrifugation and washing . Remaining liver tissue was removed by Percoll sucrose gradient centrifugation ( Mann et al . , 2010 ) . Purified eggs were cultured at 37°C/5% CO2 in Basch’s medium 169 ( Basch , 1981 ) , supplemented with 10% heat-inactivated fetal bovine serum ( FBS , Hyclone/Thermo , Logan , UT ) and 1× antibiotic-antimycotic ( Gibco , Carlsbad , CA ) for up to 10 days without significantly reducing the hatching rate of miracidia . Free-swimming miracidia were hatched in artificial pond water under bright light for 3 hr ( Samuelson et al . , 1984 ) . Unhatched eggs ( typical hatching rate ∼70–80% ) and empty egg shells were removed by centrifugation . Miracidia were transformed in vitro to mother sporocysts by exchanging pond water with sporocyst culture medium ( Ivanchenko et al . , 1999 ) . The sporocyst suspensions were co-cultured with Bge cells at 37°C/5% CO2/5% O2 . These conditions have been derived to maintain long-term cultures of sporocysts ( Bixler et al . , 2001 ) . Cercariae were obtained from Biomphalaria glabrata snails ( Schistosomiasis Resource Center ) ∼5–8 weeks post infection by exposing snails to bright light at 28°C for 1–2 hr . Cercariae and miracidia were fixed in 4% formaldehyde in pond water supplemented with 0 . 2% Triton X-100 and 1% NP-40 . Infected B . glabrata snails were fixed ( 4% formaldehyde in pond water with 0 . 2% Triton X-100 and 1% NP-40 ) at 4°C for at least 24 hr . Then the shell was crushed and removed , the snail tissue was equilibrated in 30% sucrose overnight , embedded in tissue freezing medium ( TBS ) , and cryosectioned . 30 μm sections were stained with combinations of 10 µM phalloidin conjugated to Alexa Fluor 568 , 1 µM POPO-1 iodide , 5 µM SYTOX-Green ( Invitrogen , Carlsbad , CA ) , and/or 20 µg/ml fluorescein-labeled PNA ( Vector Laboratories , Burlingame , CA ) ( Collins et al . , 2011 ) . Samples were cleared in scale A2 mounting solution ( 4M Urea , 10% glycerol , 0 . 1% Triton X-100 in PBS ) ( Hama et al . , 2011 ) . Fluorescence images were obtained on a Zeiss LSM 710 confocal microscope , with a 63× oil immersion objective ( N . A . = 1 . 4 ) . Before imaging , 10 mM ascorbic acid was added freshly to the mounting solution to prevent photobleaching . In vitro transformed sporocysts were cultured with 10 µM EdU ( Invitrogen , Carlsbad , CA ) for the indicated time periods . Following the EdU pulse , sporocysts were fixed for 30 min at room temperature or overnight at 4°C in 4% formaldehyde in Chernin’s balanced salt solution ( Chernin , 1963 ) with 0 . 2% Triton X-100 and 1% NP-40 . Fixed sporocysts were sequentially dehydrated in 50% methanol and then pure methanol . The dehydrated samples were kept at −20°C overnight , and rehydrated by exchanging methanol with 50% methanol , and then PBSTx ( PBS with 0 . 3% Triton X-100 ) . EdU incorporation was detected by click reaction with 25 µM Alexa Fluor azide conjugates ( Invitrogen ) for 20 min ( Salic and Mitchison , 2008 ) . Total RNA was purified from ∼10 , 000 freshly hatched miracidia or sporocysts 48 hr post-transformation using Trizol ( Invitrogen ) , and DNase treatment . For RNAseq , individually tagged libraries were prepared and pooled in a single lane . For each sample , ∼90 million 100 bp reads were sequenced on an Illumina HiSeq2000 at the W . M . Keck Center for Comparative and Functional Genomics , and mapped to the annotated S . mansoni genome ( version 5 ) ( Protasio et al . , 2012 ) using CLC Genomics Workbench ( CLC Bio , Aarhus , Denmark ) . Transcriptome analyses have been submitted to NCBI under the accession number GSE48282 . Comparisons between sporocyst and neoblast-enriched genes were performed using a list of 4032 transcripts whose expression was enriched in FACS-purified X1 neoblasts vs X2 neoblast progeny , and Xins differentiated cells ( Önal et al . , 2012 ) . For qPCR , total RNA was reverse transcribed to cDNA , and quantified on an Applied Biosystems Step One Plus station using GoTag qPCR reagents ( Promega , Madison , WI ) . Experiments were performed with three independent biological replicates , each containing ∼5000 animals . Relative expression levels were determined using the ΔΔCt calculation . The relative fold changes across samples measured by RNAseq were validated by qPCR of 10 randomly picked genes , giving R2 ≈ 0 . 9 . Four genes , Smp_093230 ( actin-related protein 10 ) , Smp_197220 ( 60S ribosomal protein L35 ) , Smp_089880 ( fad oxidoreductase ) , and Smp_169030 ( aspartyl-tRNA synthetase ) served as internal controls . These genes did not show life cycle stage-specific expression and their expression was not altered following RNAi . The primers used are listed in Supplementary file 1E . The fixed , dehydrated samples were stored in methanol at −20°C for up to 2 weeks without noticeable signal deterioration . The parasites were transferred to baskets ( 35-µm mesh , Intavis , Koeln , Germany ) , rehydrated , treated with 2 µg/ml proteinase K ( Invitrogen ) for 3–5 min , and post-fixed for 10 min in 4% formaldehyde in PBSTx . Hybridization and detection followed the protocol developed for adult worms ( Collins et al . , 2013 ) , except for the following modifications . The hybridization was carried out at 52°C overnight with ∼500 ng/ml antisense riboprobes , synthesized with standard in vitro transcription reactions incorporating digoxigenin-12-UTP ( Roche ) or dinitrophenol-11-UTP ( Perkin Elmer ) . Following stringency washes at 52°C , the samples were blocked with 1% casein and 7 . 5% horse serum , and incubated with anti-digoxigenin-peroxidase ( 1:1000; Roche , Indianapolis , IN ) or anti-dinitrophenol-peroxidase ( 1:250; Perkin Elmer , Waltham , MA ) at 4°C overnight . Detection was performed using tyramide signal amplification ( TSA ) with home-made tyramides ( King and Newmark , 2013 ) . For multi-color FISH , the peroxidase was quenched for 30 min in 0 . 1% sodium azide solution between sequential detections of different transcripts . Clones used for riboprobe and dsRNA synthesis were generated as described elsewhere ( Collins et al . , 2010 ) , with oligonucleotide primers listed in Supplementary file 1F . dsRNA was transcribed in vitro as described elsewhere ( Collins et al . , 2010 ) . The newly purified eggs ( ∼5000 eggs per ml ) were soaked with ∼20 µg/ml dsRNA for 8 days , with dsRNA freshly added every other day ( Figure 5A ) . Following hatching , the sporocysts were transformed and maintained in sporocyst medium supplemented with ∼20 µg/ml of dsRNA . As a negative control , parasites were soaked with a 1 . 5 kbp dsRNA derived from an irrelevant bacterial sequence ( Collins et al . , 2010 ) .	Schistosomiasis—a disease caused by parasitic flatworms known as schistosomes—affects more than 200 million people worldwide , mainly in tropical regions , and in public health importance is second only to malaria ( according to the World Health Organization ) . Chronic infection leads to damage to internal organs , and the disease is responsible for roughly 250 , 000 deaths each year . The schistosome parasite has a complex life cycle , and the worms are capable of infecting mammals during just one stage of this cycle . Infection occurs through contact with contaminated freshwater , with the infectious form of the parasite burrowing through skin . Once inside the body , the parasites mature into adults , before reproducing sexually and laying eggs that are excreted by their host back into the water supply . However , to generate the form of the parasite that can infect mammals , schistosomes must first infect an intermediate host , namely a freshwater snail . When the larval form of the parasite—which cannot infect mammals—enters the snail , the larvae undergo an unusual type of asexual embryogenesis . This results in thousands of parasites that are capable of infecting mammals . Studies suggest that a population of cells known as germinal cells are responsible for this transformation and replication process , but little is known about these cells at the molecular level . Here , Wang et al . report the gene expression profile of these cells in a species of schistosome , and use RNA-mediated silencing techniques to explore the functions of the genes . This analysis revealed that the germinal cells have a molecular signature similar to that of neoblasts—adult pluripotent stem cells found in free-living flatworms such as planarians . Neoblasts can develop into any cell type in the body , enabling planarians to repair or even replace damaged body parts . The similarity between neoblasts and germinal cells led Wang et al . to suggest that schistosomes may have evolved their parasitic life cycle partly by adapting a program of development based on stem cells in non-parasitic worms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "microbiology", "and", "infectious", "disease" ]	2013	Functional genomic characterization of neoblast-like stem cells in larval Schistosoma mansoni
T follicular helper cells ( Tfh ) are critical for the longevity and quality of antibody-mediated protection against infection . Yet few signaling pathways have been identified to be unique solely to Tfh development . ROQUIN is a post-transcriptional repressor of T cells , acting through its ROQ domain to destabilize mRNA targets important for Th1 , Th17 , and Tfh biology . Here , we report that ROQUIN has a paradoxical function on Tfh differentiation mediated by its RING domain: mice with a T cell-specific deletion of the ROQUIN RING domain have unchanged Th1 , Th2 , Th17 , and Tregs during a T-dependent response but show a profoundly defective antigen-specific Tfh compartment . ROQUIN RING signaling directly antagonized the catalytic α1 subunit of adenosine monophosphate-activated protein kinase ( AMPK ) , a central stress-responsive regulator of cellular metabolism and mTOR signaling , which is known to facilitate T-dependent humoral immunity . We therefore unexpectedly uncover a ROQUIN–AMPK metabolic signaling nexus essential for selectively promoting Tfh responses . High-affinity and long-lasting humoral immunity against infection requires controlled cross-talk between limiting CD4+CXCR5highPD1highBCL6high T follicular helper ( Tfh ) cells and immunoglobulin-maturing germinal center ( GC ) B cells in secondary lymphoid tissues ( King et al . , 2008; Victora and Nussenzweig , 2012; Nutt and Tarlinton , 2011; Ramiscal and Vinuesa , 2013 ) . As the GC largely consists of clonally diverse B cells , Tfh cells especially in narrow numbers are best at maintaining a selective pressure for B cell competition , favoring the survival of greater affinity antigen-responsive GC B cell clones ( Pratama and Vinuesa , 2014; Victora and Mesin , 2014 ) . Deregulation of Tfh cells can lead to faulty GC selection that may also seed the production of autoantibodies ( Weinstein et al . , 2012; Vinuesa et al . , 2005; Kim et al . , 2015; Linterman et al . , 2009 ) and GC-derived malignancies such as follicular lymphoma ( Rawal et al . , 2013; Klein and Dalla-Favera , 2008 ) . To date , the signals that exclusively govern Tfh cell differentiation over other T cell effector subsets remains poorly characterized . ROQUIN ( also called ROQUIN1; encoded by Rc3h1 ) acts to post-transcriptionally repress Tfh cells by binding effector T cell transcripts via its winged-helix ROQ domain ( Schuetz et al . , 2014; Tan et al . , 2014; Schlundt et al . , 2014 ) and recruiting proteins of the RNA decapping and deadenylation machinery ( Athanasopoulos et al . , 2010; Glasmacher et al . , 2010; Leppek et al . , 2013; Pratama et al . , 2013; Yu et al . , 2007; Vogel et al . , 2013 ) as well as the endoribonuclease REGNASE-1 ( Jeltsch et al . , 2014 ) . Some of its RNA targets include the Tfh-polarising Icos ( Glasmacher et al . , 2010 ) and Il6 mRNA ( Jeltsch et al . , 2014 ) as well as Ox40 ( Vogel et al . , 2013 ) and Tnf ( Pratama et al . , 2013 ) transcripts . In sanroque mice , an Rc3h1 missense point mutation , encoding for a Met199 to Arg substitution translates into a minor conformational shift in the RNA-binding ROQ domain ( Srivastava et al . , 2015 ) of ROQUIN and a loss of function in post-transcriptional repression . This leads to excessive Tfh growth and systemic autoimmunity ( Linterman et al . , 2009; Vinuesa et al . , 2005 ) . Complete ablation of ROQUIN results in unexplained perinatal lethality in C57BL/6 mice and selective deletion of ROQUIN in T cells does not lead to Tfh cell accumulation nor autoimmunity ( Bertossi et al . , 2011 ) . The latter is at least in part explained by the existence of the closely related family member ROQUIN2 ( encoded by Rc3h2 ) , which has overlapping functions with ROQUIN ( Pratama et al . , 2013; Vogel et al . , 2013 ) . The ROQUINM199R mutant protein has been proposed to act as a ‘niche-filling’ variant that has lost its RNA-regulating activity ( Pratama et al . , 2013 ) but can still localize to mRNA-regulating cytoplasmic granules to prevent the compensatory activity of ROQUIN2 . ROQUIN contains a conserved amino terminal RING finger with two conforming zinc-chelating sites ( Srivastava et al . , 2015 ) , despite an atypical aspartate as its eighth zinc ligand synonymous to RBX1 ( Kamura et al . , 1999 ) . This suggests ROQUIN may function as an E3 ubiquitin ligase ( Deshaies and Joazeiro , 2009 ) but , to date , no such enzymatic activity of the ROQUIN RING domain has been demonstrated in mammals . In vivo attempts to delineate the cellular pathways regulated by ROQUIN are made challenging due to the existence of multiple protein domains in the protein ( Figure 1—figure supplement 1a ) . The Caenorhabditis elegans ROQUIN ortholog , RLE-1 , acts through its RING domain to ubiquitinate DAF-16 , a pro-longevity forkhead box O ( FOXO ) transcription factor homolog ( Li et al . , 2007 ) . We did not find any evidence for molecular binding between ROQUIN and the fruitfly or mammalian FOXO orthologs ( Drosophila melanogaster FOXO and Mus musculus FOXO1 or FOXO3a; data not shown ) and therefore set out to understand the role of ROQUIN RING signaling in CD4+ T cell development and function by generating mice that selectively lack the ROQUIN RING zinc finger . We previously demonstrated that ROQUIN RING-deleted T cells in mice 6 days after sheep red blood cell ( SRBC ) immunization can form normal early Tfh cell responses but fail to promote optimal GC B cell reactions ( Pratama et al . , 2013 ) . Here , in mice that have developed robust Tfh-dependent GC responses toward SRBC or infected with lymphocytic choriomeningitis virus ( LCMV ) , we identify a novel and unexpected role of the ROQUIN RING domain in selectively promoting mature antigen-specific Tfh cell responses while leaving unaffected the development of other CD4+ effector T cell lineages . ROQUIN directly binds to and limits adenosine monophosphate-activated protein kinase ( AMPK ) , a tumor suppressor and central regulator of T cell glucose uptake and glycolysis ( MacIver et al . , 2011 ) . Our data indicate that loss of AMPK repression by deletion of the ROQUIN RING domain promotes stress granule persistence . This in turn cripples mTOR activity , otherwise known to play a critical role in driving CD4+ effector T cell expansion ( Delgoffe et al . , 2009; 2011 ) and T-dependent antibody responses ( Keating et al . , 2013; Zhang et al . , 2011; Gigoux et al . , 2014; De Bruyne et al . , 2015 ) . To examine the function of the ROQUIN RING domain in vivo , we generated two strains of C57BL/6 mice carrying either a germline deletion ( designated ringless; ‘rin’ allele ) or a T cell conditional deletion ( Tringless; ‘Trin’ allele ) of exon 2 in the Rc3h1 gene , which encodes the translation START codon and RING finger domain of the ROQUIN protein ( Figure 1—figure supplement 1b , c and Pratama et al . , 2013 ) . In these mice , skipping of exon 2 resulted in splicing of exon 1 to exon 3 yielding an alternative in-frame Kozak translation initiation site at Met133 ( Figure 1—figure supplement 1d , e ) . This predicted ROQUIN133-1130 protein product specifically lacks the RING domain ( Figure 1—figure supplement 1f ) . Mice homozygous for the rin allele were perinatally lethal ( Figure 1—figure supplement 1g–i ) , precluding T cell studies in intact animals . In contrast , Tringless mice were viable and showed no severe variations in thymic development and output of CD4 single positive T cells ( Figure 1—figure supplement 2a–e ) . There were also no major changes in Th1 cell differentiation in Tringless mice infected with LCMV ( Figure 1a ) , which predominantly yields LY6Chigh Th1 and LY6Clow Tfh virus-specific effector cells ( Hale et al . , 2013; Marshall et al . , 2011 ) . In Tringless animals immunized with SRBCs , the formation of Th1 , Th2 , Th17 , and regulatory T cells also remained largely unperturbed ( Figure 1—figure supplement 2f , g ) . This was mirrored in vitro with Tringless CD4+ naive T cells activated under Th1 , Th2 , Th17 , or induced Treg ( iTreg ) polarizing conditions ( Figure 1—figure supplement 2h ) displaying maximal expression of intracellular TBET , GATA3 , RORγT , and FOXP3 comparable to floxed wild-type T cell cultures ( Figure 1—figure supplement 2i ) . Surprisingly in Tringless mice , there was an overall defective Tfh cell primary response to LCMV infection ( Figure 1b–d ) and to SBRC immunization ( Figure 1—figure supplement 3a ) . ROQUIN RING-deficient T cells were also inefficient in supporting GC formation ( Figure 1e , f and Figure 1—figure supplement 3b ) , which was associated with reduced IL-21 production ( Figure 2a ) , a Tfh signature cytokine vital in supporting GC reactions ( Liu and King , 2013 ) . 10 . 7554/eLife . 08698 . 003Figure 1 . ROQUIN RING deletion in T cells preferentially controls Tfh cell formation . ( a-f ) Flow cytometric examination of mice d10 post-LCMV infection . ( a ) Proportion of LY6C+ total Th1 cells from CD4+CD44high T cells . ( b ) Identification of total Tfh cells pre-gated on CD4+CD44 high T cells . ( c ) Proportion of PD1highCXCR5high Tfh cells from CD4+CD44high T cells . ( d ) PD1highCXCR5highCD44 high Tfh cell numbers from spleen . ( e ) Proportion and ( f ) cell count of GL7 highFAShigh GC B cells in spleen . Data are pooled from three independent experiments ( n = 2–3 ) . Statistics were calculated by Student’s t-test , n . s . , not significant; p<0 . 05; p<0 . 005 . Dot symbols , individual mice; columns , median . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 00310 . 7554/eLife . 08698 . 004Figure 1—figure supplement 1 . Generation of mice with a ROQUIN RING deletion . ( a ) Secondary structure of mammalian ROQUIN protein . ( b ) Generation of ringless or Tringless mice with a deletion of Rc3h1 exon 2 when crossed to Rosa:Cre or Lck:Cre transgenic strains , respectively . White boxes , non-coding exons; black boxes , protein encoding exons; NeoR , Neomycin resistance cassette . ( c , d ) Deletion of Rc3h1 exon 2 detected in CD4+ T cells by Polymerase chain reaction ( PCR; c ) and sequencing ( d ) from cDNA . ( e ) Alignment of the mammalian Kozak sequence and alternative translational start site at Met133 in Rc3h1ringless . Solid lines , identical; dotted lines , similar . ( f ) ROQUIN protein map lacking the 14 . 5 kDa RING finger with translational rescue at Met133 . ( g , h ) Wild-type ( left ) ringless ( right ) embryos harvested at E14 ( g ) and E19 ( h ) from timed matings between ringless heterozygotes . ( i ) Embryonic genotyping and litter count from ringless heterozygous intercrosses ( n = 100 per time point ) . Statistics were calculated by Chi-square test , d . f . = 2 ( j ) Periodic acid-Schiff staining of the thoracic diaphragm from preterm animals taken at E19 . Wild-type ( left ) and ringless ( right ) are shown . Data are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 00410 . 7554/eLife . 08698 . 005Figure 1—figure supplement 2 . Phenotype of mice with a T cell-specific ROQUIN RING deletion . ( a-e ) Flow cytometric assessment of thymocyte development showing proportion of single positive ( SP ) CD4 and SP CD8 thymocytes ( a ) , total number of SP CD4 thymocytes ( b ) , proportion and count of FOXP3+CD25+ thymic regulatory T cells ( tTreg ) from SP CD4 thymocytes ( c ) , surface expression TCRβ on SP CD4 thymocytes ( d ) , and cell number of peripheral CD4+ T cell output in the spleen ( e ) . ( f , g ) d8 post-sheep red blood cell ( SRBC ) immunization of mice showing the proportion of IFNγ+ Th1 , IL4+ Th2 , and IL-17+ Th17 cells ( f ) , and the proportion of FOXP3+CD25+ Tregs ( g ) from splenic total CD4+ T cells . Data are representative of four independent experiments . ( h-i ) In vitro CD4+ T cell differentiation assay on naive T cells cultured under Th1 , Th2 , Th17 , or iTreg-polarizing conditions and analyzed d3 by flow cytometry ( h ) to measure expression of Th cell transcription factors in CD25+CD69+ activated T cells and on FOXP3+ activated T cells for iTreg cultures ( i ) . Statistics were calculated by Student’s t-test , n . s . , not significant; Dot symbols , individual mice; columns , median . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 00510 . 7554/eLife . 08698 . 006Figure 1—figure supplement 3 . Phenotype of SRBC-immunized mice with a T cell-specific ROQUIN RING deletion . ( a-c ) Flow cytometric analysis of d8 post-sheep red blood cell ( SRBC ) immunized mice showing the frequency of PD1highCXCR5high Tfh cells from total CD4+ T cells ( a ) , GL7highFAShigh germinal center ( GC ) B cells from total B220+ B cells in the spleen ( b ) compared to phosphate-buffered saline ( PBS ) injected control animals . Data are representative of four independent experiments . Statistics were calculated by Student’s t-test , p <0 . 05; Dot symbols , individual mice; columns , median . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 00610 . 7554/eLife . 08698 . 007Figure 2 . Functional competency of ROQUIN RING deleted Tfh cell responses . ( a ) Flow cytometric analysis of mice 8d after sheep red blood cell ( SRBC ) immunization showing the proportion of IL-21+CD44high effectors from total CD4+ T cells in the spleen . Data are representative of two independent experiments . ( b-i ) Flow cytometric examination of mice d10 post-lymphocytic choriomeningitis virus ( LCMV ) infection . ( b ) Proportion of IFNγ+ Th1 cells gated from total CD4+ T cells after GP61-80 peptide stimulation ex vivo . ( c ) Proportion of LY6C+ Th1 cells from virus-specific CD4+GP66-77+ T cells . ( d ) Identification of virus-specific Tfh cells pre-gated on CD4+GP66-77+ T cells . ( e ) Proportion of PD1highCXCR5high Tfh cells from virus-specific CD4+GP66-77+ T cells . ( f ) Virus-specific CD4+PD1highCXCR5highGP66-77+ Tfh cell numbers in spleen . ( g ) Representative histograms of BCL6 expression in virus-specific CD4+GP66-77+ T cells . Values included show median MFI for each genotype . ( h ) Proportion of FOXP3+ Tfr cells within the total CD4+CD44highPD1highCXCR5high Tfh gate . ( i ) Proportion of FOXP3+ Tfr cells within the virus-specific CD4+GP66-77+PD1highCXCR5high Tfh gate . Data are pooled from three independent experiments ( n = 2–3 ) . Statistics were calculated by Student’s t-test , n . s . , not significant; p<0 . 05; p<0 . 005; Dot symbols , individual mice; columns , median . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 007 By stimulating splenocytes ex vivo with GP61-80 peptide to identify virus-responsive IFNγ-producing Th1 cells ( Figure 2b ) and by examining splenic LYC6high Th1 cells amongst GP66-77+ tetramer stained T cells ( Figure 2c ) , we verified that ROQUIN RING loss did not disrupt protective Th1 responses but caused a severe abrogation of virus-specific Tfh cells during LCMV infection ( Figure 2d–f ) . Virus-specific T cells also showed significantly reduced expression of BCL6 ( Figure 2g ) , an indispensible nuclear factor for Tfh cell terminal differentiation ( Liu et al . , 2013 ) . Furthermore , we found an increased frequency of FOXP3+ T follicular regulatory ( Tfr ) cells within the total Tfh pool ( Figure 2h ) despite these Tfr cells not expressing a GP66-77 virus-specific T cell antigen receptor ( TCR; Figure 2i ) . Nonetheless , as Tfr cells are negative regulators of GC reactions ( Ramiscal and Vinuesa , 2013 ) , their abundance may indicate augmented suppression of Tfh cells and long-term B cell responses . We next sought to determine the molecular basis for the ROQUIN RING domain as a determinant in protective Tfh cell responses . Several lines of evidence implicated an involvement of ROQUIN in the negative regulation of AMPK signaling: Rc3h1ringless fetuses displayed skeletal muscle atrophy of the thoracic diaphragm ( Figure 1—figure supplement 1j ) , which is a characteristic phenotype of mice with overactive AMPK ( Sanchez et al . , 2012 ) and pointed to perinatal respiratory failure as the cause of the lethality . Also , AMPK over-expression in nematode worms has been shown to extend lifespan ( Mair et al . , 2011 ) , an observation consistent with the phenotype of worms lacking the ROQUIN ortholog RLE-1 ( Li et al . , 2007 ) . Since the AMPKα1 catalytic subunit is expressed in T cells and responds to TCR activation ( Tamas et al . , 2006 ) , we tested the possibility of ROQUIN directly binding to this subunit of AMPK ( encoded by Prkaa1 ) . Upon ectopic expression in HEK293T cells , ROQUIN colocalized with AMPKα1 diffusely or in fine cytoplasmic speckles in resting cells and within larger cytoplasmic granules upon induction of oxidative stress ( Figure 3a ) . We also observed colocalization of endogenous AMPKα1 within ROQUIN+ cytoplasmic granules in arsenite-treated primary C57BL/6 mouse embryonic fibroblasts ( MEFs ) ( Figure 3b ) with the use of an AMPKα1-specific antibody displaying no cross-reactivity toward the AMPKα2 subunit when ectopically expressed in HEK293T cells ( Figure 3—figure supplement 1a ) . Unlike the AMPKα1 subunit , ectopically expressed AMPK β and γ regulatory subunits did not associate with ROQUIN+ cytoplasmic granules , although AMPKγ2 and AMPKγ3 exhibited generally diffuse cytoplasmic distribution ( Figure 3—figure supplement 1b ) . We next determined if ROQUIN and AMPKα1 interacted by conducting in situ proximity ligation assays ( PLAs ) on primary C57BL/6 MEFs . Compared to control PLAs accounting for false interactions between endogenous AMPKα1 and non-expressed green fluorescent protein ( GFP ) detected by optimized anti-GFP immunostaining ( Figure 3—figure supplement 1c ) , we found that endogenously expressed ROQUIN and AMPKα1 proteins localized with very close molecular proximity in both resting and arsenite-stressed cells ( Figure 3c , d ) at a frequency 15-fold higher or more than weak PLA interactions previously observed between ROQUIN and AGO2 ( Srivastava et al . , 2015 ) . Moreover , we were able to coimmunoprecipitate ROQUIN and AMPKα1 when over-expressed in HEK293T cells ( Figure 3—figure supplement 1d ) or expressed endogenously in the mouse T lymphoblast line EL4 cells ( Figure 3e ) . Together with the PLAs , this indicated that ROQUIN bound specifically with the α1 subunit of AMPK and that under physiological conditions , the two proteins could form a stable complex . 10 . 7554/eLife . 08698 . 008Figure 3 . ROQUIN preferentially colocalizes and binds with the α1 subunit of AMPK . ( a ) Colocalization of V5-ROQUIN and AMPKα1-GFP ectopically expressed in resting ( top ) and 1 mM arsenite ( AS ) -treated ( bottom ) HEK293T cells . Representative of three independent experiments . ( b ) Colocalization of endogenous ROQUIN and AMPKα1 in primary ( mouse embryonic fibroblasts ) MEFs post-arsenite ( AS ) treatment . Representative of three independent experiments . ( c ) Proximity ligation assays ( PLAs ) performed on primary C57BL/6 MEFs showing interactions between endogenously expressed ROQUIN and AMPKα1 in resting cells ( ROQUIN:AMPKα1 ) and in cells stressed with 1 mM arsenite ( +AS , ROQUIN:AMPKα1 ) . Negative control PLAs ( GFP:AMPKα1 ) detecting non-expressed GFP and endogenous AMPKα1 background are also displayed . Blue , DAPI stained nuclei; Red , ligation events , Scale bar , 20 μm . Representative of three independent experiments . ( d ) Quantitative analysis of PLAs showing mean ligation events per cell ( nucleus ) for each field of view on a confocal microscope . Individual dots represent a single field of view; bar per column represents the sample mean . Statistics were calculated by one-way ANOVA with Bonferroni’s multiple comparisons test after log transformation of ratio values , n . s . , not significant; *p<0 . 0005 . ( e ) Reciprocal coimmunoprecipitation of ROQUIN and AMPKα1 endogenously expressed in EL4 cells . IB , immunoblot; IP , immunoprecipitated . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 00810 . 7554/eLife . 08698 . 009Figure 3—figure supplement 1 . Association of AMPK subunits with ROQUIN . ( a ) Detection of AMPKα1 by immunostaining primary MEFs ( as shown in main Figure 3b ) was achieved using an α1 isoform-specific immunofluorescence-grade antibody exhibiting no cross-reactivity with AMPKα2-GFP expressed in HEK293T cells . ( b ) Protein localization of V5-ROQUIN and carboxy terminal GFP fusion constructs of AMPK regulatory subunits ectopically expressed in HEK293T cells and analyzed by immunofluorescence microscopy . Data are representative of two independent experiments . ( c ) Detection of negative control PLAs ( GFP:AMPKα1; as shown in main Figure 3c ) was achieved using anti-GFP antibody titrated by immunostaining transfected HEK293T cells expressing AMPKα1-GFP fusion protein . Blue , DAPI-stained nuclei; Red , anti-GFP stain; Green , anti-AMPKα1 stain; Scale bar , 50 μm . ( d ) Reciprocal coimmunoprecipitation of V5-ROQUIN and AMPKα1-GFP over-expressed in HEK293T cells . Data are representative of four independent experiments . IB , immunoblot; IP , immunoprecipitated . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 009 To determine the functional consequence of a ROQUIN–AMPKα1 interaction , we measured AMPK activity in Tringless and wild-type T cells . In contrast to wild-type cells , phosphorylation of the AMPK target , acetyl CoA carboxylase ( ACC ) in ROQUIN RING-deficient CD4+ T cells was increased , demonstrating constitutively active AMPK activity in vitro ( Figure 4a ) and in vivo ( Figure 4b ) . Thus , ROQUIN acts through its RING domain to directly negatively regulate AMPKα1 activity in T cells . 10 . 7554/eLife . 08698 . 010Figure 4 . The ROQUIN RING finger is required for autoubiquitination and negative regulation of AMPK . ( a ) In vitro kinase assay of AMPKα in isolated CD4+ T cells during an anti-CD3 and -CD28 activation time-course . Data are pooled from two independent experiments and normalized to unstimulated wild-type ( n = 5 ) . Black columns , floxed wild-type; white columns , Tringless . Statistics were calculated by Student’s t-test , p<0 . 05 . †p <0 . 05 for wild-type at 10 min vs . wild-type at 0 and 5 min . columns , mean; error bars , s . e . m . ( b ) Phospho-blot of endogenous ACC Ser79 in resting CD4+ T cells . Representative of three independent experiments . IB , Immunoblot . ( c ) Ubiquitin immunobDot of endogenous ROQUIN immunoprecipated from EL4 cells ( d ) Immunoblot of V5-tagged ROQUIN1-1130 and ROQUIN133-1130 in transfected HEK293T cells ( left ) , endogenous ROQUIN in ringless primary MEFs ( center ) , immunoprecipitated ROQUIN in Tringless thymocytes ( right ) . ( e ) In vitro autoubiquitination assay for ROQUIN wild-type peptide ( residues 1–484 ) and RING finger deleted peptide ( residues 145–484 ) . Five consecutive lanes show the extent of ROQUIN autoubiquitination of the same in vitro reaction at 0 , 1 , 2 , 4 , and 16 h . ( f ) Cellular ubiquitination assay for full length V5-ROQUIN and RING-deleted V5-ROQUIN133-1130 ectopically expressed in HEK293T cells with HA-Ub . Data are representative of three independent experiments . IB , immunoblot; IP , immunoprecipitated . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 010 Given the important role of RING domains in driving protein substrate ubiquitination ( Deshaies and Joazeiro , 2009 ) , we next tested if the regulation of AMPK activity by ROQUIN was a result of RING-mediated AMPK ubiquitination . Absence of the ROQUIN RING domain did not alter AMPK ubiquitination ( data not shown ) . However , monoubiquitination of endogenous ROQUIN in EL4 cells was detected ( Figure 4c ) . To determine if ROQUIN monoubiquitination was dependent on the 14 . 7 kDa RING finger deleted in ROQUIN RING deficient mice ( Figure 4d ) , we tested if ROQUIN could undergo automonoubiquitination in vitro and in a cell-based ubiquitin assay . By Coomassie staining PAGE-separated peptides of in vitro ubiquitination reactions , we detected a single protein band having higher molecular weight relative to ROQUIN peptide that formed in the presence of wild-type ROQUIN1-484 and ubiquitin ( Figure 4e ) . This slowly migrating band , consistent with monoubiquitin attachment , formed at severely delayed times in the absence of the RING zinc finger . A complete absence of this higher molecular weight ROQUIN peptide modification was observed with in vitro reactions lacking ubiquitin protein . We also performed ubiquitination assays in transfected HEK293T cells and detected ubiquitin-conjugated ROQUIN by immunoprecipitation when full-length ROQUIN was over-expressed but not with expression of the ROQUIN133-1130 variant recapitulating the specific RING deletion borne by Tringless T cells ( Figure 4f ) . Together , our data show that the ROQUIN RING domain can facilitate automonoubiquitination independent of residues carboxy terminal to Asp484 . We next investigated the mechanism by which ROQUIN RING activity limits AMPK signaling . Analogous to RAPTOR inactivation within stress granules ( Thedieck et al . , 2013; Wippich et al . , 2013 ) , we hypothesized that ROQUIN localization and its ability to bind AMPK within stress granules was key to AMPK repression . We have previously shown that ROQUIN133-1130 lacking the RING domain did not coalesce with eIF3+ stress granules ( Pratama et al . , 2013 ) . To exclude the possibility that this mislocalization of RING-deficient ROQUIN was a product of over-active AMPK feedback , we investigated if AMPK hyperactivity prevented ROQUIN localizing to stress granules . Full length ROQUIN still colocalized with eIF3+ stress granules in the presence of the AMPK agonist , AICAR ( Figure 5a ) , which alone was ineffective at inducing stress granule formation ( data not shown ) . This indicates that ROQUIN133-1130 mislocalization is a direct consequence of an intrinsic lack of the RING domain . To confirm that stress granule exclusion was not a secondary effect of a structurally unstable ROQUIN1-132 deletion but rather a consequence of the loss of RING-mediated E3 ligase activity , a loss-of-function mutation of the first zinc-coordinating cysteine of the RING domain ( Cys14Ala; Figure 5b ) that typically abolishes E3 ligase activity of related RING-containing enzymes ( Fang et al . , 2001; 2000 ) was introduced into HEK293T cells . Although ROQUINC14A ectopic expression could facilitate de novo stress granule induction in the absence of arsenite treatment comparable to cells transfected with wild-type ROQUIN ( Athanasopoulos et al . , 2010 ) , we found that in response to arsenite exposure , ROQUINC14A localization to eIF3+ stress granules was significantly impaired ( Figure 5c ) . A deleted RING domain did not abrogate ROQUIN133-1130-AMPKα1 colocalization; the two proteins were detected in small aggregates most likely outside of stress granules ( Figure 5d ) . This was consistent with RING deficient ROQUIN133-1130 protein still capable of directly binding AMPKα1 ( Figure 5e ) . Together these findings indicate that ROQUIN RING signaling does not play a role in AMPK recruitment to ROQUIN but rather directs negative regulation of AMPKα1 through sequestration into stress granules following ROQUIN–AMPKα1 complex formation . 10 . 7554/eLife . 08698 . 011Figure 5 . ROQUIN RING activity controls its localization to stress granules . ( a ) Colocalization of over-expressed full length V5-tagged ROQUIN or ROQUIN133-1130 with endogenous eIF3 in HEK293T cells stressed with 1 mM arsenite ( AS ) for 1 hr with or without 2 mM AICAR . Scale bar , 50 μm . ( b ) Crystal structure of ROQUIN peptide showing amino terminal residues 6 to 75 incorporating the RING domain . Black , zinc cation; green , zinc-coordinating residue , red , zinc-coordinating Cys14 targeted for mutagenesis; yellow , zinc-chelating interaction . Data are based on structural coordinates we had previously determined ( Srivastava et al . , 2015 ) and deposited in the Protein Data Bank , accession code 4TXA . ( c ) Colocalization of over-expressed full length GFP-tagged ROQUIN or ROQUINC14A mutant with endogenous eIF3 in HEK293T cells stressed with 1 mM arsenite for 1 hr . ( d ) Colocalization of ROQUIN133-1130 with AMPKα1 when over-expressed in HEK293T cells immediately after 1 mM arsenite exposure for 1 hr . ( e ) Reciprocal coimmunoprecipitation of full length ROQUIN or ROQUIN133-1130 and AMPKα1 over-expressed in HEK293T cells . IB , immunoblot; IP , immunoprecipitated . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 011 One possible downstream effector of ROQUIN–AMPK in Tfh cells is the mechanistic Target of Rapamycin ( mTOR ) , a nutrient sensing kinase and modulator of cellular metabolism . AMPK activity directly suppresses mTORC1 signaling ( Gwinn et al . , 2008; Inoki et al . , 2003 ) , and deletion of AMPKα1 increases mTORC1 signaling in T cells ( MacIver et al . , 2011 ) . Although the role of mTOR in promoting effector CD4+ and CD8+ T cell responses is well documented ( Araki et al . , 2011; Chi , 2012 ) , mTOR signaling in Tfh cell formation , and therefore antibody responses , is incompletely understood . In this respect , we assessed mTOR function in Tringless CD4+ T cells in response to TCR and CD28 stimulation . In CD4+ T cells , we found a reduction in phosphorylated ribosomal S6 in the absence of ROQUIN RING , indicating diminished mTORC1 function ( Figure 6a ) . This effect was mild in naive CD44low T cells but accentuated in CD44 high cells . Reflecting a role for ROQUIN RING activity during early development , abated mTOR activity was also observed in ROQUIN RING deleted primary MEFs by enhanced phosphorylation of RAPTOR Ser792 ( Figure 6b ) , a target residue for AMPK-mediated inhibition . 10 . 7554/eLife . 08698 . 012Figure 6 . ROQUIN RING signaling regulates stress granule responses to promote mTOR . ( a ) Flow cytometric analysis of phospho-rS6 Ser235/236 in CD44low or CD44high anti-CD3 and anti-CD28 stimulated CD4+ T cells ( n = 4–6 ) . ( b ) Phosphoblot of ACC Ser70 and RAPTOR Ser792 in primary mouse embryonic fibroblasts ( MEFs ) recovered in complete DMEM for 3 hr after 1 hr of 1 mM arsenite treatment ( left ) . Quantitative ratios of phosphorylated RAPTOR to β-ACTIN input based in phosphoblot MFI readings ( right ) . IB , immunoblot . ( c-e ) Analysis of stress granule induction in primary MEFs analyzed by fluorescence microscopy after 1 hr of 1 mM arsenite stress treatment showing counts of eIF3+ granules per cell ( c ) , and size of individual eIF3+ granules in freshly arsenite-stressed primary MEFs based on area ( d ) and maximum feret ( e ) . ( f ) Proportion of recovering primary MEFs exhibiting cytoplasmic eIF3+ stress granules ( SG ) after arsenite-mediated stress ( n >30 per time point , with each n replicate representing a single field of view displaying 1–7 adherent cells ) . Columns , mean; error bars , s . d . ( g ) Representative micrographs displaying recovered primary MEFs at 3 hr post-arsenite stress . Scale bar , 50 μm . Statistics were calculated by Student’s t-test , n . s . , not significant; p<0 . 05; p<0 . 0005 . Data are representative of three independent double blind experiments . ( h ) Proportion of primary MEFs with eIF3+ stress granules after 1 hr of 1 mM arsenite treatment comparing wild-type and sanroque MEFs recovering in complete DMEM media . Data are representative of two independent experiments ( n >30 per time point , with each n replicate representing a single field of view displaying 1–6 adherent cells ) . Error bars , s . d . Statistics were calculated by Student’s t-test , n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 01210 . 7554/eLife . 08698 . 013Figure 6—figure supplement 1 . Stress granule sequestration of RAPTOR . ( a , b ) Colocalization of RAPTOR and eIF3+ stress granules in resting ( a ) and 1 mM arsenite ( AS ) -stressed ( b ) primary mouse embryonic fibroblasts derived from C57BL/6 timed matings . Data are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 01310 . 7554/eLife . 08698 . 014Figure 6—figure supplement 2 . AMPK controls stress granule formation and maintenance . ( a ) Proportion of C57BL/6 primary mouse embryonic fibroblasts ( MEFs ) showing cytoplasmic eIF3+ stress granules ( SG ) immediately after cellular exposure to 1 mM arsenite ( AS ) for 30 min with or without 50 μM Compound C ( CompC ) . ( b ) Proportion of primary MEFs with eIF3+ stress granules recovering in complete DMEM media after 1 hr of 1 mM AS treatment comparing cultures with or without 2 mM AICAR added during both AS treatment and recovery period . Data are representative of three independent experiments ( n >30 per time point , with each n replicate representing a single field of view displaying 1–6 adherent cells ) . Error bars , s . d . Statistics were calculated by Student’s t-test , p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 014 Stress granules are AMPK-dependent ( Hofmann et al . , 2012; Mahboubi et al . , 2015 ) cytoplasmic compartments that sequester and inactivate mTORC1 during cellular stress ( Thedieck et al . , 2013; Wippich et al . , 2013 ) . In primary MEFs , owing to their large cytoplasm and prominent stress granules , we confirmed RAPTOR localization to eIF3+ stress granules in a wild-type and Roquinringless background ( Figure 6—figure supplement 1a , b ) . We also found that arsenite-induced stress granule formation was impeded by AMPK inhibition in MEFs treated with Compound C ( Figure 6—figure supplement 2a ) . Therefore , we sought to determine if diminished mTOR signaling was associated with augmented stress granule formation or maintenance in ROQUIN RING-deficient cells . Analysis of arsenite-stressed primary MEFs by fluorescence microscopy revealed that loss of ROQUIN RING signaling did not alter stress granule induction ( Figure 6c–e ) but rather prolonged the rate of stress granule dissolution during stress recovery ( Figure 6f , g ) . A similar delay in stress granule recovery was mirrored in primary MEFs in which AMPK activity was raised upon treatment with AICAR ( Figure 6—figure supplement 2b ) . Conversely , in sanroque mutant primary MEFs expressing a ROQUIN variant incapable of regulating target mRNAs , we found that stress granule recovery post-arsenite treatment was comparable to wild-type MEFs ( Figure 6h ) . Together , these data suggest that the selective Tfh cell defect in Tringless mice may be a result of a disrupted ROQUIN–AMPK signaling axis , otherwise important in relieving stress granule inhibition of mTOR . Furthermore , ROQUIN RING-mediated stress granule subversion of mTOR activity appears to be independent of the RNA repressive functions of the ROQUIN ROQ domain . To determine if attenuated mTOR is associated with defective Tfh cell responses as observed in Tringless animals , we examined chino mice harboring a hypomorphic mutation ( chi allele ) in the Frap1 gene resulting in an Ile205Ser substitution within the HEAT repeat domain of the mTOR protein ( Figure 7a ) , the region dedicated to binding RAPTOR ( Kim et al . , 2002 ) . Unlike the in utero lethality observed in mice with complete mTOR deficiency ( Gangloff et al . , 2004; Murakami et al . , 2004 ) , chino is a viable strain that exhibits growth retardation , intact thymocyte development and output but reduced phosphorylation of ribosomal protein S6 in phorbol-12-myristate-13-acetate treated peripheral CD4+ T cells ( Daley et al . , 2013 ) . We confirmed suboptimal phosphorylation of mTOR targets 4EBP1 and S6K in chino peripheral CD4+ T cells in response to physiological TCR activation with CD28 costimulation ( Figure 7b ) . This was in contrast to significantly elevated FOXP3 expression . Thus , chino-mutant T cells represent a mild deficiency in mTOR signaling reminiscent of a partial loss-of-function in mTOR ( Zhang et al . , 2011 ) that exclusively affects extrathymic CD4+ T cell differentiation as seen in conditional T-cell-deleted Frap1 knockout mice ( Delgoffe et al . , 2009 ) . We immunized chino mice with SRBC , assessed the GC response 5 days later by flow cytometry and found that Tfh cells were severely diminished compared to wild-type controls ( Figure 7c ) . This corresponded with a reduction in the GC B cell response ( Figure 7d ) . 10 . 7554/eLife . 08698 . 015Figure 7 . mTOR signaling is required for optimal Tfh cell formation . ( a ) chino mutation causes a I205S substitution in the mTOR protein . ( b ) Flow cytometric measurements of intracellular phospho-4EBP1 Thr37/46 , phospho-S6K Thr389 , and FOXP3 in CD4+ T cells stimulated with anti-CD3 and -CD28 for 30 min ( n = 4–6 ) . MFI , mean fluorescence intensity; column , group mean , error bars , s . d . ( c , d ) chino mutants were immunized with sheep red blood cells ( SRBC ) and taken down 5 days later to analyze the proportion of PD1highCXCR5high Tfh cells from CD4+ T cells ( c ) , and the proportion of GL7highFAShigh GC B cells from B220+ B cells ( d ) in the spleen . Data are representative of three independent experiments . Statistics were calculated by Student’s t-test , p<0 . 005; *p<0 . 0005 . ( e , f ) Flow cytometric analysis 50:50 mixed LY5A wild-type:LY5B chino bone marrow chimeras d7 post-SRBC immunization showing the proportion of PD1highCXCR5high Tfh cells ( e ) , and expression of intracellular BCL6 ( f ) from the LY5A and LY5B CD4+B220− T cells . Linked dot symbols , congenic cells from same animal; MFI , mean fluorescence intensity . Data are representative of two independent experiments . Statistics were calculated by Paired Student’s t-test between congenically marked cells of the same animal , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 01510 . 7554/eLife . 08698 . 016Figure 7—figure supplement 1 . Schematic representation of ROQUIN signaling in Tfh cell ontogeny Secondary structure of the ROQUIN protein depicting the molecular function of its distinct domains in Tfh cells . The RING zinc finger mediates autoubiquitination of ROQUIN , thereby attenuating its interacting partner AMPK which otherwise facilitates stress granule formation and maintenance , and limits both mTOR signaling and Tfh cell responses . The ROQ domain and its adjacent C3H zinc finger act in synergy to destabilize mRNA targets important in T cell activation . ROQUIN2 shares overlapping mRNA regulatory functions with ROQUIN . Pro-rich , proline rich region; 2xCoil , coiled-coil domain; AMPK , Adenosine monophosphate-activated protein kinase; Tfh cell , T follicular helper cell; mTOR , Mechanistic target of rapamycin . DOI: http://dx . doi . org/10 . 7554/eLife . 08698 . 016 To determine if mTOR acts within Tfh cells , we constructed and immunized 50:50 mixed wild-type:chino bone marrow chimeras with SRBC . At d7 post-immunization , we confirmed that the percentages of mTOR mutant PD1 highCXCR5 high Tfh cells were impaired compared to their competing wild-type counterparts in the same mouse ( Figure 7e ) . This was associated with reduced BCL6 expression intrinsic to mTOR mutant CD4+ T cells ( Figure 7f ) . Our data therefore demonstrates that Tfh cells depend on intact intracellular mTOR signaling and that the chino Tfh-intrinsic phenotype closely mimics a defective ROQUIN RING deleted Tfh response . We therefore conclude that not only does mTOR act in the same polarity as the ROQUIN RING domain during Tfh cell development , but since mTOR signaling is a bone fide target of AMPK-directed inhibition ( Gwinn et al . , 2008; Inoki et al . , 2003 ) in CD4+ T cells ( Zheng et al . , 2009 ) , it is most likely that mTOR represents a molecular pathway between the ROQUIN–AMPK axis and the control of Tfh responses ( Figure 7—figure supplement 1 ) . The critical signaling requirements specific to programming Tfh cell differentiation have been under intense investigation in the past decade ( King and Sprent , 2012; Rolf et al . , 2010 ) . ROQUIN , acting through its ROQ and C3H domains , has previously been identified as a potent post-transcriptional repressor of CD4+ Tfh cells ( Glasmacher et al . , 2010; Pratama et al . , 2013; Vinuesa et al . , 2005; Vogel et al . , 2013; Yu et al . , 2007; Lee et al . , 2012 ) but has also been shown to act similarly in limiting Th1 , Th17 cells and CD8+ effector T cells ( Bertossi et al . , 2011; Jeltsch et al . , 2014; Chang et al . , 2012; Lee et al . , 2012 ) . In the present study , we highlight for the first time the cellular function of the amino terminal ROQUIN RING finger and unexpectedly its importance as a positive immunomodulator of peripheral Tfh cells exclusively . In response to SRBC immunization and also to acute LCMV primary infection , mice lacking the ROQUIN RING domain in T cells failed to optimally form mature Tfh cells that could support a robust GC response . Interestingly , Th1 , Th2 , Th17 , and Treg responses were comparable to wild-type controls in vivo and in vitro , depicting a functional uncoupling between RING signaling and ROQ-C3H activity in ROQUIN . This is consistent with our previous observations of ROQUIN RING-deficient T cells showing minimally altered expression of ICOS , a target of post-transcriptional repression ( Pratama et al . , 2013 ) . We show at the molecular level , that the RING domain of ROQUIN is required to attenuate AMPK signals . In support of this finding , adenosine metabolism has previously been linked to T-dependent antibody responses in mice with observations that antigen-specific Tfh cells displayed constitutively high surface expression of CD73 , an ecto-enzyme that catabolizes extracellular AMP into adenosine ( Iyer et al . , 2013; Conter et al . , 2014 ) . Taken together , it is possible that Tfh cells utilize a purinergic autocrine signaling pathway similarly suggested in Treg cells ( Deaglio et al . , 2007; Sitkovsky , 2009 ) , whereby CD73-generated adenosine external to Tfh cells is imported through nucleoside transporters for reversion back into cytoplasmic AMP by adenosine kinase , before ROQUIN RING-regulated activation of AMPK . Furthermore , since AMPK is an inhibitor of glycolysis and cellular growth ( Hardie et al . , 2012; Mihaylova and Shaw , 2011 ) , its activity in Tfh cells could facilitate BCL6 function , especially in transcriptionally dampening CD4+ T cell glycolysis ( Oestreich et al . , 2014 ) which is otherwise important for cell growth ( Jones and Thompson , 2007 ) . This would form the basis for why Tfh cell numbers are so tightly contained throughout a GC response , acting as a critical limiting factor for controlling the magnitude and clonal diversity of GC reactions ( Schwickert et al . , 2011; Victora and Nussenzweig , 2012; Rolf et al . , 2010 ) . At first glance , it may seem conflicting that ROQUIN RING deficiency results in unrestrained AMPK leading to crippled BCL6 expression and Tfh cell hypocellularity . However , intact ROQUIN RING signaling may be advantageous , if not critical in Tfh responses , possibly acting to secure an intricate maximal threshold of AMPK activity that is key to maintaining narrow Tfh numbers for effective GC clonal selection while allowing ample , but not excessive , Tfh support to GCs . We could not find any evidence that ROQUIN E3 ligase activity directly targets AMPK for proteosomal degradation . Instead , impaired autoubiquitination due to the absence of a functional RING domain suggests that ROQUIN E3 ligase activity can negatively regulate AMPK independent of the RNA regulatory functions of ROQUIN . Although AMPK β and γ subunits have been shown to localize to stress granules ( Mahboubi et al . , 2015 ) , our results demonstrated an inability of the ectopically expressed regulatory subunits to localize with ROQUIN+ stress granules in cells also overexpressing ROQUIN , hinting at the requirement of a different cofactor for their recruitment to stress granules . Given this and the present data demonstrating the ability of ROQUIN and AMPKα1 to bind each other and colocalize within stress granules together with previous observations of stress granule exclusion of RING-deficient ROQUIN ( Pratama et al . , 2013 ) , we propose a model whereby ROQUIN may be repressing AMPK activity via ubiquitin-dependent sequestration of the AMPKα1 subunit within stress granules and thereby promoting repression of AMPK kinase activity . This stress granule-associated regulation of AMPK may not be exclusive to ROQUIN but could involve other binding partners such as G3BP1 , which has been shown to localize with AMPKα2 in stress granules ( Mahboubi et al . , 2015 ) . Interestingly , a direct interaction has also been observed between AMPKα and G3BP1 ( Behrends et al . , 2010 ) , an integral component of stress granules that associates with ROQUIN ( Glasmacher et al . , 2010 ) . Analogous to the T cell anergy-regulating RING-type E3 ubiquitin ligases GRAIL and CBL-B that undergo autoubiquitination ( Anandasabapathy et al . , 2003; Levkowitz et al . , 1999 ) as well as targeting various T cell signaling molecules for RING-mediated ubiquitination ( Nurieva et al . , 2010; Su et al . , 2006; 2009; Lineberry et al . , 2008; Fang et al . , 2001; Fang and Liu , 2001; Jeon et al . , 2004 ) , it is likely that ROQUIN also ubiquitinates additional proteins to coordinate Tfh cell immunity . We found that ROQUIN RING deficiency results in intact thymic development and low phosphorylation levels of ribosomal S6 in a subset of peripheral CD4+ T cells , which together resembles closely chino mutant mice with impaired mTOR function ( Daley et al . , 2013 ) . Our data points to overactive AMPK as the link between the loss of ROQUIN RING activity and reduced mTOR signaling . As AMPK activity has been shown to be transiently upregulated within 5–20 min of stress induction and to decline after the appearance of stress granules ( Mahboubi et al . , 2015 ) , it is possible that the subcellular sequestration of AMPK within stress granules may represent a regulatory circuit breaker to interrupt the positive feed-forward loop that acts to shut down mTOR , mRNA translation and cell growth in response to cellular stress , AMPK induction and stress granules formation . Enhanced AMPK-mediated stress granule persistence leading to mTOR repression in the absence of ROQUIN RING signaling aligns well with a report of dampened TOR activity in eukaryotic cells during cellular stress by the transient shuttling of TOR into stress granules ( Takahara and Maeda , 2012 ) . In addition , Wippich et al . ( 2013 ) also found in HeLa cells having inactivated the stress granule inhibitory kinase DYRK3 , that stress granule longevity was the key to prolonging mTOR inhibition . Since Rc3h1 appears to be ubiquitously expressed ( Vinuesa et al . , 2005 ) , it is intriguing that the Tringless allele preferentially affects Tfh cells of the GC and no other CD4+ T cell lineage . We found in ROQUIN RING deleted CD4+ T cells that this may be a result , at least in part , of insufficient IL-21 cytokine production , which is otherwise required for optimal GC reactions and to a lesser degree , Tfh cell maintenance ( Zotos et al . , 2010; Vogelzang et al . , 2008; Linterman et al . , 2010 ) . IL-21 deficiency would also explain why a normal Tfh cell response with diminished GC B cells in Tringless mice was detected early at d6 post-SRBC immunization ( Pratama et al . , 2013 ) , but both mature GC Tfh and B cell populations were crippled at d8 in the current study . Within Tfh cells , ROQUIN RING activity may also be important for transducing stimuli downstream of a GC-specific receptor such as PD1 , which is uniquely found most highly expressed on the surface of Tfh cells in humans and mice ( Yu and Vinuesa , 2010; Kamphorst and Ahmed , 2013 ) . In CD4+ T cells , ligation of PD1 couples mTOR signals ( Francisco et al . , 2009 ) and also restricts cellular glycolysis ( Patsoukis et al . , 2015; Parry et al . , 2005 ) in a similar manner to AMPKα1 activity ( MacIver et al . , 2011; Michalek et al . , 2011 ) . It remains unclear how ROQUIN RING activity links to Tfh cell environmental stimuli , but there is evidence of ROQUIN phosphorylation by unknown kinases in human T cells ( Mayya et al . , 2009 ) . Previously , we showed that the Rc3h1 ‘sanroque’ allele encodes a ROQUINM199R mutant protein with a defective RNA-binding ROQ domain unable to repress ICOS . This , together with excessive IFNγ signaling causes aberrant accumulation of Tfh cells leading to unrestrained and pathogenic GC growth ( Lee et al . , 2012; Yu et al . , 2007 ) . The accumulation of Tfh cells in sanroque animals opposes the defective Tfh response of Tringless mice . We postulate that ROQUIN133-1130 represents a complete loss of the AMPK-regulating functions with minimal disturbance to the RNA-regulating function . This may explain why the phenotype of mice with a combined deletion of the RING domains found in both ROQUIN and that of its closely related RC3H family member ROQUIN2 ( Pratama et al . , 2013 ) is less severe than the immune deregulation of Roquin/Roquin2 double knockout mice ( Vogel et al . , 2013 ) . We have previously shown that ROQUIN2 can compensate for the RNA-regulating function of ROQUIN , both repressing overlapping mRNA targets ( Pratama et al . , 2013 ) . By contrast , the sanroque ROQUINM199R mutated protein that can still localize to stress granules and bind RNA ( Athanasopoulos et al . , 2010 ) is likely to represent a recessive ‘niche-filling’ variant that selectively inactivates the normal mRNA-regulating function of ROQUIN but preserves its assembly into the mRNA decapping complex , preventing compensatory substitution by ROQUIN2 ( Pratama et al . , 2013 ) . It is also likely that ROQUINM199R protein found in sanroque T cells retains RING finger activity to negatively regulate AMPK and promote Tfh cell development , which is compounded by the increased stability of T cell mRNAs that exacerbate Tfh accumulation and trigger autoimmunity . Indeed in mice , T cell AMPK activity has been shown to play a protective role in autoimmune models of rheumatoid arthritis and multiple sclerosis ( Nath et al . , 2009; Son et al . , 2014 ) . As a downstream target of AMPK metabolic signaling , mTOR is well known to orchestrate T cell effector differentiation and peripheral tolerance ( Chi , 2012; Araki et al . , 2011 ) . As such , deregulation of mTOR can facilitate T-dependent autoimmune disorders like systemic lupus erythematosus ( Koga et al . , 2014; Fernandez et al . , 2006; Kato and Perl , 2014 ) and multiple sclerosis ( Delgoffe et al . , 2011; Esposito et al . , 2010 ) . Although active mTOR signaling in Tfh cells has been documented ( Gigoux et al . , 2014 ) , the specific role mTOR has in Tfh cell responses remains largely uncharacterized ( Araki et al . , 2011 ) . Our data indicate that mTOR can act in concert with and downstream of ROQUIN RING signaling to support optimal Tfh cell formation . Susceptible to both ROQUIN-controlled AMPK repression and stress granule sequestration , mTOR regulation is thus integral to Tfh cell responses . Multiple studies are also in line with this model . Similar to our findings of impaired IL-21 synthesis in mTOR-attenuated Tringless CD4+ T cells , a report on Frap1 knockout T cells cultured in vitro also displayed reduced expression of IL-21 ( Delgoffe et al . , 2009 ) . Moreover , low expression and secretion of IL-21 was observed in rapamycin-treated human CD4+ T cells polarised toward the Tfh cell lineage ex vivo ( De Bruyne et al . , 2015 ) . Also in mice with reduced mTOR , T-dependent B cell proliferation , isotype switching , GC formation and antigen-specific antibody responses were significantly crippled ( Zhang et al . , 2011; Keating et al . , 2013 ) . Additional in vivo studies , however , are required to dissect the downstream signals transduced by mTOR that orchestrate Tfh immune responses . Thorough investigation of these and similar metabolic pathways including AMPK-dependent cellular bioenergetics within Tfh cells and in other GC T cell subsets ( Ramiscal and Vinuesa , 2013 ) is not only warranted but also may advance current strategies for vaccine design or reveal novel therapeutic interventions for antibody-mediated immune disorders . ROQUIN RING deleted mice ( Pratama , et al . , 2013 ) were generated by Ozgene , Australia; loxP sites that flanked or ‘floxed’ Rc3h1 exon 2 , encoding the START codon and RING motif ( Figure 1—figure supplement 1 ) , were inserted into C57BL/6 mouse embryonic stem ( ES ) cells via homologous recombination . Recombinant ES cell clones were implanted into C57BL/6 foster mothers . Heterozygote progeny were screened for germline transmission before crossing to Rosa26:Flp1 mice to remove the neo cassette . Mice harboring the floxed Rc3h1 allele ( Rc3h1lox/+ ) were then crossed to Rosa26:Cre knock-in mice for one generation . Removal of Cre expression was then achieved by a C57BL/6 backcross yielding a germline deletion of Rc3h1 exon 2 . This strain was named ringless ( rin allele ) . A conditional ROQUIN RING-deficient strain ( called Tringless ) was also generated by crossing Rc3h1lox/lox mice to Lck:Cre breeders to remove Rc3h1 exon 2 specifically in T lymphocytes ( Trin allele ) . Upon Cre-mediated excision of Rc3h1 exon 2 , rescue of in-frame protein synthesis at Met133 is expected to produce an E3 ligase defective ROQUIN mutant . Rosa26:Cre and Lck:Cre mice were maintained on a C57BL/6 background with one copy of the Cre transgene and provided by Ozgene , Australia . ENU-derived chino mutants were previously characterized ( Daley et al . , 2013 ) . To generate mixed bone marrow chimeric mice , recipient Rag1-/- mice were sublethally irradiated and reconstituted i . v . with 2 x 106 bone marrow hematopoietic stem cells . Animal experiments were approved by the Animal Experimentation Ethics Committee of the Australian National University ( Protocols J . IG . 71 . 08 , A2012/05 and A2012/53 ) and the McGill University Ethics Committee ( Protocol 7259 ) . Mice were maintained in a specific germ-free environment . Where indicated , 8 to 12 wo mice were immunized i . p . with 2 x 109 SRBC to generate a T-dependent GC response or i . p . with 2 x 105 PFU of LCMV Armstrong . The following antibodies were used in Western blots , immunoprecipitation assays and fluorescence microscopy: rabbit anti-phospho-ACC Ser79 ( Cat . 3661 , Cell Signaling ) , rabbit phospho-RAPTOR Ser792 ( Cat . 2083 , Cell Signaling ) , rabbit anti- β-ACTIN ( 13E5 , Cell Signaling ) , rabbit anti-AMPKα ( Cat . ab32047 , Abcam , UK ) , goat anti-eIF3 ( N-20 , Santa Cruz ) , mouse anti-GFP ( 7 . 1 and 13 . 1 , Roche ) , rabbit anti-GFP ( Cat . ab290 , Abcam ) , mouse anti-HA ( HA-7 , Sigma-Aldrich ) , rabbit anti-HA ( H6908 , Sigma-Aldrich ) , rabbit anti-RAPTOR ( 24C12 , Cell Signaling ) , rabbit anti-ROQUIN ( Cat . A300-514A , Bethyl Laboratories ) , mouse anti-UBIQUITIN ( P4D1 , Cell Signaling ) , mouse anti-V5 ( V5-10 , Sigma-Aldrich ) , rabbit anti-V5 ( Cat . V8137 , Sigma- Aldrich ) , mouse anti-rabbit IgG light chain ( 211-032-171 , Jackson ImmunoResearch ) , and goat anti-mouse IgG light chain ( 155-035-174 , Jackson ImmunoResearch ) . AICAR ( Calbiochem ) and Compound C ( Calbiochem ) were used according to manufacturers’ recommendations at indicated concentrations . N-terminal V5 tagged full length ROQUIN and ROQUIN133-1130 constructs have been previously described ( Pratama et al . , 2013 ) . C-terminal GFP fused ROQUIN and ROQUINC14A constructs have previously been described ( Athanasopoulos et al . , 2010 ) . GFP tagged constructs of AMPK subunits were obtained from Origene . HEK293T and EL4 cells were obtained from the ATCC and perpetuated in-house . Primary MEFs were harvested from E14 fetuses of rin/+ or san/+ pregnant females that were paired with rin/+ or san/+ males , respectively , as part of a timed mating . Where indicated in vitro stimulation of T cells was performed using anti-CD3 and anti-CD28 dual coated Dynabeads ( Invitrogen ) or for cytokine accumulation , phorbol myristate acetate ( Sigma-Aldrich ) , and ionomycin ( Sigma-Aldrich ) was used with GolgiStop ( BD Biosciences ) in RPMI 1640 medium ( Invitrogen ) supplemented with 2 mM l-glutamine ( Invitrogen ) , 100 U penicillin-streptomycin ( Invitrogen ) , 0 . 1 mM non-essential amino acids ( Invitrogen ) , 100 mM HEPES ( Sigma-Aldrich ) , 0 . 0055 mM 2-mercaptoethanol , and 10% FCS . 20 ng/mL IL-2 ( R&D Systems ) , 100 ng/mL IL-4 ( Miltenyi Biotec ) , 100 ng/mL IL-6 ( Peprotech ) , 20 ng/mL IL-12 ( Miltenyi Biotec ) , 1 ng/mL TGFβ ( R&D Systems ) , along with 1μg/mL of Biolegend antibodies anti-IL-4 , anti-IFNγ and/or anti-IL-12 were used for in vitro polarization of ( IL-2 ) Th0 , ( IL-2 , IL-12 , and anti-IL-4 ) Th1 , ( IL-2 , IL-4 , anti-IFNγ , and anti-IL-12 ) Th2 , ( IL-2 , IL-6 , TGFβ , anti-IL-4 , and anti-IFNγ ) Th17 , and ( IL-2 and TGFβ ) iTreg cultures . To stain surface markers , cells were washed and stained in ice-cold staining buffer ( 2% FCS , 0 . 1% NaN3 in PBS ) . eBioscience FOXP3 Staining Buffer Set was used for flow cytometric detection of intracellular proteins . Data were acquired by a LSRII Flow Cytometer using FACSDiva software . MEFs and HEK293T cells were prepared for fluorescence microscopy as previously described ( Athanasopoulos et al . , 2010 ) . Images were collected using an Olympus IX71 microscope with DP Controller software ( Olympus ) . CD4+ T cells were isolated from floxed wild-type or Tringless mice by MACS Microbead separation ( Miltenyi Biotec ) . AMPK activity was measured from AMPK complexes immunoprecipitated from cell lysates using anti-AMPKα antibody ( Abcam ) as previously described ( Chen et al . , 2003 ) . Detection of ACC Ser79 phosphorylation levels was also used to measure AMPK allosteric activity . Whole-cell lysates were prepared using TNE lysis buffer ( 1% NP40 , 150 mM NaCl , 20 mM Tris-base , 1 mM EDTA and Roche cOmplete EDTA-free protease inhibitory cocktail tablets all dissolved in water ) . PhosSTOP ( Roche ) was added to the TNE mix for the detection of phospho-residues . To immunoprecipitate proteins , antibody was added to pre-cleared lysates and mixed with Protein G Sepharose 4 Fast Flow ( GE Healthcare ) for 12 hr then washed . For western blotting , lysates were separated by SDS-PAGE , transferred to nitrocellulose membrane , blocked in 5% BSA Tris-buffered saline containing 0 . 05% Tween-20 , probed with primary antibodies and detected with horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibodies . MEFs were seeded on coverslips and prepared as described previously ( Srivastava et al . , 2015 ) . To induce cellular stress , 1 mM arsenite was added to cultures for 1 hr . Stains with primary antibodies were carried out using optimized conditions overnight at 4°C in a humid chamber . The primary antibodies were goat anit-AMPKα1 , clone C20 ( Santa Cruz ) at 1:75 ; with either rabbit anti-ROQUIN at 1:75 ( Cat . NB100-655 , Novus Biologicals ) or rabbit anti-GFP 1:1000 ( Cat . ab6556 , Abcam , UK ) . Images were taken on a Leica SP5 confocal microscope with a pin hole of 67 . 9 μm and an APO CS 1 . 25 UV x40 oil objective . Higher magnification images presented in Figure 3c were taken on a Leica SP5 confocal microscope with a pin-hole of 95 . 5 μm and an HCxPL APO lambda blue x63/1 . 4 oil objective . Mouse UBCH5A ( E2 ) was expressed as a GST-fusion protein and purified using standard protocols . ROQUIN constructs were also expressed as GST-fusion proteins using standard procedures except that 0 . 1 mM Zn-acetate was added to the growth media and all purification buffers . Human E1 ( His6 tagged ) was purchased from Biomol International . Bovine UBIQUITIN was purchased from Sigma-Aldrich . Ubiquitination assays were performed in 20 ml in 20 mM Tris-HCl , 50 mM NaCl , 2 mM MgCl2 , 1 mM ATP , 0 . 1 mM DTT at 25oC . Reactions were stopped by the addition of 2x SDS PAGE loading buffer and heating at 95oC for 5 min and analyzed by SDS-PAGE and Coomassie Blue staining . Typically reactions contained 0 . 1 mM E1 , 10 mM E2 , 50 mM UBIQUITIN , and 0 . 5 mg/mL ROQUIN peptide . Statistics were calculated using Prism 5 . 0a software ( GraphPad ) . Stress granule morphology and PLAs were assessed by ImageJ 1 . 46r software ( NIH ) .	The immune system protects the body from invading microbes like bacteria and viruses . Upon recognizing the presence of these microbes , cells in the immune system are activated to destroy the foreign threat and clear it from the body . A type of immune cell called T follicular helper cells ( or Tfh for short ) are formed during an infection and are essential for coordinating other immune cells to produce high-quality antibody proteins that attack the microbes . Without Tfh cells , life-long production of these protective antibodies is severely crippled , which can cause common variable immune deficiency and other serious immunodeficiency diseases . On the other hand , the body must also avoid generating excessive numbers of Tfh cells , which can lead to the production of antibodies that attack healthy cells of the body . ROQUIN is a protein that inhibits the formation of Tfh cells and other types of active T cells . A region on the protein called the ROQ domain destabilizes particular molecules of ribonucleic acid ( RNA ) that are required for these specialist T cells to form and work properly . ROQUIN belongs to a large family of enzymes that have a so-called RING domain , which is a feature that enables these enzymes to attach tags onto specific target proteins to modify their activity or stability . However , it was not known whether the RING domain of ROQUIN was active . Ramiscal et al . now address this question in mice . Unexpectedly , the experiments show that the RING domain is required to promote the formation of Tfh cells , but not other types of active T cells . This domain allows ROQUIN to repress an enzyme called AMPK , which normally blocks cell growth by regulating cell metabolism . The findings suggest that the different roles of the ROQ and RING domains allow ROQUIN to fine-tune the numbers of Tfh cells so that they remain within a safe range . In the future , these findings may aid the development of vaccines that are more efficient at generating protective Tfh cells to prevent infectious diseases .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2015	Attenuation of AMPK signaling by ROQUIN promotes T follicular helper cell formation
It was recently proposed ( Bushdid et al . , 2014 ) that humans can discriminate between at least a trillion olfactory stimuli . Here we show that this claim is the result of a fragile estimation framework capable of producing nearly any result from the reported data , including values tens of orders of magnitude larger or smaller than the one originally reported in ( Bushdid et al . , 2014 ) . Additionally , the formula used to derive this estimate is well-known to provide an upper bound , not a lower bound as reported . That is to say , the actual claim supported by the calculation is in fact that humans can discriminate at most one trillion olfactory stimuli . We conclude that there is no evidence for the original claim . The first main concern is that the estimated number of discriminable stimuli depends steeply , systematically , and non-asymptotically on choices of arbitrary experimental parameters , among them the number of subjects enrolled , the number of discrimination tests performed , and the threshold for statistical significance . We show below that the order of magnitude claim of ‘one trillion olfactory stimuli’ requires that those parameters assume a very narrow set of values . Certainly , the precise value of an estimate may change as additional data are collected , but the estimate should not change in expectation; it should not be possible to make an estimate arbitrarily large ( or small ) , simply by collecting more ( or less ) data . Similarly , the estimate itself should not become arbitrarily small or large with adjustment of a significance criterion . Estimates that scale systematically with such incidental parameter choices are considered statistically inconsistent ( Figure 1 ) . It is the inconsistency of the present estimate that produces a tremendously large space of extremely different , yet unobjectionable alternative conclusions that can be reached about the number of discriminable olfactory stimuli . 10 . 7554/eLife . 08127 . 003Figure 1 . Consistency of an estimator . An estimator is consistent if the resulting estimate asymptotically converges ( in expectation ) as sample size increases ( black line ) . Uncertainty in the estimate ( gray area ) may shrink with sample size , but the estimate itself should not systematically change with sample size , and should converge on the truth . Estimators without this property are termed inconsistent ( the blue line is a relevant example ) , and are considered unreliable , as the resulting estimate can be heavily biased by the sample size . If the estimate has a minimum and maximum allowed value ( see Equation 1 ) , an especially inconsistent estimator can even produce any estimate within that range . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 00310 . 7554/eLife . 08127 . 004Figure 1—figure supplement 1 . Fraction discriminated at which statistical significance is reached . For each possible value of the number of tests T conducted per mixture class , there is a cumulative distribution of the fraction f of those tests that will be correctly discriminated , under the null hypothesis of chance ( 13 ) responding . The choice of significance threshold α determines the fraction correct required to reject the null hypothesis , and thus count as ‘significantly discriminating’ in the framework . For a given value of α ( 0 . 05 shown here , and used in [Bushdid et al . , 2014] ) , the fraction correctly discriminated required to reach this threshold varies greatly with T . Rejecting the null hypothesis can thus be very easy or very hard depending on T ( or the number of subjects S , not shown ) , or on α . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 004 To illustrate that we can correctly recapitulate the analysis undertaken in ( Bushdid et al . , 2014 ) , Figure 2 shows our reproduction ( using raw supplementary data from [Bushdid et al . , 2014] ) of two critical figures from that paper ( Bushdid et al . , 2014 ) , from which its main conclusion was drawn . See Table 1 for definitions of parameters used here and in ( Bushdid et al . , 2014 ) . Figure 3 and Table 2 quantify the fragility of this conclusion , by generating estimates using the same framework under trivial alternative scenarios in which different numbers of subjects ( or mixtures ) were used , or different choices of statistical threshold ( α ) were used for assessing discriminability . Thus , we produced all values shown here by analyzing the data from ( Bushdid et al . , 2014 ) , using the methods described therein , and varying only parameters . Code to reproduce these and all subsequent analyses is available at http://github . com/rgerkin/trillion , documented at http://nbviewer . ipython . org/github/rgerkin/trillion/blob/master/journal . ipynb . 10 . 7554/eLife . 08127 . 005Figure 2 . Reproduction of the main result published in ( Bushdid et al . , 2014 ) , from analysis of raw data made available in supplemental materials of ( Bushdid et al . , 2014 ) . Compare to Figures 3 , 4 in that publication . ( A ) : Discriminability vs mixture overlap , expressed as a percentage of the mixture size N . From this analysis , ( Bushdid et al . , 2014 ) derives d−NN∼51% ( vertical dashed line ) as the critical value of mixture overlap at which 50% of mixtures achieve ‘significant discriminability’ . ( B ) : Estimated number of discriminable mixtures z vs mixture overlap ( expressed as a percentage of N ) allowing discrimination . The plot is obtained by regression and interpolation of results in A combined with Equation 1 , with colors corresponding to values of N as shown in A . For a value of ∼51%as derived in A , one obtains the ‘trillions’ figure reported in ( Bushdid et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 00510 . 7554/eLife . 08127 . 006Figure 2—figure supplement 1 . Reconstruction of percent correctly discriminated using raw data from ( Bushdid et al . , 2014 ) . This reproduces Figure 2B from ( Bushdid et al . , 2014 ) , and can be subsequently used to reproduce Figure 3A and ultimately Figure 3C from ( Bushdid et al . , 2014 ) . Similar reconstructions , using alternative parameter choices , were used as basis for the findings presented in Figure 3A here . Analogous reconstructions of Figures 2C , 3B , D from ( Bushdid et al . , 2014 ) ( not shown ) were used to generate Figure 3B here . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 00610 . 7554/eLife . 08127 . 010Table 1 . Definitions of parametersDOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 010zEstimated number of discriminable olfactory stimuliCNumber of distinct compounds available to make mixturesNNumber of distinct compounds in a mixtureONumber of distinct compounds shared by a mixture pairDNumber of distinct compounds in one mixture of a pair that are not shared by the other . ( D=N−O ) classAll mixture pairs with the same value of N and D . dThe value of D for which mixture pairs of a given N are more likely than not to be discriminable at a rate significantly above chance . 10 . 7554/eLife . 08127 . 007Figure 3 . The estimation framework supports nearly any alternative conclusion , including the smallest and largest estimates possible under the framework . ( A ) : Heat map showing alternative conclusions reached for different choices of T , the number of mixture pairs per class to test , and application of alternative significance threshold α for discriminability , with the data from ( Bushdid et al . , 2014 ) . Asterisks ( * ) show the parameter regime ( T = 20 mixtures , α=0 . 05 ) used in ( Bushdid et al . , 2014 ) . Other values on each axis are chosen in a geometric progression around those parameters . The contour in the lower right labeled ‘All’ demarcates a regime in which one will conclude that the largest possible number of mixture stimuli ( i . e . , all z ( d=0 ) = ( 12830 ) >1029 of them ) are discriminable ( see Equation 1 ) . The contour in the upper left labeled ‘smallest possible’ demarcates a regime in which one will conclude that the smallest possible number of stimuli are discriminable , that is , only z ( d=N=30 ) <5000 of them . The contour labeled ‘colors’ demarcates a regime in which one concludes that the number of discriminable olfactory stimuli is the same order of magnitude as the number of discriminable colors . ( B ) : Heat map similar to left , only with number of subjects on the vertical axis . A choice of α=0 . 025 is necessary to obtain the estimate that ( Bushdid et al . , 2014 ) reports for this analysis . ( C ) : Colorscale for A and B , with reference landmarks . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 00710 . 7554/eLife . 08127 . 008Figure 3—figure supplement 1 . Steep , systematic , and non-asymptotic dependence of the estimate on sample size ( S or T ) and threshold α for statistical significance . ( A ) Dependence of the estimate ( for mixtures of N = 30 ) on sample size . Black shows dependence on the number of subjects S enrolled in the study , Red shows dependence on the number of mixtures T tested per mixture class . Once the number of mixtures or subjects tested is ∼150 ( by no means an unusually large sample size ) , the conclusion that all possible ( CN ) mixtures are discriminable is guaranteed , in contradiction with experimental results . ( B ) Dependence of the estimate on the significance threshold α with ( red ) and without ( black ) a correction for multiple comparisons . ( Bushdid et al . , 2014 ) did not correct for multiple comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 00810 . 7554/eLife . 08127 . 009Table 2 . Estimates of z , the number of discriminable olfactory stimuli , for different possible parameters values , for the C = 128 , N = 30 case used in ( Bushdid et al . , 2014 ) DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 009A# Discriminable stimuli ( z ) Significance threshold ( α ) # Tests per class ( T ) 2 . 02×10120 . 05204 . 56×103†0 . 0551 . 54×1029‡0 . 051858 . 94×1030 . 001201 . 79×1040 . 0115B# Discriminable stimuli ( z ) Significance threshold ( α ) # Subjects ( S ) 3 . 81×10130 . 025264 . 56×103†0 . 02571 . 54×1029‡0 . 0251353 . 47×1070 . 001262 . 98×1050 . 0115This recapitulates selected points from Figure 3 . * Indicates that the parameter value was used in ( Bushdid et al . , 2014 ) . We assume here that new subjects perform similarly to the original subjects . Note that 4 . 56×103 ( † ) and 1 . 54×1029 ( ‡ ) are the smallest and largest possible values allowed by the framework from ( Bushdid et al . , 2014 ) . In Bushdid et al . , 2014's experimental framework , there are three sets of experiments , varying in the number of distinct molecular components N per mixture tested . We consider the N = 30 case ( without loss of generality ) for which there are ∼1029 possible olfactory stimuli , and for which the smallest possible number of discriminable stimuli is ∼4500 ( see Equation 1 below ) . Figure 3 and Table 2 thus demonstrate that ( 1 ) there is a regime of reasonable parameter choices for which one concludes that all possible olfactory stimuli ( i . e . , all ∼1029 of them ) are discriminable; and ( 2 ) there is another regime of reasonable parameter choices for which one concludes that the smallest possible number of stimuli ( i . e . , only ∼4500 ) are discriminable . The only assumption required to obtain these estimates is that performance in new subjects is similar to performance in the original subjects . The fragility of the conclusion results from the claim in ( Bushdid et al . , 2014 ) that a modest ( if very interesting ) correlation—between the discriminability of a pair of mixtures and the overlap ( fraction of shared components ) of those mixtures—is evidence that a particular degree of mixture overlap defines a boundary that partitions the discriminable from the indiscriminable in a very high-dimensional space . Below , we explore the consequences of this decision , and its implications for calculating the number of discriminable olfactory stimuli . The approach actually used in ( Bushdid et al . , 2014 ) is instead to apply a threshold not to the fraction discriminated ( explored in Figure 4—figure supplement 2 ) , but to the fraction significantly discriminable . In other words , determine for which subjects ( or alternatively , for which classes of mixtures ) the fraction discriminated is significantly greater than 13 , i . e . , for which subjects the null hypothesis of chance discrimination can be rejected . To facilitate visualization of this step , ( Bushdid et al . , 2014 ) re-plotted the summary data ( fraction correctly discriminated ) as fraction significantly discriminable ( Figure 2A ) . This view of the data provides a linear relationship between distance D and the fraction significantly discriminable , which holds across all the values of N tested . The relationship is much steeper than for fraction discriminable ( compare Figure 2 and Figure 4—figure supplement 2 ) because this hypothesis-testing step acts as a strong non-linear threshold that exaggerates otherwise small differences in the data . An arbitrary choice of threshold is required; ( Bushdid et al . , 2014 ) chose a threshold of 50% significantly discriminable , and computed d from the fraction significantly discriminable using linear regression and interpolation . Varying the threshold ( i . e . , 50% ) itself ( not shown ) , would change the computed d ( and consequently z ) , but this is not the largest issue . By introducing a hypothesis-testing step , the d derived from Figure 2 now varies systematically with the number of subjects enrolled in the study ( and the number of mixtures tested ) , and with the choice of significance criterion α . This is because each data point used to compute d becomes the binary result of a hypothesis test , each of which depends critically on sample size and test specificity . Because d is then fed into an expression ( Equation 1 ) that explodes geometrically , the result is a recipe for producing any of a range of estimates for z that one might choose . If one enlists more subjects or slackens the significance criterion , a very large ( even the largest possible ) number will be obtained . If one enlists fewer subjects or makes the significance criterion more strict , a very small ( even the smallest possible ) number will be obtained . Figure 3—figure supplement 1 shows the explicit dependence of the estimate on each of these quantities alone . Naturally , these can be varied in tandem too , with even more dramatic consequences , as described above ( Figure 3 and Table 2 ) . A hypothesis test is meant to assess the strength of evidence for or against a hypothesis ( often against a null hypothesis ) , not to make a point estimate . However , it may not be uncommon for researchers to use hypothesis testing in the manner done in ( Bushdid et al . , 2014 ) —to count the number or fraction of data points exhibiting a certain property . In many cases this may amount to a venial statistical sin with ( hopefully ) benign consequences . But that is unfortunately not the case in ( Bushdid et al . , 2014 ) , due in part to the extremely steep dependence of z on d guaranteed by Equation 1 . If one claims that an estimate is meaningful , it is fair to ask how vigorously would one have to defend a specific choice of arbitrary experimental parameters to defend a particular order-of-magnitude range around that estimate . Unfortunately , the systematic sensitivities exhibited here severely undermine the plausibility and relevance of the estimate reported in ( Bushdid et al . , 2014 ) . Due to these sensitivities , one could pick almost any number of discriminable stimuli in advance , and affirm this number using these or similar data . Ultimately , the absence of a robust d to characterize the data is an insurmountable obstacle for the framework . One might ask: what is the right way to calculate d in order to obtain a robust estimate of the number of discriminable stimuli ? Before heading down this road and devising alternative statistical approaches , it is worth first clearly articulating the assumptions of a framework in which a single variable plays such a special role . Under what conditions is it sensible to expect that plugging a single data-derived number ( d ) into Equation 1 will produce a meaningful estimate of the number of discriminable olfactory stimuli ? To gain some intuition into this , we can ask the analogous question in the simplified visual system example ( Figure 4 ) that was used as the principal motivation for the procedure . The ‘sphere packing’ calculation in this case naturally involves measuring the resolution of perception in terms of the stimulus , but its validity is not a consequence of this measurement alone . Rather , the procedure in Figure 4 is sensible because the thing we are calling an independent stimulus dimension ( wavelength ) is respected as such by perception: we encounter monotonically changing , non-redundant percepts as we move from one extreme of the stimulus space to the other . If we didn't—say , if the same percept ‘blue’ were experienced for several non-overlapping disjoint intervals—the sphere packing formulation would fall apart . We might observe that on average discriminability improves with distance , but this would not be evidence of a characteristic length scale that partitions stimulus pairs into discriminable vs indiscriminable sets . Thus the sphere-packing framework is valid only if the underlying geometry of stimulus space ( that the investigator has designed ) aligns with the geometry of perceptual space ( as implemented in neural circuitry ) . Formally , the map from stimulus space to perceptual space needs to be homeomorphic , or nearly so . See ( Meister , 2015 ) for further insight on this issue . Instead of providing evidence for this homeomorphism , it was assumed in ( Bushdid et al . , 2014 ) for the purposes of calculation that each component of the molecular library ( of size C = 128 in [Bushdid et al . , 2014] ) spanned an informative additional dimension for perception to explore: each molecule in the library is treated as an olfactory primary that is independent of all the others . This is the assumption , codified in the numerator of Equation 1 , that allows for a massive space of potential discriminable stimuli . Indeed , the guaranteed runaway growth of the numerator as molecules are added to the C-sized library was offered in ( Bushdid et al . , 2014 ) as an argument for why the reported ‘trillion’ figure is an underestimate—after all , C could always be higher . It is worthwhile to quantify the behavior of the estimate as C changes . First , the estimate depends geometrically on C , with a power law exponent of ∼30 ( Figure 5 , blue line ) . In other words , if the chemical library were doubled , the estimate z would increase by a factor of 230 under constant performance . If the component library were increased to the size of a standard flavor and fragrance catalog ( ∼2000 chemicals ) , the estimate would increase to z∼1041 , implying a unique olfactory percept for each carbon atom on earth . 10 . 7554/eLife . 08127 . 014Figure 5 . Explosive growth of the estimate z on the size ( C ) of the molecular library . The number of possible stimuli z that can be assembled by choosing N = 30distinct molecules from a library of size C increases geometrically with C ( black line ) . If a library of a different size had been used , and similar subject performance resulted , the estimated number of discriminable stimuli z would grow along a similar trajectory ( blue line ) . Even if performance deteriorated as C increased , the estimate could never fall below the red line , which represents worst-case performance ( d = N ) . This results from the combinatorial explosion inherent in Equation 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 014 Subjects' performance could become worse when mixtures are drawn from this larger , more complete library , and we acknowledge that we cannot know in advance what the newly calculated resolution d would be on the new stimulus space . In other words , as the numerator of Equation 1 increased , its denominator ( given by Equation 2 ) might conveniently grow proportionally . Let us therefore assume that with a library of sufficient size , so many mixtures become indiscriminable that the resolution becomes as poor as the framework allows , with d = N . Even in this edge case , if only mixtures differing in all components were ‘just discriminable’ , we would still calculate 1021 discriminable stimuli . If C is increased to 106 , the smallest possible number of discriminable percepts ( under the assumption of worst measurable performance , as above ) is 1061 , or 10 million trillion unique olfactory percepts for every carbon atom on earth ( Figure 5 , red line ) . One may object that the inflation of C here is an unfair critique , as the perceptual redundancy of molecules must at some point provide an important constraint on the size of the artificially constructed stimulus space . Indeed , it has been reported that as few as thirty components are required to imbue most mixtures with a common smell , even when there is no component overlap between the mixtures ( Weiss et al . , 2012 ) . But this is the essence of the problem with Equation 1: where does that point lie , and why wasn't the constraint important to consider for the original C = 128 molecular library ? Even if one takes the estimate of d to be unimpeachable , the formula used to derive z does not provide a lower bound as reported in ( Bushdid et al . , 2014 ) . This much is suggested by the worst-case behavior of Equation 1 as C grows . After all , worst case behavior should correspond to z = 1 . If one cannot discriminate anything ( maximal d ) , then there is only one percept . Examining Equation 1 more closely , we see that it is a variant of the so-called Hamming bound for constant weight codes ( MacWilliams and Sloane , 1977 ) . which is well-known to be an upper bound for an identically formulated problem in the theory of error-correcting codes . It is , as suggested in ( Bushdid et al . , 2014 ) , an estimate derived from a hypothetical sphere-packing approach to filling the stimulus space , but it is the largest possible value for the correct answer , not the smallest . Hence , according to the Hamming bound , for d=N=30 the upper bound on the number of discriminable stimuli is 4561 , and we know the correct answer to be 1 ( or 4 , depending on conventions , see the Supplemental Materials ) . Since the upper bound exceeds the correct answer , Equation 1 , while not particularly tight as an upper bound , is nonetheless not wrong , so long as we acknowledge that it is an upper and not a lower bound . The same applies for all other values of d , including the one derived from the data in ( Bushdid et al . , 2014 ) . Thus Equation 1 , as used in ( Bushdid et al . , 2014 ) , provides no insight into the lower bound for z , with a lower bound being required to overturn conventional wisdom about the number of discriminable stimuli . Instead , to obtain a lower bound one must dispense with the factor of 2 in Equation 1 , yielding Levenshtein's constant weight version of the so-called Gilbert-Varshamov bound for error-correcting codes ( [Levenshtein , 1971; MacWilliams and Sloane , 1977; Jiang and Vardy , 2004] , see Supplemental Materials ) . A plot of the lower bound obtained in this manner is shown in Figure 6B , along with the reconstructed upper bounds from ( Bushdid et al . , 2014 ) a , showing the true bounded interval for z . Intuitively , this corrected lower bound reaches z = 1 for worst-case d , implying sensibly that anosmics cannot discriminate any stimuli . In contrast , the upper bound ( reported as a lower bound in 1 ) is on the order of several thousand for worst case d , showing that it cannot be a lower bound d; this can also be confirmed in Figure 4 of ( Bushdid et al . , 2014 ) . 10 . 7554/eLife . 08127 . 015Figure 6 . Upper and lower bounds of the number of discriminable stimuli . ( A ) : Number of discriminable olfactory stimuli as a function of the estimated difference limen ( the fractional mixture overlap allowing discrimination ) . This is simply the behavior of Equation 1 as a function of d , for the three values of N used in ( Bushdid et al . , 2014 ) ; the red dot ( in both A and C ) corresponds to the value reported in ( Bushdid et al . , 2014 ) . The smallest possible estimate ( thousands of stimuli ) is indicated by the dotted line running the length of the abscissa ( note also the y-intercept ) . As described in the text and in the supplement , this graph in fact shows the behavior of the upper bound ( the so-called Hamming bound ) for the mathematical problem of sphere packing . Compare with Figure 3D in ( Bushdid et al . , 2014 ) . ( B ) : Same plot as in A , only using the lower-bound for the same calculation . ( C ) : Upper and lower bounds of the sphere packing problem for the N = 30case ( green lines from A and B , respectively . The dark gray bar shows the range of defensible estimates under the sphere-packing framework , using the d calculated in ( Bushdid et al . , 2014 ) . Using that d , the number of discriminable stimuli may be as small as ∼10 , 000 , and is guaranteed to be no larger than ∼1 trillion . Since the estimate of d is also fragile ( Figure 3 ) , the data may in fact support any value in the shaded gray area . DOI: http://dx . doi . org/10 . 7554/eLife . 08127 . 015 If one is seeking a conservative estimate of the number of discriminable stimuli in a perceptual space whose organization and intrinsic dimensionality are poorly understood , it is arguably more appropriate to use a model that accounts for the data with the smallest number of dimensions . The massive estimates possible in the framework of ( Bushdid et al . , 2014 ) are an immediate consequence of a definition of dimensionality driven by experimenter designation , not data . We therefore propose an alternative framework: use experimental data to create a working map of the perceptual space , and then apply the sphere-packing framework to that map , rather than to a map of the stimulus space . In cognitive science , psychometrics , and marketing , subject responses to stimuli are used to create maps of the underlying perceptual ( or conceptual ) representations of those stimuli . These maps are characterized by the attribute that pairs of items which are considered intuitively to be perceptually near ( rated similar or difficult to discriminate ) are nearer to one another on the map than pairs of items which are perceptually more distant ( rated dissimilar or easy to discriminate ) . There are many algorithms for generating such maps , many of which have been used before in olfaction , including variants of PCA ( Zarzo and Stanton , 2006; Khan et al . , 2007; Koulakov et al . , 2011 ) , non-negative matrix factorization ( NMF , [Castro et al . , 2013] ) , and multi-dimensional scaling ( Mamlouk et al . , 2003 ) . While there are open questions in the generation of these maps ( e . g . , how many dimensions should they have ? ) , they all have the virtue that their accuracy can be checked ( e . g . , by examining the correlation between subjects' indications of item pair dissimilarity and the distance between that pair on the map ) , and thus the maps can be improved . Developing these maps may also have the collateral benefit of revealing stimulus dimensions intrinsic to olfaction ( if any ) , which could constrain the experimental choice of a resolution to measure . Unfortunately , it is difficult if not impossible to create these maps from the data discussed here , because each mixture of a tested pair is used only once in ( Bushdid et al . , 2014 ) , in that pair alone , and never in any other pairs . Thus , there are no serial comparisons of the same mixture that could be used to anchor a stimulus on the map relative to a stimulus against which it was not directly compared experimentally . Thus , there is no way to compute distances between stimuli that do not appear together in a tested pair . In other words , the structure of the perceptual space is severely under-determined by the data . In future experiments such serial repetition of already-tested mixtures would be required to build up a data set to which the proposed method could be applied .	Scientists are interested in the number of colors , sounds and smells we can distinguish because this information can shed light onto how our brains process these senses both in health and disease . It is relatively straightforward to determine how many colors we can see or sounds we can hear because these stimuli are well defined by physical properties such as wavelength . We know the range of wavelengths that the eye can see or the ear can hear , and we can also understand how two such stimuli ( e . g . , red and blue ) are arranged perceptually ( think of a color wheel ) . It is harder , however , to do the same for smell because most ‘olfactory stimuli’ consist of mixtures of different odor molecules . Moreover , we understand much less about how olfactory stimuli are arranged perceptually . In 2014 researchers at Rockefeller University reported that humans can distinguish more than one trillion smells from one another . To calculate this number the researchers tested the ability of human subjects to discriminate between mixtures of different odor molecules . Each mixture consisted of 10 , 20 or 30 molecules selected from a chemical library of 128 different odor molecules . Since each mixture of 10 molecules could contain any 10 of the 128 molecules , more than 200 trillion combinations were possible; the number of possible combinations for the 20- and 30-molecule mixtures were even higher . The aim of the experiment was to identify—by sampling from this very large number of combinations—the number of molecules that two mixtures could have in common and still be distinguishable to the typical person . The Rockefeller team used this number and a geometrical analogy to conclude that humans could discriminate at least 1 . 72 trillion odors , which was much higher than expected from previous reports and anecdotes . Now Gerkin and Castro report that the claims made in the Rockefeller study are unsupported because of flaws in the design of the analytical framework used to make sense of the data . In particular , Gerkin and Castro report that the results are extremely sensitive to some parameters of the experimental and analytical design , such as the number of subjects tested , whereas the results of a robust analysis would not be so sensitive to such factors . By modestly varying any of these parameters it is possible to obtain almost any value for the number of smells that can be discriminated . Moreover , the geometrical analogy used set an upper bound on the final answer , rather than a lower bound: in other words , even assuming that the rest of the analysis was robust , the result should have been that humans can discriminate ‘no more than’ 1 . 72 trillion smells rather than ‘at least’ . In a separate paper Meister also reports that the 1 . 72 trillion smells claim is unjustified .	[ "Abstract", "Introduction", "Building", "the", "stimulus", "space" ]	[ "neuroscience" ]	2015	The number of olfactory stimuli that humans can discriminate is still unknown
Signals delivered by costimulatory molecules are implicated in driving T cell expansion . The requirements for these signals , however , vary from dispensable to essential in different infections . We examined the underlying mechanisms of this differential T cell costimulation dependence and found that the viral context determined the dependence on CD28/B7-mediated costimulation for expansion of naive and memory CD8+ T cells , indicating that the requirement for costimulatory signals is not imprinted . Notably , related to the high-level costimulatory molecule expression induced by lymphocytic choriomeningitis virus ( LCMV ) , CD28/B7-mediated costimulation was dispensable for accumulation of LCMV-specific CD8+ T cells because of redundancy with the costimulatory pathways induced by TNF receptor family members ( i . e . , CD27 , OX40 , and 4-1BB ) . Type I IFN signaling in viral-specific CD8+ T cells is slightly redundant with costimulatory signals . These results highlight that pathogen-specific conditions differentially and uniquely dictate the utilization of costimulatory pathways allowing shaping of effector and memory antigen-specific CD8+ T cell responses . CD8+ T cells are critical for elimination of various intracellular pathogens . By incorporating differences in TCR signal strength and duration ( signal 1 ) , the spatiotemporal availability of costimulatory molecules ( signal 2 ) and defined cytokines in the inflammatory environment ( signal 3 ) , CD8+ T cells are differentially programmed for expansion and effector cell formation resulting in considerable plasticity of the response ( Williams and Bevan , 2007; Arens and Schoenberger , 2010 ) . Costimulatory molecules augment TCR triggering but also qualitatively contribute to achieve optimal T cell expansion and differentiation ( Croft , 2003 ) . CD28 is considered as the most prominent costimulatory receptor for T cells , but signals provided by members of the TNF receptor ( TNFR ) super family such as CD27 , OX40 ( CD134 ) and 4-1BB ( CD137 ) are known to provide crucial signals as well . T cell responses seem to be differentially and contextually dependent on costimulatory interactions but the underlying mechanisms are unknown ( DeBenedette et al . , 1999; Welten et al . , 2013a; Wortzman et al . , 2013 ) . For example , the pathogen-specific CD8+ T cell response during vesicular stomatitis virus and vaccinia virus ( VV ) infection is highly driven by interactions between CD28 and the B7 molecules B7 . 1 ( CD80 ) and B7 . 2 ( CD86 ) ( Sigal et al . , 1998; Bertram et al . , 2002; Fuse et al . , 2008 ) , while in lymphocytic choriomeningitis virus ( LCMV ) infection the viral-specific CD8+ T cells seem to bypass the requirements of the CD28/B7 costimulatory pathway for primary effector T cell expansion ( Shahinian et al . , 1993; Kundig et al . , 1996; Andreasen et al . , 2000; Grujic et al . , 2010; Eberlein et al . , 2012 ) . Even within a single infection distinct requirements for costimulatory signals can be observed . In mouse cytomegalovirus ( MCMV ) , the classical ( non-inflationary ) CD8+ T cell responses are more dependent on the CD28/B7 costimulatory pathway than the so-called inflationary CD8+ T cells , which gradually accumulate at high frequencies in time ( Arens et al . , 2011b; O'Hara et al . , 2012 ) . Here we examined the mechanisms of CD8+ T cell costimulation dependency . We found that the pathogen-induced environment and not the characteristics of the viral epitopes determined the requirements of naive and of memory CD8+ T cells for CD28/B7-mediated costimulation . Remarkably , related to the induction of high costimulatory ligand expression , LCMV-specific CD8+ T cell expansion can operate in a CD28/B7 independent fashion because of redundancy with the costimulatory members of the TNFR superfamily . Furthermore , direct type I IFN signaling in viral-specific CD8+ T cells is slightly redundant with CD28/B7 and CD27/CD70-mediated costimulation . These findings demonstrate that the inflammatory environment dictates the characteristics of CD8+ T cell responses by allowing a differential utilization of stimulatory pathways . Effector CD8+ T cell formation during LCMV infection seems not to be driven by the main costimulatory CD28/B7 pathway because wild-type ( WT ) mice and mice deficient in both B7 . 1 and B7 . 2 ( Cd80/86−/− ) mount similar antigen-specific responses in magnitude , and this phenomenon is apparent after both high and low viral inoculum dosages ( Figure 1A ) . In contrast , during infection with VV or Listeria monocytogenes ( LM ) , antigen-specific CD8+ T cell responses are highly reduced in the absence of B7-mediated costimulation ( Figure 1B , C ) . CD8+ T cell responses against MCMV are dependent on B7-mediated costimulation as well , ranging from ∼sevenfold diminished responses in case of the non-inflationary M45 and M57-specific to ∼2 . 5-fold in case of the inflationary m139 and M38-specific responses ( Figure 1D ) . Effector cell differentiation of virus-specific CD8+ T cells , indicated by the downregulation of CD62L and upregulation of CD44 , also required B7-mediated costimulation in MCMV but not in LCMV infection ( Figure 1—figure supplement 1 ) . Thus , in various infections but not during LCMV infection the CD28/B7 costimulatory pathway is highly critical in driving T cell expansion . 10 . 7554/eLife . 07486 . 003Figure 1 . Differential requirements for CD28/B7-mediated costimulation in driving pathogen-specific CD8+ T cell expansion . ( A ) Wild-type ( WT ) and Cd80/86−/− mice were infected with 2 × 102 ( low dose ) or 2 × 105 ( high dose ) PFU LCMV-Armstrong . The lymphocytic choriomeningitis virus ( LCMV ) -specific CD8+ T cell response in the spleen was determined 7 days post-infection . Representative flow cytometric plots show CD3+/CD8+ cells that were stained with CD44 antibodies and MHC class I tetramers ( high dose infection ) . Percentages indicate tetramer+ cells within the CD8+ T cell population . Bar graph shows total number of splenic LCMV-specific CD8+ T cells . ( B ) Mice were infected with 2 × 105 PFU vaccinia virus ( VV ) WR and the percentage of tetramer+ cells within the CD8+ T cell population was determined in the blood 7 days post-infection . ( C ) The percentage of tetramer+ cells within the CD8+ T cell population was determined in the blood 7 days post-infection with 1 × 106 CFU LM-Quadvac . ( D ) Flow cytometric plots show a representative M45-specific tetramer staining of cells from WT and Cd80/86−/− mice at day 8 post-infection with 1 × 104 PFU mouse cytomegalovirus ( MCMV ) . Cells are gated on CD3+/CD8+ and the percentages indicate tetramer+ cells within the CD3+/CD8+ T cell population . Bar graph indicates the total number of splenic MCMV-specific CD8+ T cells . Data in bar graphs are expressed as mean + standard error of the mean ( SEM ) ( n = 5–12 mice per group ) of at least two independent experiments . Fold difference and significance ( p < 0 . 05 ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 00310 . 7554/eLife . 07486 . 004Figure 1—figure supplement 1 . Costimulatory signals program effector cell differentiation of MCMV-specific but not of LCMV-specific CD8+ T cells . WT and Cd80/86−/− mice were infected with either 2 × 105 PFU LCMV Armstrong or 1 × 104 PFU MCMV-Smith . ( A ) Representative flow cytometric plots show cell surface expression of CD44 and CD62L on total CD8+ T cells ( black ) and on GP33 and M45-specific CD8+ T cells ( blue ) 7 days post LCMV and 8 days post MCMV infection . ( B ) Bar graph indicates the percentage of CD44+/CD62L− within the MHC class I tetramer+ population , 7 days post-LCMV infection . ( C ) The percentage of CD44+/CD62L− within the tetramer+ population was determined 8 days post MCMV infection . Data in bar graphs are expressed as mean + SEM ( n = 4–8 mice per group; p < 0 . 05 ) of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 004 Next , we examined if additional triggering of the CD28/B7 costimulatory pathway is able to differentially modulate effector T cell formation . Therefore , the co-inhibitory receptor CTLA-4 that binds to B7 . 1 and B7 . 2 was blocked with antibodies during infection , which increases the availability of the B7 molecules to stimulate CD28 . Remarkably , CTLA-4 blockade during LCMV infection had no effect on T cell expansion , indicating that LCMV-specific CD8+ T cells are rather indifferent to enhanced B7-mediated signals ( Figure 2A , B ) . However , CTLA-4 blockade during MCMV infection augmented MCMV-specific CD8+ T cell responses ∼threefold in a B7-dependent manner ( Figure 2C , D ) . Thus , additional triggering of the CD28/B7 pathway is beneficial in settings in which T cell expansion is dependent on this pathway , while the enhancement of CD28/B7-mediated costimulation had no effect in conditions in which the B7 costimulatory molecules are not essential for initial T cell expansion . 10 . 7554/eLife . 07486 . 005Figure 2 . CTLA-4 blockade impacts B7-driven CD8+ T cell responses . ( A ) CTLA-4 blocking antibodies were administrated during infection with 2 × 105 PFU LCMV Armstrong in WT mice . At day 7 post-infection , the splenic LCMV-specific response was analyzed by intracellular cytokine staining . Representative flow cytometric plots show intracellular IFN-γ vs cell-surface CD8 staining after restimulation with GP33-41 peptide . The percentage of IFN-γ+ cells within the CD8+ T cell population is indicated . ( B ) Total numbers of splenic LCMV-specific CD8+ T cells are shown . ( C ) CTLA-4 interactions were abrogated by administration of blocking antibodies in WT and Cd80/86−/− mice upon infection with 1 × 104 PFU MCMV , and at day 8 post-infection the virus-specific response was analyzed by intracellular cytokine staining . Representative flow cytometric plots show intracellular IFN-γ vs CD8 staining after restimulation of splenocytes with M45985-993 peptide . The percentage of IFN-γ+ cells within the CD8+ T cell population is indicated . ( D ) Total numbers of MCMV-specific CD8+ T cells in the spleen are shown . Data in bar graphs are expressed as mean + SEM ( n = 4–5 mice per group ) of at least two independent experiments . Fold difference and significance ( p < 0 . 05 ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 005 To determine whether the characteristics of LCMV-specific epitopes define the B7-independent activation of CD8+ T cell responses , we analyzed the response to the immunodominant epitope GP33-41 of LCMV ( GP33 ) in the context of different pathogen infections . Therefore , recombinant MCMVs were generated in which the GP33 epitope was expressed within the immediate early 2 ( IE2 ) protein ( MCMV-IE2-GP33 ) or the M45 protein ( MCMV-M45-GP33 ) . The in vitro replication kinetics of MCMV-IE2-GP33 and MCMV-M45-GP33 were similar as WT virus ( Figure 3—figure supplement 1A ) . Correspondingly , in vivo infection with MCMV-IE2-GP33 induced a GP33-specific response with inflationary characteristics , as specified by a gradual increasing GP33-specific CD8+ T cell response in time with an effector memory phenotype ( Figure 3—figure supplement 1B , C ) . As determined by intracellular IFN-γ staining after restimulation ( Figure 3A , B ) or direct staining with MHC class I tetramers ( data not shown ) , the GP33-specific CD8+ T cell response elicited by both MCMV-IE2-GP33 and MCMV-M45-GP33 was dependent on B7-mediated costimulation , albeit to a higher degree when the GP33 epitope was inserted within the M45 protein . Infection with an MCMV containing the model epitope OVA257-264 ( SIINFEKL ) inserted in the M45 protein ( MCMV-M45-SIINFEKL ) resulted also in an antigen-specific T cell response that depended on B7-mediated costimulation ( Figure 3C ) , indicating that non-viral epitopes elicit similar costimulation dependent responses . Furthermore , LM expressing the LCMV GP33 epitope ( LM-GP33 ) induced GP33-specific CD8+ T cell responses that were highly dependent on B7-mediated costimulation ( Figure 3D ) . Also , upon vaccination with a synthetic long peptide ( SLP ) containing the GP33 epitope , Cd80/86−/− mice mounted a defective GP33-specific CD8+ T cell response in comparison with WT mice ( Figure 3—figure supplement 2 ) . To exclude possible effects related to differences in the TCR repertoire selection , TCR transgenic CD8+ T cells recognizing the LCMV GP33 epitope ( referred hereafter as P14 cells ) were used in different pathogenic contexts . Similar as observed for the endogenous LCMV-specific CD8+ T cell expansion , B7-mediated costimulation was dispensable for P14 cell expansion in LCMV infection . Importantly , for the expansion of P14 cells in MCMV-IE2-GP33 and LM-GP33 infection , B7-mediated signals were highly required ( Figure 3E ) , which corroborates that the inflammatory environment is predominantly determining the costimulatory requirements . Together , these data indicate that the context of viral epitope expression , rather than the intrinsic nature of the epitope or the antigen-specific CD8+ T cell population , influences the dependence on B7-mediated signals for T cell expansion . 10 . 7554/eLife . 07486 . 006Figure 3 . The context of viral epitope expression determines the requirements for B7-mediated costimulation in driving antigen-specific CD8+ T cell expansion . ( A , B ) WT and Cd80/86−/− mice were infected with 1 × 105 PFU MCMV-IE2-GP33 or MCMV-M45-GP33 , and 8 days post-infection the splenic GP33-specific CD8+ T cell response was determined by intracellular IFN-γ staining . Representative flow cytometric plots are shown and the percentage of IFN-γ+ cells within the CD8+ T cell population is indicated . Graphs indicate the total number of splenic GP33-specific CD8+ T cells . ( C ) The splenic SIINFKEL-specific CD8+ T cell response was determined by intracellular IFN-γ staining at day 8 post-infection with 1 × 105 PFU MCMV-M45-SIINFEKL . ( D ) WT and Cd80/86−/− mice were infected with 1 . 5 × 103 CFU LM-GP33 . At day 7 post-infection the splenic GP33-specific response was analyzed by intracellular IFN-γ staining upon restimulation with GP33 peptide . Representative flow cytometric plots are shown . The percentage of GP33-specific CD8+ T cells within the total CD8+ T cell population is indicated . Bar graph shows the total number of splenic GP33-specific CD8+ T cells . ( E ) 5 × 104 CD90 . 1+ Ifnar1+/+ P14 cells were adoptively transferred into WT and Cd80/86−/− mice that were subsequently infected with 2 × 105 PFU LCMV Armstrong , 1 × 105 PFU MCMV-IE2-GP33 or 1 . 5 × 103 CFU LM-GP33 . At day 7 ( LCMV , LM-GP33 ) or 8 ( MCMV-IE2-GP33 ) post-infection , the magnitude of the P14 cell response in the spleen was determined . Data in bar graphs are expressed as mean + SEM ( n = 3–5 mice per group ) of at least two independent experiments . Fold difference and significance ( p < 0 . 05 ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 00610 . 7554/eLife . 07486 . 007Figure 3—figure supplement 1 . Characteristics of recombinant MCMVs . ( A ) Replication kinetics of MCMV-IE2-GP33 , MCMV-M45-GP33 and MCMV-M45-SIINFEKL on NIH 3T3 cells . Monolayers of 3T3 cells were infected with an MOI of 0 . 1 . ( B ) WT mice were infected with 1 × 105 PFU MCMV-IE2-GP33 and the kinetics of the GP33-specific response was followed in the blood in time . ( C ) KLRG1 expression of total CD8+ T cells ( black ) and GP33+ CD8+ T cells ( blue ) is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 00710 . 7554/eLife . 07486 . 008Figure 3—figure supplement 2 . GP33-SLP vaccination is dependent on B7-mediated costimulation . WT and Cd80/86−/− mice were vaccinated s . c . at the tail base with 75 µg synthetic long peptide ( SLP ) containing the LCMV-GP33 epitope combined with 20 µg CpG in PBS . The GP33-specific CD8+ T cell response in the spleen was determined 7 days post-vaccination by intracellular cytokine staining after in vitro restimulation with GP33-41 peptide . The percentage of GP33-specific CD8+ T cells within the total CD8+ T cell population is indicated . Graph shows the total number of splenic GP33-specific CD8+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 008 Next , we examined whether the viral context imprints the costimulatory requirements for the lifespan of T cells or if memory T cells undergo secondary expansion independently of the priming context . Therefore , crisscross adoptive transfer experiments were performed with memory GP33-specific CD8+ T cells generated in different viral environments . First , memory GP33-specific CD8+ T cells were primed in the context of an LCMV infection and adoptively transferred into WT or Cd80/86−/− hosts that were subsequently infected with either LCMV or MCMV-IE2-GP33 . Comparing the GP33-specific CD8+ T cell responses upon antigenic re-challenge revealed a dispensable role for B7-mediated signals for secondary T cell expansion in an LCMV environment , but a strong requirement for these signals in an MCMV context even though these cells were primed in an LCMV environment ( Figure 4A ) . Importantly , when memory GP33-specific CD8+ T cells that depended on B7-mediated signals during priming in MCMV infection were transferred and re-challenged in an LCMV or an MCMV environment , the viral context during secondary expansion again determined the requirement on costimulatory signals ( Figure 4B ) . Together , these data indicate that during secondary T cell responses the viral context is dominant and determines the CD28/B7 costimulation dependency of virus-specific CD8+ T cells independent of the priming context . 10 . 7554/eLife . 07486 . 009Figure 4 . The infectious pathogen during antigenic re-challenge determines the requirements for CD28/B7-mediated costimulation for secondary expansion . ( A ) Experimental setup: CD45 . 1+ WT mice were infected with 2 × 105 PFU LCMV Armstrong . After 4 months GP33-specific memory CD8+ T cells were sorted and 2 × 103 cells were adoptively transferred into CD45 . 2+ WT and Cd80/86−/− mice that were subsequently infected with 2 × 105 PFU LCMV Armstrong or 1 × 105 PFU MCMV-IE2-GP33 . The total number of transferred GP33-specific CD8+ T cells was determined 6 days post re-challenge . ( B ) Similar experimental setup as described in ( A ) , except CD45 . 1+ WT mice were infected with 1 × 105 PFU MCMV-IE2-GP33 . Data in bar graphs are expressed as mean + SEM ( n = 5 mice per group ) . Fold difference and significance ( p < 0 . 05 ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 009 As the context of a viral infection determines the dependence on CD28/B7-mediated costimulation for CD8+ T cell expansion , we compared the overall composition of inflammatory mediators in LCMV and MCMV infection . Expression of the inflammation-associated cytokines IL-2 , IL-3 , IL-13 , IL-17 , GM-CSF , and TNF was not enhanced in both infections at early time points compared to naive mice ( data not shown ) . In contrast , serum levels of IFNα were particularly high in LCMV infected mice compared to the serum levels in MCMV infected mice ( Figure 5A ) . Consistent with this , at 24 hr LCMV also induced higher expression of pro-inflammatory cytokines , which have been described to be downstream of type I IFN signaling ( i . e . , Rantes , IL-6 , KC , Mip-1β and MCP-1 ) ( Teijaro et al . , 2013 ) . However , after 48 hr the concentrations of these cytokines were comparable ( Figure 5B ) . Thus , a divergent pro-inflammatory environment is induced early upon LCMV and MCMV infections . 10 . 7554/eLife . 07486 . 010Figure 5 . Influence of type I IFN signaling on the requirement of CD28/B7-mediated costimulation . WT mice were infected with 1 × 104 PFU MCMV-Smith or 2 × 105 PFU LCMV Armstrong and at indicated times post-infection serum was collected . ( A ) Levels of IFNα in serum in time are shown ( bd = below detection limit ) . ( B ) Concentrations of different pro-inflammatory cytokines as determined 24 and 48 hr post-infection . ( C ) Type I interferon receptor ( IFNAR ) blocking antibodies were administrated during LCMV infection in WT and Cd80/86−/− mice . The magnitude of the virus-specific CD8+ T cell response determined by MHC class I tetramer binding at day 7 post-infection is shown . Fold difference and significance ( p < 0 . 05 ) is indicated . ( D ) IFNα levels in serum are shown 3 days post LCMV infection . ( E ) Experimental setup: 5 × 104 CD90 . 1+ Ifnar1+/+ and Ifnar1−/− P14 cells were adoptively transferred in WT and Cd80/86−/− mice that were subsequently infected with 2 × 105 PFU LCMV Armstrong . 7 days post-infection the total numbers of splenic P14 cells was determined . Representative flow cytometric plots show gated CD3+/CD8+ T cells stained for cell surface expression of CD90 . 1 and Vα2 . Fold difference and statistical significance ( p < 0 . 05 ) between groups is indicated in the bar graphs . ( F ) Similar setup as in ( E ) except mice were infected with 1 × 105 PFU MCMV-IE2-GP33 . In addition , on day 1 and 2 , half of the mice received 1 × 105 units IFNα . 8 days post-infection the magnitude of the P14 cells in the spleen was determined . Representative flow cytometric plots show gated CD3+/CD8+ cells stained for cell surface expression of CD90 . 1 and Vα2 . Bar graph shows total number of P14 cells in WT and Cd80/86−/− mice , and fold difference and statistical significance ( p < 0 . 05 ) between groups is indicated . ( G ) Mice were vaccinated with 75 µg SLP containing the GP33 epitope in PBS . 1 × 105 units IFNα was administrated after 18 and 48 hr . At day 7 post-vaccination , GP33-specific CD8+ T cell responses were analyzed . Significance between groups is indicated ( p < 0 . 05 ) . ( H ) Experimental setup: WT mice were infected with 2 × 105 PFU LCMV Armstrong and 2 days post-infection serum was collected and transferred to mice that were infected 1 day prior with 1 × 104 PFU MCMV . The MCMV-specific CD8+ T cell response was determined 8 days post-infection by MHC class I tetramer binding . ( I ) WT and Cd80/86−/− mice were co-infected with 2 × 105 PFU LCMV Armstrong and 1 × 104 PFU MCMV , and virus-specific responses were analyzed 7 days post-infection by MHC class I tetramer binding . Fold difference and significance ( p < 0 . 05 ) is indicated . Data in all bar graphs are expressed as mean + SEM ( n = 4–8 mice per group ) of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 01010 . 7554/eLife . 07486 . 011Figure 5—figure supplement 1 . Recombinant type I IFN is functional in vitro and in vivo . ( A ) L929 cells were pre-incubated with different concentrations of recombinant IFNα2 before addition of mengovirus . After 2 days of incubation , the survival of L929 cells was determined using the MTT assay . One unit was defined as the concentration at which 50% of the cells survived . ( B ) WT mice received 1 × 105 units of IFNα2 via i . p . injection . The expression of CD69 on splenic T cells ( CD4+ and CD8+ ) and B cells ( CD19+ ) was determined 18 hr after IFNα2 administration . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 011 To determine whether the high type I IFN levels that are induced during LCMV infection substitute the CD28/B7 costimulation promoting CD8+ T cell expansion , we investigated the relationship between type I IFN signaling and B7-mediated costimulation in driving LCMV-specific CD8+ T cell expansion . Blocking antibodies for the type I IFN receptor ( IFNAR ) were administered during LCMV infection and resulted in severely diminished LCMV-specific CD8+ T cell responses in WT mice ( Figure 5C ) . IFNAR blocking antibodies administrated in Cd80/86−/− mice also severely hampered LCMV-specific responses ( Figure 5C ) . Notably , the LCMV-specific CD8+ T cell responses in WT mice with abrogated IFNAR signaling were comparable to those in IFNAR blocked Cd80/86−/− mice . Furthermore , no differences in IFNα levels were detected between WT and Cd80/86−/− mice ( Figure 5D ) . Thus , the necessity for IFNAR signaling in the induction of LCMV-specific CD8+ T cell responses does not change in the absence or presence of CD28/B7-mediated costimulation . To examine direct effects of type I IFN-mediated signaling on CD8+ T cell expansion , Ifnar1+/+ and Ifnar1−/− P14 cells were adoptively transferred in WT and costimulation deficient mice that were subsequently infected with LCMV . Ifnar1−/− P14 cells transferred to WT recipients were severely hampered in expansion compared to Ifnar1+/+ P14 cells ( Figure 5E ) , which is consistent with previous reports ( Kolumam et al . , 2005; Aichele et al . , 2006; Wiesel et al . , 2012; Crouse et al . , 2014; Xu et al . , 2014 ) and confirms that type I IFNs drive directly LCMV-specific CD8+ T cell expansion . Ifnar1+/+ P14 cells in Cd80/86−/− mice expanded vigorously and comparable to WT host mice . Importantly , Ifnar1−/− P14 cells failed to expand in Cd80/86−/− mice as well and showed a slightly weaker expansion potential as Ifnar1−/− P14 cells in WT mice ( Figure 5E ) . These data show that type I IFNs act directly on LCMV-specific CD8+ T cells , and that in the absence of this signal 3 cytokine the non-dependence of B7-mediated costimulation in driving LCMV-specific T cell expansion is to some extent altered , indicating that type I IFN signaling in expanding CD8+ T cells is slightly redundant with B7-mediated costimulation signals . Next , we examined the relationship between type I IFN signaling and the B7-mediated pathway during MCMV infection . First we tested whether MCMV-specific CD8+ T cell responses , which are driven by B7-mediated signals , are influenced by the type I IFN pathway . Adoptive transfer of Ifnar1+/+ and Ifnar1−/− P14 cells in WT mice that were subsequently infected with MCMV-IE2-GP33 resulted in profound expansion of the Ifnar1+/+ P14 cells but also of Ifnar1−/− P14 cells , although slightly diminished compared to Ifnar1+/+ P14 cells . Adoptive transfer of P14 cells in Cd80/86−/− mice resulted in hampered expansion of Ifnar1+/+ and even more so of Ifnar1−/− P14 cells , indicating that CD8+ T cells that develop during MCMV infection are to a small degree affected by type I IFN signaling ( in a somewhat redundant manner with B7-mediated costimulation ) but are most critically dependent on B7-mediated signals ( Figure 5F ) . Next , we examined if the B7-dependent MCMV-specific CD8+ T cell response can be boosted via supplementary triggering of the type I IFN pathway . We used recombinant IFNα2 that was functional both in vitro , as determined by a cytopathic effect inhibition assay ( Figure 5—figure supplement 1A ) , and in vivo as evidenced by increased expression of CD69 on lymphocytes at 18 hr upon i . p . administration ( Figure 5—figure supplement 1B ) . Addition of recombinant type I IFN on day 1 and 2 during MCMV-IE2-GP33 infection in mice that received Ifnar1+/+ and Ifnar1−/− P14 cells , caused no significant increase in the expansion of the P14 cells either transferred in WT or Cd80/86−/− mice , indicating that additional type I IFN signaling has negligible impact on B7-mediated signals that drive T cell expansion in MCMV infection ( Figure 5F ) . Administration of recombinant type I IFN during peptide vaccination , however , improved GP33-specific CD8+ T cell expansion , which indicated that IFNα is able to enhance T cell expansion in a low inflammatory context ( Figure 5G ) . To examine if the dependence of T cell expansion on B7-mediated costimulatory signals could be changed by other soluble factors than type I IFN , serum of mice that were infected for 2 days with LCMV was transferred to MCMV-infected WT and Cd80/86−/− mice . However , no differences were found in the magnitude of the MCMV-specific CD8+ T cell response ( Figure 5H ) , indicating that soluble factors in the LCMV environment do not enhance MCMV-specific CD8+ T cell expansion . To unequivocally demonstrate the uniqueness of the viral context to induce B7-mediated costimulation dependence , WT mice were co-infected with MCMV and LCMV . Remarkably , during this co-infection , MCMV-specific responses were still dependent on B7-mediated signals whereas LCMV-specific CD8+ T cells were not ( Figure 5I ) . Together , these data show that during an LCMV and MCMV infection a unique local environment is induced that principally determines the costimulatory requirements of the activated antigen-specific CD8+ T cells , and that direct type I IFN signaling in CD8+ T cells is slightly redundant with B7-mediated costimulation . To further delineate factors that could locally contribute to the CD28/B7 costimulation independence of CD8+ T cell expansion during LCMV infection , we characterized the expression of cell surface bound molecules that could impact T cell expansion . First , we examined if B7 molecules were induced upon LCMV infection . Expression of both B7 . 1 and B7 . 2 was upregulated on CD11c+ and CD11b+ cells early in infection ( Figure 6A ) . Strikingly , expression levels of B7 . 1 and B7 . 2 on these myeloid subsets were higher in LCMV infection as compared to MCMV infection . Thus , the non-dependence of B7-mediated costimulation for LCMV-specific CD8+ T cell expansion is not due to hampered expression of these costimulatory ligands during LCMV infection . 10 . 7554/eLife . 07486 . 012Figure 6 . LCMV infection induces high expression of costimulatory ligands . ( A ) Mice were infected with 2 × 105 PFU LCMV Armstrong or 1 × 104 PFU MCMV and costimulatory ligand expression was determined in the spleen . Histograms show cell surface expression of indicated costimulatory molecules on CD11b+ or CD11c+ cells at day 2 post-infection with either MCMV or LCMV . Representative staining of CD11b+ and CD11c+ cells from naive WT and Cd70/80/86−/− mice are depicted for comparison . Staining with an isotype control antibody is indicated as well . ( B ) Graphs depict mean fluorescence intensity ( MFI ) of costimulatory ligand expression on CD11b+ or CD11c+ cells in time . For each sample the MFI of the corresponding isotype control was subtracted from the MFI for each costimulatory ligand . Graphs are expressed as mean ± SEM ( n = 4 mice per group ) of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 01210 . 7554/eLife . 07486 . 013Figure 6—figure supplement 1 . Costimulatory ligands are highly induced in LCMV infection . ( A ) Mice were infected with 2 × 105 PFU LCMV Armstrong or 2 × 105 PFU VV-WR , and the expression of indicated costimulatory ligands was determined in the spleen on CD11b+ and CD11c+ cells 3 days post-infection . ( B ) IFNAR blocking antibodies were administrated during LCMV infection and the expression of indicated costimulatory ligands was determined in the spleen on CD11b+ and CD11c+ cells 3 days post-infection . Data in bar graphs are expressed as mean +SEM ( n = 4 mice per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 013 Besides costimulation via the CD28/B7 pathway , costimulatory signals can also be provided by TNFR superfamily members and their ligands including CD27/CD70 , OX40/OX40L and 4-1BB/4-1BBL . Therefore we compared the expression of the costimulatory ligands CD70 , OX40L and 4-1BBL in an LCMV and MCMV environment . Expression of both CD70 and 4-1BBL were much higher induced on CD11b+ and CD11c+ cells in LCMV infection as compared to MCMV infection ( Figure 6A , B ) . Furthermore , OX40L levels were increased in LCMV infection as well , although this expression was relatively low ( Figure 6A , B ) . Also compared to VV infection , elevated expression levels of all costimulatory ligands were observed on CD11b+ cells in the spleen in LCMV infection ( Figure 6—figure supplement 1A ) . On CD11c+ cells , B7 . 2 and 4-1BBL expression was increased in LCMV infection but the levels of B7 . 1 , CD70 and OX40L were comparable between VV and LCMV infection ( Figure 6—figure supplement 1 ) . The elevated costimulatory ligand expression levels found upon LCMV infection were partially associated with the high type I IFN levels within the LCMV-induced environment , as abrogation of type I IFN signaling , resulted to some extent in diminished costimulatory ligand expression ( Figure 6—figure supplement 1B ) . Together these data show that in LCMV infection an overall elevated expression level of costimulatory ligands is induced , which is partially induced in a type I IFN dependent manner . As multiple costimulatory molecules are highly induced during LCMV infection , we hypothesized that this might lead to a redundancy of costimulatory signals to be received by the responding T cells . The TNFR superfamily member , CD27 , is analogous to CD28 expressed on naive T cells , and binds the only known ligand CD70 . In Cd70−/− mice , no significant differences were found in the magnitude of the LCMV-specific CD8+ T cell response , indicating that the CD27/CD70 costimulatory pathway by itself has a limited or redundant role during LCMV infection ( Figure 7A ) . To investigate if CD70 and B7-mediated costimulation have overlapping roles in driving T cell expansion , we further examined LCMV-specific responses in mice genetically deficient for both CD70 and the B7 molecules . These Cd70/80/86−/− mice were viable and had no defects in the development of diverse hematopoietic cell populations ( Figure 7—figure supplement 1A–C ) . Moreover , no alterations in the TCRβ repertoire were found ( Figure 7—figure supplement 1D ) . Both GP33- and NP396-specific responses were significantly diminished in Cd70/80/86−/− mice , indicating that CD70 and B7 molecules are redundantly required for LCMV-specific CD8+ T cell expansion , and that these molecules can compensate each other ( Figure 7A ) . 10 . 7554/eLife . 07486 . 014Figure 7 . Redundant roles for costimulatory molecules in driving LCMV-specific CD8+ T cell expansion . ( A ) WT and costimulation deficient ( i . e . , Cd70−/− , Cd80/86−/− and Cd70/80/86−/− ) mice were infected with 2 × 105 PFU LCMV Armstrong . OX40L and/or 4-1BBL-mediated costimulation was abrogated by administration of blocking antibodies . The LCMV-specific CD8+ T cell response was determined 7 days post-infection using MHC class I tetramers . ( B ) The percentage of tetramer+ CD8+ T cells in the blood within the live gate at day 5 . 5 and day 7 post LCMV infection is shown as mean ± SEM . ( C ) IFNα levels in serum are shown 3 days post LCMV infection ( n = 4 mice per group ) . ( D ) OX40L and/or 4-1BBL-mediated costimulation was abrogated by administration of blocking antibodies in WT and costimulation deficient mice that were subsequently infected with 2 × 105 PFU LCMV Armstrong . The LCMV-specific CD8+ T cell response was determined 7 days post-infection using MHC class I tetramers . All responses in mice receiving blocking antibodies to costimulatory molecules were significantly decreased ( p < 0 . 05 ) compared to WT mice receiving isotype control antibodies . ( E ) The magnitude of splenic MCMV-specific CD8+ T cell pools determined by MHC class I tetramer staining after infection with 1 × 104 PFU MCMV-Smith is indicated . ( F ) The magnitude of the splenic GP33-specific CD8+ T cell response at day 8 post-infection with 1 × 105 PFU MCMV-IE2-GP33 is shown . All data in bar graphs are expressed as mean + SEM ( n = 5–12 mice per group of at least two independent experiments; p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 01410 . 7554/eLife . 07486 . 015Figure 7—figure supplement 1 . Cd70/80/86−/− mice have no defects in development of different hematopoietic populations . ( A ) The percentage of different hematopoietic populations in naive WT , Cd70−/− , Cd80/86−/− and Cd70/80/86−/− mice in spleen , blood and bone marrow ( BM ) is shown . ( B ) The total number of cells in spleen , BM and thymus is indicated . ( C ) Percentages of thymocyte subsets are shown . ( D ) The percentage of different Vβ chains within the total CD8+ T cell pool in the spleen of naive WT and costimulation deficient mice is indicated . Data in bar graphs are expressed as mean + SEM ( n = 4–5 mice per group ) of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 01510 . 7554/eLife . 07486 . 016Figure 7—figure supplement 2 . OX40L- and 4-1BBL-mediated costimulation is dispensable for primary expansion of MCMV-specific CD8+ T cells . MCMV-specific CD8+ T cell responses were determined by MHC class I tetramer staining after infection with 1 × 104 PFU MCMV-Smith and upon dual blockade of OX40L and 4-1BBL interactions . Data in bar graphs are expressed as mean + SEM ( n = 4 mice per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 016 The redundancy of the CD27/CD70 and CD28/B7 costimulatory pathways , prompted us to further define the costimulatory requirements during LCMV infection . To determine if OX40L and 4-1BBL-mediated interactions impact LCMV-specific CD8+ T cell responses , blocking antibodies were administrated . No significant differences were found when OX40L and 4-1BBL were blocked , however , when both pathways were abrogated the magnitude of the LCMV-specific CD8+ T cell response was significantly diminished ( Figure 7A , B ) . Strikingly , LCMV-specific CD8+ T cell responses were drastically decreased when OX40L/4-1BBL blockade was performed in mice lacking CD70 and B7-mediated costimulation . This diminished response was not due to defective induction of type I IFN , as IFNα levels in the serum of these mice were not substantially altered compared to WT mice ( Figure 7C ) . We further delineated the redundancy between different costimulatory molecules by additionally blocking OX40L- and/or 4-1BBL-mediated interactions in Cd70 and Cd80/86 deficient mice . Dual blockade of OX40L and 4-1BBL in Cd80/86−/− mice , and OX40L blockade in Cd70/80/86−/− mice showed comparable responses to mice in which all costimulatory pathways were abrogated , indicating that the most pronounced effects on LCMV-specific CD8+ T cell expansion are found when both B7 and OX40L-mediated interactions are abrogated ( Figure 7D ) . Together , these data indicate that virus-specific CD8+ T cell responses during LCMV infection critically depend on a plethora of costimulatory signals that are individually dispensable because they function in a highly redundant manner . To determine if costimulatory molecules were similarly working in a redundant manner in MCMV infection , WT and costimulation deficient mice were infected with MCMV . MCMV-specific CD8+ T cell responses in Cd70−/− and Cd80/86−/− mice were significantly diminished , however responses in Cd70/80/86−/− mice were even lower ( Figure 7E ) , indicating both a non-redundant and cooperative role for CD70 and B7-mediated costimulation in driving MCMV-specific T cell expansion . Similar results were obtained for GP33-specific CD8+ T cell responses using MCMV-IE2-GP33 ( Figure 7F ) . Abrogation of OX40L or 4-1BBL-mediated signals upon MCMV infection has been shown to minimally impact the initial expansion of MCMV-specific CD8+ T cells ( Humphreys et al . , 2007 , 2010 ) . Moreover , we found that upon dual blockade of OX40L and 4-1BBL-mediated interactions MCMV-specific T cell responses were not affected as well ( Figure 7—figure supplement 2 ) . These results indicate that redundancy between different costimulatory molecules is induced by the viral context . In both MCMV and LCMV infection , the virus-specific CD8+ T cell response is more affected in the absence of both CD70 and B7-mediated costimulation as compared to mice lacking only one of these costimulatory pathways . We next determined if type I IFN signaling is altered upon abrogation of dual CD70 and B7-mediated costimulation . Similar to what is found for endogenous LCMV-specific CD8+ T cell responses , Ifnar1+/+ P14 cells expanded well in WT , Cd70−/− and Cd80/86−/− mice and were to some extend hampered in expansion in Cd70/80/86−/− mice ( Figure 8A ) . The Ifnar1−/− P14 cells were rigorously hindered in their expansion , when transferred in WT mice and even more so in costimulation deficient mice . This reduced expansion of the Ifnar1−/− P14 cells in the costimulation deficient mice as compared to WT mice indicates slight redundancy of type I IFN signaling with costimulatory-driven signals in expanding CD8+ T cells . 10 . 7554/eLife . 07486 . 017Figure 8 . Type I IFN signaling in viral-specific CD8+ T cells is slightly redundant with costimulatory signals . ( A ) Schematic of experimental setup: Ifnar1+/+ and Ifnar1−/− P14 cells were adoptively transferred in WT , Cd70−/− , Cd80/86−/− and Cd70/80/86−/− mice that were subsequently infected with 2 × 105 PFU LCMV . 7 days post-infection the total numbers of P14 cells was determined in the spleen . ( B ) Similar setup as in ( A ) except mice were infected with 1 × 105 PFU MCMV-IE2-GP33 . 8 days post-infection the magnitude of the P14 cells was determined . Data in bar graphs are expressed as mean + SEM ( n = 4–8 mice per group ) and representative of two independent experiments . The fold difference and significance ( p < 0 . 05 ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07486 . 017 Furthermore , Ifnar1+/+ P14 cells were transferred to mice that were infected with MCMV-IE2-GP33 . In this setting , P14 cell expansion was critically dependent on both CD70- and B7-mediated costimulation ( Figure 8B ) . Compared to Ifnar1 proficient P14 cells , Ifnar1 deficient P14 cells had a higher degree of type I IFN dependence in the absence of costimulation , which was most pronounced when both CD70 and B7 costimulatory molecules were lacking ( Figure 8B ) . Thus , type I IFNs have a slight stimulating activity for CD8+ T cells in MCMV infection , which is more pronounced in the absence of CD70 and B7-mediated signaling , indicating that also during MCMV infection partial redundancy of type I IFN signaling with costimulation during CD8+ T cell expansion occurs . Determining the critical components required for T cell expansion in a given situation is of utmost importance for understanding resistance to virus infections and improving vaccination strategies . Using different viral models we show that the pathogen-induced environment dictates the utilization of costimulatory signals that drive CD8+ T cell expansion . Primary LCMV-specific CD8+ T cell responses have long been considered to be costimulation independent ( Shahinian et al . , 1993; Kundig et al . , 1996; Andreasen et al . , 2000; Grujic et al . , 2010; Eberlein et al . , 2012 ) . Nevertheless , the development of LCMV-specific memory CD8+ T cell formation is hampered during Cd28 or Cd80/86 deficiency ( Grujic et al . , 2010; Eberlein et al . , 2012 ) , indicating that CD28/B7-mediated costimulation occurs during LCMV infection , which is in agreement with our study . We also found that the CD27/CD70 pathway has negligible costimulatory effects for LCMV-specific CD8+ T cell expansion when solely this pathway is abrogated . This has been observed by others as well ( Matter et al . , 2005; Schildknecht et al . , 2007 ) , but recent reports suggested that blockade of the CD27/CD70 pathway can to some extend impair CD8+ T cell responses during acute LCMV infection ( Penaloza-Macmaster et al . , 2011; Munitic et al . , 2013 ) . Importantly , here we show that LCMV-specific CD8+ T cell responses are in fact critically dependent on costimulatory signals , but these signals operate in a highly redundant manner in which both members of the costimulatory CD28/B7 family and TNFR/TNF family take part . The overall expression of costimulatory ligands in the LCMV milieu exceeded the expression levels found upon an MCMV or VV infection . In this respect , it is of interest to note that abrogation of exclusively the CD28/B7 or the CD27/CD70 pathway severely hampers MCMV- and VV-specific CD8+ T cell responses ( Arens et al . , 2011b; Salek-Ardakani et al . , 2011; Welten et al . , 2013b ) , indicating that in these infections the costimulatory molecule levels are likely limited leading to non-redundant roles of costimulatory molecules . Unhampered LCMV-specific responses are observed upon dual 4-1BBL and CD28 abrogation ( DeBenedette et al . , 1999 ) and this is consistent with our data showing that multiple pathways than these have to be abrogated to observe diminished LCMV-specific CD8+ T cell responses virus-specific responses . The higher expression levels of costimulatory ligands within the LCMV environment is likely causing the redundancy amongst CD28/B7 and TNFR/TNF family members in driving LCMV-specific T cell expansion . Of interest is that even further improvement of B7-mediated signaling due to CTLA-4 blockade did not advance LCMV-specific CD8+ T cell expansion , suggesting that the observed higher expression of costimulatory molecules is at a maximal level with respect to stimulating T cells . Strong replicating VV-strains employ more costimulatory receptors as compared to weak replicating VV-strains ( Salek-Ardakani et al . , 2011 ) . Furthermore , 4-1BBL-mediated interactions are critical during severe influenza virus infections but dispensable upon a mild influenza virus ( Lin et al . , 2009 ) , indicating that the strength of the inflammatory environment dictates the employment of different costimulatory receptors . Given the higher costimulatory molecule expression , one could argue that LCMV infection elicits an elevated inflammatory milieu as compared to most other infections . Consistent with this notion is that in LCMV infection very high levels of type I IFNs are induced , which are partly responsible for the high costimulatory ligand expression . An elevated expression of costimulatory molecules in LCMV infection might also be related to a lack of immunomodulatory effects that dampen costimulatory molecule expression . During MCMV infection for example , the B7 . 1 and B7 . 2 expression in virus-infected cells is downmodulated by the virus by sophisticated immune evasion mechanism ( Loewendorf et al . , 2004; Mintern et al . , 2006; Arens et al . , 2011a ) . Perhaps related to this , is that the CD8+ T cell response to MCMV is predominantly mediated by cross-priming APCs , which are by definition not directly infected by the virus ( Torti et al . , 2011; Busche et al . , 2013 ) . Shared signaling pathways might underlie the observed redundancy among members of the costimulatory TNFR family and CD28 family . TNFR family members are known to signal via TRAF molecules , which are coupled to the activation of the NF-κB pathway via both the canonical and the non-canonical routes ( Croft , 2009 ) . CD28 is also able to signal via the NF-κB route ( Boomer and Green , 2010 ) . Another shared signaling pathway of CD28 and TNFR family members might be the c-Jun kinase pathway , which is coupled to proliferation as well ( Gravestein et al . , 1998; Skanland et al . , 2014 ) . We found redundancy between CD28 and CD27 signaling on CD8+ T cell expansion in MCMV and LCMV infection , and this has been found in influenza virus infection as well ( Hendriks et al . , 2003 ) . However , besides redundancy between CD28/B7 and TNFR/TNF families also redundancy among costimulatory TNFR family members likely happened as the response was most compromised in settings where multiple TNFR family members were targeted . The latter is consistent with observations in the influenza virus infection model , where virus-specific T cells that accumulate in the lung but not in the spleen were collectively dependent on signals mediated via a variety of TNFR family members ( Hendriks et al . , 2005 ) . We found a prominent role for the pathogenic milieu in directing CD8+ T cell responses and dictating the requirements for certain costimulatory signals . The fact that even upon LCMV and MCMV co-infection the costimulatory requirements for T cell expansion are not altered , suggest that this instruction occurs locally , likely at the level of APC-T cell interaction . The majority of the MCMV-specific CD8+ T cells is activated via cross-priming ( Torti et al . , 2011; Busche et al . , 2013 ) , and whether both direct and cross-priming occur during LCMV infection is unclear ( Freigang et al . , 2007 ) . Nevertheless CD11c+ APCs are critical for LCMV-specific CD8+ T cell priming ( Probst and van den Broek , 2005 ) . Moreover , because of different tropisms it is unlikely that MCMV and LCMV co-infect the very same cells and that the viral epitopes are presented by the same APC ( Matloubian et al . , 1993; Alexandre et al . , 2014 ) . Since APCs need to be directly activated for adequate T cell priming rather than by environmental inflammatory signals ( Kratky , 2011 ) , our data are consistent with a scenario where the two viruses activate APCs in a different manner resulting in differential provision of costimulatory signals . The enhanced costimulation during LCMV infection may besides due to stronger and distinctive ( local ) inflammation also be a consequence of longer and/or stronger antigen-presentation as compared to other viral infections . However , LCMV and MCMV are both natural mouse pathogens and infection with these viruses results in virus levels that peak around day 4 post-infection in the spleen and liver ( Buchmeier et al . , 1980; Cicin-Sain et al . , 2008 ) . Nevertheless , differential kinetics of antigen-presentation of the viral epitopes is possible . Perhaps related to our results are the observations that the pathogen-specific inflammatory environment dictates the fate of responding CD8+ T cells allowing shaping of effector and memory T cell formation ( Obar et al . , 2011; Keppler et al . , 2012; Plumlee et al . , 2013 ) . This may be connected with pathogen-specific tuning of the antigen-sensitivity of CD8+ T cells by enhancing TCR signaling ( Richer et al . , 2013 ) , the induction of distinct inflammatory cytokine levels ( Thompson et al . , 2006 ) and/or by instructing the costimulatory pathway usage ( our results ) . Although in vitro the requirements for CD28/B7-mediated costimulation can differ for primary and memory cells ( Flynn and Mullbacher , 1996 ) , we found in vivo that CD28/B7-mediated costimulation was important for the expansion of both naive and memory CD8+ T cells in MCMV infection . This is consistent with models of influenza virus , VV and murine γ-herpesvirus ( Borowski et al . , 2007; Fuse et al . , 2008 ) that require B7-mediated signals for primary and secondary expansion of virus-specific CD8+ T cells . However , the APCs that prime memory vs naive T cells might differ ( Belz et al . , 2007 ) . Type I IFNs are not required for the expansion of human memory CD8+ T cells in vitro ( Hervas-Stubbs et al . , 2010 ) . In experimental in vivo models , however , the inflammatory environment determines the signal 3 ( i . e . , type I IFN and IL-12 signaling ) dependency upon secondary infection independent of the context of priming ( Keppler and Aichele , 2011 ) . Correspondingly , we observed that the milieu of the infectious pathogen during the recall response determines the requirements for costimulatory signals as well , and suggests that the responsiveness of T cells during the initial expansion is plastic and can be modified during antigenic re-challenge . Collectively , our results highlight the importance of the inflammatory environment for both primary and secondary CD8+ T cell expansion . These findings can be beneficial for pre-clinical exploration of adoptive T cell settings , where antigen-specific T cells are expanded to large numbers . In addition , our report has important implications for prime-boost vaccination strategies , as it provides evidence for the plasticity of memory T cells that is shaped by the nature of the pathogen to generate them . C57BL/6 mice were obtained from Charles River and were used as WT mice . Cd70−/− ( Coquet et al . , 2013 ) , Cd80/86−/− ( Borriello et al . , 1997 ) and Ptprca ( Cd45 . 1 , Ly5 . 1 ) mice were bred in house to the obtained C57BL/6 background . Cd70/80/86−/− mice were generated by crossing Cd70−/− with Cd80/86−/− mice . All animals were maintained on specific pathogen free conditions at the animal facility in Leiden University Medical Center ( LUMC ) . Mice were matched for gender and were between 8-12 weeks at the start of each experiment . IFNAR proficient ( Ifnar1+/+ ) and deficient ( Ifnar1−/− ) P14 TCR transgenic mice on a CD90 . 1+ C57BL/6 background were generated by breeding as described ( Keppler et al . , 2012 ) . All animal experiments were approved by the Animal Experiments Committee of LUMC ( reference numbers: 12 , 006 , 13 , 150 , 14 , 046 and 14 , 066 ) and performed according to the recommendations and guidelines set by LUMC and by the Dutch Experiments on Animals Act that serves the implementation of ‘Guidelines on the protection of experimental animals’ by the Council of Europe . MCMV-Smith was obtained from the American Type Culture Collection ( Manassas , VA ) . Stocks were derived from salivary glands of infected BALB/c mice as described elsewhere ( Schneider et al . , 2008 ) . Viral titers were determined as described ( Welten et al . , 2013b ) . For an in vivo MCMV infection , mice were infected intraperitoneal ( i . p . ) with 1 × 104 PFU MCMV-Smith . To generate MCMV-IE2-GP33 , MCMV-M45-GP33 and MCMV-M45-SIINFEKL , nucleotide sequences encoding the GP33-41 epitope ( GP33 ) of LCMV or the SIINFEKL epitope of chicken ovalbumin were inserted by targeted mutagenesis at the C-terminus of the M45 or IE2 genes , directly in front of the stop codon . Two alanine residues in front of the peptide sequences were placed in order to enhance proteasomal cleavage . Recombinant virus was reconstituted as described elsewhere ( Dekhtiarenko et al . , 2013 ) . Mice were infected i . p . with 1 × 105 PFU MCMV-IE2-GP33 , MCMV-M45-GP33 or MCMV-M45-SIINFKEL . LCMV Armstrong was propagated on BHK cells . The titers were determined by plaque assays on Vero cells as described ( Ahmed et al . , 1984 ) . For LCMV Armstrong infection , mice were infected i . p . with 2 × 105 PFU ( high dose ) or 2 × 102 PFU ( low dose ) . For co-infection experiments , mice were infected with 2 × 105 PFU LCMV and 1 × 104 PFU MCMV-Smith . VV strain WR was purchased from the American Type Culture Collection , grown on HELA cells and quantified on VeroE6 cells as described ( Davies et al . , 2005 ) . Mice were infected i . p with 2 × 105 PFU VV . L . monocytogenes ( LM ) expressing GP33 and the attenuated LM-Quadvac strain expressing four epitopes of VV ( i . e . , A24R , C4L , K3L and B8R ) and the SIINFEKL epitope of OVA are described elsewhere ( Zenewicz et al . , 2002; Lauer et al . , 2008 ) . Mice were challenged intravenously ( i . v . ) with 1 . 5 × 103 CFU LM-GP33 or with 1 × 106 CFU LM-Quadvac . For blockade of IFNAR , mice received 1 mg of IFNAR blocking antibody ( clone MAR1-5A3; Bio X Cell , West Lebanon , NH , United States ) on day −1 and 1 post-infection . For blockade of CTLA-4 , 200 µg of αCTLA-4 ( clone UC10-4F10-11 ) was administrated on day −1 , 1 and 3 . For blockade of OX40L and 4-1BBL , 150 µg of αOX40L ( clone RM134L ) or α4-1BBL ( clone TKS-1 ) ( both Bio X Cell ) or a combination of both antibodies was administrated on day −1 , 1 and 3 post-infection . Control mice received a similar amount of a rat IgG isotype control antibody ( clone GL113 ) . All antibodies were administrated i . p . in 400 µl PBS . Splenocytes were obtained by mincing the tissue through a 70 µm nylon cell strainer ( BD Biosciences , San Jose , CA , United States ) . Blood was collected via the tail vein . Erythrocytes were lysed in a hypotonic ammonium chloride buffer . Determination of the antigen-specific T cell response by MHC class I tetramers and intracellular cytokine staining was performed as described ( Arens et al . , 2011b ) . Briefly , single-cell suspensions were incubated with fluorescently conjugated antibodies and tetramers for 30 min at 4°C . To determine the expression of costimulatory ligands , spleens were first injected with 1 mg/ml collagenase and 0 . 02 mg/ml DNAse in IMDM without FCS , after which the spleens were chopped in small pieces and incubated for 25 min at RT . Subsequently 0 . 1 M EDTA was added and cells were transferred through a 70 µm cell strainer to make single cell suspensions . Next , cells were pre-incubated with normal mouse serum and Fc-block ( clone 2 . 4G2 ) , after which biotinylated or fluorochrome conjugated antibodies were added . For analysis of intracellular cytokines , cells were restimulated for 5 hr with 1 µg/ml MHC class I restricted peptides in the presence of 1 µg/ml brefeldin A , followed by cell surface staining and intracellular staining for IFN-γ . The following fluorescently conjugated antibodies were purchased by BD Biosciences , eBioscience ( San Diego , CA , United States ) or BioLegend ( San Diego , CA , United States ) : CD3 ( V500 ) , CD8 ( A700 ) , CD11b ( eFluor450 ) , CD11c ( eFluor780 ) , CD44 ( eFluor450 ) CD45 . 1 ( FITC ) , CD62L ( eFluor780 ) , CD70 ( biotin ) , CD80 ( FITC ) , CD86 ( PE ) , CD90 . 1 ( FITC ) , IFN-γ ( APC ) , KLRG1 ( PE-Cy7 ) , OX40L ( biotin ) , Vα2 ( PE ) , 4-1BBL ( biotin ) . Fluorochrome-conjugated streptavidin ( PE , APC or Brilliant Violent 605 ) was used to detect biotinylated antibodies . Flow cytometric acquisition was performed on a BD LSR II and cells were sorted using a BD FACSAria . Data were analyzed using FlowJo software ( TreeStar , Ashland , OR , United States ) . MHC class I Db restricted tetramers for the OVA257-264 epitope ( SIINFEKL ) , the MCMV epitope M45985-993 , the LCMV epitopes GP33-41 and NP396-404 , and MHC class I Kb restricted tetramers for the MCMV epitopes M57816-824 , m139419-426 , and M38316-323 , and the VV epitopes B8R20-27 and A3L270-277 were produced as described ( Altman et al . , 1996 ) . The following class I-restricted peptides were used: M45985-993 , m139419-426 , M38316-323 , GP33-41 and NP396-404 . The following SLP containing the GP33 epitope ( underlined ) was used for vaccination: VITGIKAVYNFATCGIFALIS . Mice were vaccinated at the tail base with 75 µg SLP in PBS either combined with 20 µg CpG or supplemented with 1 × 105 units IFNα injected i . p . in 200 µl PBS at 18 and 48 hr post-vaccination . Blood was collected retro-orbitally and clotted for 30 min . Serum was collected after centrifugation and stored at −80°C until further use . Cytokines were measured in serum using a mouse Bio-Plex Pro Mouse Cytokine 23-plex immunoassay ( Bio-Rad , Herculus , CA , United States ) according to manufacturer's protocol . IFNα was measured with a mouse ProcartaPlex multiplex immunoassay ( eBioscience ) . Splenic Ifnar1+/+ and Ifnar1−/− CD90 . 1 P14 cells were enriched by negative selection of CD8+ T cells ( BD Biosciences ) and 5 × 104 cells were adoptively transferred in WT and costimulation deficient mice that were subsequently infected with either 2 × 105 PFU LCMV Armstrong or 1 × 105 PFU MCMV-IE2-GP33 . 7 days post LCMV or 8 days post MCMV infection the magnitude of P14 cells was determined . For adoptive transfer of memory GP33-specific CD8+ T cells , CD45 . 1+ congenic mice were infected with 2 × 105 PFU LCMV Armstrong . After 4 months GP33-specific memory CD8+ T cells were FACS sorted using MHC class I tetramers and 2 × 103 cells were adoptively transferred into WT and Cd80/86−/− mice that were subsequently infected with 2 × 105 PFU LCMV Armstrong or 1 × 105 PFU MCMV-IE2-GP33 . 6 days post adoptive transfer , the total number of CD45 . 1+ GP33-specific CD8+ T cells was determined . Similar experiments were performed with CD45 . 1+ congenic mice infected with 1 × 105 PFU MCMV-IE2-GP33 . For serum transfer , WT mice were infected with 2 × 105 PFU LCMV Armstrong and after 2 days , serum was collected and 150 µl was transferred i . p . to mice that were infected 1 day before with 1 × 104 PFU MCMV-Smith . 8 days post MCMV inoculation , MCMV-specific CD8+ T cell responses were determined in the spleen . DNA encoding mouse IFNα2 , the Ifna2 gene , was synthetically made and codon optimized by Geneart ( Thermo Fisher Scientific , Waltham , MA , United States ) . The gene was subcloned by Gateway technology ( Thermo Fisher Scientific ) in pDEST17 , which has an N-terminal histidine tag . After overproduction the protein was purified as described ( Franken et al . , 2000 ) and lyophilized . 2 . 5 mg of protein was resuspended in 1 ml 100 mM Tris HCl , 8 M Urea pH 8 . 0 . The dissolved protein was refolded in 50 ml 0 . 4 M L-arginine , 100 mM Tris HCl , 2 mM EDTA , 0 , 5 mM oxidized glutathione , 5 mM reduced glutathione , 5% glycerol and 0 . 5 tablet of Complete pH 8 . 0 . After 5 days of incubation at 10°C the solution was concentrated on an Ultracel 10 kD filter ( Merck Millipore ( Billerica , MA , United States ) ) . The concentrated protein was loaded on a PBS equilibrated Hi-Load 16/60 superdex 75 column . The collected peak of the protein was concentrated on the Ultracel 10 kD filter and stored with 16% glycerol at −80°C . Protein concentration was determined by Bradford and OD280 nm . The bioactivity was determined according to a protocol described elsewhere ( Seeds and Miller , 2011 ) with slight alterations . In short; L929 cells were seeded in 96-well plates in serum free RPMI and incubated at 37°C , the next day different dilutions of IFNα2 were added . The following day Mengovirus was added and after 2 days of incubation an MTT assay was performed ( Trevigen , Gaithersburg , MD , United States ) . Cell survival was determined by the following formula: ( OD570-655 sample with IFNα2 and virus/OD570-655 without virus and IFNα2 ) × 100% . One unit of IFNα2 was defined as the concentration at which 50% of the cytopathic effect was inhibited . Our batch had a bioactivity of 1 × 106 units/ml . For in vivo administration , mice received 1 × 105 units i . p . upon CMV infection or post vaccination . GraphPad Prism 6 . 0 software ( GraphPad Software , La Jolla , CA , United States ) was used for statistical analyses . To determine statistical significance between two groups an unpaired Student's t-test was performed . To evaluate significance between more than two groups , one-way ANOVA was used and values were compared to WT mice . Dunnett's post-hoc test was performed to correct for multiple comparisons . p-values <0 . 05 were considered as significant .	When the immune system detects a virus in the body it mounts a response to eliminate it . Immune cells called CD8+ T cells detect fragments of virus proteins that are presented on the surface of other immune cells . The CD8+ T cells then rapidly divide to form populations that roam the body to kill cells that are infected with the virus . Afterwards , some of the CD8+ T cells become ‘memory T cells’ , which allow the immune system to respond more rapidly if the virus returns . This means that a subsequent infection of the same virus is usually stopped before it can become severe enough for an individual to feel unwell . Vaccines take advantage of the activities of CD8+ T cells to enable a person to become ‘immune’ to a virus without having to experience the disease . Vaccines contain dead or weakened viruses that can't spread in the body , but are able to activate the CD8+ T cells . However , a vaccine may not be as effective in activating the T cells as the live virus , perhaps because it fails to trigger the production of other molecules in the host that promote T cell activation . There are many of these ‘co-stimulatory molecules’ in the body , but it is not clear exactly how they work . Now , Welten et al . show that the role of co-stimulatory molecules in the activation of CD8+ T cells depends on the type of virus and how it affects cells . Mice that were genetically engineered to lack two co-stimulatory molecules called CD80 and CD86 failed to accumulate active CD8+ T cells in response to infection with a herpes-like virus . However , if these mice were infected with a different virus called LCMV—which causes swelling of the brain and spinal cord—they produced many active CD8+ T cells to fight the infection . Welten et al . found that other co-stimulatory molecules are able to compensate for the loss of CD80 and CD86 to boost the activation of T cells in response to LCMV , but not the herpes-like virus . Further experiments showed that LCMV triggers a lot more inflammation in infected cells than the other virus . This leads to the production of many different types of co-stimulatory molecules , which ensures that if one fails to boost the activation of CD8+ T cells , another molecule can do so instead . Better understanding of how these co-stimulatory molecules work could help scientists to develop more effective vaccines in future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2015	The viral context instructs the redundancy of costimulatory pathways in driving CD8+ T cell expansion
Sexual selection is generally predicted to act more strongly on males than on females . The Darwin-Bateman paradigm predicts that this should also hold for hermaphrodites . However , measuring this strength of selection is less straightforward when both sexual functions are performed throughout the organism’s lifetime . Besides , quantifications of sexual selection are usually done during a short time window , while many animals store sperm and are long-lived . To explore whether the chosen time frame affects estimated measures of sexual selection , we recorded mating success and reproductive success over time , using a simultaneous hermaphrodite . Our results show that male sexual selection gradients are consistently positive . However , an individual’s female mating success seems to negatively affect its own male reproductive success , an effect that only becomes visible several weeks into the experiment , highlighting that the time frame is crucial for the quantification and interpretation of sexual selection measures , an insight that applies to any iteroparous mating system . Darwin defined sexual selection as selection on traits that affect mating success ( Darwin , 1871 ) . In doing so , he clearly focused on the obvious secondary sexual characters that often differ between males and females . Classical examples include antlers of deer , long and extravagantly coloured ( tail ) feathers in birds and traits of that ilk . In recent decades , this definition has been refined , most notable due to the realisation that sexual selection does not only act prior to mating – referred to as pre-copulatory sexual selection ( Darwin’s focus ) - but also after mating - post-copulatory sexual selection ( e . g . , Parker , 1970; Eberhard , 1996 ) . Investigating pre-copulatory sexual selection involves measuring mating success , which is generally defined as the number of mates that an individual copulates with ( for males MSm; for females MSf ) , and the resulting reproductive success which is measured as the number of offspring that an individual is able to produce ( for males RSm; for females RSf ) . The residual variation of the relationship between mating success and reproductive success can then be used as a quantitative proxy for post-copulatory sexual selection . These sexual selection processes are often measured , but with a strong bias towards species with separate sexes , even though simultaneous hermaphrodites are under influence of the same selective pressures ( Charnov , 1979; Nakadera and Koene , 2013; Pélissié et al . , 2012; Schärer and Pen , 2013 ) . One , now classical , study on Drosophila melanogaster by Bateman ( 1948 ) sparked the pivotal insight that the factors that limit reproductive success for males and females are different . This is referred to as the Darwin-Bateman paradigm ( or Bateman principle; reviewed in Dewsbury , 2005 ) . Due to anisogamy , for which sexual selection has been an important driving force ( reviewed in Parker and Birkhead , 2013 ) , a clear difference is found in the cost of producing male and female gametes . As a consequence , sperm production is generally not a limiting factor for males , so the number of fathered offspring ( male reproductive success , RSm; Bateman , 1948 ) depends directly on the number of mates the male can inseminate . In contrast , egg production is highly dependent on available resources and therefore limits the number of offspring the female can produce ( i . e . , female reproductive success , RSf; Bateman , 1948 ) . Even though Bateman’s experiment has been criticized based on experimental design as well as data collection and actual repeatability ( e . g . , Gowaty et al . , 2012; Gowaty et al . , 2013 ) , the basics of the Darwin-Bateman paradigm still hold ( e . g . , Janicke et al . , 2016 ) . Therefore , this paradigm has formed the basis for more formalized approaches to measuring and quantifying sexual selection , in which the difference in variance in mating success and reproductive success of males and females can be captured in different measures of sexual selection ( e . g . , Arnold , 1994; Jones , 2009; see Materials and methods ) . One insightful measure that has emerged is the sexual selection gradient ( also referred to as Bateman gradient ) , which looks at the linear relationship between mating success and reproductive success ( β; e . g . , Arnold and Duvall , 1994; Anthes et al . , 2010 ) . The steepness of the male and female Bateman gradient can inform about the strength of sexual selection , as has for example been done for polygamous red jungle fowl ( Collet et al . , 2012 ) , polyandrous rough skinned newts ( Jones , 2009 ) , and polygynous , role-reversed pipefish ( Jones et al . , 2005 ) . For species with separate sexes , the above approach is relatively straight forward , because one ‘only’ needs to regress mating success against reproductive success for each of the sexes and compare the slopes ( βmale and βfemale ) . As pointed out by Gowaty et al . ( 2012 ) , in order to obtain correct quantifications of sexual selection , independent measures of mating success and reproductive success are needed , which lacked in Bateman’s original study . As furthermore pointed out by Anthes et al . ( 2010 ) , the quantification of sexual selection is less clear-cut in simultaneous hermaphrodites , because each individual is both male and female at the same time . Hence , besides that mating success in the male role can affect the individual’s male reproductive success , it can also directly affect its female reproductive success and vice versa . These interactions between the sexual functions of the individual are referred to as cross-sex effects ( βmf and βfm; Anthes et al . , 2010 ) and trigger the legitimate question whether these cross-sex effects cause simultaneous hermaphrodites to deviate from the Darwin-Bateman paradigm . As illustrated by several recent studies , it is possible to quantify sexual selection in simultaneous hermaphrodites ( Ophryotrocha diadema: Lorenzi and Sella , 2008; Biomphalaria glabrata: Anthes et al . , 2010; Physa acuta: Pélissié et al . , 2012; Macrostomum lignano: Marie-Orleach et al . , 2016 ) , with the latter three also applying the sexual selection gradients approach outlined by Anthes et al . ( 2010 ) . Like in most of the sexual selection gradient studies on separate sexed species , these studies used a very restricted time window ( days ) within which the strength of sexual selection was estimated . However , many species are long-lived , mate many times and can store and use sperm for extended periods ( e . g . , Nakadera and Koene , 2013; Nakadera et al . , 2014a ) . Therefore , the relationship between mating success and reproductive success can be expected to change over time , especially when considering that sperm storage becomes important as soon as individuals are no longer virgin , meaning that the degree to which mating success translated into reproductive success might change ( Baena and Macías-Ordóñez , 2012; Wacker et al . , 2014; Anthes et al . , 2016 ) . Here , we present an experiment in which we quantify mating success and reproductive success , using the great pond snail Lymnaea stagnalis , over an eight week period that represents roughly a quarter of its reproductive life in nature . Our quantification of sexual selection gradients allows us to address several unresolved questions that are of general importance for understanding sexual selection . First , do these simultaneous hermaphrodites conform to the prediction that sexual selection gradients differ for the male and female function ? Second , can we detect the predicted cross-sex effects on reproductive success ? Third , do sexual selection gradients change depending on the time window of measurement ? This latter question , which can be tested given the time frame of our experiment , addresses an issue that has remained experimentally untested in any hermaphroditic species to date ( and was only addressed in one separate sexed species: Turnell and Shaw , 2015 ) . In addition , with the collected data , we can address several remaining questions that are specific to the simultaneous hermaphrodite under investigation , the pond snail L . stagnalis . We can examine whether partner availability is beneficial for offspring quality , something that is predicted based on the finding that multiple mating results in larger investment per egg in this species ( Hoffer et al . , 2012 ) . Also , by looking at the number of matings in the male and female role , we can determine whether the mating mode of this species is unilateral or relaxed reciprocal ( the latter meaning that animals tend to alternate mating roles between successive copulations , possibly with different partners; we already know it is not strict reciprocal: Koene and Ter Maat , 2005 ) . As pointed out by Anthes et al . ( 2010 ) , this information is important , as the mating mode has implications for how independently sexual selection can act on the two sexual roles of hermaphrodites . The measurements of mating activity and reproductive output in the above-mentioned Multiple partners treatment were used to quantify sexual selection by looking at the variance for each sex of one focal within a group of five snails ( see Material and methods for details ) . Due to different limits in terms of their gamete production , variance in the reproductive success of males is expected to be larger than that of females and can be captured in the variance measure I . This measure is defined as the standardized variance in relative reproductive success and its value is indicative of the opportunity for selection ( Arnold , 1994; Anthes et al . , 2010; Evans and Garcia-Gonzalez , 2016 ) . The opportunity for sexual selection , which is defined as the standardized variance in relative mating success , is captured in the variance measure Is ( Arnold , 1994; Anthes et al . , 2010 ) . Comparison of the opportunity for selection values between the male and female role ( Im and If ) shows that these values are larger for the male role ( Figure 2—figure supplement 1 and Figure 2—source data 1 ) . In addition , these values decreased over the course of the experiment . The much lower values for the opportunity for sexual selection ( Ism and Isf ) reveal a similar trend ( Figure 2—figure supplement 1 and Figure 2—source data 1 ) . Subsequently , we compared the sexual selection gradients ( β ) between and among the sexual roles . Such a gradient is the linear least-squares regression slope of sex-specific relative RS on sex-specific relative MS ( Jones , 2009; Klug et al . , 2010 ) , thus expressing the expected fitness increase achieved by mating one additional time in a specific sex role ( βm or βf ) . Because hermaphrodites express both sex functions within one body , the reproductive efforts in one reproductive function can alter reproductive fitness in the other . To deal with this non-independence of male and female reproduction , we used a multiple regression with MSm and MSf as explanatory variables on , respectively , RSf and RSm . As pointed out by Anthes et al . ( 2010 ) , this approach makes possible cross-sex effects explicit . The resulting cross-sex effects ( βmf and βfm ) describe how MS in one sex function changes RS in the other sex function . In order to make the sexual selection gradient measures comparable across time , we used relative values for both mating success and reproductive success; note that time was divided in weeks because eggs were collected at the end of each week ( hence the factor Week below ) . To analyse the sexual selection gradients from the male perspective , we used a model including the dependent variable relative RSm and the factors relative MSm , relative MSf and Week , plus their interaction , including focal identity as random , repeated factor . This analysis revealed a significant effect for the factor relative MSm on relative RSm ( F1 , 145 . 5 = 7 . 872; p = 0 . 0057 ) , while Week and the interaction term Weekrelative MSm showed no significance . This indicates that the male sexual selection gradient , βmm , is positive and does not change over time , as is also clearly reflected in the regression lines ( Figure 2; see β-values in Figure 2—figure supplement 1 , the slopes’ confidence intervals in Figure 2—source data 2 , and slope comparisons between weeks in Figure 2—source data 3 ) , revealing that continued mating in the male role assures continuous male reproductive success beyond the first week . When looking at the cross-sex effect βmf in this model , the factor relative MSf had a significant effect on relative RSm ( F1 , 145 . 7 = 29 . 956; p = 0 . 0001 ) , which is due to the significant positive relationship between the number of male matings and female reproductive success in the first week , as reflected by the significance of the interaction term Weekrelative MSf ( F7 , 136 . 1 = 4 . 159; p = 0 . 0004 ) ; an effect that disappeared afterwards ( Figure 2 and Figure 2—figure supplement 1 ) . From the female perspective , running the same model with relative RSf as dependent variable , for the female role ( βff ) , a significant relationship between female mating success and female reproductive success is found ( F1 , 140 . 6 = 5 . 101; p = 0 . 0255 ) , which seems to be caused by the negative trend lines found in the last three weeks of the experiment ( Figure 2 and Figure 2—figure supplement 1 ) . For the cross sex effect βfm the model revealed a significant effect of relative MSm on relative RSf ( F1 , 140 , 5 = 13 . 523; p = 0 . 0003 ) , which seem due to the initial positive correlation ( in Week 1 and 2; Figure 2 and Figure 2—figure supplement 1 ) , even though the interaction term Week*relative MSm was not significant . 10 . 7554/eLife . 25139 . 004Figure 2 . The relationships between male and female mating success and reproductive success . The relationships are shown for every week of the experiment . The within-sex and cross-sex sexual selection gradients are based on bivariate regressions of either reproductive success on mating success ( βmm , βff , βmf , βfm ) . Significant slopes ( p < 0 . 05 ) are indicated with a solid fitted line , a trend ( p < 0 . 10 ) is indicated with a dashed line . As shown in Figure 2—source data 3 , the slopes of different weeks do not significantly differ from each other . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 00410 . 7554/eLife . 25139 . 005Figure 2—source data 1 . The calculated values and their confidence interval ( CI ) for the opportunity for selection ( I ) and sexual selection ( Is ) for both sexual roles , indicated with subscript m or f , over the weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 00510 . 7554/eLife . 25139 . 006Figure 2—source data 2 . Slope comparisons between weeks for all the significant sexual selection gradients shown in Figures 2 and 3 . For the top right of the table this shows the results for the regressions between relative male mating success ( MSm ) versus relative male reproductive success ( RSm ) and for bottom left ( in italics ) the results are shown for the regressions between relative principal component 2 ( PC2 ) and relative male reproductive success ( RSm ) . Each cell of the table comparing two slopes indicated the t-value , degrees of freedom ( df ) and significance ( p ) . The first value column and row , respectively , indicate the sample size ( N ) , the slope ( β ) and the standard error ( SE ) of the compared lines . The cells in the Week 1 row are left empty ( - ) because this line was not significant for relative MSm versus relative RSm and could thus not be compared with the other slopes . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 00610 . 7554/eLife . 25139 . 007Figure 2—source data 3 . Slope and confidence interval of the correlations between the different sexual selection measures . The values are only given for the significant ( p < 0 . 05 , black ) and near-significant ( p < 0 . 10 , grey ) slopes shown in Figures 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 00710 . 7554/eLife . 25139 . 008Figure 2—figure supplement 1 . I-values and gradient values are shown over time , calculated for every week of the experiment . The values are shown for the relative cumulative data . For the gradients , means and standard deviations are depicted and significant slopes are indicated with an asterisk . See Figure 2—source data 1; Figure 2—source data 3 , respectively , for the confidence intervals of the I and β-values . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 008 Depending on their mating system , male and female mating success may not be fully independent in simultaneous hermaphrodites . Even in unilaterally mating species , playing both roles in a mating encounter in sequence ( i . e . , reciprocating ) could make a multiple regression analysis statistically fragile ( Mitchell-Olds and Shaw , 1987 ) . To cope with this potential problem , we follow Anthes et al . ( 2010 ) suggestion to replace MSm and MSf by their principal components ( PC ) . When comparing the outcomes of the regression analysis with the principle component analysis ( PCA ) approach , one can see that most findings were corroborated ( comparing Figures 2 and 3 ) . For proper interpretation of these graphs , it should be noted that overall mating activity , that is , the correlation component between relative MSm and relative MSf is represented by PC2; the sexual bias , that is , the relative difference between MSm and MSf , is captured by PC1 ( see Figure 3—source data 1 for details; see also Figure 2—source data 2 for slope comparisons between weeks ) . In other words , the slope βmPC2 represents βmm and βfPC2 represents βfm , and the cross-sex effects are seen in βmPC1 ( = βmf ) and βfPC1 ( = βff; Anthes et al . , 2010 ) . One crucial difference to note between the two analytical approaches is that a more female biased mating rate ( i . e . , higher PC1 values ) can negatively affect male reproductive success ( negative slope , βmPC1 ) , which is not seen in the regression analysis between relative MSf and relative RSm . Similar negative trend lines are seen for βfPC1 and βff , confirming that a more female biased mating rate may also negatively affect the individual’s female reproductive success ( in terms of offspring number , but see below ) . 10 . 7554/eLife . 25139 . 009Figure 3 . The relationships between the PCA values ( based on mating success ) and reproductive success . The relationships are shown for every week of the experiment . The within-sex and cross-sex gradients are based on bivariate regressions of either reproductive success on principal components ( βmPC2 , βfPC2 , βmPC1 , βfPC1 ) . PC1 represents the sexual bias ( the relative difference between MSm and MSf ) ; PC2 represents the overall mating activity ( the correlation component between MSm and MSf ) . Significant slopes ( p < 0 . 05 ) are indicated with a solid fitted line , a trend ( p < 0 . 10 ) is indicated with a dashed line . As shown in Figure 2—source data 3 , in italics , the slopes of different weeks do not significantly differ from each other . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 00910 . 7554/eLife . 25139 . 010Figure 3—source data 1 . Results of the principal component analysis ( PCA ) for each week of the experiment . PC1 represents the sexual bias ( the relative difference between MSm and MSf ) ; PC2 represents the overall mating activity ( the correlation component between MSm and MSf ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 010 During the whole experiment , all the mating interactions were observed . This allowed us to assess whether this species mates fully unilateral or via a form of relaxed reciprocity ( Anthes et al . , 2010 ) . When looking at the relationship between male and female mating success ( respectively , MSm and MSf ) in the Multiple partner treatment , no significant correlation emerges ( Figure 4 ) . We tested this using a GLMM with MSm and Week as factors , MSf as dependent variable and focal identity as random factor ( MSm: F1 , 147 , 6 = 1 . 314 , p = 0 . 254; Week: F7 , 144 , 6 = 12 . 421 , p < 0 . 0001; Interaction: n . s . ) . If animals had alternated roles between matings , either with the same or a different partner , this would have resulted in a positive correlation ( note that in the Single partner treatment reciprocity is enforced ) . The statistical significance of the factor Week simply reflects the cumulative nature of these data and reveals that animals keep mating as male and female throughout the experiment , as also illustrated in Figure 4 . For example , mean cumulative mating success at Week 2 averaged around 3 and at Week 8 around 12 matings for each sexual role . 10 . 7554/eLife . 25139 . 011Figure 4 . The relationship between male and female mating success . The relationship is shown for every week of the experiment . The absence of fitted lines indicates the absence of significance between the individuals’ male and female mating success . The superscripted significance letters indicated with the week numbers indicate the Tukey post-hoc differences between weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 01110 . 7554/eLife . 25139 . 012Figure 4—figure supplement 1 . The relationship between male and female reproductive success . The relationship is shown for each week of the experiment and the single significant slope ( p < 0 . 05 ) is indicated with a solid fitted line . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 012 We also looked at whether there was a clear correlation between male and female reproductive success ( respectively , RSm and RSf ) , which might be indicative of overall individual quality . While there was a slight correlation in Week 2 ( R = 0 . 41 , N = 24 , p = 0 . 046; Figure 4—figure supplement 1; which might explain the positive relationship between MSm and RSf in the first two weeks , see above ) , this relationship was absent throughout the rest of the experiment ( Figure 4—figure supplement 1 ) . This was also tested using a GLMM with RSm and Week as factors , RSf as dependent variable and focal identity as random factor ( RSm: F1 , 151 = 3 . 115 , p = 0 . 080; Week: F7 , 146 , 8 = 28 . 015 , p < 0 . 0001; Interaction: n . s . ) . Again , the statistical significance of the factor Week reflects the cumulative nature of these data and reveals that animals keep obtaining male and female reproductive success over the course of the experiment . For example , mean reproductive success at Week 2 was higher for the female than the male role , but these means lie much closer together at Week 8 , around 2000 eggs ( Figure 1 ) . The reason for male and female reproductive success not being equal is found in the fact that selfing occurs at the start of the experiment , which only contributes to RSf . In order to evaluate whether repeated mating is beneficial for offspring in this species , we assessed differences in growth , egg laying and hatching at the end of the experiment between the three treatment groups . There was no difference in body size ( shell length: ANOVA: F2 , 69 = 2 . 385 , p = 0 . 100; body weight: ANOVA: F2 , 61 = 1 . 441 , p = 0 . 245 ) between the focals of the different treatments . Also , we found no difference in number of eggs between the masses produced by the focals of the different treatments at the end of the experiment ( ANOVA: F2 , 63 = 0 . 155 , p = 0 . 857 ) . Because we did not follow the development of all the egg masses that were laid on the final day of the experiment , we verified whether there was a difference in the number of eggs that we followed per treatment , but this was not the case ( ANOVA: F2 , 31 = 0 . 168 , p = 0 . 846 ) . The overall proportion of hatching success and development scored after 14 days differed between the treatments ( χ22 = 473 . 245 , p < 0 . 0001 ) . Using nonparametric multiple comparisons , we found that in the Multiple partners treatment significantly more offspring had reached hatching than in the No partner treatment ( Wilcoxon: Z = −2 . 609 , p = 0 . 009 ) , while the Single partner treatment did not differ significantly from either treatment ( Figure 5A ) . When looking at the proportion of undeveloped offspring , the Single partner and No partner treatments differ significantly from each other ( Wilcoxon: Z = 2 . 060 , p = 0 . 039 ) , while the difference between the Multiple partners and No partner treatment shows a trend ( Wilcoxon: Z = 1 . 714 , p = 0 . 086; Figure 5B ) . 10 . 7554/eLife . 25139 . 013Figure 5 . Hatching success of eggs collected at the end of the experimental period . The proportion of undeveloped ( A ) and hatched ( B ) eggs is shown for each of the three treatments . The different letters indicate significant differences based on Wilcoxon multiple comparisons ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25139 . 013 The study that we present here is the first to look in detail at the effect of repeated mating on reproductive success over time in a simultaneous hermaphrodite . By using a simultaneous hermaphrodite , we could answer several unresolved questions that are relevant for the understanding of sexual selection in general . Firstly , we showed that the potential gain in reproductive success is consistently positive via the male function . Secondly , the existence of cross-sex effects was supported by the sexual selection gradients , emphasizing that their effects are important in the long run . Thirdly , our data clearly showed that sexual selection gradients change over time , which is important for the interpretation of these values . Fourthly , we showed that in this pond snail , repeated mating was beneficial for the development of offspring . Finally , our data confirmed that the investigated hermaphrodite mates unilateral , without any conditional role alternation . In the following , we will briefly discuss each of these conclusions and place these in the larger context of their general implications for sexual selection . While it becomes clear from the above that it is very informative to run such an experiment for a longer time , this approach also emphasises that it does complicate the interpretation of the sexual selection measures . In short-term experiments , and ( imposed ) semelparous situations , one can separate pre- from post-copulatory components of sexual selection by looking at the residual variance in reproductive success , i . e . the variance not explained by mating success ( this residual can be explained by post-copulatory processes: Rose et al . , 2013a , 2013b; Pélissié et al . , 2014; Janicke et al . , 2015 ) . This is no longer straight-forward in a dataset from a longer running experiment where everyone has mated repeatedly with everyone else and sperm from those previous matings is still in storage . So , even though our data show that at the end of the experiment 19% to 52% of the variance in reproductive success ( RS ) is explained by mating success ( MS; which is essentially pre-copulatory ) , post-copulatory processes , such as sperm storage and sperm competition , have had time to act on reproductive success ( and are thus partially included ) . Hence , one can still attribute the remaining variance to post-copulatory processes , but this no longer captures all this variance . For example , age now becomes a confounding factor and cannot be fully excluded . These considerations are of biologically relevance because they apply to any species , separate sexed or hermaphroditic , that mates multiple times over an extended reproductive period and stores sperm before producing eggs . The only studies that have been able to deal with this issue partly ( though not the factor age ) , so far , are the ones by Pélissié et al . ( 2014 ) and Turnell and Shaw ( 2015 ) . They achieved this by taking mating order into account , which was facilitated by the short duration of their experiments , and their findings largely corroborate what we observe early on in our experiment . To also be able to take the effect of age into account , a similar experiment to the one presented here would need to be run in which the virgin snails ( and their mating partners ) have different standardized ages at the start of the eight-week experiment . Our data reveal that repeated mating , especially with several different partners is beneficial for the hatching success of the offspring . This is especially true when the Multiple partner treatment is compared to the No partner treatment . Hence , this could also suggest that self fertilization , as occurred in the No partner treatment , is detrimental to offspring . Similar effects of selfing have been found for many snail species ( e . g . , Escobar et al . , 2011 ) , and are corroborated by the higher number of undeveloped eggs in that treatment . Note that in this case the non-significant difference between the No partner and Multiple partners treatments , although they seem very different , is due to both sample size and the non-parametric test that needed to be used . Interestingly , results so far had not indicated very clear negative effects of selfing for L . stagnalis ( Escobar et al . , 2011; Puurtinen et al . , 2007 ) , so the fact that we do observe it here may reflect that we looked at the development of offspring in more detail , or that such effects only become apparent in the long run . In addition , while the Single partner treatment differs from neither of the other treatments , it is suggestive of a gradient in which repeated mating may also be beneficial for hatching success when it occurs with the same partner ( but to properly answer this question , a Single copulation treatment would need to be included in a follow-up study ) . Notwithstanding , our data are the first to show that multiple mating does offer a benefit for the development of eggs of L . stagnalis , and are supported by previous work that showed that such eggs are heavier ( Hoffer et al . , 2012 ) . These findings are also in line with work on other species ( e . g . , Callosobruchus maculatus: Power and Holman , 2014 ) . It remains to be investigated whether this effect on offspring has any further implications on Lymnaea’s growth and survival . Multiple mating has already been shown to preferentially occur with different partners ( Koene and Ter Maat , 2007 ) and behavioural studies imply that this species does not alternate mating roles conditionally ( i . e . , they can swap sexual roles within one mating interaction , but this is not obligatory: Koene and Ter Maat , 2005 ) . This is of importance since it has implications for whether sexual selection can act independently on the two sexual roles of a hermaphrodite ( Anthes et al . , 2010 ) ; see also Arnold , 1994 ) . Given that our current data reveal no relation between MSm and MSf , this supports that the mating mode of this species is unilateral ( no exchange of sex roles ) . Hence , male and female strategies can probably be optimized independently in this species . The lack of a clear correlation between RSm and RSf also indicates that an individual successful in the male role is not necessarily successful in the female role . The latter is also supported by our finding that reproductive success can be gained constantly via the male function , as shown by the significant male sexual selection gradients ( βmm ) throughout the experiment . Given that this mating system seems largely male-driven , we expected to also see cross-sex effects . These effects should become visible as negative gradients when looking at male or female reproductive success ( respectively , βmf or βmPC1 and βfm or βfPC1 ) . Such negative effects emerged more clearly from the PCA approach , and only near the end of the experimental period ( βfPC1 ) . This may explain why such cross-sex effects were not found in earlier studies that lasted only for several days ( Anthes et al . , 2010; Pélissié et al . , 2012; Pélissié et al . , 2014 ) . Moreover , our results highlight that cross-sex effects do have an impact on the reproductive success of this hermaphroditic animal . The cross-sex effects that we observe can potentially be explained by the reported negative effects of seminal fluid proteins on both sexual functions ( Koene et al . , 2010; Nakadera et al . , 2014b ) , but this remains to be tested directly . Interestingly , we did not find any indication for a trade-off between investing in the two sexes , which would have resulted in a negative βfm ( although a trend is seen in the last weeks of the experiment for βfPC1 ) . Based on sex allocation theory , one might have expected that increased mating success as a male trades off with female reproductive success ( βfm ) , or vice versa ( βmf; Anthes et al . , 2010; Charnov , 1979; Schärer , 2009 ) . As mentioned above , other studies on different simultaneous hermaphrodites did not find any cross-sex effects , while they did find that the mating system is mainly driven by the male function ( B . glabrata: Anthes et al . , 2010 and P . acuta: Pélissié et al . , 2012 , Pélissié et al . , 2014; but see Janicke et al . , 2015 ) . An important difference between those studies and our study is that while they followed individuals for 3 to 5 days , we followed them for 56 days ( 8 weeks ) . This also allowed us to evaluate the effect of the chosen time frame on the results of such studies . Our data showed that from the start , the main effect , the significant βmm , is already captured ( also with the PCA approach: βmPC2 ) . This relationship became statistically stronger over time . In contrast , the negative effects only emerged much later into the experiment , which indicates that such effects on reproductive success may only be detectable in the longer run . Hence , short term experiments only take a snap shot of reproductive success ( of virgins ) . To conclude , we showed that L . stagnalis has a unilateral mating system and that sperm donors gain most reproductive success from repeated mating , even though they seem to lose some reproductive success in the long run ( a cross-sex effect from mating frequently as a female ) . Our data also do reveal that these sperm recipients benefit indirectly from repeated mating , since their egg development and hatching success is higher . Our experiment therefore showed that the experimental time frame is very important for the quantification and interpretation of sexual selection measures , an insight that applies to any mating system with multiple mating . The basommatophoran pond snail L . stagnalis is a species common to the Holarctic region , and resides in ponds , ditches , and lakes . At the mass culture facility at VU University , a laboratory population of L . stagnalis has been maintained on running , low-copper water for more than 50 years ( Van Der Steen et al . , 1969 ) . Snails are kept under a 12L:12D photoperiod . Each month , egg masses laid within a 24 hr time frame are raised to become the next generation . Snails are alternatingly fed fish food flakes ( TetraPhyll GmbH . ) and broad leaf lettuce . At a shell length of about 18 mm , around the age of two months , individuals begin to copulate , soon followed by production of egg masses . L . stagnalis has a mixed mating system with high outcrossing rates despite low self-fertilization depression ( Nakadera et al . , 2014a , Nakadera et al . , 2017; Puurtinen et al . , 2007; Cain , 1956; Coutellec and Caquet , 2011; Koene et al . , 2009 ) . A single copulation interaction is unilateral , meaning that one partner performs the male role and the other the female role . After an initial copulation role-alternation can take place ( Koene and Ter Maat , 2005 ) and also chain-copulations can be observed in groups . Mating rates increase with population density ( Koene and Ter Maat , 2007 ) and copulation can be easily observed in the laboratory ( Van Duivenboden and Ter Maat , 1988; De Boer et al . , 1996 ) . L . stagnalis is a promiscuous species and can store received sperm ( allosperm ) for about 2 months ( 62 days: Nakadera et al . , 2014a ) and use it seemingly random when fertilizing eggs ( Koene et al . , 2009 ) . Pond snails are highly fecund and usually lay between 100–300 eggs per week in 1–4 egg masses , depending on mating conditions ( Hoffer et al . , 2012; Van Duivenboden et al . , 1985 ) . Two hundred immature snails ( shell length < 15 mm ) , that hatched from multiple egg masses laid on the same day ( ±24 hr ) , were collected from our mass culture . They were individually housed in perforated plastic jars in a laminar-flow basin ( 20 ± 1°C ) to let them reach maturity . A bee tag was glued to their shell for identification purposes . For the duration of the maturation period and the experiment , 19 . 6 cm2 of lettuce was provided daily per animal . During this maturation period , a clean jar was provided weekly , and egg laying capability was checked . At 14 weeks after hatching ( shell length ~30 mm ) all virgins were confirmed to be laying self-fertilized eggs . After this confirmation , all snails were sedated with 50 mM MgCl2 and , using fine surgical scissors ( World Precision Instruments , Inc . , Saratosa , USA ) , a small part of their foot was cut off for genotyping purposes ( see next section ) . The quantification of mating success ( MS ) and reproductive success ( RS ) started when animals - still virgin - were 110 days old , at which time they had had plenty of time to fully recover from tissue sampling . The experiment included three treatments: Multiple partners ( 25 groups of 5 snails each ) , Single partner ( 25 groups of 2 snails each ) , and No partner ( 25 single snails that remained virgins ) . The Single partner and No partner treatments were included in the experiment to test for potential effects of repeated mating with different partners in the Multiple partners treatment . For the snails in groups , we made sure that the focal had a microsatellite genotype ( see Genotyping protocol ) that could be unequivocally distinguished from the other snails in its group . The genotypes of the non-focal individuals necessarily overlapped because we found three alleles at this locus within our experimental population . During eight weeks , for each treatment mating activity was observed twice a week for 7 hr on the first and fourth day of each week ( Figure 1 ) . The snails only had access to mating partners during these 7 hr mating trials , when they were together in a jar . During this time , the volume of water per individual was set at 100 ml , so that snail density was equal among treatments ( 500 ml for Multiple partner groups , 200 ml for Single partner pairs , and 100 ml for No partner virgins ) . The rest of the time all snails were kept in their isolation jars ( Figure 1A ) . Within all groups , copulation and mating role was noted for each individual ( 175 snails in total all of which mated more than once in the male and female role ) , hence we had complete observational data based on which we could calculate each individual’s MS . We used cumulative number of matings , not mates , because this species mates frequently and within the first weeks all individuals have already mated at least once with each group member in both sexual roles , hence maximizing number of mates . Thus , due to the restricted number of different mates available , Bateman gradients could not be quantified as these are defined based on mate identity . By looking at mating frequency we thus determined sexual selection gradients rather than Bateman gradients sensu stricto ( see also Anthes et al . , 2010; Collet et al . , 2014; Fritzsche and Arnqvis , 2013; Marie-Orleach et al . , 2016 ) . Egg masses laid during mating trials were removed from the container and placed in the isolation jar of the mother . Each week , egg masses were collected from the isolation jars , measured to the nearest 0 . 5 mm , and placed in 10 ml vials , one for every individual . The egg numbers of 24 random egg masses per treatment were counted , making it possible to estimate the number of eggs in an egg mass of length x for each week . After counting , egg masses were returned to their vials and then freeze-dried . The total dry weight of all the egg masses was determined on a microbalance ( type 1712 MP8 , Sartorius ) to the nearest 0 . 01 mg . In the first , second , fourth , sixth and eighth week , one egg mass per individual ( if any ) from the Multiple partners treatment was allowed to develop until the embryo was large enough for genotyping ( at 9–10 days after egg laying ) . Then , these masses were freeze dried , weighed , and stored at −20°C until offspring genotyping . For all snails , shell length and body weight were measured on day 15 and day 57 after the start of the behavioural observations . Growth in L . stagnalis follows a sigmoid curve , and the experimental period started while the snails had entered the asymptotic phase , thus restricting potential budget effects . Egg masses laid on day 57 ( the day after the eighth week; the end of the experiment ) were placed in Petri dishes with 15 ml of water each , which was refreshed every other day for two weeks . After 14 days , for each egg mass the number of undeveloped eggs , early embryos , late embryos and hatchlings were counted under a stereo microscope . For genotyping the experimental animals , total genomic DNA was extracted by crushing tissue samples in 100 µl 50 mM NaOH in a 1 . 5 ml vial , vortexed and left standing for 10 min . After digestion of the connective tissue the solution was neutralized by adding 10 µl 1 M TRIS-HCl of pH 8 . 0 ( protocol adapted from Meeker et al . , 2007 ) . After centrifugation at 14000 rpm for 10 min , the supernatant was transferred to a clean vial . The precipitate containing the tissue debris was discarded . PCR amplification of the A16 microsatellite locus ( Knott et al . , 2003; please note that the reverse primer is displayed in 3'−5' orientation in that publication ) was performed in 25 µl reaction mixture containing 5 . 0 µl of 5x PCR buffer , 1 . 5 µl of 25 mM MgCl2 , 2 . 0 µl of 10 mM dNTP’s , 1 µl of the 5 µM forward and reverse primer each , 0 . 2 µl GO-taq polymerase ( Promega ) , plus 0 . 02 µl proofreading polymerase ( pfu , Promega ) and 9 . 3 µl H2O ( Sigma ) . Lastly , 5 µl of genomic DNA sample was added to the reaction mixture . The PCR amplification protocol consisted of an initial denaturation at 95°C for 5 min , followed by 35 cycles of 95°C for 15 s , 55°C for 45 s , and 72°C for 60 s , with a final extension period of 72°C for 5 min in a thermocycler ( MJ Research PTC-100 ) . A volume of 16 µl amplification product was added to 4 µl loading dye ( Elchrom Scientific ) which was then denatured at 95°C for 5 min , and chilled on ice . Spreadex EL 600 Wide Mini Gels ( S-2 × 25 slots , Elchrom Scientific ) were submerged in a buffer solution ( 55°C , 0 . 8x TAE ) , and slots were carefully filled with PCR product , including one slot for each half gel for a 250 bp DNA ladder . Electrophoresis was performed at 120 V for 165 min , with a second loading PCR product after 45 min . Gels were stained in 150 ml 0 . 25x TAE buffer containing 15 µl Syber Gold for 45 min . All gels were photographed and snails were visually genotyped by two persons , without inconsistencies . Genotyping of offspring was performed on single eggs . Because the high fecundity of L . stagnalis ( ~100–300 eggs per week per individual ) ruled out complete genotyping of all offspring , we genotyped between 10 and 24 ( 16 on average ) randomly selected offspring per developed egg mass of focal and non-focal individuals ( see also Experimental design ) . Sixteen embryonic snails , recognized by their dark colour , were removed randomly from an egg mass and put singly in a well of a 96-wells PCR-plate . When the plate was full ( 6 × 16 offspring ) the tissue was crushed in 50 µl 50 mM NaOH , incubated at room temperature for 10 min , and neutralized with 5 µl 1 M TRIS-HCl of pH 8 . After centrifugation , supernatant was either stored at −20°C or used for amplification directly . The PCR and electrophoresis protocol was identical to the one mentioned above . For all focal individuals , reproductive success in the female role ( RSf ) was expressed as the number of eggs produced by the focal , and was calculated on a weekly basis . Male reproductive success ( RSm ) of the focals was estimated based on the genotyping data of the A16 microsatellite for the random subset of eggs as described above ( see Genotyping protocol ) . Because we had incomplete paternity sampling ( i . e . , not all eggs from each individual were genotyped; see also Mobley and Jones , 2013 ) , we used the observed paternity and overall proportion of female matings of each mate within each group to estimate paternity for the non-genotyped egg masses . We entered this proportion and the actual number of fathered offspring ( determined by genotyping ) into a generalized linear model ( GLM ) . We used a binomial distribution and a logit link function ( logistic regression ) , corrected for overdispersion and used the number of genotyped offspring as a weighing factor ( GLM fit: χ21 = 64 . 156 , p < 0 . 0001 ) . This predicted paternity share was then multiplied by the total number of offspring produced by the mating partners in each group , thus resulting in an estimate of RSm for each focal ( see Pélissié et al . , 2012 ) . All statistical procedures were performed using JMP 9 . 0 . 0 ( SAS ) . Due to different limits in terms of their gamete production ( as explained above ) , variance in the reproductive success of males is expected to be larger than that of females and can be captured in the variance measure I . This measure is defined as the standardized variance in relative reproductive success and its value is indicative of the opportunity for selection ( Gowaty et al . , 2003; Jones , 2009 ) . The opportunity for sexual selection , which is defined as the standardized variance in relative mating success , is captured in the variance measure Is ( Arnold , 1994; Anthes et al . , 2010 ) . In contrast to such opportunity values , a real measure of sexual selection can be obtained by looking at the relationship between mating success and reproductive success , which generally results in a steeper regression line for males than females . The slope of such a linear regression line is referred to as the Bateman or sexual selection gradient ( β; e . g . , Arnold and Duvall , 1994; Anthes et al . , 2010 , Anthes et al . , 2016 ) . We first calculated the opportunity for ( overall ) selection , I , by dividing RS’s variance by its squared mean . Likewise , we calculate the opportunity for sexual selection , Is , by dividing MS’s variance by its squared mean . Given that we are dealing with a simultaneous hermaphrodite , these can be calculated both for the male ( Im , Ism ) and female ( If , Isf ) role ( see Lorenzi and Sella , 2008; Shuster and Wade , 2003 ) . Subsequently , we calculated the other important , and often used , measure of sexual selection , the sexual selection gradient ( β ) . As already explained in the Results section , this is the linear least-squares regression slope of sex-specific relative RS on sex-specific relative MS ( Jones , 2009; Klug et al . , 2010 ) . To deal with the non-independence of male and female reproduction in these hermaphrodites , we used a multiple regression with MSm and MSf as explanatory variables on , respectively , RSf and RSm . For simultaneous hermaphrodites , depending on their mating system , male and female mating success may not be fully independent . Even in unilaterally mating species , reciprocity in mating ( i . e . , playing both roles in a mating encounter ) could make a multiple regression analysis statistically fragile ( Wacker et al . , 2014 ) . We followed Anthes et al . ( 2010 ) suggestion to replace MSm and MSf by their principal components ( PC ) to cope with this potential problem . This approach results in two completely independent new variables ( PC1 and PC2 ) that represent overall mating activity and the sex bias in mating , not necessarily in that order . Finally , we investigated the effect of time on the above measures by comparing the measures of sexual selection over time in the experiment using a GLMM . So far , experiments were generally only performed over a short time frame , as pointed out earlier; e . g . Pélissié et al . , 2012 , Pélissié et al . , 2014; Anthes et al . , 2010; Collet et al . , 2012; Rose et al . , 2013a; but see Turnell and Shaw , 2015 ) .	Many factors affect an organism’s ability to survive and reproduce . These factors are often called “selection pressures” and include the availability of food and shelter , conditions in the environment such as temperature , and the presence of diseases and predators . Males and females experience different selection pressures so they often evolve to look different – consider , for example , the male deer’s antlers and the peacock’s colourful tail feathers . Such traits arise from a phenomenon called sexual selection , the selection pressures that act on an organism’s ability to obtain a mate . Measuring sexual selection is not only of interest to scientists looking to understand how evolutionary processes work; it also has wider applications , including in wildlife conservation . For instance , knowing which cues are important for successful reproduction could help efforts to breed endangered animals in captivity and stop them from going extinct . Scientists study sexual selection in a species by measuring how successful males and females are at mating and reproducing . Past studies have found that a female’s reproductive success mainly depends on there being enough resources available for her to produce eggs , while a male’s success depends on him getting access to these eggs . However , most research into sexual selection has been on species with separate sexes . It is more difficult to measure sexual selection in species – like snails and slugs – where each individual is male and female at the same time . As such , it is not clear if reproductive success in these species , which are known as simultaneous hermaphrodites , depends on the same factors as those species with separate sexes . To address this , Hoffer et al . measured sexual selection in the great pond snail Lymnaea stagnalis , a simultaneous hermaphrodite . Most studies estimate sexual selection based on measurements taken over several days . Instead , Hoffer et al . observed the great pond snail over a period of eight weeks , which is about a quarter of its reproductive life . The experiments showed that mating multiple times , especially with multiple partners , overall improves the development of the snail’s offspring . The male part of the great pond snail gains the most reproductive success from repeated mating , whereas the female part may in fact be negatively affected . These negative effects were only seen several weeks into the experiment , and so they show that sexual selection pressures change over time . Future research is needed to determine what causes the negative effects on the female part of the great pond snail . Overall , these findings stress the need for careful consideration of the time frame over which future measurements of sexual selection take place , not just in hermaphrodites , but in all species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2017	Sexual selection gradients change over time in a simultaneous hermaphrodite
The vascular pathogen Verticillium dahliae infects the roots of plants to cause Verticillium wilt . The molecular mechanisms underlying V . dahliae virulence and host resistance remain elusive . Here , we demonstrate that a secretory protein , VdSCP41 , functions as an intracellular effector that promotes V . dahliae virulence . The Arabidopsis master immune regulators CBP60g and SARD1 and cotton GhCBP60b are targeted by VdSCP41 . VdSCP41 binds the C-terminal portion of CBP60g to inhibit its transcription factor activity . Further analyses reveal a transcription activation domain within CBP60g that is required for VdSCP41 targeting . Mutations in both CBP60g and SARD1 compromise Arabidopsis resistance against V . dahliae and partially impair VdSCP41-mediated virulence . Moreover , virus-induced silencing of GhCBP60b compromises cotton resistance to V . dahliae . This work uncovers a virulence strategy in which the V . dahliae secretory protein VdSCP41 directly targets plant transcription factors to inhibit immunity , and reveals CBP60g , SARD1 and GhCBP60b as crucial components governing V . dahliae resistance . The vascular pathogen Verticillium dahliae infects a broad range of plants and causes devastating diseases . Verticillium dahliae can survive in the form of microsclerotia in soil for over ten years ( Schnathorst , 1981 ) . During V . dahliae colonization , microsclerotia germinate and develop hyphae that , upon perception of plant roots , adhere tightly to the root surface ( Zhao et al . , 2014 ) . A few hyphae form hyphopodia at the infection site and further differentiate into penetration pegs that penetrate into plant cells and colonize the vascular tissue ( Fradin and Thomma , 2006; Schnathorst , 1981; Vallad and Subbarao , 2008; Zhao et al . , 2014Zhao et al . , 2016 ) . Although a few key steps mediating these infection processes have been elucidated , the molecular mechanisms underlying V . dahliae virulence remain largely unknown . Progress has been made in isolating virulence factors that are crucial for V . dahliae virulence . Several genes that regulate V . dahliae development have been characterized as contributing to virulence . VDH1 and VdGARP1 , which regulate microsclerotial development , are required for V . dahliae virulence in cotton plants ( Gao et al . , 2010; Klimes and Dobinson , 2006 ) . VdRac1 and VdPKAC1 regulate both the development and the pathogenicity of the fungus in host plants ( Tian et al . , 2015; Tzima et al . , 2010 , 2012 ) . VdEG1 , VdEG3 , VdCYP1 , and VdSNF1 also function as factors that contribute to V . dahliae virulence as it infects cotton ( Gui et al . , 2017; Tzima et al . , 2011; Zhang et al . , 2016 ) . In addition to development-associated virulence factors , fungal pathogens deliver effectors that act as virulence factors , thereby inhibiting host defense and promoting pathogenesis . Bacterial and oomycete pathogens also deliver such effectors . The race 1 strain-specific effector Ave1 contributes to V . dahliae virulence on tomato plants not carrying Ve1 ( de Jonge et al . , 2012 ) . Necrosis and ethylene-inducing peptide 1 ( Nep1 ) -like proteins ( NLPs ) ( NLP1 and NLP2 ) secreted by V . dahliae strain JR2 are required for its pathogenicity in tomato and Arabidopsis plants ( Santhanam et al . , 2013 ) . VdSge1 encodes a transcriptional regulator that controls the expression of six putative effector genes . Deletion of VdSge1 in V . dahliae significantly impairs its pathogenicity in tomato , suggesting an important role for secreted effectors in suppressing host immunity in V . dahliae ( Santhanam and Thomma , 2013 ) . Nevertheless , only two secreted effectors of V . dahliae have been indicated to function inside the plant cell to modulate host immunity to date . VdIsc1 , a V . dahliae effector lacking a known signal peptide , is thought to be delivered into host cells to hydrolyze a salicylic acid ( SA ) precursor and thereby inhibit salicylate metabolism ( Liu et al . , 2014 ) . The small cysteine-containing protein ( SCP ) VdSCP7 translocates into the nucleus of plant cells to either suppress or induce defense in plants through unknown mechanisms ( Zhang et al . , 2017 ) . Plants are equipped with immune components to counteract V . dahliae virulence . Tomato Ve1 has been identified as an effective resistance locus that recognizes Ave1 secreted by Race 1 strains ( Fradin et al . , 2009; Schaible et al . , 1951 ) . Genetic analyses indicated that EDS1 , NDR1 and SERK3/BAK1 are required for Ve1-mediated resistance in both tomato and Arabidopsis ( Fradin et al . , 2009 , 2011 ) . Studies in cotton also showed that GhBAK1 and GhNDR1 are crucial components in regulating defense against V . dahliae ( Gao et al . , 2013b; Gao et al . , 2011 ) , demonstrating that both NDR1 and SERK3/BAK1 are required in a conserved mechanism for defense against V . dahliae in these plants . GbWRKY1 , GhSSN , GbERF , GhMLP28 , GhMKK2 and GbNRX1 have also been shown to be required for cotton resistance against V . dahliae ( Gao et al . , 2011; Li et al . , 2014a , 2016; Qin et al . , 2004; Sun et al . , 2014; Yang et al . , 2015 ) . A comparative proteomic analysis indicated the involvement of both brassinosteroids and jasmonic acid signaling pathways in the regulation of cotton resistance to V . dahliae ( Gao et al . , 2013a ) . Although several regulators have been identified , the mechanisms through which plants defend against V . dahliae remain obscure , and further investigation is required to isolate more host immune components governing V . dahliae resistance . Targeting key immune components is a common strategy employed by pathogenic effectors to promote pathogenicity ( Boller and He , 2009; Cui et al . , 2009; Dou and Zhou , 2012 ) ; thus , the screening of host proteins targeted by pathogenic effectors provides an efficient way to identify crucial host defense components . The calmodulin-binding proteins ( CBPs ) function to bind calmodulin and thus to transduce calcium signals ( Bouché et al . , 2005 ) . The plant-specific CALMODULIN BINDING PROTEIN 60 ( CBP60 ) protein family contains eight family members , including CBP60a–g and SYSTEMIC ACQUIRED RESISTANCE DEFICIENT 1 ( SARD1 ) ( Bouché et al . , 2005 ) . CBP60g and SARD1 are two closely related members that function partially redundantly in both SA signaling and bacterial resistance ( Zhang et al . , 2010; Wang et al . , 2011 ) . CBP60g contains a calmodulin-binding domain ( CBD ) that is essential for its function in defense , whereas SARD1 does not bind calmodulin ( Wang et al . , 2009Wang et al . , 2011; Zhang et al . , 2010 ) . Both CBP60g and SARD1 function as transcription factors that directly bind to the promoters of genes that control SA synthesis , such as Isochorismate Synthase 1 ( ICS1 ) , ENHANCED DISEASE SUSCEPTIBILITY 5 ( EDS5 ) , and NON-EXPRESSOR OF PATHOGENESIS RELATED GENES1 ( NPR1 ) ( Dong , 2004; Nawrath et al . , 2002; Sun et al . , 2015; Wildermuth et al . , 2001 ) . Moreover , ChIP-seq analyses have revealed that CBP60g and SARD1 directly bind to the promoters of a number of genes , thereby regulating pathogen-associated molecular patterns ( PAMPs ) -triggered immunity ( PTI ) , effector-triggered immunity ( ETI ) and systemic acquired resistance ( SAR ) ( Sun et al . , 2015 ) , indicating their broad role in the regulation of plant immunity . In this study , we identified VdSCP41 as a virulence effector that suppresses plant immunity induced by PAMPs . VdSCP41 interacts with Arabidopsis CBP60g and SARD1 and modulates their transcription factor activity . The contribution of VdSCP41 to V . dahliae virulence is significantly reduced during the infection of the cbp60g-1/sard1-1 double mutant . GhCBP60b , the closest protein homolog of CBP60g in cotton , is also targeted by VdSCP41 and contributes to cotton resistance against V . dahliae . Taken together , our findings revealed that CBP60g , SARD1 and GhCBP60b are novel components that govern V . dahliae resistance and that these proteins are modulated by a secretory effector VdSCP41 . Secretome analyses revealed more than 700 potential secreted proteins in V . dahliae ( Klosterman et al . , 2011 ) , but to date , relatively few of these secreted proteins have been characterizes as carrying virulence function . We performed a reverse genetic screen to identify secreted proteins that are crucial for V . dahliae virulence . Using the newly developed USER-ATMT-DS binary vector ( Wang et al . , 2016 ) , we constructed 56 gene deletion mutants in V . dahliae strain 592 ( V592 ) , each of which targets an individual potentially secreted protein . The resulting mutants were subjected to virulence assessment in host plants , including upland cotton ( Gossypium hirsutum ) and the model plant Arabidopsis . A mutant carrying a targeted deletion of VdSCP41 ( VdΔscp41 ) was isolated ( Figure 1A ) and shown to display significantly reduced virulence compared with WT strain V592 . VdSCP41 encodes a hypothetical protein in the secretome of the V . dahliae Ls . 17 strain ( VdLs . 17 ) ( Klosterman et al . , 2011 ) . The expression of this gene is significantly upregulated at 2 days post-inoculation ( dpi ) in Arabidopsis ( Figure 1—figure supplement 1 ) , suggesting a putative function of VdSCP41 in V . dahliae infection . Targeted gene deletion of VdSCP41 in the VdΔscp41 mutant was verified by southern blotting ( Figure 1B ) . The VdΔscp41 mutant displayed much weaker disease symptoms than WT V592 in both upland cotton ( Figure 1C ) and Arabidopsis ( Figure 1D ) . Disease index analyses indicated significantly reduced virulence of the VdΔscp41 mutant compared with that of V592 in both upland cotton and Arabidopsis ( Figure 1E–F , Figure 1—source data 1 ) . The reduced virulence of the VdΔscp41 mutant was restored upon complementation with GFP-tagged VdSCP41 ( VdΔscp41/VdSCP41-GFP ) ( Figure 1C–F ) . Thus , VdSCP41 functions as a virulence effector that contributes to V . dahliae virulence on host plants . In V . dahliae , some signal-peptide-containing SCPs are delivered to the septin-organized hyphal neck , which develops from the base of the hyphopodia and functions as a fungus–host penetration interface for the dynamic delivery of secretory proteins ( Zhou et al . , 2017 ) . VdSCP41 contains an N-terminal signal peptide predicted by SignalP ( http://www . cbs . dtu . dk/services/SignalP/ ) ( Figure 2—figure supplement 1A ) . Therefore , we examined the localization of VdSCP41 in V . dahliae . GFP-tagged VdSCP41 ( VdSCP41-GFP ) was found to localize to the base of the hyphopodium and showed ring signals surrounding the hyphal neck when the Vd∆scp41 mutant strain complemented with VdSCP41-GFP ( Vd∆scp41/VdSCP41-GFP ) was cultured on a cellophane membrane for hyphopodium induction ( Figure 2A ) . By contrast , VdSCP41 lacking signal peptide ( ΔspVdSCP41-GFP ) showed diffused signal at the base of the hyphopodium without clear ring signals surrounding the hyphal neck . The results demonstrate that VdSCP41-GFP was delivered to the penetration interface for secretion . Nuclear localization signal ( NLS ) sequence prediction ( http://nls-mapper . iab . keio . ac . jp/cgi-bin/NLS_Mapper_form . cgi ) identified a potential NLS in VdSCP41 ( Figure 2—figure supplement 1A ) . We therefore took advantage of this NLS to examine the putative translocation of V . dahliae-delivered VdSCP41 into the nucleus of plant cells . The Vd∆scp41/VdSCP41-GFP and GFP-expressing V592 ( V592-GFP ) strains were separately inoculated onto onion epidermal cells . Although GFP fluorescence from the V592-GFP strain was observed in conidial spores , VdSCP41-GFP secreted by V . dahliae was capable of translocating into plant cells , and of localizing to the nucleus in addition to the pericellular space of onion epidermal cells ( Figure 2—figure supplement 1B ) . The potential NLS sequence of SCP41 was mutated ( SCP41-nls-GFP ) and then complemented into Vd∆scp41 to construct the VdΔscp41/VdSCP41-nls-GFP strain . In contrast to VdSCP41-GFP , VdSCP41-nls-GFP failed to translocate into the nucleus of onion epidermal cells ( Figure 2—figure supplement 1B ) . These results suggested that VdSCP41 delivered by V . dahliae translocates into the nucleus of the plant cell , which requires the NLS predicted within its sequence . We next transiently expressed mCherry-tagged VdSCP41 in plants to further verify its nuclear localization in plant cells . As the signal peptide located at the N terminus may guide secreted proteins into the plant extracellular space in some cases , we fused mCherry to VdSCP41 both with and without ( ΔspVdSCP41 ) the signal peptide in order to analyze subcellular localization . VdSCP41-mCherry and ΔspVdSCP41-mCherry were individually transiently expressed in either Arabidopsis protoplasts or in Nicotiana benthamiana ( N . b . ) leaves . mCherry fluorescence imaging revealed that both VdSCP41 and ΔspVdSCP41 localized to the nucleus in Arabidopsis cells ( Figure 2B ) . The protein expression level of VdSCP41-mCherry and ΔspVdSCP41-mCherry was detected by immunoblot ( Figure 2—figure supplement 1C ) . Similar nuclear localization was observed for both VdSCP41-mCherry and ΔspVdSCP41-mCherry in N . b . cells ( Figure 2—figure supplement 1D ) . These results are consistent with the nuclear localization of the V . dahliae-delivered VdSCP41 in onion epidermal cells . It is believed that the initial function of a fungal effector protein is to suppress PTI . To investigate whether VdSCP41 is capable of inhibiting plant immunity when it is directly expressed in plants , Arabidopsis transgenic lines expressing ΔspVdSCP41 were constructed and assessed for PAMP-induced defense responses . Flg22 is the best-characterized PAMP derived from a bacterial pathogen , and it induces the expression of PTI-responsive genes in wildtype ( WT ) Arabidopsis plants . We observed reduced induction of flg22-induced ICS1 in two independent VdSCP41-expressing lines ( Figure 2—figure supplement 1E ) , suggesting inhibition of PTI conferred by VdSCP41 . NLP proteins derived from bacterial , oomycete and fungal organisms have recently been characterized as PAMPs . A conserved 20-amino-acid peptide ( nlp20 ) within NLP represents the active immunogenic motif that induces PTI in plants ( Albert et al . , 2015; Böhm et al . , 2014; Ottmann et al . , 2009 ) . We previously reported the immune-inducing activity of VdNLP1 and VdNLP2 derived from V592 in N . b . , Arabidopsis , and cotton plants ( Zhou et al . , 2012 ) . A corresponding peptide located in VdNLP2 derived from V592 ( nlp20Vd2 ) was synthesized and shown to cause upregulation of cotton PR genes ( Du et al . , 2017 ) , indicating the immunogenic activity of this peptide in cotton . Treatment of Arabidopsis with nlp20Vd2 significantly induced ICS1 and FMO1 expression ( Figure 2C–D , Figure 2—source data 1 ) , indicating that nlp20Vd2 also exhibits the immunogenic activity characteristic of a PAMP in Arabidopsis . In transgenic lines expressing VdSCP41 , but not transgenic lines expressing NLS mutated VdSCP41 ( ΔspVdSCP41-nls ) , suppression of nlp20Vd2-induced ICS1 ( Figure 2C ) and FMO1 ( Figure 2D ) was observed , suggesting that VdSCP41 expression in plants inhibits nlp20Vd2-triggered immunity . ICS1 encodes the key enzyme controlling SA production and is required for pathogen-induced SA accumulation ( Wildermuth et al . , 2001 ) . We next quantified SA production in response to a nonpathogenic bacterial pathogen , Pst DC3000 hrcC– , in both WT and ΔspVdSCP41-expressing plants . Consistent with the suppression of PAMP-induced ICS1 expression , transgenic lines expressing ΔspVdSCP41 accumulated less free SA in response to Pst hrcC– than did WT plants ( Figure 2—figure supplement 2A ) . Inoculation of Arabidopsis with the WT V . dahliae strain V592 induced the expression of ICS1 and FMO1 ( Figure 2—figure supplement 2B–C ) . This induced expression was further enhanced when the plants were inoculated with the VdΔscp41 mutant instead of V592 ( Figure 2—figure supplement 2B–C ) , indicating that VdSCP41 suppresses the induced expression of ICS1 and FMO1 during V . dahliae infection . Taken together , these results reveal an inhibitory role of VdSCP41 in modulating plant immunity . Modulating the activity of plant immune components is a strategy commonly used by effectors to suppress host immunity . To explore the virulence mechanisms employed by VdSCP41 in inhibiting plant immunity , we next searched for plant components that are targeted by VdSCP41 . ΔspVdSCP41 was fused with a 3 × FLAG tag and transiently expressed in Arabidopsis protoplasts , before the VdSCP41-containing protein complexes were purified . Protein lysates were immuno-precipitated using anti-FLAG-conjugated beads and subjected to tandem mass spectrometry . The plant CBP CBP60g was identified as a candidate interactor of VdSCP41 ( Supplementary file 1 ) . CBP60g was then fused with a 3 × HA tag and used for reverse co-immunoprecipitation ( Co-IP ) analysis to verify its interaction with VdSCP41 . FLAG-tagged VdSCP41 was transfected , either alone or together with HA-tagged CBP60g , into Arabidopsis protoplasts for transient expression . Anti-HA IP followed by an anti-FLAG immunoblot revealed that VdSCP41 was co-purified with CBP60g from plant cells ( Figure 3—figure supplement 1A ) . VdSCP41 was further divided into an N-terminal portion ( VdSCP41N , amino acids 1–213 ) and a C-terminal portion ( VdSCP41C , amino acids 163-end ) , which exhibited an overlap of 50 amino acids , which were used to test interactions with CBP60g . Co-IP analysis revealed that VdSCP41C is sufficient for interaction with CBP60g ( Figure 3A ) . Quantitative luciferase complementation imaging assays were performed to further verify the interaction between VdSCP41 and CBP60g in N . b . . Co-expression of the N-terminal region of luciferase ( NLuc ) -tagged BIK1 and the C-terminal region of luciferase ( CLuc ) -tagged XLG2 driven by the35S promoter was performed as a positive interaction control ( Liang et al . , 2016 ) . The co-expression of NLuc-VdSCP41 and CLuc-CBP60g driven by the 35S promoter in N . b . resulted in much higher luciferase activity than did the co-expression of NLuc-VdSCP41 and CLuc-XLG2 or of CLuc-CBP60g and NLuc-BIK1 ( Figure 3B–C , Figure 3—source data 1 ) , confirming the interaction between VdSCP41 and CBP60g in the plants . The expression levels of the NLuc- and CLuc-fusion proteins were further detected by immunoblotting ( Figure 3—figure supplement 1B ) . The nuclear localization of VdSCP41 prompted us to examine whether CBP60g co-localizes with VdSCP41 in the nucleus . GFP-tagged CBP60g mainly localized in the nucleus when it was expressed alone in N . b . cells ( Figure 3—figure supplement 1C ) . GFP-tagged CBP60g was co-expressed with mCherry-tagged VdSCP41 without signal peptide ( ΔspVdSCP41-mCherry ) in N . b . leaves through Agrobacterium-mediated transient expression . An overlay of the results of GFP and mCherry fluorescence imaging indicated co-localization of ΔspVdSCP41 and CBP60g in the nucleus of N . b . cells ( Figure 3D ) . The protein expression level of GFP-CBP60g , ΔspVdSCP41-mCherry and ΔspVdSCP41-nls-mCherry was detected by immunoblot with the indicated antibodies ( Figure 3—figure supplement 1D ) . In addition to co-localization , co-expression of ΔspVdSCP41-mCherry significantly increased the nuclear accumulation of CBP60g-GFP ( Figure 3D ) , which was not observed when ΔspVdSCP41-nls–mCherry with a mutated NLS was co-expressed with CBP60g-GFP ( Figure 3D ) . CBP60g was induced by pathogen and PAMP treatments and was required for full resistance against bacterial pathogens ( Wang et al . , 2011; Zhang et al . , 2010 ) . A closely related protein in the CBP family , SARD1 , functions partially redundantly with CBP60g in bacterial resistance ( Sun et al . , 2015; Wang et al . , 2011 ) . We also detected an interaction between VdSCP41 and SARD1 by Co-IP in Arabidopsis protoplasts ( Figure 3—figure supplement 2A ) . VdSCP41 co-localized with SARD1 in N . b . leaves and increased its nuclear accumulation ( Figure 3—figure supplement 2B ) , indicating similar targeting of SARD1 by VdSCP41 in N . b . . It is unlikely to be an interaction between CBP60g and SARD1 because the luciferase complementation assay did not show interaction between CLuc-tagged CBP60g and NLuc-tagged SARD1 ( Figure 3—figure supplement 3 ) . Taken together , the results described above demonstrated that both CBP60g and SARD1 are targeted by VdSCP41 . CBP60g encodes a plant-specific transcription factor that regulates the expression of a number of defense-related genes . The fact that CBP60g functions as a master transcription regulator ( Sun et al . , 2015; Wang et al . , 2011 ) prompted us to examine whether the targeting of CBP60g by VdSCP41 affects the induction of its target genes . Dual reporter analyses revealed that CBP60g expression in Arabidopsis protoplasts significantly enhanced the expression of ICS1::LUC or FMO1::LUC ( firefly luciferase ) ( Figure 4A–B , Figure 4—source data 1 ) . Co-expression of VdSCP41 inhibited CBP60g-induced ICS1::LUC ( Figure 4A ) and FMO1::LUC ( Figure 4B ) expression , whereas the co-expression of VdSCP41N , which is unable to bind CBP60g , was impaired in this inhibition ( Figure 4A–B ) . CBP60g was next divided into an N-terminal portion ( CBP60gN , amino acids 1–361 ) containing its DNA-binding domain and a C-terminal portion ( CBP60gC , amino acids 211-end ) lacking the functional DNA-binding domain ( Zhang et al . , 2010 ) , which exhibited an overlap of 150 amino acids . CBP60gN and CBP60gC were then used to test interactions with VdSCP41 . Co-IP analysis showed that VdSCP41 binds to CBP60gC , but not CBP60gN ( Figure 4C ) . The results indicated that VdSCP41 binds the C-terminal portion of CBP60g to interfere with its activity . To test whether VdSCP41 targeting directly affects the DNA-binding activity of CBP60g , recombinant GST-tagged CBP60g and His-tagged VdSCP41C were purified and used for electrophoretic mobility shift assays ( EMSAs ) . GST-CBP60g showed a specific binding to a 60-bp DNA fragment ( ICS1 promoter probe ) within the ICS1 promoter , which is reduced by the addition of unlabelled probe , as previously reported ( Zhang et al . , 2010 ) ( Figure 4D ) . The preincubation of VdSCP41C with CBP60g significantly reduced the DNA-binding activity of CBP60g , whereas a soluble fragment of His-tagged VdSCP41 which does not contain the C-terminal portion ( VdSCP41100-163 ) did not ( Figure 4D ) . Another His-tagged V . dahliae protein ( VDAG_01962 ) also did not affect the DNA-binding activity of CBP60g ( Figure 4D ) . Coexpression of VdSCP41 did not lead to cleavage or mobility shift of CBP60g ( Figure 4—figure supplement 1 ) , suggesting that VdSCP41 is unlikely to act as a protease to target CBP60g . The results proved that binding of VdSCP41C to CBP60g directly inhibits the DNA-binding activity of CBP60g . The binding of VdSCP41 to the C-terminal portion of CBP60g prompted us to test the role of CBP60gC in CBP60g-mediated gene activation . Compared to the full induction of ICS1::LUC or FMO1::LUC by CBP60g ( Figure 5A , Figure 5—source data 1 ) , the deletion of the C-terminal portion ( ΔC-CBP60g ) dramatically compromised its activity to induce both ICS1::LUC and FMO1::LUC ( Figure 5A ) , indicating that CBP60gC is required for CBP60g-mediated gene activation . The results suggest that putative transcription activator activity may be harbored within the CBP60gC . The basic helix-loop-helix ( bHLH ) transcription factor MYC2 directly binds to the G-box-like ( CANNTG ) element ( Dombrecht et al . , 2007; Godoy et al . , 2011; Lian et al . , 2017 ) via its bHLH domain to regulate the expression of its target genes , such as the TERPENE SYNTHASE gene 10 ( TPS10 ) ( Li et al . , 2014b ) . We therefore took advantage of bHLH-mediated binding to the TPS10 promoter to examine the putative transcription activator activity of CBP60gC . A fragment within the C-terminal portion of CBP60g ( amino acids 211–440 ) activated TPS10::LUC reporter when it was fused with the bHLH domain of MYC2 ( bHLH-CBP60g211-440 ) ( Figure 5B , Figure 5—source data 1 ) rather than bHLHMYC2 alone , indicating that the CBP60g211-440 harbors transcription activator activity . Moreover , Co-IP analysis showed that CBP60g211-end but not CBP60g441-end co-purified with VdSCP41 , indicating the requirement for CBP60g211-440 for binding to VdSCP41 ( Figure 5C ) . Thus , CBP60gC harbors a transcription activation domain , CBP60g211-440 , that is required for VdSCP41 targeting . The ΔC-CBP60g was further co-transfected with CBP60g , together with ICS1::LUC or FMO1::LUC . Dual reporter analyses indicated that co-expression of ΔC-CBP60g significantly suppressed CBP60g-induced ICS1::LUC and FMO1::LUC ( Figure 5A ) , suggesting a dominant-negative effect of ΔC-CBP60g on the activity of CBP60g . SARD1 functions both redundantly and differentially with CBP60g ( Sun et al . , 2015; Wang et al . , 2011 ) . The finding that VdSCP41 targeted both CBP60g and SARD1 prompted us to assess the roles of CBP60g and SARD1 in resistance to V . dahliae . The WT and cbp60g-1/sard1-1 double mutant plants were inoculated with V592 for the assessment of disease symptoms . As shown in Figure 6 , the cbp60g-1/sard1-1 double mutant displayed compromised resistance compared with the WT plants , demonstrating a contribution of CBP60g and SARD1 to V . dahliae resistance . We next investigated the requirement for CBP60g and SARD1 for VdSCP41-mediated virulence . The WT and cbp60g-1/sard1-1 double mutant plants were inoculated with the VdΔscp41 mutant . When compared with V592 , the VdΔscp41 mutant displayed reduced virulence on the WT plants . The reduced virulence arising from VdSCP41 deletion was partially impaired in the cbp60g-1/sard1-1 double mutant plants compared with that in the WT plants ( Figure 6 , Figure 6—source data 1 ) . The results indicated that both CBP60g and SARD1 are required for full virulence conferred by VdSCP41 . However , we still observed reduced virulence of the VdΔscp41 mutant compared with that of V592 on the cbp60g-1/sard1-1 double mutant plants , suggesting the existence of additional targets for VdSCP41 during V . dahliae infection . As in Arabidopsis , we observed similar compromised virulence of the VdΔscp41 mutant in upland cotton ( Figure 1E ) compared with V592 . The results prompted us to test the putative VdSCP41 targeting of GhCBP60b , the closest protein homolog of CBP60g and SARD1 in cotton . Co-IP analysis revealed an interaction between VdSCP41 and GhCBP60b ( Figure 7A ) . To examine the subcellular localization of GhCBP60b , GhCBP60b-GFP was constructed for transient expression in N . b . leaves . GhCBP60b-GFP localized to the nucleus of N . b . cells , and co-expression of GhCBP60b-GFP with VdSCP41-mCherry revealed co-localization of VdSCP41 and GhCBP60b in the nucleus of N . b . cells ( Figure 7B ) , suggesting conserved targeting of Arabidopsis CBP60g and SARD1 and of cotton GhCBP60b by VdSCP41 . To further examine whether GhCBP60b functions in resistance to V . dahliae in cotton , we generated a virus-induced gene silencing ( VIGS ) vector ( Liu et al . , 2002 ) targeting GhCBP60b ( pTRV2-GhCBP60b ) . Cotton plants were infiltrated with pTRV1 together with pTRV2 or pTRV2-GhCBP60b and further inoculated with V . dahliae . When compared with pTRV2 , pTRV2-GhCBP60b-infiltrated cotton plants exhibited higher ratios of wilting ( Figure 7C , Figure 7—source data 1 ) and more severe disease symptoms ( Figure 7D–E , Figure 7—source data 1 ) , indicating a role for GhCBP60b in cotton resistance to V . dahliae . The reduced expression of GhCBP60b in pTRV2-GhCBP60b-infiltrated plants was verified using RT-PCR ( Figure 7F , Figure 7—source data 1 ) . The results support the targeting of GhCBP60b by VdSCP41 for virulence during V . dahliae infection in cotton . To suppress plant defense actively or to modulate host physiology to benefit pathogenic fitness , oomycete and fungal pathogens deliver hundreds of effectors through specialized intracellular fungal structures , such as haustoria and infection hyphaea , into the host apoplastic space or directly into plant cells ( Lo Presti et al . , 2015 ) . Although haustoria have not been observed during infection , V . dahliae develops a penetration peg from a hyphopodium when infecting plant roots ( Zhao et al . , 2016 ) . During V . dahliae infection , the penetration peg that developed from the hyphopodium further develops a hyphal neck and forms a septin ring that partitions the hyphopodium and invasive hyphae . This septin-organized apparatus functions as a fungus–host interface for the dynamic delivery of secretory proteins , such as SCPs ( Zhao et al . , 2016Zhou et al . , 2017 ) . To date , relatively few V . dahliae-secreted effectors have been characterized as functioning inside host cells to modulate host immunity . Here , we provided evidence of the secretion and translocation of VdSCP41 , which is secreted by V . dahliae , into plant cells . VdSCP41 contains a signal peptide and localizes to the base of the hyphopodium to form a septin-like ring during infection ( Figure 2A ) , indicating that it is secreted via the septin-organized apparatus at the fungus–host interface . The localization of the VdSCP41 delivered by V . dahliae at the nucleus of onion epidermal cells ( Figure 2—figure supplement 1B ) suggested the translocation of VdSCP41 into plant cells . A few conserved motifs have been indicated to serve as signals for the uptake of fungal or oomycete effectors into plant cells . For example , a few effectors secreted by cereal powdery mildew and rust pathogens possess a Y/F/WxC motif that serves as a signal for translocation into the plant cell ( Godfrey et al . , 2010; Spanu et al . , 2010 ) . A set of oomycete effectors contain a RXLR-dEER motif that appears to assist in targeting effectors into plant cells ( Dou et al . , 2008; Whisson et al . , 2007 ) . Another set of oomycete effectors possess a FLAK motif for translocation ( Schornack et al . , 2010 ) . However , VdSCP41 lacks any of the above conserved motifs , and the mechanisms that assist the uptake of VdSCP41 into plant cells remain unclear . CBP60g and SARD1 are master regulators in immunity . ChIP-seq analyses have allowed the identification of a large number of target genes for CBP60g and SARD1 and support a model in which CBP60g and SARD1 accumulate in the plant nucleus and act as master regulators , which activate both positive and negative immune regulators ( Sun et al . , 2015; Wang et al . , 2009; Zhang et al . , 2010 ) . Genetic evidence revealed that CBP60g and SARD1 are positive immune regulators required for immune responses and bacterial resistance ( Wang et al . , 2009; Zhang et al . , 2010 ) . We demonstrated that VdSCP41 targets CBP60g and SARD1 and interferes with their activity , thus supporting the virulence function of VdSCP41 for immune suppression . Consistent with the inhibition of CBP60g and SARD1 activity , we observed less pathogen-induced SA accumulation in transgenic lines that expressed VdSCP41 than in WT plants ( Figure 2—figure supplement 2A ) . As CBP60g and SARD1 also bind directly to the promoters of both ALD1 and SARD4 ( Sun et al . , 2018 ) ( two major genes encoding pipecolic acid ( Pip ) biosynthesis enzymes ) to regulate Pip biosynthesis , a potential contribution of Pip to V . dahliae resistance will be of interest for further investigation . We showed that VdSCP41 binds the C-terminal portion of CBP60g to interfere with its transcription factor activity ( Figure 4A–B ) . Coexpression of VdSCP41 did not lead to cleavage or mobility shift of CBP60g ( Figure 4—figure supplement 1 ) , suggesting that VdSCP41 is unlikely to act as a protease to target CBP60g . In addition , CBP60g211-440 , a domain within the C-terminal portion of CBP60g , harbors transcription activator activity ( Figure 5B ) and is required for interaction with VdSCP41 ( Figure 5C ) . It is likely that CBP60gC functions to promote transcriptional activation by recruiting additional activators , and that binding of VdSCP41 interrupts either the activity of this domain or the recruitment of associated activators via this domain . The dominant-negative function of ΔC-CBP60g ( Figure 5A ) supports a dominant-negative effect on CBP60g activity arising from increased nuclear accumulation of VdSCP41-impaired CBP60g . Thus , VdSCP41-mediated over-accumulation of CBP60g provides an additional strategy to further interfere with CBP60g activity as a transcription factor . The results suggest a novel virulence strategy in which a pathogenic effector directly targets host transcription factors to interfere with their activity and to modulate plant immunity . However , it is equally possible that the increased nuclear accumulation of CBP60g results from feed-back regulation as a result of impaired CBP60g function . CBP60g and SARD1 are key components of the SA signaling pathway , which serves as an attractive target for bacterial and fungal effectors ( DebRoy et al . , 2004; Djamei et al . , 2011; Nomura et al . , 2011 ) . Targeting of CBP60g and SARD1 by VdSCP41 may therefore interfere with plant resistance to V . dahliae by manipulating SA signaling . On the other hand , ChIP-seq analyses identified a number of SA-independent regulators that are directly targeted by CBP60g and SARD1 ( Sun et al . , 2015 ) , revealing broader and SA-independent functions of CBP60g and SARD1 in immunity . It is likely that signaling pathways other than SA are also targeted by VdSCP41 through CBP60g and SARD1 to interfere with plant immunity . Consistent with this assumption , we showed that nlp20Vd2 functions to trigger the induction of both SA-dependent and SA-independent defense genes during V . dahliae infection ( Figure 2C–D ) . Furthermore , VdSCP41 expression suppressed the induction of both SA-dependent ICS1 and SA-independent FMO1 by nlp20Vd2 ( Figure 2C–D ) . Modulation of CBP60g and SARD1 may result in interference with their function as master transcription factors in PTI and in other defense pathways . Taken together , our results support a model in which PAMPs from V . dahliae , such as NLPs and chitins , are recognized by plants to induce the expression of CBP60g and SARD1 , which subsequently regulate the expression of a number of immune regulators to defend against pathogen infection ( Figure 8A ) . VdSCP41 secreted by V . dahliae functions as an intracellular effector that targets the transcription activator domain of CBPs , interrupting the function of this domain , to interfere with their transcription factor activity , and thus modulates both SA-dependent and SA-independent regulators to inhibit plant immunity against V . dahliae ( Figure 8B ) . The Verticillium dahliae strain V592 ( Gao et al . , 2010 ) was used in this study . Verticillium dahliae strains were grown on potato dextrose agar ( PDA ) medium at 25°C in the dark . To collect conidia , the mycelial plugs were cultured in potato dextrose broth ( PDB ) liquid medium at 25°C with shaking at 200 rpm for 3–5 days . Cotton plants ( 'Xinluzao No . 16' ) were used for virulence assessment in this study ( Zhou et al . , 2017 ) . Arabidopsis thaliana plants used in this study include Col-0 ( wild-type ) and the cbp60g-1/sard1-1 mutant ( Zhang et al . , 2010 ) . VdSCP41 was amplified from the V592 cDNA and cloned into pCambia1300-35S-FLAG ( Zhang et al . , 2010 ) to construct Arabidopsis transgenic lines expressing VdSCP41 . Antibodies used in this study include anti-FLAG ( RRID:AB_259529 ) , anti-HA ( RRID:AB_514506 ) , anti-CLuc ( RRID:AB_439707 ) , anti-GFP ( RRID:AB_390913 ) , and anti-mCherry ( Easybio , BE2026 ) . The upstream and downstream flanking sequences were PCR amplified from V592 genomic DNA and cloned into a pGKO-HPT vector ( Wang et al . , 2016 ) . The resulting construct was transformed into Agrobacterium tumefaciens EHA105 , and used for A . tumefaciens-mediated transformation ( ATMT ) to generate the VdΔscp41 mutant strain ( Wang et al . , 2016 ) . The genomic region of VdSCP41 , including 1 . 5 kb upstream from the start codon , was amplified and cloned into a pNat-Tef-TrpC vector ( Zhou et al . , 2017 ) to generate a construct for complementation . The resulting construct was transformed into Agrobacterium tumefaciens EHA105 , and used for ATMT to generate a VdΔscp41/VdSCP41-GFP strain . Primers used in this study are listed in Supplementary file 2 . Arabidopsis protoplasts isolated from 10 gram leaves were transfected with ΔspVdSCP41-FLAG , ΔspVdSCP45-FLAG , or empty vector for protein expression . Transfected protoplasts were collected and total protein was extracted with extraction buffer containing 50 mM HEPES ( pH 7 . 5 ) , 150 mM KCl , 1 mM EDTA , 1 mM DTT , 0 . 2% Triton X-100 , and 1 × proteinase inhibitor cocktail . Total protein was incubated with 50 μl anti-FLAG agarose beads ( Sigma ) for 12 hr at 4°C . The immunocomplex was washed three times using the buffer described above and eluted with 100 μl 1 μg/μl 3 × FLAG peptide . The eluted proteins were run 10 mm into the separating gel and stained with Proteo Silver stain kit ( Sigma ) . Total protein was destained and digested in-gel with sequencing grade trypsin ( 10 ng/mL trypsin , 50 mM ammonium bicarbonate [pH 8 . 0] ) overnight . Peptides were sequentially extracted with 5% formic acid/50% acetonitrile and 0 . 1% formic acid/75% acetonitrile and concentrated to 20 μl . The extracted peptides were separated by an analytical capillary column packed with 5 mm spherical C18 reversed-phase material . The eluted peptides were sprayed into a LTQ mass spectrometer ( Thermo Fisher Scientific ) equipped with a nano-ESI ion source . The mass spectrometer was operated in data-dependent mode with one MS scan followed by five MS/MS scans for each cycle . Database searches were performed on an in-house Mascot server ( Matrix Science Ltd . , London , UK ) against the IPI ( International Protein Index ) Arabidopsis protein database . Five-week-old Arabidopsis plants were used for protoplast isolation . pUC-35S-VdSCP41-FLAG , or its variant constructs , was co-transfected with pUC-35S-CBP60g-HA , or its variant constructs , pUC-35S-SARD1-HA or pUC-35S-GhCBP60b-HA , into Arabidopsis protoplasts . Total protein was extracted with extraction buffer . For anti-HA IP , total protein was incubated with 2 μg of anti-HA antibody together with protein A agarose at 4°C for 4 hr . The agarose beads were collected and boiled for 5 min with 1 × protein loading buffer . Immunoprecipitates were separated by 10% SDS-PAGE , and the presence of VdSCP41-FLAG or its variants , CBP60g-HA or its variants , SARD1-HA , or GhCBP60b-HA was detected by anti-FLAG or anti-HA immunoblot . SA was extracted and measured following the method described previously ( Sun et al . , 2015 ) . Around 0 . 3 grams of leaf tissue , collected from 4-week-old plants , was ground into powder in liquid nitrogen . Plant leaves were infiltrated with or without Pst DC3000 hrcC– ( OD600 = 0 . 1 ) 12 hr before sample collection . Three samples were analysed for each treatment . The samples were extracted with 0 . 8 mL 90% methanol and sonicated for 15 min , and the supernatant was transferred into a new tube . The pellet was re-extracted with 0 . 5 mL of 100% methanol , and the supernatant was combined with the first-step supernatant and dried by vacuum . The pellet was resuspended in 500 μL 0 . 1 M sodium acetate ( pH 5 . 5 ) in 10% methanol . An equal volume of 10% TCA was added and the samples were vortexed and sonicated for 5 min . After centrifugation , the supernatant was extracted three times with 0 . 5 ml of extraction buffer ( ethylacetate/cyclopentane/isopropanol:100/99/1 by volume ) . After spinning , the organic phases were collected and dried by vacuum . The samples were then dissolved in 250 μL 100% methanol and filtered through a 0 . 22 μm filter . The samples were then assayed by HPLC-MS/MS analysis on a AB SCIEX QTRAP 4500 system ( AB SCIEX , Foster , CA , USA ) . Agrobacterium tumefaciens GV3101 strain carrying CLuc- or NLuc-tagged consctruct were infiltrated into leaves of 4-week-old N . b . . LUC activity in leaves was examined 2 days post infiltration . N . b . leaves were treated with 1 μM luciferin and kept in the dark for 5 min to quench the fluorescence . LUC image was captured by CCD imaging apparatus , and the quantitative LUC activity was determined by microplate luminometer . Expression of CLuc-tagged proteins or NLuc-tagged proteins was detected by anti-CLuc or anti-HA western blot , respectively . The full-length CBP60g was cloned into the pGEX-6p-1 vector . VdSCP41C , VdSCP41100-163 or VDAG_01962 was cloned into the pET-28a vector . The resulting constructs were transformed into E . coli BL21 ( DE3 ) competent cells . BL21 ( DE3 ) strains containing the expression vectors were cultured and induced by isopropylthio-galactoside ( IPTG ) at 16°C for 16 hr . GST-tagged full-length CBP60g , His-tagged VdSCP41C or VdSCP41100-163 , and His-tagged VDAG_01962 recombinant protein were affinity purified and used for EMSAs . A 60-bp probe within the DNA fragment 7 ( Zhang et al . , 2010 ) in the ICS1 promoter were labeled with [γ-32P]ATP using T4 polynucleotide kinase . Binding reactions were carried out in a 20 μl volume of reaction buffer ( 10 mM Tris-HCl [pH 7 . 5] , 50 mM KCl , 1 mM DTT , 1 μl 50 ng/μl poly[dI-dC] ) for 30 min at room temperature . Labeled DNA probe ( 2 fmol ) was incubated with 4 μg; GST-CBP60g . 150 × unlabeled DNA probe was used for competition . 1 . 5 μg; His-tagged VdSCP41C163-end , VdSCP41100-163 , or VDAG_01962 was preincubated with GST-CBP60g for 30 min at room temperature before DNA binding . The reaction was stopped by adding DNA loading buffer and the samples were separated by a 5% native PAGE gel . After electrophoresis , the gel was autoradiographed . Total RNA was isolated with the TRIzol reagent ( Invitrogen ) and used for cDNA synthesis with SuperScript III First-Strand Synthesis System for RT-PCR ( Invitrogen ) following the instructions provided by the manufacturer . The quantitative PCR was performed with the SYBR Premix Ex Taq kit ( TaKaRa ) following standard protocols . Arabidopsis Col-0 or transgenic plants were treated with H2O , 1 μM flg22 or nlp20Vd2 ( Du et al . , 2017 ) as indicated for 3 hr . RNA was isolated and used for RT-PCR analysis for the expression of ICS1 and FMO1 . AtTUB4 was used as internal control . To detect gene expression in cotton plants , leaves from the cotton plants were collected 14 days post Agrobacterium infiltration . Quantitative RT-PCR was performed as described above . Gossypium hirsutum HISTONE3 was used as internal control . For RT-PCR analyses of VdSCP41 , the V592 or VdΔscp41 strain was incubated with the roots of 7-day-old Arabidopsis Col-0 plants for 2 days . Conidia were collected for RNA isolation and RT-PCR analysis to analyze the expression of VdSCP41 . For RT-PCR analyses of V . dahliae-infected Arabidopsis plants , Arabidopsis plants infected with or without the V592 or VdΔscp41 strain were collected for RNA isolation and RT-PCR analysis for the expression of ICS1 and FMO1 . VdELF1 was used as internal control for VdSCP41 . Agrobacterium tumefaciens EHA105 strain carrying pCambia1300-35S-CBP60g-GFP , pCambia1300-35S-SARD1-GFP , or pCambia1300-35S-GhCBP60b-GFP was infiltrated alone , or together with the A . tumefaciens EHA105 strain carrying pCambia1300-35S-VdSCP41-mCherry ( or its variant mutants ) into leaves of 4-week-old N . b . GFP and mCherry fluorescence were observed with Leica SP8 confocal microscopy 3 days post infiltration . The intensity of fluorescent signals was determined by Image J software . For fluorescence microscopy in Arabidopsis , Arabidopsis protoplasts were transfected with 35S-CBP60g-GFP alone or together with 35S-VdSCP41-mCherry ( or its variant mutants ) . The protoplasts were incubated overnight under faint light before GFP and mCherry fluorescence were observed . To examine the subcellular localization of VdSCP41 , conidia were cultured on cellophane and incubated for 3–9 days before observation by microscopy . The pieces of cellophane with mycelium were collected and observed as described ( Zhou et al . , 2017 ) . The plasma membrane of the fungi was stained with FM4-64 ( red ) . Arabidopsis protoplasts were co-transfected with ICS1::LUC or FMO1::LUC and 35S::RLUC ( Renilla luciferase ) alone , or together with VdSCP41 , CBP60g , or their variants . 12 hr after transfection , the protein of transfected protoplasts was isolated , and the LUC activity was determined by using a Dual-Luciferase Reporter system ( Promega ) according to the manufacture’s instructions . The VIGS was performed as described previously ( Gao and Shan , 2013 ) . Cotton plants were grown at 23–25°C in the growth room until two cotyledons had emerged . A 465-bp fragment of GhCBP60b cDNA was PCR amplified from G . hirsutum and cloned into pTRV2 plasmid ( Liu et al . , 2002 ) . The Agrobacterium strain carrying pTRV1 , together with the Agrobacterium strain carrying pTRV2 or pTRV2-GhCBP60b , was infiltrated into the cotyledons of the cotton plants . The cotton plants were root-dip-inoculated with V . dahliae V592 2 weeks post Agrobacterium infiltration . Cotton or Arabidopsis plants were infected by the root-dip inoculation method ( Gao et al . , 2010 ) . A conidial suspension of 107/ml from the indicated strain was used as the inoculum . The disease grade was classified as follows: Grade 0 ( no symptoms ) , 1 ( 0–25% wilted leaves ) , 2 ( 25–50% ) , 3 ( 50–75% ) and 4 ( 75–100% ) . The disease index was calculated as 100 × ( sum [number of plants × disease grade] ) / ( [total number of plants] × [maximal disease grade] ) ( Xu et al . , 2014 ) . The onion epidermis infection assay was performed as described ( Zhang et al . , 2017 ) . A conidial suspension of 107/ml from the V592-GFP , VdΔscp41/SCP41-GFP or VdΔscp41/SCP41-nls-GFP strain was inoculated onto the inner layer of onion epidermal cells and incubated on 1% water agar plates for 3–5 days before confocal imaging .	Like animals , plants have an immune system to protect themselves from disease . When a plant detects a disease-causing microbe , proteins that serve as master regulators of its immune system activate defense-related genes . Yet some microbes can overcome these defenses and successfully infect plants . Verticillium dahliae is a fungus , found in soil , that infects the roots of many plants – including cotton , tomatoes and potatoes . Infection by this fungus causes the leaves to curl and discolor , and the plant to wilt . The V . dahliae fungus releases , or secretes , nearly 800 proteins during an infection . Yet it remains unknown if and how any of these proteins help the fungus to infect plants . A better understanding of how V . dahliae impairs plant immunity to infect its hosts could give insights into ways to improve plant resistance against this fungus . Now , Qin et al . show that a secreted protein called VdSCP41 promotes V . dahliae infection in both cotton and Arabidopsis plants . Further experiments showed that after leaving the fungus , VdSCP41 enters into the plant’s own cells . Protein-protein interaction and biochemical studies then indicated VdSCP41 associates with a master immune regulator in Arabidopsis called CBP60g . This interaction interferes with CBP60g’s ability to activate the defense-related genes . Now that this role for VdSCP41 has been confirmed , the next step would be to see if targeting it would make plants more resistant to this fungus . One approach would be to genetically engineer plants so that they can specifically shut down , or ‘silence’ , the fungal gene that encodes for this protein . Further experiments are required to see whether using this technique – known as host-induced gene silencing ( or HIGS for short ) – against VdSCP41would enhance plant resistance to V . dahliae . If it does prove effective , this approach may eventually reduce the need for chemical pesticides to protect crop plants .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "microbiology", "and", "infectious", "disease" ]	2018	The plant-specific transcription factors CBP60g and SARD1 are targeted by a Verticillium secretory protein VdSCP41 to modulate immunity
In amyotrophic lateral sclerosis ( ALS ) and animal models of ALS , including SOD1-G93A mice , disassembly of the neuromuscular synapse precedes motor neuron loss and is sufficient to cause a decline in motor function that culminates in lethal respiratory paralysis . We treated SOD1-G93A mice with an agonist antibody to MuSK , a receptor tyrosine kinase essential for maintaining neuromuscular synapses , to determine whether increasing muscle retrograde signaling would slow nerve terminal detachment from muscle . The agonist antibody , delivered after disease onset , slowed muscle denervation , promoting motor neuron survival , improving motor system output , and extending the lifespan of SOD1-G93A mice . These findings suggest a novel therapeutic strategy for ALS , using an antibody format with clinical precedence , which targets a pathway essential for maintaining attachment of nerve terminals to muscle . Amyotrophic lateral sclerosis ( ALS ) is a neurodegenerative disease that progresses relentlessly from a subtle decline in motor function to lethal respiratory paralysis within a few years of diagnosis ( Pasinelli and Brown , 2006; Taylor et al . , 2016 ) . The disease can be familial and caused by dominant mutations in one of several genes , including SOD1 , C9orf72 , TDP43 , and FUS ( Taylor et al . , 2016 ) . More commonly , however , the disease is idiopathic . Although motor neuron cell death is a hallmark feature of ALS , the loss of neuromuscular synapses occurs prior to the loss of motor neurons and is the primary cause of motor paralysis in both familial and sporadic forms of ALS ( Fischer et al . , 2004; Schaefer et al . , 2005; Pun et al . , 2006 ) . The detachment of motor nerve terminals and withdrawal of motor axons has received less attention than the later loss of motor neurons , but therapeutic approaches designed to preserve neuromuscular synapses have the potential to maintain motor function , especially during the early phases of disease , and provide benefit to the quality of life for patient and family . Transgenic mice bearing dominant mutations in the human SOD1 gene , including SOD1-G93A mice , recapitulate the hallmark features of ALS and provide the most thoroughly studied animal model for ALS ( Vinsant et al . , 2013a; Vinsant et al . , 2013b ) . Moreover , because detachment of motor nerve terminals is the primary cause for paralysis in SOD1-G93A mice , SOD1-G93A mice represent a clinically relevant model for ALS . The signaling pathways that control attachment of motor axon terminals to muscle are only beginning to be understood , but two genes , Lrp4 and Musk , expressed by muscle , play important roles . Lrp4 , a member of the LDL receptor family , is the muscle receptor for the critical neuronal ligand , Agrin ( Kim et al . , 2008; Zhang et al . , 2008 ) . Upon binding Agrin , Lrp4 associates with MuSK , a receptor tyrosine kinase , stimulating MuSK and leading to anchoring and enhanced expression of critical postsynaptic proteins , including Lrp4 ( Burden et al . , 2013 ) . Clustered Lrp4 then signals back to motor axons to stimulate their attachment and differentiation ( Yumoto et al . , 2012 ) . Recessive mutations in Agrin , Lrp4 or Musk cause congenital myasthenia , a group of neuromuscular disorders , distinct from ALS , which compromise the structure and function of neuromuscular synapses and lead to muscle weakness and fatigue ( Engel et al . , 2015 ) . Moreover , autoantibodies to Agrin , Lrp4 , or MuSK cause myasthenia gravis ( MG ) , which is likewise distinct from ALS ( Gilhus and Verschuuren , 2015 ) . In MuSK MG , the pathogenic antibodies are usually directed to the first Ig-like domain in MuSK and reduce MuSK phosphorylation by impairing binding between Lrp4 and MuSK ( Huijbers et al . , 2013; Koneczny et al . , 2013 ) . Although defects in the MuSK signaling pathway are not associated with ALS , increasing MuSK gene expression stabilizes neuromuscular synapses in SOD1-G93A mice , reducing the extent of muscle denervation and improving motor function ( Pérez-García and Burden , 2012 ) . However , these experiments used transgenic mice to modestly increase MuSK expression from muscle , beginning during early development , several months prior to disease onset . Therefore , the therapeutic potential of increasing MuSK signaling as a strategy to reduce denervation and improve motor function in patients diagnosed with ALS remained unclear . Here , we sought to determine whether a pharmacological approach to increase MuSK activity in vivo would preserve neuromuscular synapses in SOD1-G93A mice when dosing was initiated after disease onset . This type of approach would have substantially improved potential for translation to ALS patients without the complex requirements for gene therapy ( Miyoshi et al . , 2017 ) . A previous study identified twenty-one single chain antibodies ( scFvs ) that recognize mouse MuSK and raised the idea that a subset of these antibodies may function as MuSK agonists in vivo ( Xie et al . , 1997 ) . We studied the activity of two antibodies , #13 and #22 , reported to stimulate MuSK in cultured myotubes , as well as antibody #21 , reported to bind but not stimulate MuSK . We confirmed that antibodies #13 and #22 , re-engineered as human IgG1 molecules , stimulated MuSK tyrosine phosphorylation in the C2 mouse muscle cell line ( Figure 1A ) , whereas antibody #21 , as well as a control IgG1 antibody to ragweed pollen , failed to stimulate MuSK phosphorylation ( Figure 1A ) . Agrin stimulates MuSK phosphorylation by binding Lrp4 , which promotes association between Lrp4 and MuSK , requiring the first of three Ig-like domains in MuSK ( Zhang et al . , 2011 ) . In contrast to the Agrin-dependent mechanism for activating MuSK , the agonist antibody binds the Fz-like domain in MuSK , force-dimerizing and stimulating MuSK phosphorylation , independent of Lrp4 ( Figure 1B , C and Figure 1—figure supplement 1 ) . Importantly , the Fz-like domain is dispensable for synapse formation in mice ( Remédio et al . , 2016 ) . To determine whether agonist antibody #13 could engage MuSK in vivo , we intraperitoneally ( IP ) injected varying amounts of the MuSK agonist antibody on a human IgG1 backbone , or a control human IgG1 antibody to ragweed pollen , into wild type mice . Several days later , we stained whole mounts of the diaphragm muscle to determine whether the agonist antibody engaged MuSK at the synapse . Figure 2A shows that neuromuscular synapses were labeled specifically by the MuSK agonist antibody . MuSK staining was evident as early as 3 days ( Figure 2Aiv-vi ) , and staining persisted for at least 7 days after the single injection ( Figure 2A , vii-ix ) . The organization of AChRs and nerve terminals appeared normal ( Figure 2A , ii , v , viii ) , indicating that the MuSK agonist antibody did not disturb major features of synaptic differentiation . Moreover , visual observation of the antibody-injected mice did not reveal overt behavioral abnormalities , indicating that the MuSK agonist antibody was well tolerated by the mice . Two mg/kg of the agonist antibody was sufficient to saturate MuSK labeling at the synapse ( Figure 2B ) and increase MuSK tyrosine phosphorylation in vivo ( Figure 1—figure supplement 1 ) . We measured the pharmacokinetic properties of the injected antibody and found that the half-life of the injected antibody in blood was ~12 days ( Figure 2C ) . The antibody exhibited linear clearance for 21 days after antibody injection , indicating that exposure could be maintained over several weeks . In addition , these results demonstrated that the mouse immune system did not recognize and clear the antibody , which contained a human Fc region , from the circulation over this three-week time period ( Figure 2C ) . We studied female and male SOD1-G93A mice , on a C57BL/6 background , with 21–26 copies of the human SOD1-G93A gene ( Figure 3—figure supplement 1 ) . In SOD1-G93A mice , denervation of limb muscles begins at P50 , whereas denervation of the diaphragm muscle begins a month later ( Pun et al . , 2006; Rocha et al . , 2013 ) . Because denervation of the diaphragm muscle is responsible for lethal respiratory paralysis , we focused our analysis on innervation of this muscle . We first quantified the extent of innervation in the diaphragm muscle at P90 by staining for nerve terminals and postsynaptic AChRs , which remain even at denervated synaptic sites ( Figure 3A ) . Denervation was evident in SOD1-G93A mice as early as P90 ( Figure 3B , C ) . From P90 to P110 , the extent of full innervation , defined as perfect apposition of nerve terminals and the AChR-rich postsynaptic membrane , decreased from 77 . 3% to 18 . 1% in female and from 53 . 1% to 16 . 1% in male SOD1-G93A mice ( Figure 3B ) . Likewise , the extent of complete denervation increased from 2 . 3% to 41% in female and from 16 . 7% to 24 . 4% in male SOD1-G93A mice over this twenty-day period ( Figure 3C ) . SOD1-G93A mice were injected with the MuSK agonist antibody at P90 . Because the antibody had a half-life of 12 days and 2 mg/kg of antibody saturated MuSK at the synapse ( Figure 2B , C ) , we injected SOD1-G93A mice with 10 mg/kg of agonist antibody , ensuring that the antibody concentration in blood would remain at saturating levels for MuSK-binding over the 20 day period . We found that a single dose of the MuSK agonist antibody increased the number of fully innervated synapses by 2 . 7- and 2 . 5-fold in female and male SOD1-G93A mice , respectively , and decreased the number of fully denervated synapses by 3 . 7- and 2 . 3-fold in female and male SOD1-G93A mice , respectively ( Figure 3B , C ) . These findings demonstrated that the MuSK agonist antibody , introduced after disease onset , decreased motor axon withdrawal from the diaphragm muscle . To determine whether the MuSK agonist antibody could preserve neuromuscular synapses over a longer time period , we chronically dosed SOD1-G93A mice . To avoid host recognition and clearance of the antibody during chronic exposure , we used a MuSK #13 antibody on a murine IgG2a backbone that also lacked effector functions ( Lo et al . , 2017 ) . The ability of this ‘reverse chimera’ to bind and stimulate MuSK was similar to the antibody with a human IgG backbone ( Figure 4—figure supplement 1 ) . Moreover , the ‘reverse chimera’ had a half-life similar to the human agonist antibody in vivo ( Figure 4—figure supplement 2 ) . SOD1-G93A mice were injected with 10 mg/kg of the reverse chimera agonist antibody at P90 and every 24 days thereafter , and we sacrificed chronically injected mice every 24 days to quantify innervation of the diaphragm muscle ( Figure 4A ) . Because 2 mg/kg of antibody saturated MuSK at the synapse and because the antibody had a 11 day half-life in blood , this dosing schedule ensured that saturating levels of the MuSK agonist antibody were maintained at all times ( Figure 4—figure supplement 2 ) . In SOD1-G93A mice injected with a control antibody to GP120 , synaptic loss continued to decline from P114 through P162 , so that only 11% of the synapses were fully innervated at P162 ( Figure 4B ) . This progressive loss was halted by injection of the MuSK agonist antibody , as the number of fully innervated synapses was largely unchanged ( 40–50% ) from P114 to P162 in SOD1-G93A mice injected with the MuSK agonist antibody ( Figure 4B ) . Similarly , the number of fully denervated synapses continued to increase from P114 through P162 in SOD1-G93A mice injected with the control antibody , whereas this progressive increase was prevented by the MuSK agonist antibody ( Figure 4C ) . These findings indicate that the MuSK agonist antibody prevented further synaptic loss and preserved synapses for at least 50 days after signs of denervation and disease were evident in SOD1-G93A mice . During disease progression , synapses transition through a partially innervated phase , when only a portion of the AChR-rich postsynaptic membrane is apposed by motor nerve terminals ( Figure 4—figure supplement 3 ) . Although the number of partially innervated synapses was similar in SOD1-G93A mice injected with the control or MuSK agonist antibody ( Figure 4—figure supplement 3 ) , the extent of nerve terminal coverage was 34% greater at partially innervated synapses in mice injected with the MuSK agonist antibody ( Figure 4—figure supplement 3 ) . Thus , the MuSK agonist antibody increased both full innervation as well as nerve terminal coverage at partially innervated synapses in SOD1-G93A mice . To determine whether maintaining neuromuscular synapses led to improved motor system output , we used an ex-vivo phrenic nerve/diaphragm muscle preparation to measure the compound muscle action potentials ( CMAPs ) , following phrenic nerve stimulation . We studied SOD1-G93A mice three to four weeks prior to end-stage ( Figure 5 ) . We stimulated the phrenic nerve to the diaphragm muscle and recorded CMAPs , which elicit muscle contraction ( Figure 5—figure supplement 1 ) . We found no significant difference in the amplitude of the first CMAP between SOD1-G93A mice injected with the MuSK agonist antibody or the control antibody to GP120 ( anti-GP120-treated males: 5 . 95 ± 1 . 14 mV; anti-MuSK-treated males: 5 . 93 ± 0 . 52 mV; anti-GP120-treated females: 4 . 95 ± 0 . 54 mV; anti-MuSK-treated females: 5 . 81 ± 0 . 63 mV ) . We next measured the reliability of synaptic transmission at the neuromuscular junction by repetitively stimulating the phrenic nerve at a physiological frequency ( 20 Hz ) . We found a rapid and severe decline in the amplitude of the CMAP , indicative of synaptic dysfunction and denervation , in SOD1-G93A mice chronically injected with the control antibody to GP120 . In contrast , the decline in CMAP amplitude was far less severe in SOD1-G93A mice treated with the MuSK agonist antibody , demonstrating that the MuSK agonist antibody improved neuromuscular function ( Figure 5 ) . Moreover , repetitive stimulation of the phrenic nerve at a more challenging frequency ( 50 Hz ) led to frequent failures to elicit a CMAP in SOD1-G93A mice injected with the control antibody to GP120 . Such failures were less frequent in SOD1-G93A mice injected with the MuSK agonist antibody , similar to wild type mice ( Figure 5 ) . These CMAP failures are likely due to presynaptic mechanisms , such as conduction block or impaired neurotransmitter release , rather than the inability of motor end plates to generate an action potential . In either case , the maintenance of neuromuscular synapses , stimulated by the MuSK agonist antibody , led to improved reliability of synaptic transmission and output of the critically important diaphragm muscle in SOD1-G93A mice . We next assessed whether preserving neuromuscular synapses in SOD1-G93A mice reduced motor neuron death . During embryonic development motor neuron death is regulated by innervation and reduced when motor neurons make additional synapses with muscle ( Hollyday and Hamburger , 1976; Tanaka and Landmesser , 1986; Landmesser , 1992 ) , whereas survival of adult motor neurons is less dependent upon muscle innervation ( Lowrie and Vrbová , 1992 ) . We quantified the number of motor neurons , stained for choline acetyltransferase ( ChAT ) , in the lumbar spinal cord of SOD1-G93A mice injected chronically either with the control antibody to GP120 or the MuSK agonist antibody ( Figure 6A ) . The MuSK agonist antibody increased the number of motor neurons by 31% to 57% at P138 ( Figure 6B ) , during the peak period of motor neuron cell death in SOD1-G93A mice when approximately half of spinal motor neurons have been lost ( Vinsant et al . , 2013a ) . These findings demonstrate that increasing retrograde signaling after disease onset not only preserves neuromuscular synapses but also promotes survival of spinal motor neurons in SOD1-G93A mice . Denervation of the diaphragm muscle is responsible for lethal respiratory paralysis in SOD1-G93A mice and ALS . We therefore asked whether maintaining neuromuscular synapses and improving output of the diaphragm muscle extended the lifespan of SOD1-G93A mice . Female SOD1-G93A mice injected with the control antibody to GP120 had an average lifespan of 169 days ( see Materials and methods ) , whereas male SOD1-G93A mice injected with the control antibody had an average lifespan of 157 . 5 days ( Figure 6C , D ) . Chronic injection with the MuSK agonist antibody prolonged survival of female and male SOD1-G93A mice by 7 ( p<0 . 05 ) and 10 days ( p<0 . 001 ) , respectively ( Figure 6C , D ) . Thus , the MuSK agonist antibody , introduced after disease onset , slowed the disassembly of neuromuscular synapses , improved motor output of the diaphragm muscle and extended the lifespan of SOD1-G93A mice . ALS is a devastating disease that progresses in a relentless manner from detachment of motor nerve terminals to lethal respiratory paralysis within several years of diagnosis . Currently , there is an unmet need for therapies that significantly alter the course of disease . Here , we describe a therapeutic approach designed to slow the loss of motor innervation to muscle by targeting a well-defined molecule and mechanism for forming and maintaining neuromuscular synapses . We show that an agonist antibody to MuSK , introduced after disease onset , decreases muscle denervation , improves motor system output , reduces motor neuron loss and extends survival in an aggressive mouse model of ALS . If this strategy , described here for an aggressive mouse model of ALS , were similarly successful in preserving innervation in sporadic and familial ALS , this therapeutic approach would have the potential to improve the quality of life for ALS patients , and as such warrants further study . Anti-sense RNA directed toward SOD1 is currently being tested as a promising therapeutic for ALS caused by mutations in SOD1 ( Miller et al . , 2013 ) . A similar approach may ultimately be effective for other dominant , familial forms of ALS ( Reddy and Miller , 2015; van Zundert and Brown , 2017 ) . However , >80% of ALS patients are diagnosed with sporadic ALS , so strategies to inactivate a single culprit gene are not tenable for most cases of ALS . Instead , multiple , concurrent therapeutic interventions that effectively address the pathology and symptoms of ALS will likely be necessary to alter the course of disease ( Brown and Al-Chalabi , 2017 ) . Because synaptic loss and muscle denervation are common to sporadic as well as familial forms of ALS , the approach described here has the potential to be effective for both forms of ALS . Moreover , increasing MuSK activity and retrograde signaling may also slow the deterioration of neuromuscular synapses in other neuromuscular diseases and during aging ( Engel et al . , 2015; Gilhus and Verschuuren , 2015; Valdez et al . , 2012; Poort et al . , 2016 ) . Consistent with this idea , adenoviral expression of Dok-7 , an inside-outside activator of MuSK , not only extends longevity of SOD1-G93A mice but also provides benefit in other mouse models of neuromuscular disease , including congenital myasthenia and Emery-Dreifuss muscular dystrophy ( Miyoshi et al . , 2017; Arimura et al . , 2014 ) . Further , there is increasing evidence that synaptic loss occurs early during disease progression in other neurodegenerative diseases , such as Alzheimer’s disease , Huntington’s disease , Parkinson’s disease and Frontotemporal dementia and Spinal Muscular Atrophy ( Henstridge et al . , 2016 ) , so similar strategies , designed to preserve synapses , may slow progression in these diseases as well . Our proof of concept experiments were designed to determine whether boosting retrograde signaling in vivo with the MuSK agonist antibody might slow motor axon withdrawal and muscle denervation in SOD1-G93A mice . As such , we introduced the MuSK agonist antibody after denervation was already evident , during the early phase of denervation in female SOD1-G93A mice and mid-phase in male SOD1-G93A mice , but before SOD1-G93A mice exhibited overt and severe deficits in limb motor function . This timing for delivery of the MuSK agonist antibody may be pertinent and significant for ALS , as denervation is the cause of muscle fibrillations , an early clinical sign in ALS . Because MuSK-dependent retrograde signaling is likely to act focally on nerve terminals and axons that are near the postsynaptic membrane and to be less effective in promoting regeneration of axons that have fully withdrawn , early delivery of a MuSK agonist is likely to be more effective than later delivery in ALS . However , ALS is a diagnosis of exclusion , leading to delays in diagnosis . Nonetheless , even at late stages of disease , a majority of synapses in SOD1-G93A mice are partially innervated , and the MuSK agonist antibody improved nerve terminal coverage at these partially innervated synapses . These findings suggest that the MuSK agonist antibody may also be effective if introduced later during disease . However , because overt motor deficits become evident in SOD1-G93A mice only a month before death , this aggressive mouse model of ALS may not be the optimal and most informative model to infer whether later introduction of the MuSK agonist antibody can stabilize synapses and slow motor dysfunction in ALS . The loss of motor neurons during embryonic development is regulated , at least in part by synapse formation ( Hollyday and Hamburger , 1976; Tanaka and Landmesser , 1986; Landmesser , 1992 ) . The increased survival of motor neurons in MuSK agonist antibody-injected SOD1-G93A mice indicates that adult motor neurons can likewise receive trophic support from muscle . Thus , preserving neuromuscular synapses not only maintains the essential attachment of nerve to muscle but also provides the added benefit of promoting motor neuron survival . Although we used an agonist antibody to MuSK to stimulate retrograde signaling from muscle , one can envisage other approaches to stimulate MuSK or enhance retrograde signaling in order to maintain attachment of motor axons to muscle . The MuSK agonist antibody is effective at maintaining neuromuscular synapses in SOD1-G93A mice up to P162 , but within the next week , synapses are lost , and the mice die . Because the MuSK agonist antibody is designed to maintain neuromuscular synapses and does not directly target or address the underlying cause of the disease and other pathologies in SOD1-G93A mice and ALS , the benefit of increasing retrograde signaling from muscle to nerve and promoting nerve terminal attachment is limited . Nonetheless , although the antibody cannot override the many pathological pathways that occur in the motor neuron and in non-neuronal cells , this therapeutic approach has a potent effect on the course of disease , reducing synaptic loss , improving motor output and extending the lifespan of SOD1-G93A mice longer than riluzole , the long-standing FDA approved treatment for ALS ( Jablonski et al . , 2014 ) . Motor neuron cell death is a critical feature in ALS , but elimination of Bax , which prevents apoptotic cell death , fails to preserve neuromuscular synapses and increases survival of SOD1-G93A mice by only 20 days ( Gould et al . , 2006 ) . Together with our studies , these findings give credence to the idea that combinatorial therapeutic interventions , including those that preserve neuromuscular synapses , will be necessary to fully address the complex pathology and symptoms of ALS and contribute to an improved quality of life for patient and family . The investigators were blinded from knowing whether mice were treated with the MuSK agonist or control antibody while acquiring and initially analyzing data . Data are presented as mean ±SEM . Statistical comparisons between groups were analyzed using an unpaired , two-tailed Student’s t-test , log-rank test ( survival ) , linear regression ( CMAPs ) , or two-way ANOVA ( failures ) . Statistical analyses were conducted using Prism 7 . 0 software ( GraphPad Software ) . The number ( n ) of mice used to calculate the mean , SEM values and the confidence limits ( p values ) are indicated in the figure legends . The copy number of the human SOD1-G93A gene was routinely quantified by TaqMan real-time PCR and normalized to GAPDH ( Life Technologies Assay# Mm00186822_cn ) . All mice included in this study had 21–26 copies of hSOD1-G93A ( Figure 3—figure supplement 1 ) . DietGel 76A ( ClearH20 ) was placed on the cage floor so that mice had ready access to nourishment . Others have measured the lifespan SOD1-G93A mice by placing mice on their side and sacrificing mice if they were unable to right themselves in 15 s . Because we were concerned that this assay reported on limb muscle function and may not be temporally aligned with the time of death , we used a variant assay , which provided an accurate measure of longevity . When mice were unable to right themselves to eat or drink over the course of several hours , they invariably succumbed within a day; we defined this time as disease end-point and sacrificed mice at this time . Mice were housed and maintained according to Institutional Animal Use and Care Committee ( IACUC ) guidelines . Diaphragm muscles were stained with Alexa 594-conjugated α-bungarotoxin ( α-BGT ) ( Life Technologies , Carlsbad , CA ) to mark AChRs and rabbit antibodies to Neurofilament-L ( SYnaptic Systems , Goettingen , Germany ) and Synapsin 1/2 ( SYynaptic Systems , Goettingen , Germany ) to label axons and nerve terminals , as described previously ( Jaworski and Burden , 2006; Friese et al . , 2007 ) . At fully innervated synapses , nerve terminal staining completely overlapped with postsynaptic AChRs , whereas nerve terminals were absent from original synaptic sites , marked by AChRs , at fully denervated synapses . At partially innervated synapses , nerve terminals occupied only a portion of the postsynaptic membrane . We examined a minimum of 50 synapses in the diaphragm muscle from each mouse and designated each synapse as fully innervated , partially innervated , or fully denervated . At each partially innervated synapse , the percentage of AChR-stained postsynaptic membrane that was apposed by Synapsin-stained nerve terminals was quantified using Volocity imaging software ( PerkinElmer , Waltham , MA ) . To visualize and quantify staining of the agonist antibody , containing human Fc , at the neuromuscular junction , we used an Alexa 647-conjugated anti-human secondary ( Life Technologies , Carlsbad , CA ) . Whole mounts of muscles were imaged with a Zeiss LSM800 confocal microscope , and the fluorescent signal was quantified as described previously ( Jaworski and Burden , 2006; Friese et al . , 2007 ) . Spinal cords were dissected from mice perfused with 4% formaldehyde . Frozen sections ( 20 μm ) of the lumbar region were stained with antibodies to choline acetyltransferase ( ChAT ) ( AB144P-200UL from Millipore , Billerica , MA ) . We defined motor neurons as cells in the ventral horn of the lumbar spinal cord that were positive for ChAT , excluding ChAT-positive preganglionic and Pitx2-positive neurons . We only counted ChAT-stained cells with a clearly defined nucleus in order to avoid double-counting motor neurons in multiple sections . We analyzed ~10 sections , evenly spaced in the lumbar enlargement , which together contained >50 motor neurons in each mouse . Chimeric antibodies were produced by transferring cDNAs encoding the variable regions of MuSK agonist antibody #13 to expression vectors containing the mouse kappa and IgG2a constant region . MuSK agonist antibodies were produced in CHO cells and purified by Protein A and size exclusion chromatography . The activity of the reverse chimera antibody for stimulating clustering of AChRs in C2 myotubes was similar to that for the human agonist antibody to MuSK ( Figure 4—figure supplement 1 ) . Fab fragments were prepared by protease digestion of human IgG1 followed by removal of uncleaved IgG and Fc fragments on a Protein A Sepharose column and size exclusion chromatography . We used a solid-phase binding assay to measure binding between the MuSK agonist antibody and the extracellular ( ecto ) region ( E22 to T494 ) , the first three Ig-like domains ( E22 to I103 ) or the Frizzled-like domain ( D312 to K456 ) from mouse MuSK ( Zhang et al . , 2011 ) . Maxisorp plates were coated with MuSK agonist antibody #13 ( 5μg/ml ) , and subsequently incubated with 8-His-tagged MuSK proteins , followed by a horseradish peroxidase ( HRP ) conjugated antibody to 8-His . Bound HRP was quantified by measuring HRP activity ( Thermo scientific#34028 ) . C2C12 muscle cells were purchased from the ATCC and were not tested for mycoplasma prior to use . These C2C12 muscle cells , as well as immortalized wild type or Lrp4 mutant muscle cells , were differentiated and treated with either neural Agrin ( 1 nM ) or antibodies ( 10 nM ) . MuSK was immunoprecipitated from lysates , and MuSK expression and MuSK tyrosine phosphorylation were measured by probing Western blots , as described previously ( Herbst and Burden , 2000 ) . C2C12 cells were grown in 24-well culture plates in DMEM with 10% fetal bovine serum ( FBS ) until myoblasts were 70% confluent . Myoblasts were then allowed to differentiate into myotubes by replacing the FBS with 2% horse serum . After 7 days , the cultures were treated for 16 hr with varying concentrations of rcMuSK antibody #13 or a Fab from antibody MuSK #13 . Cells were fixed in 4% paraformaldehyde and stained with Alexa 488 conjugated- α-BGT . Two to four images were collected from each well , and the number of AChR clusters was analyzed using imageJ software . Neural Agrin ( 10 nM ) ( R and D Systems , Minneapolis , MN ) was used as a positive control for AChR clustering ( data not shown ) . Hind-limb muscles were denervated by cutting the sciatic nerve , as described previously ( Simon et al . , 1992 ) . Four days after denervation , mice were injected with MuSK agonist antibody #13 , and we measured MuSK expression and MuSK tyrosine phosphorylation 3 days later . MuSK and Dok-7 were immunoprecipitated from lysates , and their expression levels were determined by Western blotting ( Herbst and Burden , 2000; Hallock et al . , 2010 ) . MuSK tyrosine phosphorylation was measured by probing Western blots with antibody 4G10 , as described previously ( Herbst and Burden , 2000; Hallock et al . , 2010 ) . To assess the function of neuromuscular junction in the mouse diaphragm muscle ( Figure 5—figure supplement 1 ) , we developed an ex vivo phrenic nerve-diaphragm preparation . We studied the diaphragm muscle from ~P140 male and ~P150 female mice , which is three to four weeks prior to end-stage , respectively . We did not use the in vivo preparation , described by others ( Lepore et al . , 2011 ) , because we were concerned that in vivo stimulation of the phrenic nerve , at moderate to high frequencies , would lead to variable and unreliable CMAP recordings , likely due to changes in the electrode position caused by muscle contraction . Moreover , a related method , reported to record from the mouse diaphragm muscle , uses a surface recording electrode , and likely monitors the activity of multiple thoracic muscles ( Martin et al . , 2015 ) . Thus , following anesthesia with 5% isoflurane , mice were decapitated , and the diaphragm muscle , together with the phrenic nerve , was quickly isolated and transferred to a customized recording chamber . The chamber was perfused continuously with oxygenated ( 95% O2 and 5% CO2 ) artificial cerebrospinal fluid solution ( 128 . 25 mM NaCl , 4 mM KCl , 0 . 58 mM NaH2PO4 , 21 mM NaHCO3 , 30 mM D-glucose , 1 . 5 mM CaCl2 , and 1 mM MgSO4 ) at a rate of ~10 ml/min at room temperature ( ~20–24°C ) . The phrenic nerve that innervates the left hemi-diaphragm muscle was stimulated by drawing the distal part of the left phrenic nerve into a suction electrode ( Figure 5—figure supplement 1 ) . We validated proper positioning of the stimulating electrode by visually inspecting muscle contractions following stimulation of the phrenic nerve . EMG activity was recorded using a suction electrode placed in the upper left quadrant of the muscle , 1 mm toward the costal side of the main intramuscular nerve and endplate zone in the middle of the muscle . A light suction was applied to the recording electrode to secure a tight seal between the tip of the electrode and the muscle fibers . In this manner , damage to the diaphragm muscle was avoided , which was confirmed by observing muscle contractions during stimulation . The phrenic nerve was stimulated with square pulses ( 0 . 2 ms in duration ) at several frequencies ( 1 Hz to 50 Hz ) for 60 s . The intensity of stimulation was progressively increased from the threshold , defined as the minimum response in three out of five trials , until the CMAP reached a maximal response . The stimulation intensity was set at twice the intensity required for the maximal response to ensure a supra-maximal intensity of stimulation ( 25µA to 200µA ) . Recordings were accepted for analysis only when the CMAP amplitude ( peak-to-peak ) was unchanged following 1 Hz stimulation . The amplitudes of the evoked CMAPs at higher frequencies were expressed as a percentage of the first evoked CMAP for the entire duration of stimulation . Recordings were fed to an A/D interface ( Digidata 1440A , Molecular Devices ) and acquired with Clampex ( v10 . 2 , Molecular Devices ) at a sampling rate of 50 kHz . Data were analyzed off-line using Clampfit ( v10 . 2 , Molecular Devices ) . We defined CMAP failures as the absence of an evoked response discernable from the background noise recorded prior to the stimulation .	Amyotrophic lateral sclerosis – often shortened to ALS – is a disease that starts with difficulties moving and progresses to paralysis of many muscles , including those used for breathing . The disease is usually lethal , with patients rarely surviving more than a few years after diagnosis . There is no cure or effective treatment for the disease . It begins with the breakdown of the connections , or synapses , between the muscles and the nerve cells that connect with them . After this , the nerve cell itself breaks down . Many therapeutic approaches have focused on attempts to prevent the nerve cells from dying , but few target the initial degeneration of the synapse . Cantor et al . asked if intervening when the synapse has already begun to break down could slow the progression of the disease in mice with ALS . Their approach involved using an antibody to bind to a receptor protein called MuSK , which plays an important role in maintaining the synapse between muscle and nerve cell . The antibody boosted the receptor’s activity , helping to preserve synapses , including those that connect nerve cells to the diaphragm muscle . The experiments showed that the antibody treatment led to fewer synapses breaking down , and kept more of the nerve cells alive . Healthier connections between the nervous system and the diaphragm improved the function of this muscle . As a result , the mice given the antibody treatment had a slightly extended lifespan , compared with those given no treatment . The findings suggest a possible new way to develop treatments for ALS , which could be used in combination with other therapies , such as those aimed at improving the health of the nerve cells . Together , this could improve quality of life for the majority of patients with ALS . Similar strategies could be used to develop treatments to preserve synapses in other neurodegenerative diseases , such as Alzheimer’s , Parkinson’s and Huntington’s disease , as well as some kinds of dementia . Preserving synapses early on , before the significant loss of nerve cells , could help to slow the progression of these diseases , improve the patients' quality of life and extend their lifespans too .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Preserving neuromuscular synapses in ALS by stimulating MuSK with a therapeutic agonist antibody

Subsets and Splits

No community queries yet

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